xref: /openbmc/linux/include/rdma/ib_verbs.h (revision 95777591)
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/dma-mapping.h>
45 #include <linux/kref.h>
46 #include <linux/list.h>
47 #include <linux/rwsem.h>
48 #include <linux/workqueue.h>
49 #include <linux/irq_poll.h>
50 #include <uapi/linux/if_ether.h>
51 #include <net/ipv6.h>
52 #include <net/ip.h>
53 #include <linux/string.h>
54 #include <linux/slab.h>
55 #include <linux/netdevice.h>
56 #include <linux/refcount.h>
57 #include <linux/if_link.h>
58 #include <linux/atomic.h>
59 #include <linux/mmu_notifier.h>
60 #include <linux/uaccess.h>
61 #include <linux/cgroup_rdma.h>
62 #include <uapi/rdma/ib_user_verbs.h>
63 #include <rdma/restrack.h>
64 #include <uapi/rdma/rdma_user_ioctl.h>
65 #include <uapi/rdma/ib_user_ioctl_verbs.h>
66 
67 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
68 
69 struct ib_umem_odp;
70 
71 extern struct workqueue_struct *ib_wq;
72 extern struct workqueue_struct *ib_comp_wq;
73 extern struct workqueue_struct *ib_comp_unbound_wq;
74 
75 union ib_gid {
76 	u8	raw[16];
77 	struct {
78 		__be64	subnet_prefix;
79 		__be64	interface_id;
80 	} global;
81 };
82 
83 extern union ib_gid zgid;
84 
85 enum ib_gid_type {
86 	/* If link layer is Ethernet, this is RoCE V1 */
87 	IB_GID_TYPE_IB        = 0,
88 	IB_GID_TYPE_ROCE      = 0,
89 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
90 	IB_GID_TYPE_SIZE
91 };
92 
93 #define ROCE_V2_UDP_DPORT      4791
94 struct ib_gid_attr {
95 	struct net_device	*ndev;
96 	struct ib_device	*device;
97 	union ib_gid		gid;
98 	enum ib_gid_type	gid_type;
99 	u16			index;
100 	u8			port_num;
101 };
102 
103 enum rdma_node_type {
104 	/* IB values map to NodeInfo:NodeType. */
105 	RDMA_NODE_IB_CA 	= 1,
106 	RDMA_NODE_IB_SWITCH,
107 	RDMA_NODE_IB_ROUTER,
108 	RDMA_NODE_RNIC,
109 	RDMA_NODE_USNIC,
110 	RDMA_NODE_USNIC_UDP,
111 };
112 
113 enum {
114 	/* set the local administered indication */
115 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
116 };
117 
118 enum rdma_transport_type {
119 	RDMA_TRANSPORT_IB,
120 	RDMA_TRANSPORT_IWARP,
121 	RDMA_TRANSPORT_USNIC,
122 	RDMA_TRANSPORT_USNIC_UDP
123 };
124 
125 enum rdma_protocol_type {
126 	RDMA_PROTOCOL_IB,
127 	RDMA_PROTOCOL_IBOE,
128 	RDMA_PROTOCOL_IWARP,
129 	RDMA_PROTOCOL_USNIC_UDP
130 };
131 
132 __attribute_const__ enum rdma_transport_type
133 rdma_node_get_transport(enum rdma_node_type node_type);
134 
135 enum rdma_network_type {
136 	RDMA_NETWORK_IB,
137 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
138 	RDMA_NETWORK_IPV4,
139 	RDMA_NETWORK_IPV6
140 };
141 
142 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
143 {
144 	if (network_type == RDMA_NETWORK_IPV4 ||
145 	    network_type == RDMA_NETWORK_IPV6)
146 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
147 
148 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
149 	return IB_GID_TYPE_IB;
150 }
151 
152 static inline enum rdma_network_type
153 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
154 {
155 	if (attr->gid_type == IB_GID_TYPE_IB)
156 		return RDMA_NETWORK_IB;
157 
158 	if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
159 		return RDMA_NETWORK_IPV4;
160 	else
161 		return RDMA_NETWORK_IPV6;
162 }
163 
164 enum rdma_link_layer {
165 	IB_LINK_LAYER_UNSPECIFIED,
166 	IB_LINK_LAYER_INFINIBAND,
167 	IB_LINK_LAYER_ETHERNET,
168 };
169 
170 enum ib_device_cap_flags {
171 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
172 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
173 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
174 	IB_DEVICE_RAW_MULTI			= (1 << 3),
175 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
176 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
177 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
178 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
179 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
180 	/* Not in use, former INIT_TYPE		= (1 << 9),*/
181 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
182 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
183 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
184 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
185 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
186 
187 	/*
188 	 * This device supports a per-device lkey or stag that can be
189 	 * used without performing a memory registration for the local
190 	 * memory.  Note that ULPs should never check this flag, but
191 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
192 	 * which will always contain a usable lkey.
193 	 */
194 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
195 	/* Reserved, old SEND_W_INV		= (1 << 16),*/
196 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
197 	/*
198 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
199 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
200 	 * messages and can verify the validity of checksum for
201 	 * incoming messages.  Setting this flag implies that the
202 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
203 	 */
204 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
205 	IB_DEVICE_UD_TSO			= (1 << 19),
206 	IB_DEVICE_XRC				= (1 << 20),
207 
208 	/*
209 	 * This device supports the IB "base memory management extension",
210 	 * which includes support for fast registrations (IB_WR_REG_MR,
211 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
212 	 * also be set by any iWarp device which must support FRs to comply
213 	 * to the iWarp verbs spec.  iWarp devices also support the
214 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
215 	 * stag.
216 	 */
217 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
218 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
219 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
220 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
221 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
222 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
223 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
224 	/*
225 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
226 	 * support execution of WQEs that involve synchronization
227 	 * of I/O operations with single completion queue managed
228 	 * by hardware.
229 	 */
230 	IB_DEVICE_CROSS_CHANNEL			= (1 << 27),
231 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
232 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
233 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
234 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
235 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
236 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
237 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
238 	IB_DEVICE_RDMA_NETDEV_OPA_VNIC		= (1ULL << 35),
239 	/* The device supports padding incoming writes to cacheline. */
240 	IB_DEVICE_PCI_WRITE_END_PADDING		= (1ULL << 36),
241 };
242 
243 enum ib_signature_prot_cap {
244 	IB_PROT_T10DIF_TYPE_1 = 1,
245 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
246 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
247 };
248 
249 enum ib_signature_guard_cap {
250 	IB_GUARD_T10DIF_CRC	= 1,
251 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
252 };
253 
254 enum ib_atomic_cap {
255 	IB_ATOMIC_NONE,
256 	IB_ATOMIC_HCA,
257 	IB_ATOMIC_GLOB
258 };
259 
260 enum ib_odp_general_cap_bits {
261 	IB_ODP_SUPPORT		= 1 << 0,
262 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
263 };
264 
265 enum ib_odp_transport_cap_bits {
266 	IB_ODP_SUPPORT_SEND	= 1 << 0,
267 	IB_ODP_SUPPORT_RECV	= 1 << 1,
268 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
269 	IB_ODP_SUPPORT_READ	= 1 << 3,
270 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
271 };
272 
273 struct ib_odp_caps {
274 	uint64_t general_caps;
275 	struct {
276 		uint32_t  rc_odp_caps;
277 		uint32_t  uc_odp_caps;
278 		uint32_t  ud_odp_caps;
279 	} per_transport_caps;
280 };
281 
282 struct ib_rss_caps {
283 	/* Corresponding bit will be set if qp type from
284 	 * 'enum ib_qp_type' is supported, e.g.
285 	 * supported_qpts |= 1 << IB_QPT_UD
286 	 */
287 	u32 supported_qpts;
288 	u32 max_rwq_indirection_tables;
289 	u32 max_rwq_indirection_table_size;
290 };
291 
292 enum ib_tm_cap_flags {
293 	/*  Support tag matching on RC transport */
294 	IB_TM_CAP_RC		    = 1 << 0,
295 };
296 
297 struct ib_tm_caps {
298 	/* Max size of RNDV header */
299 	u32 max_rndv_hdr_size;
300 	/* Max number of entries in tag matching list */
301 	u32 max_num_tags;
302 	/* From enum ib_tm_cap_flags */
303 	u32 flags;
304 	/* Max number of outstanding list operations */
305 	u32 max_ops;
306 	/* Max number of SGE in tag matching entry */
307 	u32 max_sge;
308 };
309 
310 struct ib_cq_init_attr {
311 	unsigned int	cqe;
312 	int		comp_vector;
313 	u32		flags;
314 };
315 
316 enum ib_cq_attr_mask {
317 	IB_CQ_MODERATE = 1 << 0,
318 };
319 
320 struct ib_cq_caps {
321 	u16     max_cq_moderation_count;
322 	u16     max_cq_moderation_period;
323 };
324 
325 struct ib_dm_mr_attr {
326 	u64		length;
327 	u64		offset;
328 	u32		access_flags;
329 };
330 
331 struct ib_dm_alloc_attr {
332 	u64	length;
333 	u32	alignment;
334 	u32	flags;
335 };
336 
337 struct ib_device_attr {
338 	u64			fw_ver;
339 	__be64			sys_image_guid;
340 	u64			max_mr_size;
341 	u64			page_size_cap;
342 	u32			vendor_id;
343 	u32			vendor_part_id;
344 	u32			hw_ver;
345 	int			max_qp;
346 	int			max_qp_wr;
347 	u64			device_cap_flags;
348 	int			max_send_sge;
349 	int			max_recv_sge;
350 	int			max_sge_rd;
351 	int			max_cq;
352 	int			max_cqe;
353 	int			max_mr;
354 	int			max_pd;
355 	int			max_qp_rd_atom;
356 	int			max_ee_rd_atom;
357 	int			max_res_rd_atom;
358 	int			max_qp_init_rd_atom;
359 	int			max_ee_init_rd_atom;
360 	enum ib_atomic_cap	atomic_cap;
361 	enum ib_atomic_cap	masked_atomic_cap;
362 	int			max_ee;
363 	int			max_rdd;
364 	int			max_mw;
365 	int			max_raw_ipv6_qp;
366 	int			max_raw_ethy_qp;
367 	int			max_mcast_grp;
368 	int			max_mcast_qp_attach;
369 	int			max_total_mcast_qp_attach;
370 	int			max_ah;
371 	int			max_fmr;
372 	int			max_map_per_fmr;
373 	int			max_srq;
374 	int			max_srq_wr;
375 	int			max_srq_sge;
376 	unsigned int		max_fast_reg_page_list_len;
377 	u16			max_pkeys;
378 	u8			local_ca_ack_delay;
379 	int			sig_prot_cap;
380 	int			sig_guard_cap;
381 	struct ib_odp_caps	odp_caps;
382 	uint64_t		timestamp_mask;
383 	uint64_t		hca_core_clock; /* in KHZ */
384 	struct ib_rss_caps	rss_caps;
385 	u32			max_wq_type_rq;
386 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
387 	struct ib_tm_caps	tm_caps;
388 	struct ib_cq_caps       cq_caps;
389 	u64			max_dm_size;
390 };
391 
392 enum ib_mtu {
393 	IB_MTU_256  = 1,
394 	IB_MTU_512  = 2,
395 	IB_MTU_1024 = 3,
396 	IB_MTU_2048 = 4,
397 	IB_MTU_4096 = 5
398 };
399 
400 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
401 {
402 	switch (mtu) {
403 	case IB_MTU_256:  return  256;
404 	case IB_MTU_512:  return  512;
405 	case IB_MTU_1024: return 1024;
406 	case IB_MTU_2048: return 2048;
407 	case IB_MTU_4096: return 4096;
408 	default: 	  return -1;
409 	}
410 }
411 
412 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
413 {
414 	if (mtu >= 4096)
415 		return IB_MTU_4096;
416 	else if (mtu >= 2048)
417 		return IB_MTU_2048;
418 	else if (mtu >= 1024)
419 		return IB_MTU_1024;
420 	else if (mtu >= 512)
421 		return IB_MTU_512;
422 	else
423 		return IB_MTU_256;
424 }
425 
426 enum ib_port_state {
427 	IB_PORT_NOP		= 0,
428 	IB_PORT_DOWN		= 1,
429 	IB_PORT_INIT		= 2,
430 	IB_PORT_ARMED		= 3,
431 	IB_PORT_ACTIVE		= 4,
432 	IB_PORT_ACTIVE_DEFER	= 5
433 };
434 
435 enum ib_port_width {
436 	IB_WIDTH_1X	= 1,
437 	IB_WIDTH_2X	= 16,
438 	IB_WIDTH_4X	= 2,
439 	IB_WIDTH_8X	= 4,
440 	IB_WIDTH_12X	= 8
441 };
442 
443 static inline int ib_width_enum_to_int(enum ib_port_width width)
444 {
445 	switch (width) {
446 	case IB_WIDTH_1X:  return  1;
447 	case IB_WIDTH_2X:  return  2;
448 	case IB_WIDTH_4X:  return  4;
449 	case IB_WIDTH_8X:  return  8;
450 	case IB_WIDTH_12X: return 12;
451 	default: 	  return -1;
452 	}
453 }
454 
455 enum ib_port_speed {
456 	IB_SPEED_SDR	= 1,
457 	IB_SPEED_DDR	= 2,
458 	IB_SPEED_QDR	= 4,
459 	IB_SPEED_FDR10	= 8,
460 	IB_SPEED_FDR	= 16,
461 	IB_SPEED_EDR	= 32,
462 	IB_SPEED_HDR	= 64
463 };
464 
465 /**
466  * struct rdma_hw_stats
467  * @lock - Mutex to protect parallel write access to lifespan and values
468  *    of counters, which are 64bits and not guaranteeed to be written
469  *    atomicaly on 32bits systems.
470  * @timestamp - Used by the core code to track when the last update was
471  * @lifespan - Used by the core code to determine how old the counters
472  *   should be before being updated again.  Stored in jiffies, defaults
473  *   to 10 milliseconds, drivers can override the default be specifying
474  *   their own value during their allocation routine.
475  * @name - Array of pointers to static names used for the counters in
476  *   directory.
477  * @num_counters - How many hardware counters there are.  If name is
478  *   shorter than this number, a kernel oops will result.  Driver authors
479  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
480  *   in their code to prevent this.
481  * @value - Array of u64 counters that are accessed by the sysfs code and
482  *   filled in by the drivers get_stats routine
483  */
484 struct rdma_hw_stats {
485 	struct mutex	lock; /* Protect lifespan and values[] */
486 	unsigned long	timestamp;
487 	unsigned long	lifespan;
488 	const char * const *names;
489 	int		num_counters;
490 	u64		value[];
491 };
492 
493 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
494 /**
495  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
496  *   for drivers.
497  * @names - Array of static const char *
498  * @num_counters - How many elements in array
499  * @lifespan - How many milliseconds between updates
500  */
501 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
502 		const char * const *names, int num_counters,
503 		unsigned long lifespan)
504 {
505 	struct rdma_hw_stats *stats;
506 
507 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
508 			GFP_KERNEL);
509 	if (!stats)
510 		return NULL;
511 	stats->names = names;
512 	stats->num_counters = num_counters;
513 	stats->lifespan = msecs_to_jiffies(lifespan);
514 
515 	return stats;
516 }
517 
518 
519 /* Define bits for the various functionality this port needs to be supported by
520  * the core.
