xref: /openbmc/linux/include/rdma/ib_verbs.h (revision 62ede7779904bc75bdd84f1ff0016113956ce3b4)
1 /*
2  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
3  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
4  * Copyright (c) 2004 Intel Corporation.  All rights reserved.
5  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
6  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
7  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
9  *
10  * This software is available to you under a choice of one of two
11  * licenses.  You may choose to be licensed under the terms of the GNU
12  * General Public License (GPL) Version 2, available from the file
13  * COPYING in the main directory of this source tree, or the
14  * OpenIB.org BSD license below:
15  *
16  *     Redistribution and use in source and binary forms, with or
17  *     without modification, are permitted provided that the following
18  *     conditions are met:
19  *
20  *      - Redistributions of source code must retain the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer.
23  *
24  *      - Redistributions in binary form must reproduce the above
25  *        copyright notice, this list of conditions and the following
26  *        disclaimer in the documentation and/or other materials
27  *        provided with the distribution.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36  * SOFTWARE.
37  */
38 
39 #if !defined(IB_VERBS_H)
40 #define IB_VERBS_H
41 
42 #include <linux/types.h>
43 #include <linux/device.h>
44 #include <linux/mm.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
54 #include <net/ipv6.h>
55 #include <net/ip.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/netdevice.h>
59 
60 #include <linux/if_link.h>
61 #include <linux/atomic.h>
62 #include <linux/mmu_notifier.h>
63 #include <linux/uaccess.h>
64 #include <linux/cgroup_rdma.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 
67 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
68 
69 extern struct workqueue_struct *ib_wq;
70 extern struct workqueue_struct *ib_comp_wq;
71 
72 union ib_gid {
73 	u8	raw[16];
74 	struct {
75 		__be64	subnet_prefix;
76 		__be64	interface_id;
77 	} global;
78 };
79 
80 extern union ib_gid zgid;
81 
82 enum ib_gid_type {
83 	/* If link layer is Ethernet, this is RoCE V1 */
84 	IB_GID_TYPE_IB        = 0,
85 	IB_GID_TYPE_ROCE      = 0,
86 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
87 	IB_GID_TYPE_SIZE
88 };
89 
90 #define ROCE_V2_UDP_DPORT      4791
91 struct ib_gid_attr {
92 	enum ib_gid_type	gid_type;
93 	struct net_device	*ndev;
94 };
95 
96 enum rdma_node_type {
97 	/* IB values map to NodeInfo:NodeType. */
98 	RDMA_NODE_IB_CA 	= 1,
99 	RDMA_NODE_IB_SWITCH,
100 	RDMA_NODE_IB_ROUTER,
101 	RDMA_NODE_RNIC,
102 	RDMA_NODE_USNIC,
103 	RDMA_NODE_USNIC_UDP,
104 };
105 
106 enum {
107 	/* set the local administered indication */
108 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
109 };
110 
111 enum rdma_transport_type {
112 	RDMA_TRANSPORT_IB,
113 	RDMA_TRANSPORT_IWARP,
114 	RDMA_TRANSPORT_USNIC,
115 	RDMA_TRANSPORT_USNIC_UDP
116 };
117 
118 enum rdma_protocol_type {
119 	RDMA_PROTOCOL_IB,
120 	RDMA_PROTOCOL_IBOE,
121 	RDMA_PROTOCOL_IWARP,
122 	RDMA_PROTOCOL_USNIC_UDP
123 };
124 
125 __attribute_const__ enum rdma_transport_type
126 rdma_node_get_transport(enum rdma_node_type node_type);
127 
128 enum rdma_network_type {
129 	RDMA_NETWORK_IB,
130 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
131 	RDMA_NETWORK_IPV4,
132 	RDMA_NETWORK_IPV6
133 };
134 
135 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
136 {
137 	if (network_type == RDMA_NETWORK_IPV4 ||
138 	    network_type == RDMA_NETWORK_IPV6)
139 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
140 
141 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
142 	return IB_GID_TYPE_IB;
143 }
144 
145 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
146 							    union ib_gid *gid)
147 {
148 	if (gid_type == IB_GID_TYPE_IB)
149 		return RDMA_NETWORK_IB;
150 
151 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
152 		return RDMA_NETWORK_IPV4;
153 	else
154 		return RDMA_NETWORK_IPV6;
155 }
156 
157 enum rdma_link_layer {
158 	IB_LINK_LAYER_UNSPECIFIED,
159 	IB_LINK_LAYER_INFINIBAND,
160 	IB_LINK_LAYER_ETHERNET,
161 };
162 
163 enum ib_device_cap_flags {
164 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
165 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
166 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
167 	IB_DEVICE_RAW_MULTI			= (1 << 3),
168 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
169 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
170 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
171 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
172 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
173 	IB_DEVICE_INIT_TYPE			= (1 << 9),
174 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
175 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
176 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
177 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
178 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
179 
180 	/*
181 	 * This device supports a per-device lkey or stag that can be
182 	 * used without performing a memory registration for the local
183 	 * memory.  Note that ULPs should never check this flag, but
184 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
185 	 * which will always contain a usable lkey.