521  */
522 /* Management                           0x00000FFF */
523 #define RDMA_CORE_CAP_IB_MAD            0x00000001
524 #define RDMA_CORE_CAP_IB_SMI            0x00000002
525 #define RDMA_CORE_CAP_IB_CM             0x00000004
526 #define RDMA_CORE_CAP_IW_CM             0x00000008
527 #define RDMA_CORE_CAP_IB_SA             0x00000010
528 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
529 
530 /* Address format                       0x000FF000 */
531 #define RDMA_CORE_CAP_AF_IB             0x00001000
532 #define RDMA_CORE_CAP_ETH_AH            0x00002000
533 #define RDMA_CORE_CAP_OPA_AH            0x00004000
534 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
535 
536 /* Protocol                             0xFFF00000 */
537 #define RDMA_CORE_CAP_PROT_IB           0x00100000
538 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
539 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
540 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
541 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
542 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
543 
544 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
545 					| RDMA_CORE_CAP_PROT_ROCE     \
546 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
547 
548 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
549 					| RDMA_CORE_CAP_IB_MAD \
550 					| RDMA_CORE_CAP_IB_SMI \
551 					| RDMA_CORE_CAP_IB_CM  \
552 					| RDMA_CORE_CAP_IB_SA  \
553 					| RDMA_CORE_CAP_AF_IB)
554 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
555 					| RDMA_CORE_CAP_IB_MAD  \
556 					| RDMA_CORE_CAP_IB_CM   \
557 					| RDMA_CORE_CAP_AF_IB   \
558 					| RDMA_CORE_CAP_ETH_AH)
559 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
560 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
561 					| RDMA_CORE_CAP_IB_MAD  \
562 					| RDMA_CORE_CAP_IB_CM   \
563 					| RDMA_CORE_CAP_AF_IB   \
564 					| RDMA_CORE_CAP_ETH_AH)
565 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
566 					| RDMA_CORE_CAP_IW_CM)
567 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
568 					| RDMA_CORE_CAP_OPA_MAD)
569 
570 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
571 
572 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
573 
574 struct ib_port_attr {
575 	u64			subnet_prefix;
576 	enum ib_port_state	state;
577 	enum ib_mtu		max_mtu;
578 	enum ib_mtu		active_mtu;
579 	int			gid_tbl_len;
580 	unsigned int		ip_gids:1;
581 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
582 	u32			port_cap_flags;
583 	u32			max_msg_sz;
584 	u32			bad_pkey_cntr;
585 	u32			qkey_viol_cntr;
586 	u16			pkey_tbl_len;
587 	u32			sm_lid;
588 	u32			lid;
589 	u8			lmc;
590 	u8			max_vl_num;
591 	u8			sm_sl;
592 	u8			subnet_timeout;
593 	u8			init_type_reply;
594 	u8			active_width;
595 	u8			active_speed;
596 	u8                      phys_state;
597 	u16			port_cap_flags2;
598 };
599 
600 enum ib_device_modify_flags {
601 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
602 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
603 };
604 
605 #define IB_DEVICE_NODE_DESC_MAX 64
606 
607 struct ib_device_modify {
608 	u64	sys_image_guid;
609 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
610 };
611 
612 enum ib_port_modify_flags {
613 	IB_PORT_SHUTDOWN		= 1,
614 	IB_PORT_INIT_TYPE		= (1<<2),
615 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
616 	IB_PORT_OPA_MASK_CHG		= (1<<4)
617 };
618 
619 struct ib_port_modify {
620 	u32	set_port_cap_mask;
621 	u32	clr_port_cap_mask;
622 	u8	init_type;
623 };
624 
625 enum ib_event_type {
626 	IB_EVENT_CQ_ERR,
627 	IB_EVENT_QP_FATAL,
628 	IB_EVENT_QP_REQ_ERR,
629 	IB_EVENT_QP_ACCESS_ERR,
630 	IB_EVENT_COMM_EST,
631 	IB_EVENT_SQ_DRAINED,
632 	IB_EVENT_PATH_MIG,
633 	IB_EVENT_PATH_MIG_ERR,
634 	IB_EVENT_DEVICE_FATAL,
635 	IB_EVENT_PORT_ACTIVE,
636 	IB_EVENT_PORT_ERR,
637 	IB_EVENT_LID_CHANGE,
638 	IB_EVENT_PKEY_CHANGE,
639 	IB_EVENT_SM_CHANGE,
640 	IB_EVENT_SRQ_ERR,
641 	IB_EVENT_SRQ_LIMIT_REACHED,
642 	IB_EVENT_QP_LAST_WQE_REACHED,
643 	IB_EVENT_CLIENT_REREGISTER,
644 	IB_EVENT_GID_CHANGE,
645 	IB_EVENT_WQ_FATAL,
646 };
647 
648 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
649 
650 struct ib_event {
651 	struct ib_device	*device;
652 	union {
653 		struct ib_cq	*cq;
654 		struct ib_qp	*qp;
655 		struct ib_srq	*srq;
656 		struct ib_wq	*wq;
657 		u8		port_num;
658 	} element;
659 	enum ib_event_type	event;
660 };
661 
662 struct ib_event_handler {
663 	struct ib_device *device;
664 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
665 	struct list_head  list;
666 };
667 
668 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
669 	do {							\
670 		(_ptr)->device  = _device;			\
671 		(_ptr)->handler = _handler;			\
672 		INIT_LIST_HEAD(&(_ptr)->list);			\
673 	} while (0)
674 
675 struct ib_global_route {
676 	const struct ib_gid_attr *sgid_attr;
677 	union ib_gid	dgid;
678 	u32		flow_label;
679 	u8		sgid_index;
680 	u8		hop_limit;
681 	u8		traffic_class;
682 };
683 
684 struct ib_grh {
685 	__be32		version_tclass_flow;
686 	__be16		paylen;
687 	u8		next_hdr;
688 	u8		hop_limit;
689 	union ib_gid	sgid;
690 	union ib_gid	dgid;
691 };
692 
693 union rdma_network_hdr {
694 	struct ib_grh ibgrh;
695 	struct {
696 		/* The IB spec states that if it's IPv4, the header
697 		 * is located in the last 20 bytes of the header.
698 		 */
699 		u8		reserved[20];
700 		struct iphdr	roce4grh;
701 	};
702 };
703 
704 #define IB_QPN_MASK		0xFFFFFF
705 
706 enum {
707 	IB_MULTICAST_QPN = 0xffffff
708 };
709 
710 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
711 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
712 
713 enum ib_ah_flags {
714 	IB_AH_GRH	= 1
715 };
716 
717 enum ib_rate {
718 	IB_RATE_PORT_CURRENT = 0,
719 	IB_RATE_2_5_GBPS = 2,
720 	IB_RATE_5_GBPS   = 5,
721 	IB_RATE_10_GBPS  = 3,
722 	IB_RATE_20_GBPS  = 6,
723 	IB_RATE_30_GBPS  = 4,
724 	IB_RATE_40_GBPS  = 7,
725 	IB_RATE_60_GBPS  = 8,
726 	IB_RATE_80_GBPS  = 9,
727 	IB_RATE_120_GBPS = 10,
728 	IB_RATE_14_GBPS  = 11,
729 	IB_RATE_56_GBPS  = 12,
730 	IB_RATE_112_GBPS = 13,
731 	IB_RATE_168_GBPS = 14,
732 	IB_RATE_25_GBPS  = 15,
733 	IB_RATE_100_GBPS = 16,
734 	IB_RATE_200_GBPS = 17,
735 	IB_RATE_300_GBPS = 18,
736 	IB_RATE_28_GBPS  = 19,
737 	IB_RATE_50_GBPS  = 20,
738 	IB_RATE_400_GBPS = 21,
739 	IB_RATE_600_GBPS = 22,
740 };
741 
742 /**
743  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
744  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
745  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
746  * @rate: rate to convert.
747  */
748 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
749 
750 /**
751  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
752  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
753  * @rate: rate to convert.
754  */
755 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
756 
757 
758 /**
759  * enum ib_mr_type - memory region type
760  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
761  *                            normal registration
762  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
763  *                            signature operations (data-integrity
764  *                            capable regions)
765  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
766  *                            register any arbitrary sg lists (without
767  *                            the normal mr constraints - see
768  *                            ib_map_mr_sg)
769  */
770 enum ib_mr_type {
771 	IB_MR_TYPE_MEM_REG,
772 	IB_MR_TYPE_SIGNATURE,
773 	IB_MR_TYPE_SG_GAPS,
774 };
775 
776 /**
777  * Signature types
778  * IB_SIG_TYPE_NONE: Unprotected.
779  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
780  */
781 enum ib_signature_type {
782 	IB_SIG_TYPE_NONE,
783 	IB_SIG_TYPE_T10_DIF,
784 };
785 
786 /**
787  * Signature T10-DIF block-guard types
788  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
789  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
790  */
791 enum ib_t10_dif_bg_type {
792 	IB_T10DIF_CRC,
793 	IB_T10DIF_CSUM
794 };
795 
796 /**
797  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
798  *     domain.
799  * @bg_type: T10-DIF block guard type (CRC|CSUM)
800  * @pi_interval: protection information interval.
801  * @bg: seed of guard computation.
802  * @app_tag: application tag of guard block
803  * @ref_tag: initial guard block reference tag.
804  * @ref_remap: Indicate wethear the reftag increments each block
805  * @app_escape: Indicate to skip block check if apptag=0xffff
806  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
807  * @apptag_check_mask: check bitmask of application tag.
808  */
809 struct ib_t10_dif_domain {
810 	enum ib_t10_dif_bg_type bg_type;
811 	u16			pi_interval;
812 	u16			bg;
813 	u16			app_tag;
814 	u32			ref_tag;
815 	bool			ref_remap;
816 	bool			app_escape;
817 	bool			ref_escape;
818 	u16			apptag_check_mask;
819 };
820 
821 /**
822  * struct ib_sig_domain - Parameters for signature domain
823  * @sig_type: specific signauture type
824  * @sig: union of all signature domain attributes that may
825  *     be used to set domain layout.
826  */
827 struct ib_sig_domain {
828 	enum ib_signature_type sig_type;
829 	union {
830 		struct ib_t10_dif_domain dif;
831 	} sig;
832 };
833 
834 /**
835  * struct ib_sig_attrs - Parameters for signature handover operation
836  * @check_mask: bitmask for signature byte check (8 bytes)
837  * @mem: memory domain layout desciptor.
838  * @wire: wire domain layout desciptor.
839  */
840 struct ib_sig_attrs {
841 	u8			check_mask;
842 	struct ib_sig_domain	mem;
843 	struct ib_sig_domain	wire;
844 };
845 
846 enum ib_sig_err_type {
847 	IB_SIG_BAD_GUARD,
848 	IB_SIG_BAD_REFTAG,
849 	IB_SIG_BAD_APPTAG,
850 };
851 
852 /**
853  * Signature check masks (8 bytes in total) according to the T10-PI standard:
854  *  -------- -------- ------------
855  * | GUARD  | APPTAG |   REFTAG   |
856  * |  2B    |  2B    |    4B      |
857  *  -------- -------- ------------
858  */
859 enum {
860 	IB_SIG_CHECK_GUARD	= 0xc0,
861 	IB_SIG_CHECK_APPTAG	= 0x30,
862 	IB_SIG_CHECK_REFTAG	= 0x0f,
863 };
864 
865 /**
866  * struct ib_sig_err - signature error descriptor
867  */
868 struct ib_sig_err {
869 	enum ib_sig_err_type	err_type;
870 	u32			expected;
871 	u32			actual;
872 	u64			sig_err_offset;
873 	u32			key;
874 };
875 
876 enum ib_mr_status_check {
877 	IB_MR_CHECK_SIG_STATUS = 1,
878 };
879 
880 /**
881  * struct ib_mr_status - Memory region status container
882  *
883  * @fail_status: Bitmask of MR checks status. For each
884  *     failed check a corresponding status bit is set.
885  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
886  *     failure.
887  */
888 struct ib_mr_status {
889 	u32		    fail_status;
890 	struct ib_sig_err   sig_err;
891 };
892 
893 /**
894  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
895  * enum.
896  * @mult: multiple to convert.
897  */
898 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
899 
900 enum rdma_ah_attr_type {
901 	RDMA_AH_ATTR_TYPE_UNDEFINED,
902 	RDMA_AH_ATTR_TYPE_IB,
903 	RDMA_AH_ATTR_TYPE_ROCE,
904 	RDMA_AH_ATTR_TYPE_OPA,
905 };
906 
907 struct ib_ah_attr {
908 	u16			dlid;
909 	u8			src_path_bits;
910 };
911 
912 struct roce_ah_attr {
913 	u8			dmac[ETH_ALEN];
914 };
915 
916 struct opa_ah_attr {
917 	u32			dlid;
918 	u8			src_path_bits;
919 	bool			make_grd;
920 };
921 
922 struct rdma_ah_attr {
923 	struct ib_global_route	grh;
924 	u8			sl;
925 	u8			static_rate;
926 	u8			port_num;
927 	u8			ah_flags;
928 	enum rdma_ah_attr_type type;
929 	union {
930 		struct ib_ah_attr ib;
931 		struct roce_ah_attr roce;
932 		struct opa_ah_attr opa;
933 	};
934 };
935 
936 enum ib_wc_status {
937 	IB_WC_SUCCESS,
938 	IB_WC_LOC_LEN_ERR,
939 	IB_WC_LOC_QP_OP_ERR,
940 	IB_WC_LOC_EEC_OP_ERR,
941 	IB_WC_LOC_PROT_ERR,
942 	IB_WC_WR_FLUSH_ERR,
943 	IB_WC_MW_BIND_ERR,
944 	IB_WC_BAD_RESP_ERR,
945 	IB_WC_LOC_ACCESS_ERR,
946 	IB_WC_REM_INV_REQ_ERR,
947 	IB_WC_REM_ACCESS_ERR,
948 	IB_WC_REM_OP_ERR,
949 	IB_WC_RETRY_EXC_ERR,
950 	IB_WC_RNR_RETRY_EXC_ERR,
951 	IB_WC_LOC_RDD_VIOL_ERR,
952 	IB_WC_REM_INV_RD_REQ_ERR,
953 	IB_WC_REM_ABORT_ERR,
954 	IB_WC_INV_EECN_ERR,
955 	IB_WC_INV_EEC_STATE_ERR,
956 	IB_WC_FATAL_ERR,
957 	IB_WC_RESP_TIMEOUT_ERR,
958 	IB_WC_GENERAL_ERR
959 };
960 
961 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
962 
963 enum ib_wc_opcode {
964 	IB_WC_SEND,
965 	IB_WC_RDMA_WRITE,
966 	IB_WC_RDMA_READ,
967 	IB_WC_COMP_SWAP,
968 	IB_WC_FETCH_ADD,
969 	IB_WC_LSO,
970 	IB_WC_LOCAL_INV,
971 	IB_WC_REG_MR,
972 	IB_WC_MASKED_COMP_SWAP,
973 	IB_WC_MASKED_FETCH_ADD,
974 /*
975  * Set value of IB_WC_RECV so consumers can test if a completion is a
976  * receive by testing (opcode & IB_WC_RECV).
977  */
978 	IB_WC_RECV			= 1 << 7,
979 	IB_WC_RECV_RDMA_WITH_IMM
980 };
981 
982 enum ib_wc_flags {
983 	IB_WC_GRH		= 1,
984 	IB_WC_WITH_IMM		= (1<<1),
985 	IB_WC_WITH_INVALIDATE	= (1<<2),
986 	IB_WC_IP_CSUM_OK	= (1<<3),
987 	IB_WC_WITH_SMAC		= (1<<4),
988 	IB_WC_WITH_VLAN		= (1<<5),
989 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
990 };
991 
992 struct ib_wc {
993 	union {
994 		u64		wr_id;
995 		struct ib_cqe	*wr_cqe;
996 	};
997 	enum ib_wc_status	status;
998 	enum ib_wc_opcode	opcode;
999 	u32			vendor_err;
1000 	u32			byte_len;
1001 	struct ib_qp	       *qp;
1002 	union {
1003 		__be32		imm_data;
1004 		u32		invalidate_rkey;
1005 	} ex;
1006 	u32			src_qp;
1007 	u32			slid;
1008 	int			wc_flags;
1009 	u16			pkey_index;
1010 	u8			sl;
1011 	u8			dlid_path_bits;
1012 	u8			port_num;	/* valid only for DR SMPs on switches */
1013 	u8			smac[ETH_ALEN];
1014 	u16			vlan_id;
1015 	u8			network_hdr_type;
1016 };
1017 
1018 enum ib_cq_notify_flags {
1019 	IB_CQ_SOLICITED			= 1 << 0,
1020 	IB_CQ_NEXT_COMP			= 1 << 1,
1021 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1022 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
1023 };
1024 
1025 enum ib_srq_type {
1026 	IB_SRQT_BASIC,
1027 	IB_SRQT_XRC,
1028 	IB_SRQT_TM,
1029 };
1030 
1031 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1032 {
1033 	return srq_type == IB_SRQT_XRC ||
1034 	       srq_type == IB_SRQT_TM;
1035 }
1036 
1037 enum ib_srq_attr_mask {
1038 	IB_SRQ_MAX_WR	= 1 << 0,
1039 	IB_SRQ_LIMIT	= 1 << 1,
1040 };
1041 
1042 struct ib_srq_attr {
1043 	u32	max_wr;
1044 	u32	max_sge;
1045 	u32	srq_limit;
1046 };
1047 
1048 struct ib_srq_init_attr {
1049 	void		      (*event_handler)(struct ib_event *, void *);
1050 	void		       *srq_context;
1051 	struct ib_srq_attr	attr;
1052 	enum ib_srq_type	srq_type;
1053 
1054 	struct {
1055 		struct ib_cq   *cq;
1056 		union {
1057 			struct {
1058 				struct ib_xrcd *xrcd;
1059 			} xrc;
1060 
1061 			struct {
1062 				u32		max_num_tags;
1063 			} tag_matching;
1064 		};
1065 	} ext;
1066 };
1067 
1068 struct ib_qp_cap {
1069 	u32	max_send_wr;
1070 	u32	max_recv_wr;
1071 	u32	max_send_sge;
1072 	u32	max_recv_sge;
1073 	u32	max_inline_data;
1074 
1075 	/*
1076 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1077 	 * ib_create_qp() will calculate the right amount of neededed WRs
1078 	 * and MRs based on this.
1079 	 */
1080 	u32	max_rdma_ctxs;
1081 };
1082 
1083 enum ib_sig_type {
1084 	IB_SIGNAL_ALL_WR,
1085 	IB_SIGNAL_REQ_WR
1086 };
1087 
1088 enum ib_qp_type {
1089 	/*
1090 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1091 	 * here (and in that order) since the MAD layer uses them as
1092 	 * indices into a 2-entry table.