186 	 */
187 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
188 	IB_DEVICE_RESERVED /* old SEND_W_INV */	= (1 << 16),
189 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
190 	/*
191 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
192 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
193 	 * messages and can verify the validity of checksum for
194 	 * incoming messages.  Setting this flag implies that the
195 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
196 	 */
197 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
198 	IB_DEVICE_UD_TSO			= (1 << 19),
199 	IB_DEVICE_XRC				= (1 << 20),
200 
201 	/*
202 	 * This device supports the IB "base memory management extension",
203 	 * which includes support for fast registrations (IB_WR_REG_MR,
204 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
205 	 * also be set by any iWarp device which must support FRs to comply
206 	 * to the iWarp verbs spec.  iWarp devices also support the
207 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
208 	 * stag.
209 	 */
210 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
211 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
212 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
213 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
214 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
215 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
216 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
217 	/*
218 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
219 	 * support execution of WQEs that involve synchronization
220 	 * of I/O operations with single completion queue managed
221 	 * by hardware.
222 	 */
223 	IB_DEVICE_CROSS_CHANNEL		= (1 << 27),
224 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
225 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
226 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
227 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
228 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
229 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
230 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
231 	IB_DEVICE_RDMA_NETDEV_OPA_VNIC		= (1ULL << 35),
232 };
233 
234 enum ib_signature_prot_cap {
235 	IB_PROT_T10DIF_TYPE_1 = 1,
236 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
237 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
238 };
239 
240 enum ib_signature_guard_cap {
241 	IB_GUARD_T10DIF_CRC	= 1,
242 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
243 };
244 
245 enum ib_atomic_cap {
246 	IB_ATOMIC_NONE,
247 	IB_ATOMIC_HCA,
248 	IB_ATOMIC_GLOB
249 };
250 
251 enum ib_odp_general_cap_bits {
252 	IB_ODP_SUPPORT		= 1 << 0,
253 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
254 };
255 
256 enum ib_odp_transport_cap_bits {
257 	IB_ODP_SUPPORT_SEND	= 1 << 0,
258 	IB_ODP_SUPPORT_RECV	= 1 << 1,
259 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
260 	IB_ODP_SUPPORT_READ	= 1 << 3,
261 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
262 };
263 
264 struct ib_odp_caps {
265 	uint64_t general_caps;
266 	struct {
267 		uint32_t  rc_odp_caps;
268 		uint32_t  uc_odp_caps;
269 		uint32_t  ud_odp_caps;
270 	} per_transport_caps;
271 };
272 
273 struct ib_rss_caps {
274 	/* Corresponding bit will be set if qp type from
275 	 * 'enum ib_qp_type' is supported, e.g.