1093 	 */
1094 	IB_QPT_SMI,
1095 	IB_QPT_GSI,
1096 
1097 	IB_QPT_RC,
1098 	IB_QPT_UC,
1099 	IB_QPT_UD,
1100 	IB_QPT_RAW_IPV6,
1101 	IB_QPT_RAW_ETHERTYPE,
1102 	IB_QPT_RAW_PACKET = 8,
1103 	IB_QPT_XRC_INI = 9,
1104 	IB_QPT_XRC_TGT,
1105 	IB_QPT_MAX,
1106 	IB_QPT_DRIVER = 0xFF,
1107 	/* Reserve a range for qp types internal to the low level driver.
1108 	 * These qp types will not be visible at the IB core layer, so the
1109 	 * IB_QPT_MAX usages should not be affected in the core layer
1110 	 */
1111 	IB_QPT_RESERVED1 = 0x1000,
1112 	IB_QPT_RESERVED2,
1113 	IB_QPT_RESERVED3,
1114 	IB_QPT_RESERVED4,
1115 	IB_QPT_RESERVED5,
1116 	IB_QPT_RESERVED6,
1117 	IB_QPT_RESERVED7,
1118 	IB_QPT_RESERVED8,
1119 	IB_QPT_RESERVED9,
1120 	IB_QPT_RESERVED10,
1121 };
1122 
1123 enum ib_qp_create_flags {
1124 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1125 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1126 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1127 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1128 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1129 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1130 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1131 	/* FREE					= 1 << 7, */
1132 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1133 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1134 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1135 	IB_QP_CREATE_PCI_WRITE_END_PADDING	= 1 << 11,
1136 	/* reserve bits 26-31 for low level drivers' internal use */
1137 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1138 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1139 };
1140 
1141 /*
1142  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1143  * callback to destroy the passed in QP.
1144  */
1145 
1146 struct ib_qp_init_attr {
1147 	/* Consumer's event_handler callback must not block */
1148 	void                  (*event_handler)(struct ib_event *, void *);
1149 
1150 	void		       *qp_context;
1151 	struct ib_cq	       *send_cq;
1152 	struct ib_cq	       *recv_cq;
1153 	struct ib_srq	       *srq;
1154 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1155 	struct ib_qp_cap	cap;
1156 	enum ib_sig_type	sq_sig_type;
1157 	enum ib_qp_type		qp_type;
1158 	u32			create_flags;
1159 
1160 	/*
1161 	 * Only needed for special QP types, or when using the RW API.
1162 	 */
1163 	u8			port_num;
1164 	struct ib_rwq_ind_table *rwq_ind_tbl;
1165 	u32			source_qpn;
1166 };
1167 
1168 struct ib_qp_open_attr {
1169 	void                  (*event_handler)(struct ib_event *, void *);
1170 	void		       *qp_context;
1171 	u32			qp_num;
1172 	enum ib_qp_type		qp_type;
1173 };
1174 
1175 enum ib_rnr_timeout {
1176 	IB_RNR_TIMER_655_36 =  0,
1177 	IB_RNR_TIMER_000_01 =  1,
1178 	IB_RNR_TIMER_000_02 =  2,
1179 	IB_RNR_TIMER_000_03 =  3,
1180 	IB_RNR_TIMER_000_04 =  4,
1181 	IB_RNR_TIMER_000_06 =  5,
1182 	IB_RNR_TIMER_000_08 =  6,
1183 	IB_RNR_TIMER_000_12 =  7,
1184 	IB_RNR_TIMER_000_16 =  8,
1185 	IB_RNR_TIMER_000_24 =  9,
1186 	IB_RNR_TIMER_000_32 = 10,
1187 	IB_RNR_TIMER_000_48 = 11,
1188 	IB_RNR_TIMER_000_64 = 12,
1189 	IB_RNR_TIMER_000_96 = 13,
1190 	IB_RNR_TIMER_001_28 = 14,
1191 	IB_RNR_TIMER_001_92 = 15,
1192 	IB_RNR_TIMER_002_56 = 16,
1193 	IB_RNR_TIMER_003_84 = 17,
1194 	IB_RNR_TIMER_005_12 = 18,
1195 	IB_RNR_TIMER_007_68 = 19,
1196 	IB_RNR_TIMER_010_24 = 20,
1197 	IB_RNR_TIMER_015_36 = 21,
1198 	IB_RNR_TIMER_020_48 = 22,
1199 	IB_RNR_TIMER_030_72 = 23,
1200 	IB_RNR_TIMER_040_96 = 24,
1201 	IB_RNR_TIMER_061_44 = 25,
1202 	IB_RNR_TIMER_081_92 = 26,
1203 	IB_RNR_TIMER_122_88 = 27,
1204 	IB_RNR_TIMER_163_84 = 28,
1205 	IB_RNR_TIMER_245_76 = 29,
1206 	IB_RNR_TIMER_327_68 = 30,
1207 	IB_RNR_TIMER_491_52 = 31
1208 };
1209 
1210 enum ib_qp_attr_mask {
1211 	IB_QP_STATE			= 1,
1212 	IB_QP_CUR_STATE			= (1<<1),
1213 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1214 	IB_QP_ACCESS_FLAGS		= (1<<3),
1215 	IB_QP_PKEY_INDEX		= (1<<4),
1216 	IB_QP_PORT			= (1<<5),
1217 	IB_QP_QKEY			= (1<<6),
1218 	IB_QP_AV			= (1<<7),
1219 	IB_QP_PATH_MTU			= (1<<8),
1220 	IB_QP_TIMEOUT			= (1<<9),
1221 	IB_QP_RETRY_CNT			= (1<<10),
1222 	IB_QP_RNR_RETRY			= (1<<11),
1223 	IB_QP_RQ_PSN			= (1<<12),
1224 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1225 	IB_QP_ALT_PATH			= (1<<14),
1226 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1227 	IB_QP_SQ_PSN			= (1<<16),
1228 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1229 	IB_QP_PATH_MIG_STATE		= (1<<18),
1230 	IB_QP_CAP			= (1<<19),
1231 	IB_QP_DEST_QPN			= (1<<20),
1232 	IB_QP_RESERVED1			= (1<<21),
1233 	IB_QP_RESERVED2			= (1<<22),
1234 	IB_QP_RESERVED3			= (1<<23),
1235 	IB_QP_RESERVED4			= (1<<24),
1236 	IB_QP_RATE_LIMIT		= (1<<25),
1237 };
1238 
1239 enum ib_qp_state {
1240 	IB_QPS_RESET,
1241 	IB_QPS_INIT,
1242 	IB_QPS_RTR,
1243 	IB_QPS_RTS,
1244 	IB_QPS_SQD,
1245 	IB_QPS_SQE,
1246 	IB_QPS_ERR
1247 };
1248 
1249 enum ib_mig_state {
1250 	IB_MIG_MIGRATED,
1251 	IB_MIG_REARM,
1252 	IB_MIG_ARMED
1253 };
1254 
1255 enum ib_mw_type {
1256 	IB_MW_TYPE_1 = 1,
1257 	IB_MW_TYPE_2 = 2
1258 };
1259 
1260 struct ib_qp_attr {
1261 	enum ib_qp_state	qp_state;
1262 	enum ib_qp_state	cur_qp_state;
1263 	enum ib_mtu		path_mtu;
1264 	enum ib_mig_state	path_mig_state;
1265 	u32			qkey;
1266 	u32			rq_psn;
1267 	u32			sq_psn;
1268 	u32			dest_qp_num;
1269 	int			qp_access_flags;
1270 	struct ib_qp_cap	cap;
1271 	struct rdma_ah_attr	ah_attr;
1272 	struct rdma_ah_attr	alt_ah_attr;
1273 	u16			pkey_index;
1274 	u16			alt_pkey_index;
1275 	u8			en_sqd_async_notify;
1276 	u8			sq_draining;
1277 	u8			max_rd_atomic;
1278 	u8			max_dest_rd_atomic;
1279 	u8			min_rnr_timer;
1280 	u8			port_num;
1281 	u8			timeout;
1282 	u8			retry_cnt;
1283 	u8			rnr_retry;
1284 	u8			alt_port_num;
1285 	u8			alt_timeout;
1286 	u32			rate_limit;
1287 };
1288 
1289 enum ib_wr_opcode {
1290 	/* These are shared with userspace */
1291 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1292 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1293 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1294 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1295 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1296 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1297 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1298 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1299 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1300 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1301 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1302 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1303 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1304 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1305 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1306 
1307 	/* These are kernel only and can not be issued by userspace */
1308 	IB_WR_REG_MR = 0x20,
1309 	IB_WR_REG_SIG_MR,
1310 
1311 	/* reserve values for low level drivers' internal use.
1312 	 * These values will not be used at all in the ib core layer.
1313 	 */
1314 	IB_WR_RESERVED1 = 0xf0,
1315 	IB_WR_RESERVED2,
1316 	IB_WR_RESERVED3,
1317 	IB_WR_RESERVED4,
1318 	IB_WR_RESERVED5,
1319 	IB_WR_RESERVED6,
1320 	IB_WR_RESERVED7,
1321 	IB_WR_RESERVED8,
1322 	IB_WR_RESERVED9,
1323 	IB_WR_RESERVED10,
1324 };
1325 
1326 enum ib_send_flags {
1327 	IB_SEND_FENCE		= 1,
1328 	IB_SEND_SIGNALED	= (1<<1),
1329 	IB_SEND_SOLICITED	= (1<<2),
1330 	IB_SEND_INLINE		= (1<<3),
1331 	IB_SEND_IP_CSUM		= (1<<4),
1332 
1333 	/* reserve bits 26-31 for low level drivers' internal use */
1334 	IB_SEND_RESERVED_START	= (1 << 26),
1335 	IB_SEND_RESERVED_END	= (1 << 31),
1336 };
1337 
1338 struct ib_sge {
1339 	u64	addr;
1340 	u32	length;
1341 	u32	lkey;
1342 };
1343 
1344 struct ib_cqe {
1345 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1346 };
1347 
1348 struct ib_send_wr {
1349 	struct ib_send_wr      *next;
1350 	union {
1351 		u64		wr_id;
1352 		struct ib_cqe	*wr_cqe;
1353 	};
1354 	struct ib_sge	       *sg_list;
1355 	int			num_sge;
1356 	enum ib_wr_opcode	opcode;
1357 	int			send_flags;
1358 	union {
1359 		__be32		imm_data;
1360 		u32		invalidate_rkey;
1361 	} ex;
1362 };
1363 
1364 struct ib_rdma_wr {
1365 	struct ib_send_wr	wr;
1366 	u64			remote_addr;
1367 	u32			rkey;
1368 };
1369 
1370 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1371 {
1372 	return container_of(wr, struct ib_rdma_wr, wr);
1373 }
1374 
1375 struct ib_atomic_wr {
1376 	struct ib_send_wr	wr;
1377 	u64			remote_addr;
1378 	u64			compare_add;
1379 	u64			swap;
1380 	u64			compare_add_mask;
1381 	u64			swap_mask;
1382 	u32			rkey;
1383 };
1384 
1385 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1386 {
1387 	return container_of(wr, struct ib_atomic_wr, wr);
1388 }
1389 
1390 struct ib_ud_wr {
1391 	struct ib_send_wr	wr;
1392 	struct ib_ah		*ah;
1393 	void			*header;
1394 	int			hlen;
1395 	int			mss;
1396 	u32			remote_qpn;
1397 	u32			remote_qkey;
1398 	u16			pkey_index; /* valid for GSI only */
1399 	u8			port_num;   /* valid for DR SMPs on switch only */
1400 };
1401 
1402 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1403 {
1404 	return container_of(wr, struct ib_ud_wr, wr);
1405 }
1406 
1407 struct ib_reg_wr {
1408 	struct ib_send_wr	wr;
1409 	struct ib_mr		*mr;
1410 	u32			key;
1411 	int			access;
1412 };
1413 
1414 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1415 {
1416 	return container_of(wr, struct ib_reg_wr, wr);
1417 }
1418 
1419 struct ib_sig_handover_wr {
1420 	struct ib_send_wr	wr;
1421 	struct ib_sig_attrs    *sig_attrs;
1422 	struct ib_mr	       *sig_mr;
1423 	int			access_flags;
1424 	struct ib_sge	       *prot;
1425 };
1426 
1427 static inline const struct ib_sig_handover_wr *
1428 sig_handover_wr(const struct ib_send_wr *wr)
1429 {
1430 	return container_of(wr, struct ib_sig_handover_wr, wr);
1431 }
1432 
1433 struct ib_recv_wr {
1434 	struct ib_recv_wr      *next;
1435 	union {
1436 		u64		wr_id;
1437 		struct ib_cqe	*wr_cqe;
1438 	};
1439 	struct ib_sge	       *sg_list;
1440 	int			num_sge;
1441 };
1442 
1443 enum ib_access_flags {
1444 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1445 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1446 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1447 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1448 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1449 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1450 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1451 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1452 
1453 	IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1454 };
1455 
1456 /*
1457  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1458  * are hidden here instead of a uapi header!
1459  */
1460 enum ib_mr_rereg_flags {
1461 	IB_MR_REREG_TRANS	= 1,
1462 	IB_MR_REREG_PD		= (1<<1),
1463 	IB_MR_REREG_ACCESS	= (1<<2),
1464 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1465 };
1466 
1467 struct ib_fmr_attr {
1468 	int	max_pages;
1469 	int	max_maps;
1470 	u8	page_shift;
1471 };
1472 
1473 struct ib_umem;
1474 
1475 enum rdma_remove_reason {
1476 	/*
1477 	 * Userspace requested uobject deletion or initial try
1478 	 * to remove uobject via cleanup. Call could fail
1479 	 */
1480 	RDMA_REMOVE_DESTROY,
1481 	/* Context deletion. This call should delete the actual object itself */
1482 	RDMA_REMOVE_CLOSE,
1483 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1484 	RDMA_REMOVE_DRIVER_REMOVE,
1485 	/* uobj is being cleaned-up before being committed */
1486 	RDMA_REMOVE_ABORT,
1487 };
1488 
1489 struct ib_rdmacg_object {
1490 #ifdef CONFIG_CGROUP_RDMA
1491 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1492 #endif
1493 };
1494 
1495 struct ib_ucontext {
1496 	struct ib_device       *device;
1497 	struct ib_uverbs_file  *ufile;
1498 	/*
1499 	 * 'closing' can be read by the driver only during a destroy callback,
1500 	 * it is set when we are closing the file descriptor and indicates
1501 	 * that mm_sem may be locked.
1502 	 */
1503 	bool closing;
1504 
1505 	bool cleanup_retryable;
1506 
1507 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1508 	void (*invalidate_range)(struct ib_umem_odp *umem_odp,
1509 				 unsigned long start, unsigned long end);
1510 	struct mutex per_mm_list_lock;
1511 	struct list_head per_mm_list;
1512 #endif
1513 
1514 	struct ib_rdmacg_object	cg_obj;
1515 	/*
1516 	 * Implementation details of the RDMA core, don't use in drivers:
1517 	 */
1518 	struct rdma_restrack_entry res;
1519 };
1520 
1521 struct ib_uobject {
1522 	u64			user_handle;	/* handle given to us by userspace */
1523 	/* ufile & ucontext owning this object */
1524 	struct ib_uverbs_file  *ufile;
1525 	/* FIXME, save memory: ufile->context == context */
1526 	struct ib_ucontext     *context;	/* associated user context */
1527 	void		       *object;		/* containing object */
1528 	struct list_head	list;		/* link to context's list */
1529 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1530 	int			id;		/* index into kernel idr */
1531 	struct kref		ref;
1532 	atomic_t		usecnt;		/* protects exclusive access */
1533 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1534 
1535 	const struct uverbs_api_object *uapi_object;
1536 };
1537 
1538 struct ib_udata {
1539 	const void __user *inbuf;
1540 	void __user *outbuf;
1541 	size_t       inlen;
1542 	size_t       outlen;
1543 };
1544 
1545 struct ib_pd {
1546 	u32			local_dma_lkey;
1547 	u32			flags;
1548 	struct ib_device       *device;
1549 	struct ib_uobject      *uobject;
1550 	atomic_t          	usecnt; /* count all resources */
1551 
1552 	u32			unsafe_global_rkey;
1553 
1554 	/*
1555 	 * Implementation details of the RDMA core, don't use in drivers:
1556 	 */
1557 	struct ib_mr	       *__internal_mr;
1558 	struct rdma_restrack_entry res;
1559 };
1560 
1561 struct ib_xrcd {
1562 	struct ib_device       *device;
1563 	atomic_t		usecnt; /* count all exposed resources */
1564 	struct inode	       *inode;
1565 
1566 	struct mutex		tgt_qp_mutex;
1567 	struct list_head	tgt_qp_list;
1568 };
1569 
1570 struct ib_ah {
1571 	struct ib_device	*device;
1572 	struct ib_pd		*pd;
1573 	struct ib_uobject	*uobject;
1574 	const struct ib_gid_attr *sgid_attr;
1575 	enum rdma_ah_attr_type	type;
1576 };
1577 
1578 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1579 
1580 enum ib_poll_context {
1581 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1582 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1583 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1584 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1585 };
1586 
1587 struct ib_cq {
1588 	struct ib_device       *device;
1589 	struct ib_uobject      *uobject;
1590 	ib_comp_handler   	comp_handler;
1591 	void                  (*event_handler)(struct ib_event *, void *);
1592 	void                   *cq_context;
1593 	int               	cqe;
1594 	atomic_t          	usecnt; /* count number of work queues */
1595 	enum ib_poll_context	poll_ctx;
1596 	struct ib_wc		*wc;
1597 	union {
1598 		struct irq_poll		iop;
1599 		struct work_struct	work;
1600 	};
1601 	struct workqueue_struct *comp_wq;
1602 	/*
1603 	 * Implementation details of the RDMA core, don't use in drivers:
1604 	 */
1605 	struct rdma_restrack_entry res;
1606 };
1607 
1608 struct ib_srq {
1609 	struct ib_device       *device;
1610 	struct ib_pd	       *pd;
1611 	struct ib_uobject      *uobject;
1612 	void		      (*event_handler)(struct ib_event *, void *);
1613 	void		       *srq_context;
1614 	enum ib_srq_type	srq_type;
1615 	atomic_t		usecnt;
1616 
1617 	struct {
1618 		struct ib_cq   *cq;
1619 		union {
1620 			struct {
1621 				struct ib_xrcd *xrcd;
1622 				u32		srq_num;
1623 			} xrc;
1624 		};
1625 	} ext;
1626 };
1627 
1628 enum ib_raw_packet_caps {
1629 	/* Strip cvlan from incoming packet and report it in the matching work
1630 	 * completion is supported.
1631 	 */
1632 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1633 	/* Scatter FCS field of an incoming packet to host memory is supported.
1634 	 */
1635 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1636 	/* Checksum offloads are supported (for both send and receive). */
1637 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1638 	/* When a packet is received for an RQ with no receive WQEs, the
1639 	 * packet processing is delayed.
1640 	 */
1641 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1642 };
1643 
1644 enum ib_wq_type {
1645 	IB_WQT_RQ
1646 };
1647 
1648 enum ib_wq_state {
1649 	IB_WQS_RESET,
1650 	IB_WQS_RDY,
1651 	IB_WQS_ERR
1652 };
1653 
1654 struct ib_wq {
1655 	struct ib_device       *device;
1656 	struct ib_uobject      *uobject;
1657 	void		    *wq_context;
1658 	void		    (*event_handler)(struct ib_event *, void *);
1659 	struct ib_pd	       *pd;
1660 	struct ib_cq	       *cq;
1661 	u32		wq_num;
1662 	enum ib_wq_state       state;
1663 	enum ib_wq_type	wq_type;
1664 	atomic_t		usecnt;
1665 };
1666 
1667 enum ib_wq_flags {
1668 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1669 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1670 	IB_WQ_FLAGS_DELAY_DROP		= 1 << 2,
1671 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1672 };
1673 
1674 struct ib_wq_init_attr {
1675 	void		       *wq_context;
1676 	enum ib_wq_type	wq_type;
1677 	u32		max_wr;
1678 	u32		max_sge;
1679 	struct	ib_cq	       *cq;
1680 	void		    (*event_handler)(struct ib_event *, void *);
1681 	u32		create_flags; /* Use enum ib_wq_flags */
1682 };
1683 
1684 enum ib_wq_attr_mask {
1685 	IB_WQ_STATE		= 1 << 0,
1686 	IB_WQ_CUR_STATE		= 1 << 1,
1687 	IB_WQ_FLAGS		= 1 << 2,
1688 };
1689 
1690 struct ib_wq_attr {
1691 	enum	ib_wq_state	wq_state;
1692 	enum	ib_wq_state	curr_wq_state;
1693 	u32			flags; /* Use enum ib_wq_flags */
1694 	u32			flags_mask; /* Use enum ib_wq_flags */
1695 };
1696 
1697 struct ib_rwq_ind_table {
1698 	struct ib_device	*device;
1699 	struct ib_uobject      *uobject;
1700 	atomic_t		usecnt;
1701 	u32		ind_tbl_num;
1702 	u32		log_ind_tbl_size;
1703 	struct ib_wq	**ind_tbl;
1704 };
1705 
1706 struct ib_rwq_ind_table_init_attr {
1707 	u32		log_ind_tbl_size;
1708 	/* Each entry is a pointer to Receive Work Queue */
1709 	struct ib_wq	**ind_tbl;
1710 };
1711 
1712 enum port_pkey_state {
1713 	IB_PORT_PKEY_NOT_VALID = 0,
1714 	IB_PORT_PKEY_VALID = 1,
1715 	IB_PORT_PKEY_LISTED = 2,
1716 };
1717 
1718 struct ib_qp_security;
1719 
1720 struct ib_port_pkey {
1721 	enum port_pkey_state	state;
1722 	u16			pkey_index;
1723 	u8			port_num;
1724 	struct list_head	qp_list;
1725 	struct list_head	to_error_list;
1726 	struct ib_qp_security  *sec;
1727 };
1728 
1729 struct ib_ports_pkeys {
1730 	struct ib_port_pkey	main;
1731 	struct ib_port_pkey	alt;
1732 };
1733 
1734 struct ib_qp_security {
1735 	struct ib_qp	       *qp;
1736 	struct ib_device       *dev;
1737 	/* Hold this mutex when changing port and pkey settings. */
1738 	struct mutex		mutex;
1739 	struct ib_ports_pkeys  *ports_pkeys;
1740 	/* A list of all open shared QP handles.  Required to enforce security
1741 	 * properly for all users of a shared QP.
1742 	 */
1743 	struct list_head        shared_qp_list;
1744 	void                   *security;
1745 	bool			destroying;
1746 	atomic_t		error_list_count;
1747 	struct completion	error_complete;
1748 	int			error_comps_pending;
1749 };
1750 
1751 /*
1752  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1753  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1754  */
1755 struct ib_qp {
1756 	struct ib_device       *device;
1757 	struct ib_pd	       *pd;
1758 	struct ib_cq	       *send_cq;
1759 	struct ib_cq	       *recv_cq;
1760 	spinlock_t		mr_lock;
1761 	int			mrs_used;
1762 	struct list_head	rdma_mrs;
1763 	struct list_head	sig_mrs;
1764 	struct ib_srq	       *srq;
1765 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1766 	struct list_head	xrcd_list;
1767 
1768 	/* count times opened, mcast attaches, flow attaches */
1769 	atomic_t		usecnt;
1770 	struct list_head	open_list;
1771 	struct ib_qp           *real_qp;
1772 	struct ib_uobject      *uobject;
1773 	void                  (*event_handler)(struct ib_event *, void *);
1774 	void		       *qp_context;
1775 	/* sgid_attrs associated with the AV's */
1776 	const struct ib_gid_attr *av_sgid_attr;
1777 	const struct ib_gid_attr *alt_path_sgid_attr;
1778 	u32			qp_num;
1779 	u32			max_write_sge;
1780 	u32			max_read_sge;
1781 	enum ib_qp_type		qp_type;
1782 	struct ib_rwq_ind_table *rwq_ind_tbl;
1783 	struct ib_qp_security  *qp_sec;
1784 	u8			port;
1785 
1786 	/*
1787 	 * Implementation details of the RDMA core, don't use in drivers:
1788 	 */
1789 	struct rdma_restrack_entry     res;
1790 };
1791 
1792 struct ib_dm {
1793 	struct ib_device  *device;
1794 	u32		   length;
1795 	u32		   flags;
1796 	struct ib_uobject *uobject;
1797 	atomic_t	   usecnt;
1798 };
1799 
1800 struct ib_mr {
1801 	struct ib_device  *device;
1802 	struct ib_pd	  *pd;
1803 	u32		   lkey;
1804 	u32		   rkey;
1805 	u64		   iova;
1806 	u64		   length;
1807 	unsigned int	   page_size;
1808 	bool		   need_inval;
1809 	union {
1810 		struct ib_uobject	*uobject;	/* user */
1811 		struct list_head	qp_entry;	/* FR */
1812 	};
1813 
1814 	struct ib_dm      *dm;
1815 
1816 	/*
1817 	 * Implementation details of the RDMA core, don't use in drivers:
1818 	 */
1819 	struct rdma_restrack_entry res;
1820 };
1821 
1822 struct ib_mw {
1823 	struct ib_device	*device;
1824 	struct ib_pd		*pd;
1825 	struct ib_uobject	*uobject;
1826 	u32			rkey;
1827 	enum ib_mw_type         type;
1828 };
1829 
1830 struct ib_fmr {
1831 	struct ib_device	*device;
1832 	struct ib_pd		*pd;
1833 	struct list_head	list;
1834 	u32			lkey;
1835 	u32			rkey;
1836 };
1837 
1838 /* Supported steering options */
1839 enum ib_flow_attr_type {
1840 	/* steering according to rule specifications */
1841 	IB_FLOW_ATTR_NORMAL		= 0x0,
1842 	/* default unicast and multicast rule -
1843 	 * receive all Eth traffic which isn't steered to any QP
1844 	 */
1845 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1846 	/* default multicast rule -
1847 	 * receive all Eth multicast traffic which isn't steered to any QP
1848 	 */
1849 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1850 	/* sniffer rule - receive all port traffic */
1851 	IB_FLOW_ATTR_SNIFFER		= 0x3
1852 };
1853 
1854 /* Supported steering header types */
1855 enum ib_flow_spec_type {
1856 	/* L2 headers*/
1857 	IB_FLOW_SPEC_ETH		= 0x20,
1858 	IB_FLOW_SPEC_IB			= 0x22,
1859 	/* L3 header*/
1860 	IB_FLOW_SPEC_IPV4		= 0x30,
1861 	IB_FLOW_SPEC_IPV6		= 0x31,
1862 	IB_FLOW_SPEC_ESP                = 0x34,
1863 	/* L4 headers*/
1864 	IB_FLOW_SPEC_TCP		= 0x40,
1865 	IB_FLOW_SPEC_UDP		= 0x41,
1866 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1867 	IB_FLOW_SPEC_GRE		= 0x51,
1868 	IB_FLOW_SPEC_MPLS		= 0x60,
1869 	IB_FLOW_SPEC_INNER		= 0x100,
1870 	/* Actions */
1871 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1872 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1873 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1874 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1875 };
1876 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1877 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1878 
1879 /* Flow steering rule priority is set according to it's domain.
1880  * Lower domain value means higher priority.
1881  */
1882 enum ib_flow_domain {
1883 	IB_FLOW_DOMAIN_USER,
1884 	IB_FLOW_DOMAIN_ETHTOOL,
1885 	IB_FLOW_DOMAIN_RFS,
1886 	IB_FLOW_DOMAIN_NIC,
1887 	IB_FLOW_DOMAIN_NUM /* Must be last */
1888 };
1889 
1890 enum ib_flow_flags {
1891 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1892 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1893 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1894 };
1895 
1896 struct ib_flow_eth_filter {
1897 	u8	dst_mac[6];
1898 	u8	src_mac[6];
1899 	__be16	ether_type;
1900 	__be16	vlan_tag;
1901 	/* Must be last */
1902 	u8	real_sz[0];
1903 };
1904 
1905 struct ib_flow_spec_eth {
1906 	u32			  type;
1907 	u16			  size;
1908 	struct ib_flow_eth_filter val;
1909 	struct ib_flow_eth_filter mask;
1910 };
1911 
1912 struct ib_flow_ib_filter {
1913 	__be16 dlid;
1914 	__u8   sl;
1915 	/* Must be last */
1916 	u8	real_sz[0];
1917 };
1918 
1919 struct ib_flow_spec_ib {
1920 	u32			 type;
1921 	u16			 size;
1922 	struct ib_flow_ib_filter val;
1923 	struct ib_flow_ib_filter mask;
1924 };
1925 
1926 /* IPv4 header flags */
1927 enum ib_ipv4_flags {
1928 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1929 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1930 				    last have this flag set */
1931 };
1932 
1933 struct ib_flow_ipv4_filter {
1934 	__be32	src_ip;
1935 	__be32	dst_ip;
1936 	u8	proto;
1937 	u8	tos;
1938 	u8	ttl;
1939 	u8	flags;
1940 	/* Must be last */
1941 	u8	real_sz[0];
1942 };
1943 
1944 struct ib_flow_spec_ipv4 {
1945 	u32			   type;
1946 	u16			   size;
1947 	struct ib_flow_ipv4_filter val;
1948 	struct ib_flow_ipv4_filter mask;
1949 };
1950 
1951 struct ib_flow_ipv6_filter {
1952 	u8	src_ip[16];
1953 	u8	dst_ip[16];
1954 	__be32	flow_label;
1955 	u8	next_hdr;
1956 	u8	traffic_class;
1957 	u8	hop_limit;
1958 	/* Must be last */
1959 	u8	real_sz[0];
1960 };
1961 
1962 struct ib_flow_spec_ipv6 {
1963 	u32			   type;
1964 	u16			   size;
1965 	struct ib_flow_ipv6_filter val;
1966 	struct ib_flow_ipv6_filter mask;
1967 };
1968 
1969 struct ib_flow_tcp_udp_filter {
1970 	__be16	dst_port;
1971 	__be16	src_port;
1972 	/* Must be last */
1973 	u8	real_sz[0];
1974 };
1975 
1976 struct ib_flow_spec_tcp_udp {
1977 	u32			      type;
1978 	u16			      size;
1979 	struct ib_flow_tcp_udp_filter val;
1980 	struct ib_flow_tcp_udp_filter mask;
1981 };
1982 
1983 struct ib_flow_tunnel_filter {
1984 	__be32	tunnel_id;
1985 	u8	real_sz[0];
1986 };
1987 
1988 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1989  * the tunnel_id from val has the vni value
1990  */
1991 struct ib_flow_spec_tunnel {
1992 	u32			      type;
1993 	u16			      size;
1994 	struct ib_flow_tunnel_filter  val;
1995 	struct ib_flow_tunnel_filter  mask;
1996 };
1997 
1998 struct ib_flow_esp_filter {
1999 	__be32	spi;
2000 	__be32  seq;
2001 	/* Must be last */
2002 	u8	real_sz[0];
2003 };
2004 
2005 struct ib_flow_spec_esp {
2006 	u32                           type;
2007 	u16			      size;
2008 	struct ib_flow_esp_filter     val;
2009 	struct ib_flow_esp_filter     mask;
2010 };
2011 
2012 struct ib_flow_gre_filter {
2013 	__be16 c_ks_res0_ver;
2014 	__be16 protocol;
2015 	__be32 key;
2016 	/* Must be last */
2017 	u8	real_sz[0];
2018 };
2019 
2020 struct ib_flow_spec_gre {
2021 	u32                           type;
2022 	u16			      size;
2023 	struct ib_flow_gre_filter     val;
2024 	struct ib_flow_gre_filter     mask;
2025 };
2026 
2027 struct ib_flow_mpls_filter {
2028 	__be32 tag;
2029 	/* Must be last */
2030 	u8	real_sz[0];
2031 };
2032 
2033 struct ib_flow_spec_mpls {
2034 	u32                           type;
2035 	u16			      size;
2036 	struct ib_flow_mpls_filter     val;
2037 	struct ib_flow_mpls_filter     mask;
2038 };
2039 
2040 struct ib_flow_spec_action_tag {
2041 	enum ib_flow_spec_type	      type;
2042 	u16			      size;
2043 	u32                           tag_id;
2044 };
2045 
2046 struct ib_flow_spec_action_drop {
2047 	enum ib_flow_spec_type	      type;
2048 	u16			      size;
2049 };
2050 
2051 struct ib_flow_spec_action_handle {
2052 	enum ib_flow_spec_type	      type;
2053 	u16			      size;
2054 	struct ib_flow_action	     *act;
2055 };
2056 
2057 enum ib_counters_description {
2058 	IB_COUNTER_PACKETS,
2059 	IB_COUNTER_BYTES,
2060 };
2061 
2062 struct ib_flow_spec_action_count {
2063 	enum ib_flow_spec_type type;
2064 	u16 size;
2065 	struct ib_counters *counters;
2066 };
2067 
2068 union ib_flow_spec {
2069 	struct {
2070 		u32			type;
2071 		u16			size;
2072 	};
2073 	struct ib_flow_spec_eth		eth;
2074 	struct ib_flow_spec_ib		ib;
2075 	struct ib_flow_spec_ipv4        ipv4;
2076 	struct ib_flow_spec_tcp_udp	tcp_udp;
2077 	struct ib_flow_spec_ipv6        ipv6;
2078 	struct ib_flow_spec_tunnel      tunnel;
2079 	struct ib_flow_spec_esp		esp;
2080 	struct ib_flow_spec_gre		gre;
2081 	struct ib_flow_spec_mpls	mpls;
2082 	struct ib_flow_spec_action_tag  flow_tag;
2083 	struct ib_flow_spec_action_drop drop;
2084 	struct ib_flow_spec_action_handle action;
2085 	struct ib_flow_spec_action_count flow_count;
2086 };
2087 
2088 struct ib_flow_attr {
2089 	enum ib_flow_attr_type type;
2090 	u16	     size;
2091 	u16	     priority;
2092 	u32	     flags;
2093 	u8	     num_of_specs;
2094 	u8	     port;
2095 	union ib_flow_spec flows[];
2096 };
2097 
2098 struct ib_flow {
2099 	struct ib_qp		*qp;
2100 	struct ib_device	*device;
2101 	struct ib_uobject	*uobject;
2102 };
2103 
2104 enum ib_flow_action_type {
2105 	IB_FLOW_ACTION_UNSPECIFIED,
2106 	IB_FLOW_ACTION_ESP = 1,
2107 };
2108 
2109 struct ib_flow_action_attrs_esp_keymats {
2110 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2111 	union {
2112 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2113 	} keymat;
2114 };
2115 
2116 struct ib_flow_action_attrs_esp_replays {
2117 	enum ib_uverbs_flow_action_esp_replay			protocol;
2118 	union {
2119 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2120 	} replay;
2121 };
2122 
2123 enum ib_flow_action_attrs_esp_flags {
2124 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2125 	 * This is done in order to share the same flags between user-space and
2126 	 * kernel and spare an unnecessary translation.