276 	 * supported_qpts |= 1 << IB_QPT_UD
277 	 */
278 	u32 supported_qpts;
279 	u32 max_rwq_indirection_tables;
280 	u32 max_rwq_indirection_table_size;
281 };
282 
283 enum ib_cq_creation_flags {
284 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
285 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
286 };
287 
288 struct ib_cq_init_attr {
289 	unsigned int	cqe;
290 	int		comp_vector;
291 	u32		flags;
292 };
293 
294 struct ib_device_attr {
295 	u64			fw_ver;
296 	__be64			sys_image_guid;
297 	u64			max_mr_size;
298 	u64			page_size_cap;
299 	u32			vendor_id;
300 	u32			vendor_part_id;
301 	u32			hw_ver;
302 	int			max_qp;
303 	int			max_qp_wr;
304 	u64			device_cap_flags;
305 	int			max_sge;
306 	int			max_sge_rd;
307 	int			max_cq;
308 	int			max_cqe;
309 	int			max_mr;
310 	int			max_pd;
311 	int			max_qp_rd_atom;
312 	int			max_ee_rd_atom;
313 	int			max_res_rd_atom;
314 	int			max_qp_init_rd_atom;
315 	int			max_ee_init_rd_atom;
316 	enum ib_atomic_cap	atomic_cap;
317 	enum ib_atomic_cap	masked_atomic_cap;
318 	int			max_ee;
319 	int			max_rdd;
320 	int			max_mw;
321 	int			max_raw_ipv6_qp;
322 	int			max_raw_ethy_qp;
323 	int			max_mcast_grp;
324 	int			max_mcast_qp_attach;
325 	int			max_total_mcast_qp_attach;
326 	int			max_ah;
327 	int			max_fmr;
328 	int			max_map_per_fmr;
329 	int			max_srq;
330 	int			max_srq_wr;
331 	int			max_srq_sge;
332 	unsigned int		max_fast_reg_page_list_len;
333 	u16			max_pkeys;
334 	u8			local_ca_ack_delay;
335 	int			sig_prot_cap;
336 	int			sig_guard_cap;
337 	struct ib_odp_caps	odp_caps;
338 	uint64_t		timestamp_mask;
339 	uint64_t		hca_core_clock; /* in KHZ */
340 	struct ib_rss_caps	rss_caps;
341 	u32			max_wq_type_rq;
342 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
343 };
344 
345 enum ib_mtu {
346 	IB_MTU_256  = 1,
347 	IB_MTU_512  = 2,
348 	IB_MTU_1024 = 3,
349 	IB_MTU_2048 = 4,
350 	IB_MTU_4096 = 5
351 };
352 
353 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
354 {
355 	switch (mtu) {
356 	case IB_MTU_256:  return  256;
357 	case IB_MTU_512:  return  512;
358 	case IB_MTU_1024: return 1024;
359 	case IB_MTU_2048: return 2048;
360 	case IB_MTU_4096: return 4096;
361 	default: 	  return -1;
362 	}
363 }
364 
365 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
366 {
367 	if (mtu >= 4096)
368 		return IB_MTU_4096;
369 	else if (mtu >= 2048)
370 		return IB_MTU_2048;
371 	else if (mtu >= 1024)
372 		return IB_MTU_1024;
373 	else if (mtu >= 512)
374 		return IB_MTU_512;
375 	else
376 		return IB_MTU_256;
377 }
378 
379 enum ib_port_state {
380 	IB_PORT_NOP		= 0,
381 	IB_PORT_DOWN		= 1,
382 	IB_PORT_INIT		= 2,
383 	IB_PORT_ARMED		= 3,
384 	IB_PORT_ACTIVE		= 4,
385 	IB_PORT_ACTIVE_DEFER	= 5
386 };
387 
388 enum ib_port_cap_flags {
389 	IB_PORT_SM				= 1 <<  1,
390 	IB_PORT_NOTICE_SUP			= 1 <<  2,
391 	IB_PORT_TRAP_SUP			= 1 <<  3,
392 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
393 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
394 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
395 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
396 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
397 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
398 	IB_PORT_SM_DISABLED			= 1 << 10,
399 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
400 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
401 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
402 	IB_PORT_CM_SUP				= 1 << 16,
403 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
404 	IB_PORT_REINIT_SUP			= 1 << 18,
405 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
406 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
407 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
408 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
409 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
410 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
411 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
412 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
413 };
414 
415 enum ib_port_width {
416 	IB_WIDTH_1X	= 1,
417 	IB_WIDTH_4X	= 2,
418 	IB_WIDTH_8X	= 4,
419 	IB_WIDTH_12X	= 8
420 };
421 
422 static inline int ib_width_enum_to_int(enum ib_port_width width)
423 {
424 	switch (width) {
425 	case IB_WIDTH_1X:  return  1;
426 	case IB_WIDTH_4X:  return  4;
427 	case IB_WIDTH_8X:  return  8;
428 	case IB_WIDTH_12X: return 12;
429 	default: 	  return -1;
430 	}
431 }
432 
433 enum ib_port_speed {
434 	IB_SPEED_SDR	= 1,
435 	IB_SPEED_DDR	= 2,
436 	IB_SPEED_QDR	= 4,
437 	IB_SPEED_FDR10	= 8,
438 	IB_SPEED_FDR	= 16,
439 	IB_SPEED_EDR	= 32,
440 	IB_SPEED_HDR	= 64
441 };
442 
443 /**
444  * struct rdma_hw_stats
445  * @timestamp - Used by the core code to track when the last update was
446  * @lifespan - Used by the core code to determine how old the counters
447  *   should be before being updated again.  Stored in jiffies, defaults
448  *   to 10 milliseconds, drivers can override the default be specifying
449  *   their own value during their allocation routine.