2127 	 */
2128 
2129 	/* Kernel flags */
2130 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2131 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2132 };
2133 
2134 struct ib_flow_spec_list {
2135 	struct ib_flow_spec_list	*next;
2136 	union ib_flow_spec		spec;
2137 };
2138 
2139 struct ib_flow_action_attrs_esp {
2140 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2141 	struct ib_flow_action_attrs_esp_replays		*replay;
2142 	struct ib_flow_spec_list			*encap;
2143 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2144 	 * Value of 0 is a valid value.
2145 	 */
2146 	u32						esn;
2147 	u32						spi;
2148 	u32						seq;
2149 	u32						tfc_pad;
2150 	/* Use enum ib_flow_action_attrs_esp_flags */
2151 	u64						flags;
2152 	u64						hard_limit_pkts;
2153 };
2154 
2155 struct ib_flow_action {
2156 	struct ib_device		*device;
2157 	struct ib_uobject		*uobject;
2158 	enum ib_flow_action_type	type;
2159 	atomic_t			usecnt;
2160 };
2161 
2162 struct ib_mad_hdr;
2163 struct ib_grh;
2164 
2165 enum ib_process_mad_flags {
2166 	IB_MAD_IGNORE_MKEY	= 1,
2167 	IB_MAD_IGNORE_BKEY	= 2,
2168 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2169 };
2170 
2171 enum ib_mad_result {
2172 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2173 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2174 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2175 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2176 };
2177 
2178 struct ib_port_cache {
2179 	u64		      subnet_prefix;
2180 	struct ib_pkey_cache  *pkey;
2181 	struct ib_gid_table   *gid;
2182 	u8                     lmc;
2183 	enum ib_port_state     port_state;
2184 };
2185 
2186 struct ib_cache {
2187 	rwlock_t                lock;
2188 	struct ib_event_handler event_handler;
2189 	struct ib_port_cache   *ports;
2190 };
2191 
2192 struct iw_cm_verbs;
2193 
2194 struct ib_port_immutable {
2195 	int                           pkey_tbl_len;
2196 	int                           gid_tbl_len;
2197 	u32                           core_cap_flags;
2198 	u32                           max_mad_size;
2199 };
2200 
2201 /* rdma netdev type - specifies protocol type */
2202 enum rdma_netdev_t {
2203 	RDMA_NETDEV_OPA_VNIC,
2204 	RDMA_NETDEV_IPOIB,
2205 };
2206 
2207 /**
2208  * struct rdma_netdev - rdma netdev
2209  * For cases where netstack interfacing is required.
2210  */
2211 struct rdma_netdev {
2212 	void              *clnt_priv;
2213 	struct ib_device  *hca;
2214 	u8                 port_num;
2215 
2216 	/*
2217 	 * cleanup function must be specified.
2218 	 * FIXME: This is only used for OPA_VNIC and that usage should be
2219 	 * removed too.
2220 	 */
2221 	void (*free_rdma_netdev)(struct net_device *netdev);
2222 
2223 	/* control functions */
2224 	void (*set_id)(struct net_device *netdev, int id);
2225 	/* send packet */
2226 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2227 		    struct ib_ah *address, u32 dqpn);
2228 	/* multicast */
2229 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2230 			    union ib_gid *gid, u16 mlid,
2231 			    int set_qkey, u32 qkey);
2232 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2233 			    union ib_gid *gid, u16 mlid);
2234 };
2235 
2236 struct rdma_netdev_alloc_params {
2237 	size_t sizeof_priv;
2238 	unsigned int txqs;
2239 	unsigned int rxqs;
2240 	void *param;
2241 
2242 	int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2243 				      struct net_device *netdev, void *param);
2244 };
2245 
2246 struct ib_port_pkey_list {
2247 	/* Lock to hold while modifying the list. */
2248 	spinlock_t                    list_lock;
2249 	struct list_head              pkey_list;
2250 };
2251 
2252 struct ib_counters {
2253 	struct ib_device	*device;
2254 	struct ib_uobject	*uobject;
2255 	/* num of objects attached */
2256 	atomic_t	usecnt;
2257 };
2258 
2259 struct ib_counters_read_attr {
2260 	u64	*counters_buff;
2261 	u32	ncounters;
2262 	u32	flags; /* use enum ib_read_counters_flags */
2263 };
2264 
2265 struct uverbs_attr_bundle;
2266 
2267 /**
2268  * struct ib_device_ops - InfiniBand device operations
2269  * This structure defines all the InfiniBand device operations, providers will
2270  * need to define the supported operations, otherwise they will be set to null.
2271  */
2272 struct ib_device_ops {
2273 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2274 			 const struct ib_send_wr **bad_send_wr);
2275 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2276 			 const struct ib_recv_wr **bad_recv_wr);
2277 	void (*drain_rq)(struct ib_qp *qp);
2278 	void (*drain_sq)(struct ib_qp *qp);
2279 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2280 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2281 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2282 	int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2283 	int (*post_srq_recv)(struct ib_srq *srq,
2284 			     const struct ib_recv_wr *recv_wr,
2285 			     const struct ib_recv_wr **bad_recv_wr);
2286 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2287 			   u8 port_num, const struct ib_wc *in_wc,
2288 			   const struct ib_grh *in_grh,
2289 			   const struct ib_mad_hdr *in_mad, size_t in_mad_size,
2290 			   struct ib_mad_hdr *out_mad, size_t *out_mad_size,
2291 			   u16 *out_mad_pkey_index);
2292 	int (*query_device)(struct ib_device *device,
2293 			    struct ib_device_attr *device_attr,
2294 			    struct ib_udata *udata);
2295 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2296 			     struct ib_device_modify *device_modify);
2297 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2298 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2299 						     int comp_vector);
2300 	int (*query_port)(struct ib_device *device, u8 port_num,
2301 			  struct ib_port_attr *port_attr);
2302 	int (*modify_port)(struct ib_device *device, u8 port_num,
2303 			   int port_modify_mask,
2304 			   struct ib_port_modify *port_modify);
2305 	/**
2306 	 * The following mandatory functions are used only at device
2307 	 * registration.  Keep functions such as these at the end of this
2308 	 * structure to avoid cache line misses when accessing struct ib_device
2309 	 * in fast paths.
2310 	 */
2311 	int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2312 				  struct ib_port_immutable *immutable);
2313 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2314 					       u8 port_num);
2315 	/**
2316 	 * When calling get_netdev, the HW vendor's driver should return the
2317 	 * net device of device @device at port @port_num or NULL if such
2318 	 * a net device doesn't exist. The vendor driver should call dev_hold
2319 	 * on this net device. The HW vendor's device driver must guarantee
2320 	 * that this function returns NULL before the net device has finished
2321 	 * NETDEV_UNREGISTER state.
2322 	 */
2323 	struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2324 	/**
2325 	 * rdma netdev operation
2326 	 *
2327 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2328 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2329 	 */
2330 	struct net_device *(*alloc_rdma_netdev)(
2331 		struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2332 		const char *name, unsigned char name_assign_type,
2333 		void (*setup)(struct net_device *));
2334 
2335 	int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2336 				      enum rdma_netdev_t type,
2337 				      struct rdma_netdev_alloc_params *params);
2338 	/**
2339 	 * query_gid should be return GID value for @device, when @port_num
2340 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2341 	 * is RoCE link layer.
2342 	 */
2343 	int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2344 			 union ib_gid *gid);
2345 	/**
2346 	 * When calling add_gid, the HW vendor's driver should add the gid
2347 	 * of device of port at gid index available at @attr. Meta-info of
2348 	 * that gid (for example, the network device related to this gid) is
2349 	 * available at @attr. @context allows the HW vendor driver to store
2350 	 * extra information together with a GID entry. The HW vendor driver may
2351 	 * allocate memory to contain this information and store it in @context
2352 	 * when a new GID entry is written to. Params are consistent until the
2353 	 * next call of add_gid or delete_gid. The function should return 0 on
2354 	 * success or error otherwise. The function could be called
2355 	 * concurrently for different ports. This function is only called when
2356 	 * roce_gid_table is used.
2357 	 */
2358 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2359 	/**
2360 	 * When calling del_gid, the HW vendor's driver should delete the
2361 	 * gid of device @device at gid index gid_index of port port_num
2362 	 * available in @attr.
2363 	 * Upon the deletion of a GID entry, the HW vendor must free any
2364 	 * allocated memory. The caller will clear @context afterwards.
2365 	 * This function is only called when roce_gid_table is used.
2366 	 */
2367 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2368 	int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2369 			  u16 *pkey);
2370 	struct ib_ucontext *(*alloc_ucontext)(struct ib_device *device,
2371 					      struct ib_udata *udata);
2372 	int (*dealloc_ucontext)(struct ib_ucontext *context);
2373 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2374 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2375 	struct ib_pd *(*alloc_pd)(struct ib_device *device,
2376 				  struct ib_ucontext *context,
2377 				  struct ib_udata *udata);
2378 	int (*dealloc_pd)(struct ib_pd *pd);
2379 	struct ib_ah *(*create_ah)(struct ib_pd *pd,
2380 				   struct rdma_ah_attr *ah_attr, u32 flags,
2381 				   struct ib_udata *udata);
2382 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2383 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2384 	int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2385 	struct ib_srq *(*create_srq)(struct ib_pd *pd,
2386 				     struct ib_srq_init_attr *srq_init_attr,
2387 				     struct ib_udata *udata);
2388 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2389 			  enum ib_srq_attr_mask srq_attr_mask,
2390 			  struct ib_udata *udata);
2391 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2392 	int (*destroy_srq)(struct ib_srq *srq);
2393 	struct ib_qp *(*create_qp)(struct ib_pd *pd,
2394 				   struct ib_qp_init_attr *qp_init_attr,
2395 				   struct ib_udata *udata);
2396 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2397 			 int qp_attr_mask, struct ib_udata *udata);
2398 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2399 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2400 	int (*destroy_qp)(struct ib_qp *qp);
2401 	struct ib_cq *(*create_cq)(struct ib_device *device,
2402 				   const struct ib_cq_init_attr *attr,
2403 				   struct ib_ucontext *context,
2404 				   struct ib_udata *udata);
2405 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2406 	int (*destroy_cq)(struct ib_cq *cq);
2407 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2408 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2409 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2410 				     u64 virt_addr, int mr_access_flags,
2411 				     struct ib_udata *udata);
2412 	int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2413 			     u64 virt_addr, int mr_access_flags,
2414 			     struct ib_pd *pd, struct ib_udata *udata);
2415 	int (*dereg_mr)(struct ib_mr *mr);
2416 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2417 				  u32 max_num_sg);
2418 	int (*advise_mr)(struct ib_pd *pd,
2419 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2420 			 struct ib_sge *sg_list, u32 num_sge,
2421 			 struct uverbs_attr_bundle *attrs);
2422 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2423 			 unsigned int *sg_offset);
2424 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2425 			       struct ib_mr_status *mr_status);
2426 	struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2427 				  struct ib_udata *udata);
2428 	int (*dealloc_mw)(struct ib_mw *mw);
2429 	struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2430 				    struct ib_fmr_attr *fmr_attr);
2431 	int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2432 			    u64 iova);
2433 	int (*unmap_fmr)(struct list_head *fmr_list);
2434 	int (*dealloc_fmr)(struct ib_fmr *fmr);
2435 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2436 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2437 	struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2438 				      struct ib_ucontext *ucontext,
2439 				      struct ib_udata *udata);
2440 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2441 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2442 				       struct ib_flow_attr *flow_attr,
2443 				       int domain, struct ib_udata *udata);
2444 	int (*destroy_flow)(struct ib_flow *flow_id);
2445 	struct ib_flow_action *(*create_flow_action_esp)(
2446 		struct ib_device *device,
2447 		const struct ib_flow_action_attrs_esp *attr,
2448 		struct uverbs_attr_bundle *attrs);
2449 	int (*destroy_flow_action)(struct ib_flow_action *action);
2450 	int (*modify_flow_action_esp)(
2451 		struct ib_flow_action *action,
2452 		const struct ib_flow_action_attrs_esp *attr,
2453 		struct uverbs_attr_bundle *attrs);
2454 	int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2455 				 int state);
2456 	int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2457 			     struct ifla_vf_info *ivf);
2458 	int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2459 			    struct ifla_vf_stats *stats);
2460 	int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2461 			   int type);
2462 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2463 				   struct ib_wq_init_attr *init_attr,
2464 				   struct ib_udata *udata);
2465 	int (*destroy_wq)(struct ib_wq *wq);
2466 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2467 			 u32 wq_attr_mask, struct ib_udata *udata);
2468 	struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2469 		struct ib_device *device,
2470 		struct ib_rwq_ind_table_init_attr *init_attr,
2471 		struct ib_udata *udata);
2472 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2473 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2474 				  struct ib_ucontext *context,
2475 				  struct ib_dm_alloc_attr *attr,
2476 				  struct uverbs_attr_bundle *attrs);
2477 	int (*dealloc_dm)(struct ib_dm *dm);
2478 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2479 				   struct ib_dm_mr_attr *attr,
2480 				   struct uverbs_attr_bundle *attrs);
2481 	struct ib_counters *(*create_counters)(
2482 		struct ib_device *device, struct uverbs_attr_bundle *attrs);
2483 	int (*destroy_counters)(struct ib_counters *counters);
2484 	int (*read_counters)(struct ib_counters *counters,
2485 			     struct ib_counters_read_attr *counters_read_attr,
2486 			     struct uverbs_attr_bundle *attrs);
2487 	/**
2488 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2489 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2490 	 *   core when the device is removed.  A lifespan of -1 in the return
2491 	 *   struct tells the core to set a default lifespan.
2492 	 */
2493 	struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2494 						u8 port_num);
2495 	/**
2496 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2497 	 * @index - The index in the value array we wish to have updated, or
2498 	 *   num_counters if we want all stats updated
2499 	 * Return codes -
2500 	 *   < 0 - Error, no counters updated
2501 	 *   index - Updated the single counter pointed to by index
2502 	 *   num_counters - Updated all counters (will reset the timestamp
2503 	 *     and prevent further calls for lifespan milliseconds)
2504 	 * Drivers are allowed to update all counters in leiu of just the
2505 	 *   one given in index at their option
2506 	 */
2507 	int (*get_hw_stats)(struct ib_device *device,
2508 			    struct rdma_hw_stats *stats, u8 port, int index);
2509 };
2510 
2511 struct ib_device {
2512 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2513 	struct device                *dma_device;
2514 	struct ib_device_ops	     ops;
2515 	char                          name[IB_DEVICE_NAME_MAX];
2516 
2517 	struct list_head              event_handler_list;
2518 	spinlock_t                    event_handler_lock;
2519 
2520 	rwlock_t			client_data_lock;
2521 	struct list_head              core_list;
2522 	/* Access to the client_data_list is protected by the client_data_lock
2523 	 * rwlock and the lists_rwsem read-write semaphore
2524 	 */
2525 	struct list_head              client_data_list;
2526 
2527 	struct ib_cache               cache;
2528 	/**
2529 	 * port_immutable is indexed by port number
2530 	 */
2531 	struct ib_port_immutable     *port_immutable;
2532 
2533 	int			      num_comp_vectors;
2534 
2535 	struct ib_port_pkey_list     *port_pkey_list;
2536 
2537 	struct iw_cm_verbs	     *iwcm;
2538 
2539 	struct module               *owner;
2540 	struct device                dev;
2541 	/* First group for device attributes,
2542 	 * Second group for driver provided attributes (optional).
2543 	 * It is NULL terminated array.
2544 	 */
2545 	const struct attribute_group	*groups[3];
2546 
2547 	struct kobject			*ports_kobj;
2548 	struct list_head             port_list;
2549 
2550 	enum {
2551 		IB_DEV_UNINITIALIZED,
2552 		IB_DEV_REGISTERED,
2553 		IB_DEV_UNREGISTERED
2554 	}                            reg_state;
2555 
2556 	int			     uverbs_abi_ver;
2557 	u64			     uverbs_cmd_mask;
2558 	u64			     uverbs_ex_cmd_mask;
2559 
2560 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2561 	__be64			     node_guid;
2562 	u32			     local_dma_lkey;
2563 	u16                          is_switch:1;
2564 	u8                           node_type;
2565 	u8                           phys_port_cnt;
2566 	struct ib_device_attr        attrs;
2567 	struct attribute_group	     *hw_stats_ag;
2568 	struct rdma_hw_stats         *hw_stats;
2569 
2570 #ifdef CONFIG_CGROUP_RDMA
2571 	struct rdmacg_device         cg_device;
2572 #endif
2573 
2574 	u32                          index;
2575 	/*
2576 	 * Implementation details of the RDMA core, don't use in drivers
2577 	 */
2578 	struct rdma_restrack_root     res;
2579 
2580 	const struct uapi_definition   *driver_def;
2581 	enum rdma_driver_id		driver_id;
2582 
2583 	/*
2584 	 * Positive refcount indicates that the device is currently
2585 	 * registered and cannot be unregistered.
2586 	 */
2587 	refcount_t refcount;
2588 	struct completion unreg_completion;
2589 };
2590 
2591 struct ib_client {
2592 	char  *name;
2593 	void (*add)   (struct ib_device *);
2594 	void (*remove)(struct ib_device *, void *client_data);
2595 
2596 	/* Returns the net_dev belonging to this ib_client and matching the
2597 	 * given parameters.
2598 	 * @dev:	 An RDMA device that the net_dev use for communication.
2599 	 * @port:	 A physical port number on the RDMA device.
2600 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2601 	 * @gid:	 A GID that the net_dev uses to communicate.
2602 	 * @addr:	 An IP address the net_dev is configured with.
2603 	 * @client_data: The device's client data set by ib_set_client_data().
2604 	 *
2605 	 * An ib_client that implements a net_dev on top of RDMA devices
2606 	 * (such as IP over IB) should implement this callback, allowing the
2607 	 * rdma_cm module to find the right net_dev for a given request.
2608 	 *
2609 	 * The caller is responsible for calling dev_put on the returned
2610 	 * netdev. */
2611 	struct net_device *(*get_net_dev_by_params)(
2612 			struct ib_device *dev,
2613 			u8 port,
2614 			u16 pkey,
2615 			const union ib_gid *gid,
2616 			const struct sockaddr *addr,
2617 			void *client_data);
2618 	struct list_head list;
2619 };
2620 
2621 struct ib_device *ib_alloc_device(size_t size);
2622 void ib_dealloc_device(struct ib_device *device);
2623 
2624 void ib_get_device_fw_str(struct ib_device *device, char *str);
2625 
2626 int ib_register_device(struct ib_device *device, const char *name,
2627 		       int (*port_callback)(struct ib_device *, u8,
2628 					    struct kobject *));
2629 void ib_unregister_device(struct ib_device *device);
2630 
2631 int ib_register_client   (struct ib_client *client);
2632 void ib_unregister_client(struct ib_client *client);
2633 
2634 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2635 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2636 			 void *data);
2637 void ib_set_device_ops(struct ib_device *device,
2638 		       const struct ib_device_ops *ops);
2639 
2640 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2641 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2642 		      unsigned long pfn, unsigned long size, pgprot_t prot);
2643 int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2644 			struct vm_area_struct *vma, struct page *page,
2645 			unsigned long size);
2646 #else
2647 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
2648 				    struct vm_area_struct *vma,
2649 				    unsigned long pfn, unsigned long size,
2650 				    pgprot_t prot)
2651 {
2652 	return -EINVAL;
2653 }
2654 static inline int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2655 				struct vm_area_struct *vma, struct page *page,
2656 				unsigned long size)
2657 {
2658 	return -EINVAL;
2659 }
2660 #endif
2661 
2662 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2663 {
2664 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2665 }
2666 
2667 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2668 {
2669 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2670 }
2671 
2672 static inline bool ib_is_buffer_cleared(const void __user *p,
2673 					size_t len)
2674 {
2675 	bool ret;
2676 	u8 *buf;
2677 
2678 	if (len > USHRT_MAX)
2679 		return false;
2680 
2681 	buf = memdup_user(p, len);
2682 	if (IS_ERR(buf))
2683 		return false;
2684 
2685 	ret = !memchr_inv(buf, 0, len);
2686 	kfree(buf);
2687 	return ret;
2688 }
2689 
2690 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2691 				       size_t offset,
2692 				       size_t len)
2693 {
2694 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
2695 }
2696 
2697 /**
2698  * ib_is_destroy_retryable - Check whether the uobject destruction
2699  * is retryable.
2700  * @ret: The initial destruction return code
2701  * @why: remove reason
2702  * @uobj: The uobject that is destroyed
2703  *
2704  * This function is a helper function that IB layer and low-level drivers
2705  * can use to consider whether the destruction of the given uobject is
2706  * retry-able.
2707  * It checks the original return code, if it wasn't success the destruction
2708  * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2709  * the remove reason. (i.e. why).
2710  * Must be called with the object locked for destroy.
2711  */
2712 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2713 					   struct ib_uobject *uobj)
2714 {
2715 	return ret && (why == RDMA_REMOVE_DESTROY ||
2716 		       uobj->context->cleanup_retryable);
2717 }
2718 
2719 /**
2720  * ib_destroy_usecnt - Called during destruction to check the usecnt
2721  * @usecnt: The usecnt atomic
2722  * @why: remove reason
2723  * @uobj: The uobject that is destroyed
2724  *
2725  * Non-zero usecnts will block destruction unless destruction was triggered by
2726  * a ucontext cleanup.
2727  */
2728 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2729 				    enum rdma_remove_reason why,
2730 				    struct ib_uobject *uobj)
2731 {
2732 	if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2733 		return -EBUSY;
2734 	return 0;
2735 }
2736 
2737 /**
2738  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2739  * contains all required attributes and no attributes not allowed for
2740  * the given QP state transition.
2741  * @cur_state: Current QP state
2742  * @next_state: Next QP state
2743  * @type: QP type
2744  * @mask: Mask of supplied QP attributes
2745  *
2746  * This function is a helper function that a low-level driver's
2747  * modify_qp method can use to validate the consumer's input.  It
2748  * checks that cur_state and next_state are valid QP states, that a
2749  * transition from cur_state to next_state is allowed by the IB spec,
2750  * and that the attribute mask supplied is allowed for the transition.
2751  */
2752 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2753 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
2754 
2755 void ib_register_event_handler(struct ib_event_handler *event_handler);
2756 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2757 void ib_dispatch_event(struct ib_event *event);
2758 
2759 int ib_query_port(struct ib_device *device,
2760 		  u8 port_num, struct ib_port_attr *port_attr);
2761 
2762 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2763 					       u8 port_num);
2764 
2765 /**
2766  * rdma_cap_ib_switch - Check if the device is IB switch
2767  * @device: Device to check
2768  *
2769  * Device driver is responsible for setting is_switch bit on
2770  * in ib_device structure at init time.
2771  *
2772  * Return: true if the device is IB switch.
2773  */
2774 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2775 {
2776 	return device->is_switch;
2777 }
2778 
2779 /**
2780  * rdma_start_port - Return the first valid port number for the device
2781  * specified
2782  *
2783  * @device: Device to be checked
2784  *
2785  * Return start port number
2786  */
2787 static inline u8 rdma_start_port(const struct ib_device *device)
2788 {
2789 	return rdma_cap_ib_switch(device) ? 0 : 1;
2790 }
2791 
2792 /**
2793  * rdma_end_port - Return the last valid port number for the device
2794  * specified
2795  *
2796  * @device: Device to be checked
2797  *
2798  * Return last port number
2799  */
2800 static inline u8 rdma_end_port(const struct ib_device *device)
2801 {
2802 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2803 }
2804 
2805 static inline int rdma_is_port_valid(const struct ib_device *device,
2806 				     unsigned int port)
2807 {
2808 	return (port >= rdma_start_port(device) &&
2809 		port <= rdma_end_port(device));
2810 }
2811 
2812 static inline bool rdma_is_grh_required(const struct ib_device *device,
2813 					u8 port_num)
2814 {
2815 	return device->port_immutable[port_num].core_cap_flags &
2816 		RDMA_CORE_PORT_IB_GRH_REQUIRED;
2817 }
2818 
2819 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2820 {
2821 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2822 }
2823 
2824 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2825 {
2826 	return device->port_immutable[port_num].core_cap_flags &
2827 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2828 }
2829 
2830 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2831 {
2832 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2833 }
2834 
2835 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2836 {
2837 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2838 }
2839 
2840 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2841 {
2842 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2843 }
2844 
2845 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2846 {
2847 	return rdma_protocol_ib(device, port_num) ||
2848 		rdma_protocol_roce(device, port_num);
2849 }
2850 
2851 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2852 {
2853 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2854 }
2855 
2856 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2857 {
2858 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2859 }
2860 
2861 /**
2862  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2863  * Management Datagrams.
2864  * @device: Device to check
2865  * @port_num: Port number to check
2866  *
2867  * Management Datagrams (MAD) are a required part of the InfiniBand
2868  * specification and are supported on all InfiniBand devices.  A slightly
2869  * extended version are also supported on OPA interfaces.
2870  *
2871  * Return: true if the port supports sending/receiving of MAD packets.
2872  */
2873 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2874 {
2875 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2876 }
2877 
2878 /**
2879  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2880  * Management Datagrams.
2881  * @device: Device to check
2882  * @port_num: Port number to check
2883  *
2884  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2885  * datagrams with their own versions.  These OPA MADs share many but not all of
2886  * the characteristics of InfiniBand MADs.
2887  *
2888  * OPA MADs differ in the following ways:
2889  *
2890  *    1) MADs are variable size up to 2K
2891  *       IBTA defined MADs remain fixed at 256 bytes
2892  *    2) OPA SMPs must carry valid PKeys
2893  *    3) OPA SMP packets are a different format
2894  *
2895  * Return: true if the port supports OPA MAD packet formats.
2896  */
2897 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2898 {
2899 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2900 		== RDMA_CORE_CAP_OPA_MAD;
2901 }
2902 
2903 /**
2904  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2905  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2906  * @device: Device to check
2907  * @port_num: Port number to check
2908  *
2909  * Each InfiniBand node is required to provide a Subnet Management Agent
2910  * that the subnet manager can access.  Prior to the fabric being fully
2911  * configured by the subnet manager, the SMA is accessed via a well known
2912  * interface called the Subnet Management Interface (SMI).  This interface
2913  * uses directed route packets to communicate with the SM to get around the
2914  * chicken and egg problem of the SM needing to know what's on the fabric
2915  * in order to configure the fabric, and needing to configure the fabric in
2916  * order to send packets to the devices on the fabric.  These directed
2917  * route packets do not need the fabric fully configured in order to reach
2918  * their destination.  The SMI is the only method allowed to send
2919  * directed route packets on an InfiniBand fabric.
2920  *
2921  * Return: true if the port provides an SMI.
2922  */
2923 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2924 {
2925 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2926 }
2927 
2928 /**
2929  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2930  * Communication Manager.
2931  * @device: Device to check
2932  * @port_num: Port number to check
2933  *
2934  * The InfiniBand Communication Manager is one of many pre-defined General
2935  * Service Agents (GSA) that are accessed via the General Service
2936  * Interface (GSI).  It's role is to facilitate establishment of connections
2937  * between nodes as well as other management related tasks for established
2938  * connections.
2939  *
2940  * Return: true if the port supports an IB CM (this does not guarantee that
2941  * a CM is actually running however).
2942  */
2943 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2944 {
2945 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2946 }
2947 
2948 /**
2949  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2950  * Communication Manager.
2951  * @device: Device to check
2952  * @port_num: Port number to check
2953  *
2954  * Similar to above, but specific to iWARP connections which have a different
2955  * managment protocol than InfiniBand.
2956  *
2957  * Return: true if the port supports an iWARP CM (this does not guarantee that
2958  * a CM is actually running however).
2959  */
2960 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2961 {
2962 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2963 }
2964 
2965 /**
2966  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2967  * Subnet Administration.
2968  * @device: Device to check
2969  * @port_num: Port number to check
2970  *
2971  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2972  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2973  * fabrics, devices should resolve routes to other hosts by contacting the
2974  * SA to query the proper route.
2975  *
2976  * Return: true if the port should act as a client to the fabric Subnet
2977  * Administration interface.  This does not imply that the SA service is
2978  * running locally.
2979  */
2980 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2981 {
2982 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2983 }
2984 
2985 /**
2986  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2987  * Multicast.
2988  * @device: Device to check
2989  * @port_num: Port number to check
2990  *
2991  * InfiniBand multicast registration is more complex than normal IPv4 or
2992  * IPv6 multicast registration.  Each Host Channel Adapter must register
2993  * with the Subnet Manager when it wishes to join a multicast group.  It
2994  * should do so only once regardless of how many queue pairs it subscribes
2995  * to this group.  And it should leave the group only after all queue pairs
2996  * attached to the group have been detached.
2997  *
2998  * Return: true if the port must undertake the additional adminstrative
2999  * overhead of registering/unregistering with the SM and tracking of the
3000  * total number of queue pairs attached to the multicast group.
3001  */
3002 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3003 {
3004 	return rdma_cap_ib_sa(device, port_num);
3005 }
3006 
3007 /**
3008  * rdma_cap_af_ib - Check if the port of device has the capability
3009  * Native Infiniband Address.
3010  * @device: Device to check
3011  * @port_num: Port number to check
3012  *
3013  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3014  * GID.  RoCE uses a different mechanism, but still generates a GID via
3015  * a prescribed mechanism and port specific data.
3016  *
3017  * Return: true if the port uses a GID address to identify devices on the
3018  * network.
3019  */
3020 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3021 {
3022 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
3023 }
3024 
3025 /**
3026  * rdma_cap_eth_ah - Check if the port of device has the capability
3027  * Ethernet Address Handle.
3028  * @device: Device to check
3029  * @port_num: Port number to check
3030  *
3031  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3032  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3033  * port.  Normally, packet headers are generated by the sending host
3034  * adapter, but when sending connectionless datagrams, we must manually
3035  * inject the proper headers for the fabric we are communicating over.
3036  *
3037  * Return: true if we are running as a RoCE port and must force the
3038  * addition of a Global Route Header built from our Ethernet Address
3039  * Handle into our header list for connectionless packets.
3040  */
3041 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3042 {
3043 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
3044 }
3045 
3046 /**
3047  * rdma_cap_opa_ah - Check if the port of device supports
3048  * OPA Address handles
3049  * @device: Device to check
3050  * @port_num: Port number to check
3051  *
3052  * Return: true if we are running on an OPA device which supports
3053  * the extended OPA addressing.
3054  */
3055 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3056 {
3057 	return (device->port_immutable[port_num].core_cap_flags &
3058 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3059 }
3060 
3061 /**
3062  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3063  *
3064  * @device: Device
3065  * @port_num: Port number
3066  *
3067  * This MAD size includes the MAD headers and MAD payload.  No other headers
3068  * are included.
3069  *
3070  * Return the max MAD size required by the Port.  Will return 0 if the port
3071  * does not support MADs
3072  */
3073 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3074 {
3075 	return device->port_immutable[port_num].max_mad_size;
3076 }
3077 
3078 /**
3079  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3080  * @device: Device to check
3081  * @port_num: Port number to check
3082  *
3083  * RoCE GID table mechanism manages the various GIDs for a device.
3084  *
3085  * NOTE: if allocating the port's GID table has failed, this call will still
3086  * return true, but any RoCE GID table API will fail.
3087  *
3088  * Return: true if the port uses RoCE GID table mechanism in order to manage
3089  * its GIDs.
3090  */
3091 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3092 					   u8 port_num)
3093 {
3094 	return rdma_protocol_roce(device, port_num) &&
3095 		device->ops.add_gid && device->ops.del_gid;
3096 }
3097 
3098 /*
3099  * Check if the device supports READ W/ INVALIDATE.
3100  */
3101 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3102 {
3103 	/*
3104 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3105 	 * has support for it yet.
3106 	 */
3107 	return rdma_protocol_iwarp(dev, port_num);
3108 }
3109 
3110 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3111 			 int state);
3112 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3113 		     struct ifla_vf_info *info);
3114 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3115 		    struct ifla_vf_stats *stats);
3116 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3117 		   int type);
3118 
3119 int ib_query_pkey(struct ib_device *device,
3120 		  u8 port_num, u16 index, u16 *pkey);
3121 
3122 int ib_modify_device(struct ib_device *device,
3123 		     int device_modify_mask,
3124 		     struct ib_device_modify *device_modify);
3125 
3126 int ib_modify_port(struct ib_device *device,
3127 		   u8 port_num, int port_modify_mask,
3128 		   struct ib_port_modify *port_modify);
3129 
3130 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3131 		u8 *port_num, u16 *index);
3132 
3133 int ib_find_pkey(struct ib_device *device,
3134 		 u8 port_num, u16 pkey, u16 *index);
3135 
3136 enum ib_pd_flags {
3137 	/*
3138 	 * Create a memory registration for all memory in the system and place
3139 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3140 	 * ULPs to avoid the overhead of dynamic MRs.
3141 	 *
3142 	 * This flag is generally considered unsafe and must only be used in
3143 	 * extremly trusted environments.  Every use of it will log a warning
3144 	 * in the kernel log.
3145 	 */
3146 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3147 };
3148 
3149 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3150 		const char *caller);
3151 #define ib_alloc_pd(device, flags) \
3152 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3153 void ib_dealloc_pd(struct ib_pd *pd);
3154 
3155 enum rdma_create_ah_flags {
3156 	/* In a sleepable context */
3157 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3158 };
3159 
3160 /**
3161  * rdma_create_ah - Creates an address handle for the given address vector.