450  * @name - Array of pointers to static names used for the counters in
451  *   directory.
452  * @num_counters - How many hardware counters there are.  If name is
453  *   shorter than this number, a kernel oops will result.  Driver authors
454  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
455  *   in their code to prevent this.
456  * @value - Array of u64 counters that are accessed by the sysfs code and
457  *   filled in by the drivers get_stats routine
458  */
459 struct rdma_hw_stats {
460 	unsigned long	timestamp;
461 	unsigned long	lifespan;
462 	const char * const *names;
463 	int		num_counters;
464 	u64		value[];
465 };
466 
467 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
468 /**
469  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
470  *   for drivers.
471  * @names - Array of static const char *
472  * @num_counters - How many elements in array
473  * @lifespan - How many milliseconds between updates
474  */
475 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
476 		const char * const *names, int num_counters,
477 		unsigned long lifespan)
478 {
479 	struct rdma_hw_stats *stats;
480 
481 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
482 			GFP_KERNEL);
483 	if (!stats)
484 		return NULL;
485 	stats->names = names;
486 	stats->num_counters = num_counters;
487 	stats->lifespan = msecs_to_jiffies(lifespan);
488 
489 	return stats;
490 }
491 
492 
493 /* Define bits for the various functionality this port needs to be supported by
494  * the core.
495  */
496 /* Management                           0x00000FFF */
497 #define RDMA_CORE_CAP_IB_MAD            0x00000001
498 #define RDMA_CORE_CAP_IB_SMI            0x00000002
499 #define RDMA_CORE_CAP_IB_CM             0x00000004
500 #define RDMA_CORE_CAP_IW_CM             0x00000008
501 #define RDMA_CORE_CAP_IB_SA             0x00000010
502 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
503 
504 /* Address format                       0x000FF000 */
505 #define RDMA_CORE_CAP_AF_IB             0x00001000
506 #define RDMA_CORE_CAP_ETH_AH            0x00002000
507 #define RDMA_CORE_CAP_OPA_AH            0x00004000
508 
509 /* Protocol                             0xFFF00000 */
510 #define RDMA_CORE_CAP_PROT_IB           0x00100000
511 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
512 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
513 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
514 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
515 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
516 
517 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
518 					| RDMA_CORE_CAP_IB_MAD \
519 					| RDMA_CORE_CAP_IB_SMI \
520 					| RDMA_CORE_CAP_IB_CM  \
521 					| RDMA_CORE_CAP_IB_SA  \
522 					| RDMA_CORE_CAP_AF_IB)
523 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
524 					| RDMA_CORE_CAP_IB_MAD  \
525 					| RDMA_CORE_CAP_IB_CM   \
526 					| RDMA_CORE_CAP_AF_IB   \
527 					| RDMA_CORE_CAP_ETH_AH)
528 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
529 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
530 					| RDMA_CORE_CAP_IB_MAD  \
531 					| RDMA_CORE_CAP_IB_CM   \
532 					| RDMA_CORE_CAP_AF_IB   \
533 					| RDMA_CORE_CAP_ETH_AH)