3162  * @pd: The protection domain associated with the address handle.
3163  * @ah_attr: The attributes of the address vector.
3164  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3165  *
3166  * The address handle is used to reference a local or global destination
3167  * in all UD QP post sends.
3168  */
3169 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3170 			     u32 flags);
3171 
3172 /**
3173  * rdma_create_user_ah - Creates an address handle for the given address vector.
3174  * It resolves destination mac address for ah attribute of RoCE type.
3175  * @pd: The protection domain associated with the address handle.
3176  * @ah_attr: The attributes of the address vector.
3177  * @udata: pointer to user's input output buffer information need by
3178  *         provider driver.
3179  *
3180  * It returns 0 on success and returns appropriate error code on error.
3181  * The address handle is used to reference a local or global destination
3182  * in all UD QP post sends.
3183  */
3184 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3185 				  struct rdma_ah_attr *ah_attr,
3186 				  struct ib_udata *udata);
3187 /**
3188  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3189  *   work completion.
3190  * @hdr: the L3 header to parse
3191  * @net_type: type of header to parse
3192  * @sgid: place to store source gid
3193  * @dgid: place to store destination gid
3194  */
3195 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3196 			      enum rdma_network_type net_type,
3197 			      union ib_gid *sgid, union ib_gid *dgid);
3198 
3199 /**
3200  * ib_get_rdma_header_version - Get the header version
3201  * @hdr: the L3 header to parse
3202  */
3203 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3204 
3205 /**
3206  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3207  *   work completion.
3208  * @device: Device on which the received message arrived.
3209  * @port_num: Port on which the received message arrived.
3210  * @wc: Work completion associated with the received message.
3211  * @grh: References the received global route header.  This parameter is
3212  *   ignored unless the work completion indicates that the GRH is valid.
3213  * @ah_attr: Returned attributes that can be used when creating an address
3214  *   handle for replying to the message.
3215  * When ib_init_ah_attr_from_wc() returns success,
3216  * (a) for IB link layer it optionally contains a reference to SGID attribute
3217  * when GRH is present for IB link layer.
3218  * (b) for RoCE link layer it contains a reference to SGID attribute.
3219  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3220  * attributes which are initialized using ib_init_ah_attr_from_wc().
3221  *
3222  */
3223 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3224 			    const struct ib_wc *wc, const struct ib_grh *grh,
3225 			    struct rdma_ah_attr *ah_attr);
3226 
3227 /**
3228  * ib_create_ah_from_wc - Creates an address handle associated with the
3229  *   sender of the specified work completion.
3230  * @pd: The protection domain associated with the address handle.
3231  * @wc: Work completion information associated with a received message.
3232  * @grh: References the received global route header.  This parameter is
3233  *   ignored unless the work completion indicates that the GRH is valid.
3234  * @port_num: The outbound port number to associate with the address.
3235  *
3236  * The address handle is used to reference a local or global destination
3237  * in all UD QP post sends.
3238  */
3239 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3240 				   const struct ib_grh *grh, u8 port_num);
3241 
3242 /**
3243  * rdma_modify_ah - Modifies the address vector associated with an address
3244  *   handle.
3245  * @ah: The address handle to modify.
3246  * @ah_attr: The new address vector attributes to associate with the
3247  *   address handle.
3248  */
3249 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3250 
3251 /**
3252  * rdma_query_ah - Queries the address vector associated with an address
3253  *   handle.
3254  * @ah: The address handle to query.
3255  * @ah_attr: The address vector attributes associated with the address
3256  *   handle.
3257  */
3258 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3259 
3260 enum rdma_destroy_ah_flags {
3261 	/* In a sleepable context */
3262 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3263 };
3264 
3265 /**
3266  * rdma_destroy_ah - Destroys an address handle.
3267  * @ah: The address handle to destroy.
3268  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3269  */
3270 int rdma_destroy_ah(struct ib_ah *ah, u32 flags);
3271 
3272 /**
3273  * ib_create_srq - Creates a SRQ associated with the specified protection
3274  *   domain.
3275  * @pd: The protection domain associated with the SRQ.
3276  * @srq_init_attr: A list of initial attributes required to create the
3277  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
3278  *   the actual capabilities of the created SRQ.
3279  *
3280  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3281  * requested size of the SRQ, and set to the actual values allocated
3282  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
3283  * will always be at least as large as the requested values.
3284  */
3285 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3286 			     struct ib_srq_init_attr *srq_init_attr);
3287 
3288 /**
3289  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3290  * @srq: The SRQ to modify.
3291  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3292  *   the current values of selected SRQ attributes are returned.
3293  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3294  *   are being modified.
3295  *
3296  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3297  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3298  * the number of receives queued drops below the limit.
3299  */
3300 int ib_modify_srq(struct ib_srq *srq,
3301 		  struct ib_srq_attr *srq_attr,
3302 		  enum ib_srq_attr_mask srq_attr_mask);
3303 
3304 /**
3305  * ib_query_srq - Returns the attribute list and current values for the
3306  *   specified SRQ.
3307  * @srq: The SRQ to query.
3308  * @srq_attr: The attributes of the specified SRQ.
3309  */
3310 int ib_query_srq(struct ib_srq *srq,
3311 		 struct ib_srq_attr *srq_attr);
3312 
3313 /**
3314  * ib_destroy_srq - Destroys the specified SRQ.
3315  * @srq: The SRQ to destroy.
3316  */
3317 int ib_destroy_srq(struct ib_srq *srq);
3318 
3319 /**
3320  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3321  * @srq: The SRQ to post the work request on.
3322  * @recv_wr: A list of work requests to post on the receive queue.
3323  * @bad_recv_wr: On an immediate failure, this parameter will reference
3324  *   the work request that failed to be posted on the QP.
3325  */
3326 static inline int ib_post_srq_recv(struct ib_srq *srq,
3327 				   const struct ib_recv_wr *recv_wr,
3328 				   const struct ib_recv_wr **bad_recv_wr)
3329 {
3330 	const struct ib_recv_wr *dummy;
3331 
3332 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3333 					      bad_recv_wr ? : &dummy);
3334 }
3335 
3336 /**
3337  * ib_create_qp - Creates a QP associated with the specified protection
3338  *   domain.
3339  * @pd: The protection domain associated with the QP.
3340  * @qp_init_attr: A list of initial attributes required to create the
3341  *   QP.  If QP creation succeeds, then the attributes are updated to
3342  *   the actual capabilities of the created QP.
3343  */
3344 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3345 			   struct ib_qp_init_attr *qp_init_attr);
3346 
3347 /**
3348  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3349  * @qp: The QP to modify.
3350  * @attr: On input, specifies the QP attributes to modify.  On output,
3351  *   the current values of selected QP attributes are returned.
3352  * @attr_mask: A bit-mask used to specify which attributes of the QP
3353  *   are being modified.
3354  * @udata: pointer to user's input output buffer information
3355  *   are being modified.
3356  * It returns 0 on success and returns appropriate error code on error.
3357  */
3358 int ib_modify_qp_with_udata(struct ib_qp *qp,
3359 			    struct ib_qp_attr *attr,
3360 			    int attr_mask,
3361 			    struct ib_udata *udata);
3362 
3363 /**
3364  * ib_modify_qp - Modifies the attributes for the specified QP and then
3365  *   transitions the QP to the given state.
3366  * @qp: The QP to modify.
3367  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3368  *   the current values of selected QP attributes are returned.
3369  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3370  *   are being modified.
3371  */
3372 int ib_modify_qp(struct ib_qp *qp,
3373 		 struct ib_qp_attr *qp_attr,
3374 		 int qp_attr_mask);
3375 
3376 /**
3377  * ib_query_qp - Returns the attribute list and current values for the
3378  *   specified QP.
3379  * @qp: The QP to query.
3380  * @qp_attr: The attributes of the specified QP.
3381  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3382  * @qp_init_attr: Additional attributes of the selected QP.
3383  *
3384  * The qp_attr_mask may be used to limit the query to gathering only the
3385  * selected attributes.
3386  */
3387 int ib_query_qp(struct ib_qp *qp,
3388 		struct ib_qp_attr *qp_attr,
3389 		int qp_attr_mask,
3390 		struct ib_qp_init_attr *qp_init_attr);
3391 
3392 /**
3393  * ib_destroy_qp - Destroys the specified QP.
3394  * @qp: The QP to destroy.
3395  */
3396 int ib_destroy_qp(struct ib_qp *qp);
3397 
3398 /**
3399  * ib_open_qp - Obtain a reference to an existing sharable QP.
3400  * @xrcd - XRC domain
3401  * @qp_open_attr: Attributes identifying the QP to open.
3402  *
3403  * Returns a reference to a sharable QP.
3404  */
3405 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3406 			 struct ib_qp_open_attr *qp_open_attr);
3407 
3408 /**
3409  * ib_close_qp - Release an external reference to a QP.
3410  * @qp: The QP handle to release
3411  *
3412  * The opened QP handle is released by the caller.  The underlying
3413  * shared QP is not destroyed until all internal references are released.
3414  */
3415 int ib_close_qp(struct ib_qp *qp);
3416 
3417 /**
3418  * ib_post_send - Posts a list of work requests to the send queue of
3419  *   the specified QP.
3420  * @qp: The QP to post the work request on.
3421  * @send_wr: A list of work requests to post on the send queue.
3422  * @bad_send_wr: On an immediate failure, this parameter will reference
3423  *   the work request that failed to be posted on the QP.
3424  *
3425  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3426  * error is returned, the QP state shall not be affected,
3427  * ib_post_send() will return an immediate error after queueing any
3428  * earlier work requests in the list.
3429  */
3430 static inline int ib_post_send(struct ib_qp *qp,
3431 			       const struct ib_send_wr *send_wr,
3432 			       const struct ib_send_wr **bad_send_wr)
3433 {
3434 	const struct ib_send_wr *dummy;
3435 
3436 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3437 }
3438 
3439 /**
3440  * ib_post_recv - Posts a list of work requests to the receive queue of
3441  *   the specified QP.
3442  * @qp: The QP to post the work request on.
3443  * @recv_wr: A list of work requests to post on the receive queue.
3444  * @bad_recv_wr: On an immediate failure, this parameter will reference
3445  *   the work request that failed to be posted on the QP.
3446  */
3447 static inline int ib_post_recv(struct ib_qp *qp,
3448 			       const struct ib_recv_wr *recv_wr,
3449 			       const struct ib_recv_wr **bad_recv_wr)
3450 {
3451 	const struct ib_recv_wr *dummy;
3452 
3453 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3454 }
3455 
3456 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3457 			    int nr_cqe, int comp_vector,
3458 			    enum ib_poll_context poll_ctx, const char *caller);
3459 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3460 	__ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3461 
3462 void ib_free_cq(struct ib_cq *cq);
3463 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3464 
3465 /**
3466  * ib_create_cq - Creates a CQ on the specified device.
3467  * @device: The device on which to create the CQ.
3468  * @comp_handler: A user-specified callback that is invoked when a
3469  *   completion event occurs on the CQ.
3470  * @event_handler: A user-specified callback that is invoked when an
3471  *   asynchronous event not associated with a completion occurs on the CQ.
3472  * @cq_context: Context associated with the CQ returned to the user via
3473  *   the associated completion and event handlers.
3474  * @cq_attr: The attributes the CQ should be created upon.
3475  *
3476  * Users can examine the cq structure to determine the actual CQ size.
3477  */
3478 struct ib_cq *__ib_create_cq(struct ib_device *device,
3479 			     ib_comp_handler comp_handler,
3480 			     void (*event_handler)(struct ib_event *, void *),
3481 			     void *cq_context,
3482 			     const struct ib_cq_init_attr *cq_attr,
3483 			     const char *caller);
3484 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3485 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3486 
3487 /**
3488  * ib_resize_cq - Modifies the capacity of the CQ.
3489  * @cq: The CQ to resize.
3490  * @cqe: The minimum size of the CQ.
3491  *
3492  * Users can examine the cq structure to determine the actual CQ size.
3493  */
3494 int ib_resize_cq(struct ib_cq *cq, int cqe);
3495 
3496 /**
3497  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3498  * @cq: The CQ to modify.
3499  * @cq_count: number of CQEs that will trigger an event
3500  * @cq_period: max period of time in usec before triggering an event
3501  *
3502  */
3503 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3504 
3505 /**
3506  * ib_destroy_cq - Destroys the specified CQ.
3507  * @cq: The CQ to destroy.
3508  */
3509 int ib_destroy_cq(struct ib_cq *cq);
3510 
3511 /**
3512  * ib_poll_cq - poll a CQ for completion(s)
3513  * @cq:the CQ being polled
3514  * @num_entries:maximum number of completions to return
3515  * @wc:array of at least @num_entries &struct ib_wc where completions
3516  *   will be returned
3517  *
3518  * Poll a CQ for (possibly multiple) completions.  If the return value
3519  * is < 0, an error occurred.  If the return value is >= 0, it is the
3520  * number of completions returned.  If the return value is
3521  * non-negative and < num_entries, then the CQ was emptied.
3522  */
3523 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3524 			     struct ib_wc *wc)
3525 {
3526 	return cq->device->ops.poll_cq(cq, num_entries, wc);
3527 }
3528 
3529 /**
3530  * ib_req_notify_cq - Request completion notification on a CQ.
3531  * @cq: The CQ to generate an event for.
3532  * @flags:
3533  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3534  *   to request an event on the next solicited event or next work
3535  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3536  *   may also be |ed in to request a hint about missed events, as
3537  *   described below.
3538  *
3539  * Return Value:
3540  *    < 0 means an error occurred while requesting notification
3541  *   == 0 means notification was requested successfully, and if
3542  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3543  *        were missed and it is safe to wait for another event.  In
3544  *        this case is it guaranteed that any work completions added
3545  *        to the CQ since the last CQ poll will trigger a completion
3546  *        notification event.
3547  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3548  *        in.  It means that the consumer must poll the CQ again to
3549  *        make sure it is empty to avoid missing an event because of a
3550  *        race between requesting notification and an entry being
3551  *        added to the CQ.  This return value means it is possible
3552  *        (but not guaranteed) that a work completion has been added
3553  *        to the CQ since the last poll without triggering a
3554  *        completion notification event.
3555  */
3556 static inline int ib_req_notify_cq(struct ib_cq *cq,
3557 				   enum ib_cq_notify_flags flags)
3558 {
3559 	return cq->device->ops.req_notify_cq(cq, flags);
3560 }
3561 
3562 /**
3563  * ib_req_ncomp_notif - Request completion notification when there are
3564  *   at least the specified number of unreaped completions on the CQ.
3565  * @cq: The CQ to generate an event for.
3566  * @wc_cnt: The number of unreaped completions that should be on the
3567  *   CQ before an event is generated.
3568  */
3569 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3570 {
3571 	return cq->device->ops.req_ncomp_notif ?
3572 		cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3573 		-ENOSYS;
3574 }
3575 
3576 /**
3577  * ib_dma_mapping_error - check a DMA addr for error
3578  * @dev: The device for which the dma_addr was created
3579  * @dma_addr: The DMA address to check
3580  */
3581 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3582 {
3583 	return dma_mapping_error(dev->dma_device, dma_addr);
3584 }
3585 
3586 /**
3587  * ib_dma_map_single - Map a kernel virtual address to DMA address
3588  * @dev: The device for which the dma_addr is to be created
3589  * @cpu_addr: The kernel virtual address
3590  * @size: The size of the region in bytes
3591  * @direction: The direction of the DMA
3592  */
3593 static inline u64 ib_dma_map_single(struct ib_device *dev,
3594 				    void *cpu_addr, size_t size,
3595 				    enum dma_data_direction direction)
3596 {
3597 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3598 }
3599 
3600 /**
3601  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3602  * @dev: The device for which the DMA address was created
3603  * @addr: The DMA address
3604  * @size: The size of the region in bytes
3605  * @direction: The direction of the DMA
3606  */
3607 static inline void ib_dma_unmap_single(struct ib_device *dev,
3608 				       u64 addr, size_t size,
3609 				       enum dma_data_direction direction)
3610 {
3611 	dma_unmap_single(dev->dma_device, addr, size, direction);
3612 }
3613 
3614 /**
3615  * ib_dma_map_page - Map a physical page to DMA address
3616  * @dev: The device for which the dma_addr is to be created
3617  * @page: The page to be mapped
3618  * @offset: The offset within the page
3619  * @size: The size of the region in bytes
3620  * @direction: The direction of the DMA
3621  */
3622 static inline u64 ib_dma_map_page(struct ib_device *dev,
3623 				  struct page *page,
3624 				  unsigned long offset,
3625 				  size_t size,
3626 					 enum dma_data_direction direction)
3627 {
3628 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3629 }
3630 
3631 /**
3632  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3633  * @dev: The device for which the DMA address was created
3634  * @addr: The DMA address
3635  * @size: The size of the region in bytes
3636  * @direction: The direction of the DMA
3637  */
3638 static inline void ib_dma_unmap_page(struct ib_device *dev,
3639 				     u64 addr, size_t size,
3640 				     enum dma_data_direction direction)
3641 {
3642 	dma_unmap_page(dev->dma_device, addr, size, direction);
3643 }
3644 
3645 /**
3646  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3647  * @dev: The device for which the DMA addresses are to be created
3648  * @sg: The array of scatter/gather entries
3649  * @nents: The number of scatter/gather entries
3650  * @direction: The direction of the DMA
3651  */
3652 static inline int ib_dma_map_sg(struct ib_device *dev,
3653 				struct scatterlist *sg, int nents,
3654 				enum dma_data_direction direction)
3655 {
3656 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3657 }
3658 
3659 /**
3660  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3661  * @dev: The device for which the DMA addresses were created
3662  * @sg: The array of scatter/gather entries
3663  * @nents: The number of scatter/gather entries
3664  * @direction: The direction of the DMA
3665  */
3666 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3667 				   struct scatterlist *sg, int nents,
3668 				   enum dma_data_direction direction)
3669 {
3670 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
3671 }
3672 
3673 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3674 				      struct scatterlist *sg, int nents,
3675 				      enum dma_data_direction direction,
3676 				      unsigned long dma_attrs)
3677 {
3678 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3679 				dma_attrs);
3680 }
3681 
3682 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3683 					 struct scatterlist *sg, int nents,
3684 					 enum dma_data_direction direction,
3685 					 unsigned long dma_attrs)
3686 {
3687 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3688 }
3689 /**
3690  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3691  * @dev: The device for which the DMA addresses were created
3692  * @sg: The scatter/gather entry
3693  *
3694  * Note: this function is obsolete. To do: change all occurrences of
3695  * ib_sg_dma_address() into sg_dma_address().