534 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
535 					| RDMA_CORE_CAP_IW_CM)
536 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
537 					| RDMA_CORE_CAP_OPA_MAD)
538 
539 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
540 
541 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
542 
543 struct ib_port_attr {
544 	u64			subnet_prefix;
545 	enum ib_port_state	state;
546 	enum ib_mtu		max_mtu;
547 	enum ib_mtu		active_mtu;
548 	int			gid_tbl_len;
549 	u32			port_cap_flags;
550 	u32			max_msg_sz;
551 	u32			bad_pkey_cntr;
552 	u32			qkey_viol_cntr;
553 	u16			pkey_tbl_len;
554 	u32			sm_lid;
555 	u32			lid;
556 	u8			lmc;
557 	u8			max_vl_num;
558 	u8			sm_sl;
559 	u8			subnet_timeout;
560 	u8			init_type_reply;
561 	u8			active_width;
562 	u8			active_speed;
563 	u8                      phys_state;
564 	bool			grh_required;
565 };
566 
567 enum ib_device_modify_flags {
568 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
569 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
570 };
571 
572 #define IB_DEVICE_NODE_DESC_MAX 64
573 
574 struct ib_device_modify {
575 	u64	sys_image_guid;
576 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
577 };
578 
579 enum ib_port_modify_flags {
580 	IB_PORT_SHUTDOWN		= 1,
581 	IB_PORT_INIT_TYPE		= (1<<2),
582 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
583 	IB_PORT_OPA_MASK_CHG		= (1<<4)
584 };
585 
586 struct ib_port_modify {
587 	u32	set_port_cap_mask;
588 	u32	clr_port_cap_mask;
589 	u8	init_type;
590 };
591 
592 enum ib_event_type {
593 	IB_EVENT_CQ_ERR,
594 	IB_EVENT_QP_FATAL,
595 	IB_EVENT_QP_REQ_ERR,
596 	IB_EVENT_QP_ACCESS_ERR,
597 	IB_EVENT_COMM_EST,
598 	IB_EVENT_SQ_DRAINED,
599 	IB_EVENT_PATH_MIG,
600 	IB_EVENT_PATH_MIG_ERR,
601 	IB_EVENT_DEVICE_FATAL,
602 	IB_EVENT_PORT_ACTIVE,
603 	IB_EVENT_PORT_ERR,
604 	IB_EVENT_LID_CHANGE,
605 	IB_EVENT_PKEY_CHANGE,
606 	IB_EVENT_SM_CHANGE,
607 	IB_EVENT_SRQ_ERR,
608 	IB_EVENT_SRQ_LIMIT_REACHED,
609 	IB_EVENT_QP_LAST_WQE_REACHED,
610 	IB_EVENT_CLIENT_REREGISTER,
611 	IB_EVENT_GID_CHANGE,
612 	IB_EVENT_WQ_FATAL,
613 };
614 
615 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
616 
617 struct ib_event {
618 	struct ib_device	*device;
619 	union {
620 		struct ib_cq	*cq;
621 		struct ib_qp	*qp;
622 		struct ib_srq	*srq;
623 		struct ib_wq	*wq;
624 		u8		port_num;
625 	} element;
626 	enum ib_event_type	event;
627 };
628 
629 struct ib_event_handler {
630 	struct ib_device *device;
631 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
632 	struct list_head  list;
633 };
634 
635 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
636 	do {							\
637 		(_ptr)->device  = _device;			\
638 		(_ptr)->handler = _handler;			\
639 		INIT_LIST_HEAD(&(_ptr)->list);			\
640 	} while (0)
641 
642 struct ib_global_route {
643 	union ib_gid	dgid;
644 	u32		flow_label;
645 	u8		sgid_index;
646 	u8		hop_limit;
647 	u8		traffic_class;
648 };
649 
650 struct ib_grh {
651 	__be32		version_tclass_flow;
652 	__be16		paylen;
653 	u8		next_hdr;
654 	u8		hop_limit;
655 	union ib_gid	sgid;
656 	union ib_gid	dgid;
657 };
658 
659 union rdma_network_hdr {
660 	struct ib_grh ibgrh;
661 	struct {
662 		/* The IB spec states that if it's IPv4, the header
663 		 * is located in the last 20 bytes of the header.
664 		 */
665 		u8		reserved[20];
666 		struct iphdr	roce4grh;
667 	};
668 };
669 
670 #define IB_QPN_MASK		0xFFFFFF
671 
672 enum {
673 	IB_MULTICAST_QPN = 0xffffff
674 };
675 
676 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
677 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
678 
679 enum ib_ah_flags {
680 	IB_AH_GRH	= 1
681 };
682 
683 enum ib_rate {
684 	IB_RATE_PORT_CURRENT = 0,
685 	IB_RATE_2_5_GBPS = 2,
686 	IB_RATE_5_GBPS   = 5,
687 	IB_RATE_10_GBPS  = 3,
688 	IB_RATE_20_GBPS  = 6,
689 	IB_RATE_30_GBPS  = 4,
690 	IB_RATE_40_GBPS  = 7,
691 	IB_RATE_60_GBPS  = 8,
692 	IB_RATE_80_GBPS  = 9,
693 	IB_RATE_120_GBPS = 10,
694 	IB_RATE_14_GBPS  = 11,
695 	IB_RATE_56_GBPS  = 12,
696 	IB_RATE_112_GBPS = 13,
697 	IB_RATE_168_GBPS = 14,
698 	IB_RATE_25_GBPS  = 15,
699 	IB_RATE_100_GBPS = 16,
700 	IB_RATE_200_GBPS = 17,
701 	IB_RATE_300_GBPS = 18
702 };
703 
704 /**
705  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
706  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
707  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
708  * @rate: rate to convert.
709  */
710 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
711 
712 /**
713  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
714  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
715  * @rate: rate to convert.
716  */
717 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
718 
719 
720 /**
721  * enum ib_mr_type - memory region type
722  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
723  *                            normal registration
724  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
725  *                            signature operations (data-integrity
726  *                            capable regions)
727  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
728  *                            register any arbitrary sg lists (without
729  *                            the normal mr constraints - see
730  *                            ib_map_mr_sg)
731  */
732 enum ib_mr_type {
733 	IB_MR_TYPE_MEM_REG,
734 	IB_MR_TYPE_SIGNATURE,
735 	IB_MR_TYPE_SG_GAPS,
736 };
737 
738 /**
739  * Signature types
740  * IB_SIG_TYPE_NONE: Unprotected.
741  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
742  */
743 enum ib_signature_type {
744 	IB_SIG_TYPE_NONE,
745 	IB_SIG_TYPE_T10_DIF,
746 };
747 
748 /**
749  * Signature T10-DIF block-guard types
750  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
751  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
752  */
753 enum ib_t10_dif_bg_type {
754 	IB_T10DIF_CRC,
755 	IB_T10DIF_CSUM
756 };
757 
758 /**
759  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
760  *     domain.
761  * @bg_type: T10-DIF block guard type (CRC|CSUM)
762  * @pi_interval: protection information interval.
763  * @bg: seed of guard computation.
764  * @app_tag: application tag of guard block
765  * @ref_tag: initial guard block reference tag.
766  * @ref_remap: Indicate wethear the reftag increments each block
767  * @app_escape: Indicate to skip block check if apptag=0xffff
768  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
769  * @apptag_check_mask: check bitmask of application tag.
770  */
771 struct ib_t10_dif_domain {
772 	enum ib_t10_dif_bg_type bg_type;
773 	u16			pi_interval;
774 	u16			bg;
775 	u16			app_tag;
776 	u32			ref_tag;
777 	bool			ref_remap;
778 	bool			app_escape;
779 	bool			ref_escape;
780 	u16			apptag_check_mask;
781 };
782 
783 /**
784  * struct ib_sig_domain - Parameters for signature domain
785  * @sig_type: specific signauture type
786  * @sig: union of all signature domain attributes that may
787  *     be used to set domain layout.
788  */
789 struct ib_sig_domain {
790 	enum ib_signature_type sig_type;
791 	union {
792 		struct ib_t10_dif_domain dif;
793 	} sig;
794 };
795 
796 /**
797  * struct ib_sig_attrs - Parameters for signature handover operation
798  * @check_mask: bitmask for signature byte check (8 bytes)
799  * @mem: memory domain layout desciptor.
800  * @wire: wire domain layout desciptor.