3696  */
3697 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3698 				    struct scatterlist *sg)
3699 {
3700 	return sg_dma_address(sg);
3701 }
3702 
3703 /**
3704  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3705  * @dev: The device for which the DMA addresses were created
3706  * @sg: The scatter/gather entry
3707  *
3708  * Note: this function is obsolete. To do: change all occurrences of
3709  * ib_sg_dma_len() into sg_dma_len().
3710  */
3711 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3712 					 struct scatterlist *sg)
3713 {
3714 	return sg_dma_len(sg);
3715 }
3716 
3717 /**
3718  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3719  * @dev: The device for which the DMA address was created
3720  * @addr: The DMA address
3721  * @size: The size of the region in bytes
3722  * @dir: The direction of the DMA
3723  */
3724 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3725 					      u64 addr,
3726 					      size_t size,
3727 					      enum dma_data_direction dir)
3728 {
3729 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3730 }
3731 
3732 /**
3733  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3734  * @dev: The device for which the DMA address was created
3735  * @addr: The DMA address
3736  * @size: The size of the region in bytes
3737  * @dir: The direction of the DMA
3738  */
3739 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3740 						 u64 addr,
3741 						 size_t size,
3742 						 enum dma_data_direction dir)
3743 {
3744 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3745 }
3746 
3747 /**
3748  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3749  * @dev: The device for which the DMA address is requested
3750  * @size: The size of the region to allocate in bytes
3751  * @dma_handle: A pointer for returning the DMA address of the region
3752  * @flag: memory allocator flags
3753  */
3754 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3755 					   size_t size,
3756 					   dma_addr_t *dma_handle,
3757 					   gfp_t flag)
3758 {
3759 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3760 }
3761 
3762 /**
3763  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3764  * @dev: The device for which the DMA addresses were allocated
3765  * @size: The size of the region
3766  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3767  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3768  */
3769 static inline void ib_dma_free_coherent(struct ib_device *dev,
3770 					size_t size, void *cpu_addr,
3771 					dma_addr_t dma_handle)
3772 {
3773 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3774 }
3775 
3776 /**
3777  * ib_dereg_mr - Deregisters a memory region and removes it from the
3778  *   HCA translation table.
3779  * @mr: The memory region to deregister.
3780  *
3781  * This function can fail, if the memory region has memory windows bound to it.
3782  */
3783 int ib_dereg_mr(struct ib_mr *mr);
3784 
3785 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3786 			  enum ib_mr_type mr_type,
3787 			  u32 max_num_sg);
3788 
3789 /**
3790  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3791  *   R_Key and L_Key.
3792  * @mr - struct ib_mr pointer to be updated.
3793  * @newkey - new key to be used.
3794  */
3795 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3796 {
3797 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3798 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3799 }
3800 
3801 /**
3802  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3803  * for calculating a new rkey for type 2 memory windows.
3804  * @rkey - the rkey to increment.
3805  */
3806 static inline u32 ib_inc_rkey(u32 rkey)
3807 {
3808 	const u32 mask = 0x000000ff;
3809 	return ((rkey + 1) & mask) | (rkey & ~mask);
3810 }
3811 
3812 /**
3813  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3814  * @pd: The protection domain associated with the unmapped region.
3815  * @mr_access_flags: Specifies the memory access rights.
3816  * @fmr_attr: Attributes of the unmapped region.
3817  *
3818  * A fast memory region must be mapped before it can be used as part of
3819  * a work request.
3820  */
3821 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3822 			    int mr_access_flags,
3823 			    struct ib_fmr_attr *fmr_attr);
3824 
3825 /**
3826  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3827  * @fmr: The fast memory region to associate with the pages.
3828  * @page_list: An array of physical pages to map to the fast memory region.
3829  * @list_len: The number of pages in page_list.
3830  * @iova: The I/O virtual address to use with the mapped region.
3831  */
3832 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3833 				  u64 *page_list, int list_len,
3834 				  u64 iova)
3835 {
3836 	return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
3837 }
3838 
3839 /**
3840  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3841  * @fmr_list: A linked list of fast memory regions to unmap.
3842  */
3843 int ib_unmap_fmr(struct list_head *fmr_list);
3844 
3845 /**
3846  * ib_dealloc_fmr - Deallocates a fast memory region.
3847  * @fmr: The fast memory region to deallocate.
3848  */
3849 int ib_dealloc_fmr(struct ib_fmr *fmr);
3850 
3851 /**
3852  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3853  * @qp: QP to attach to the multicast group.  The QP must be type
3854  *   IB_QPT_UD.
3855  * @gid: Multicast group GID.
3856  * @lid: Multicast group LID in host byte order.
3857  *
3858  * In order to send and receive multicast packets, subnet
3859  * administration must have created the multicast group and configured
3860  * the fabric appropriately.  The port associated with the specified
3861  * QP must also be a member of the multicast group.
3862  */
3863 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3864 
3865 /**
3866  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3867  * @qp: QP to detach from the multicast group.
3868  * @gid: Multicast group GID.
3869  * @lid: Multicast group LID in host byte order.
3870  */
3871 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3872 
3873 /**
3874  * ib_alloc_xrcd - Allocates an XRC domain.
3875  * @device: The device on which to allocate the XRC domain.
3876  * @caller: Module name for kernel consumers
3877  */
3878 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3879 #define ib_alloc_xrcd(device) \
3880 	__ib_alloc_xrcd((device), KBUILD_MODNAME)
3881 
3882 /**
3883  * ib_dealloc_xrcd - Deallocates an XRC domain.
3884  * @xrcd: The XRC domain to deallocate.
3885  */
3886 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3887 
3888 static inline int ib_check_mr_access(int flags)
3889 {
3890 	/*
3891 	 * Local write permission is required if remote write or
3892 	 * remote atomic permission is also requested.
3893 	 */
3894 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3895 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3896 		return -EINVAL;
3897 
3898 	return 0;
3899 }
3900 
3901 static inline bool ib_access_writable(int access_flags)
3902 {
3903 	/*
3904 	 * We have writable memory backing the MR if any of the following
3905 	 * access flags are set.  "Local write" and "remote write" obviously
3906 	 * require write access.  "Remote atomic" can do things like fetch and
3907 	 * add, which will modify memory, and "MW bind" can change permissions
3908 	 * by binding a window.
3909 	 */
3910 	return access_flags &
3911 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
3912 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
3913 }
3914 
3915 /**
3916  * ib_check_mr_status: lightweight check of MR status.
3917  *     This routine may provide status checks on a selected
3918  *     ib_mr. first use is for signature status check.
3919  *
3920  * @mr: A memory region.
3921  * @check_mask: Bitmask of which checks to perform from
3922  *     ib_mr_status_check enumeration.
3923  * @mr_status: The container of relevant status checks.
3924  *     failed checks will be indicated in the status bitmask
3925  *     and the relevant info shall be in the error item.
3926  */
3927 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3928 		       struct ib_mr_status *mr_status);
3929 
3930 /**
3931  * ib_device_try_get: Hold a registration lock
3932  * device: The device to lock
3933  *
3934  * A device under an active registration lock cannot become unregistered. It
3935  * is only possible to obtain a registration lock on a device that is fully
3936  * registered, otherwise this function returns false.
3937  *
3938  * The registration lock is only necessary for actions which require the
3939  * device to still be registered. Uses that only require the device pointer to
3940  * be valid should use get_device(&ibdev->dev) to hold the memory.
3941  *
3942  */
3943 static inline bool ib_device_try_get(struct ib_device *dev)
3944 {
3945 	return refcount_inc_not_zero(&dev->refcount);
3946 }
3947 
3948 void ib_device_put(struct ib_device *device);
3949 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3950 					    u16 pkey, const union ib_gid *gid,
3951 					    const struct sockaddr *addr);
3952 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3953 			   struct ib_wq_init_attr *init_attr);
3954 int ib_destroy_wq(struct ib_wq *wq);
3955 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3956 		 u32 wq_attr_mask);
3957 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3958 						 struct ib_rwq_ind_table_init_attr*
3959 						 wq_ind_table_init_attr);
3960 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3961 
3962 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3963 		 unsigned int *sg_offset, unsigned int page_size);
3964 
3965 static inline int
3966 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3967 		  unsigned int *sg_offset, unsigned int page_size)
3968 {
3969 	int n;
3970 
3971 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3972 	mr->iova = 0;
3973 
3974 	return n;
3975 }
3976 
3977 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3978 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3979 
3980 void ib_drain_rq(struct ib_qp *qp);
3981 void ib_drain_sq(struct ib_qp *qp);
3982 void ib_drain_qp(struct ib_qp *qp);
3983 
3984 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3985 
3986 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3987 {
3988 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3989 		return attr->roce.dmac;
3990 	return NULL;
3991 }
3992 
3993 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3994 {
3995 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3996 		attr->ib.dlid = (u16)dlid;
3997 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3998 		attr->opa.dlid = dlid;
3999 }
4000 
4001 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4002 {
4003 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4004 		return attr->ib.dlid;
4005 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4006 		return attr->opa.dlid;
4007 	return 0;
4008 }
4009 
4010 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4011 {
4012 	attr->sl = sl;
4013 }
4014 
4015 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4016 {
4017 	return attr->sl;
4018 }
4019 
4020 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4021 					 u8 src_path_bits)
4022 {
4023 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4024 		attr->ib.src_path_bits = src_path_bits;
4025 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4026 		attr->opa.src_path_bits = src_path_bits;
4027 }
4028 
4029 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4030 {
4031 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4032 		return attr->ib.src_path_bits;
4033 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4034 		return attr->opa.src_path_bits;
4035 	return 0;
4036 }
4037 
4038 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4039 					bool make_grd)
4040 {
4041 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4042 		attr->opa.make_grd = make_grd;
4043 }
4044 
4045 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4046 {
4047 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4048 		return attr->opa.make_grd;
4049 	return false;
4050 }
4051 
4052 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4053 {
4054 	attr->port_num = port_num;
4055 }
4056 
4057 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4058 {
4059 	return attr->port_num;
4060 }
4061 
4062 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4063 					   u8 static_rate)
4064 {
4065 	attr->static_rate = static_rate;
4066 }
4067 
4068 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4069 {
4070 	return attr->static_rate;
4071 }
4072 
4073 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4074 					enum ib_ah_flags flag)
4075 {
4076 	attr->ah_flags = flag;
4077 }
4078 
4079 static inline enum ib_ah_flags
4080 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4081 {
4082 	return attr->ah_flags;
4083 }
4084 
4085 static inline const struct ib_global_route
4086 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4087 {
4088 	return &attr->grh;
4089 }
4090 
4091 /*To retrieve and modify the grh */
4092 static inline struct ib_global_route
4093 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4094 {
4095 	return &attr->grh;
4096 }
4097 
4098 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4099 {
4100 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4101 
4102 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4103 }
4104 
4105 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4106 					     __be64 prefix)
4107 {
4108 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4109 
4110 	grh->dgid.global.subnet_prefix = prefix;
4111 }
4112 
4113 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4114 					    __be64 if_id)
4115 {
4116 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4117 
4118 	grh->dgid.global.interface_id = if_id;
4119 }
4120 
4121 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4122 				   union ib_gid *dgid, u32 flow_label,
4123 				   u8 sgid_index, u8 hop_limit,
4124 				   u8 traffic_class)
4125 {
4126 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4127 
4128 	attr->ah_flags = IB_AH_GRH;
4129 	if (dgid)
4130 		grh->dgid = *dgid;
4131 	grh->flow_label = flow_label;
4132 	grh->sgid_index = sgid_index;
4133 	grh->hop_limit = hop_limit;
4134 	grh->traffic_class = traffic_class;
4135 	grh->sgid_attr = NULL;
4136 }
4137 
4138 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4139 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4140 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4141 			     const struct ib_gid_attr *sgid_attr);
4142 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4143 		       const struct rdma_ah_attr *src);
4144 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4145 			  const struct rdma_ah_attr *new);
4146 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4147 
4148 /**
4149  * rdma_ah_find_type - Return address handle type.
4150  *
4151  * @dev: Device to be checked
4152  * @port_num: Port number
4153  */
4154 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4155 						       u8 port_num)
4156 {
4157 	if (rdma_protocol_roce(dev, port_num))
4158 		return RDMA_AH_ATTR_TYPE_ROCE;
4159 	if (rdma_protocol_ib(dev, port_num)) {
4160 		if (rdma_cap_opa_ah(dev, port_num))
4161 			return RDMA_AH_ATTR_TYPE_OPA;
4162 		return RDMA_AH_ATTR_TYPE_IB;
4163 	}
4164 
4165 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4166 }
4167 
4168 /**
4169  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4170  *     In the current implementation the only way to get
4171  *     get the 32bit lid is from other sources for OPA.
4172  *     For IB, lids will always be 16bits so cast the
4173  *     value accordingly.
4174  *
4175  * @lid: A 32bit LID
4176  */
4177 static inline u16 ib_lid_cpu16(u32 lid)
4178 {
4179 	WARN_ON_ONCE(lid & 0xFFFF0000);
4180 	return (u16)lid;
4181 }
4182 
4183 /**
4184  * ib_lid_be16 - Return lid in 16bit BE encoding.
4185  *
4186  * @lid: A 32bit LID
4187  */
4188 static inline __be16 ib_lid_be16(u32 lid)
4189 {
4190 	WARN_ON_ONCE(lid & 0xFFFF0000);
4191 	return cpu_to_be16((u16)lid);
4192 }
4193 
4194 /**
4195  * ib_get_vector_affinity - Get the affinity mappings of a given completion
4196  *   vector
4197  * @device:         the rdma device
4198  * @comp_vector:    index of completion vector
4199  *
4200  * Returns NULL on failure, otherwise a corresponding cpu map of the
4201  * completion vector (returns all-cpus map if the device driver doesn't
4202  * implement get_vector_affinity).
4203  */
4204 static inline const struct cpumask *
4205 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4206 {
4207 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4208 	    !device->ops.get_vector_affinity)
4209 		return NULL;
4210 
4211 	return device->ops.get_vector_affinity(device, comp_vector);
4212 
4213 }
4214 
4215 /**
4216  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4217  * and add their gids, as needed, to the relevant RoCE devices.
4218  *
4219  * @device:         the rdma device
4220  */
4221 void rdma_roce_rescan_device(struct ib_device *ibdev);
4222 
4223 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4224 
4225 
4226 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4227 
4228 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4229 				     enum rdma_netdev_t type, const char *name,
4230 				     unsigned char name_assign_type,
4231 				     void (*setup)(struct net_device *));
4232 
4233 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4234 		     enum rdma_netdev_t type, const char *name,
4235 		     unsigned char name_assign_type,
4236 		     void (*setup)(struct net_device *),
4237 		     struct net_device *netdev);
4238 
4239 /**
4240  * rdma_set_device_sysfs_group - Set device attributes group to have
4241  *				 driver specific sysfs entries at
4242  *				 for infiniband class.
4243  *
4244  * @device:	device pointer for which attributes to be created
4245  * @group:	Pointer to group which should be added when device
4246  *		is registered with sysfs.
4247  * rdma_set_device_sysfs_group() allows existing drivers to expose one
4248  * group per device to have sysfs attributes.
4249  *
4250  * NOTE: New drivers should not make use of this API; instead new device
4251  * parameter should be exposed via netlink command. This API and mechanism
4252  * exist only for existing drivers.
4253  */
4254 static inline void
4255 rdma_set_device_sysfs_group(struct ib_device *dev,
4256 			    const struct attribute_group *group)
4257 {
4258 	dev->groups[1] = group;
4259 }
4260 
4261 #endif /* IB_VERBS_H */
4262