801  */
802 struct ib_sig_attrs {
803 	u8			check_mask;
804 	struct ib_sig_domain	mem;
805 	struct ib_sig_domain	wire;
806 };
807 
808 enum ib_sig_err_type {
809 	IB_SIG_BAD_GUARD,
810 	IB_SIG_BAD_REFTAG,
811 	IB_SIG_BAD_APPTAG,
812 };
813 
814 /**
815  * struct ib_sig_err - signature error descriptor
816  */
817 struct ib_sig_err {
818 	enum ib_sig_err_type	err_type;
819 	u32			expected;
820 	u32			actual;
821 	u64			sig_err_offset;
822 	u32			key;
823 };
824 
825 enum ib_mr_status_check {
826 	IB_MR_CHECK_SIG_STATUS = 1,
827 };
828 
829 /**
830  * struct ib_mr_status - Memory region status container
831  *
832  * @fail_status: Bitmask of MR checks status. For each
833  *     failed check a corresponding status bit is set.
834  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
835  *     failure.
836  */
837 struct ib_mr_status {
838 	u32		    fail_status;
839 	struct ib_sig_err   sig_err;
840 };
841 
842 /**
843  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
844  * enum.
845  * @mult: multiple to convert.
846  */
847 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
848 
849 enum rdma_ah_attr_type {
850 	RDMA_AH_ATTR_TYPE_IB,
851 	RDMA_AH_ATTR_TYPE_ROCE,
852 	RDMA_AH_ATTR_TYPE_OPA,
853 };
854 
855 struct ib_ah_attr {
856 	u16			dlid;
857 	u8			src_path_bits;
858 };
859 
860 struct roce_ah_attr {
861 	u8			dmac[ETH_ALEN];
862 };
863 
864 struct opa_ah_attr {
865 	u32			dlid;
866 	u8			src_path_bits;
867 };
868 
869 struct rdma_ah_attr {
870 	struct ib_global_route	grh;
871 	u8			sl;
872 	u8			static_rate;
873 	u8			port_num;
874 	u8			ah_flags;
875 	enum rdma_ah_attr_type type;
876 	union {
877 		struct ib_ah_attr ib;
878 		struct roce_ah_attr roce;
879 		struct opa_ah_attr opa;
880 	};
881 };
882 
883 enum ib_wc_status {
884 	IB_WC_SUCCESS,
885 	IB_WC_LOC_LEN_ERR,
886 	IB_WC_LOC_QP_OP_ERR,
887 	IB_WC_LOC_EEC_OP_ERR,
888 	IB_WC_LOC_PROT_ERR,
889 	IB_WC_WR_FLUSH_ERR,
890 	IB_WC_MW_BIND_ERR,
891 	IB_WC_BAD_RESP_ERR,
892 	IB_WC_LOC_ACCESS_ERR,
893 	IB_WC_REM_INV_REQ_ERR,
894 	IB_WC_REM_ACCESS_ERR,
895 	IB_WC_REM_OP_ERR,
896 	IB_WC_RETRY_EXC_ERR,
897 	IB_WC_RNR_RETRY_EXC_ERR,
898 	IB_WC_LOC_RDD_VIOL_ERR,
899 	IB_WC_REM_INV_RD_REQ_ERR,
900 	IB_WC_REM_ABORT_ERR,
901 	IB_WC_INV_EECN_ERR,
902 	IB_WC_INV_EEC_STATE_ERR,
903 	IB_WC_FATAL_ERR,
904 	IB_WC_RESP_TIMEOUT_ERR,
905 	IB_WC_GENERAL_ERR
906 };
907 
908 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
909 
910 enum ib_wc_opcode {
911 	IB_WC_SEND,
912 	IB_WC_RDMA_WRITE,
913 	IB_WC_RDMA_READ,
914 	IB_WC_COMP_SWAP,
915 	IB_WC_FETCH_ADD,
916 	IB_WC_LSO,
917 	IB_WC_LOCAL_INV,
918 	IB_WC_REG_MR,
919 	IB_WC_MASKED_COMP_SWAP,
920 	IB_WC_MASKED_FETCH_ADD,
921 /*
922  * Set value of IB_WC_RECV so consumers can test if a completion is a
923  * receive by testing (opcode & IB_WC_RECV).
924  */
925 	IB_WC_RECV			= 1 << 7,
926 	IB_WC_RECV_RDMA_WITH_IMM
927 };
928 
929 enum ib_wc_flags {
930 	IB_WC_GRH		= 1,
931 	IB_WC_WITH_IMM		= (1<<1),
932 	IB_WC_WITH_INVALIDATE	= (1<<2),
933 	IB_WC_IP_CSUM_OK	= (1<<3),
934 	IB_WC_WITH_SMAC		= (1<<4),
935 	IB_WC_WITH_VLAN		= (1<<5),
936 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
937 };
938 
939 struct ib_wc {
940 	union {
941 		u64		wr_id;
942 		struct ib_cqe	*wr_cqe;
943 	};
944 	enum ib_wc_status	status;
945 	enum ib_wc_opcode	opcode;
946 	u32			vendor_err;
947 	u32			byte_len;
948 	struct ib_qp	       *qp;
949 	union {
950 		__be32		imm_data;
951 		u32		invalidate_rkey;
952 	} ex;
953 	u32			src_qp;
954 	int			wc_flags;
955 	u16			pkey_index;
956 	u32			slid;
957 	u8			sl;
958 	u8			dlid_path_bits;
959 	u8			port_num;	/* valid only for DR SMPs on switches */
960 	u8			smac[ETH_ALEN];
961 	u16			vlan_id;
962 	u8			network_hdr_type;
963 };
964 
965 enum ib_cq_notify_flags {
966 	IB_CQ_SOLICITED			= 1 << 0,
967 	IB_CQ_NEXT_COMP			= 1 << 1,
968 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
969 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
970 };
971 
972 enum ib_srq_type {
973 	IB_SRQT_BASIC,
974 	IB_SRQT_XRC
975 };
976 
977 enum ib_srq_attr_mask {
978 	IB_SRQ_MAX_WR	= 1 << 0,
979 	IB_SRQ_LIMIT	= 1 << 1,
980 };
981 
982 struct ib_srq_attr {
983 	u32	max_wr;
984 	u32	max_sge;
985 	u32	srq_limit;
986 };
987 
988 struct ib_srq_init_attr {
989 	void		      (*event_handler)(struct ib_event *, void *);
990 	void		       *srq_context;
991 	struct ib_srq_attr	attr;
992 	enum ib_srq_type	srq_type;
993 
994 	union {
995 		struct {
996 			struct ib_xrcd *xrcd;
997 			struct ib_cq   *cq;
998 		} xrc;
999 	} ext;
1000 };
1001 
1002 struct ib_qp_cap {
1003 	u32	max_send_wr;
1004 	u32	max_recv_wr;
1005 	u32	max_send_sge;
1006 	u32	max_recv_sge;
1007 	u32	max_inline_data;
1008 
1009 	/*
1010 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1011 	 * ib_create_qp() will calculate the right amount of neededed WRs
1012 	 * and MRs based on this.
1013 	 */
1014 	u32	max_rdma_ctxs;
1015 };
1016 
1017 enum ib_sig_type {
1018 	IB_SIGNAL_ALL_WR,
1019 	IB_SIGNAL_REQ_WR
1020 };
1021 
1022 enum ib_qp_type {
1023 	/*
1024 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1025 	 * here (and in that order) since the MAD layer uses them as
1026 	 * indices into a 2-entry table.
1027 	 */
1028 	IB_QPT_SMI,
1029 	IB_QPT_GSI,
1030 
1031 	IB_QPT_RC,
1032 	IB_QPT_UC,
1033 	IB_QPT_UD,
1034 	IB_QPT_RAW_IPV6,
1035 	IB_QPT_RAW_ETHERTYPE,
1036 	IB_QPT_RAW_PACKET = 8,
1037 	IB_QPT_XRC_INI = 9,
1038 	IB_QPT_XRC_TGT,
1039 	IB_QPT_MAX,
1040 	/* Reserve a range for qp types internal to the low level driver.
1041 	 * These qp types will not be visible at the IB core layer, so the
1042 	 * IB_QPT_MAX usages should not be affected in the core layer
1043 	 */
1044 	IB_QPT_RESERVED1 = 0x1000,
1045 	IB_QPT_RESERVED2,
1046 	IB_QPT_RESERVED3,
1047 	IB_QPT_RESERVED4,
1048 	IB_QPT_RESERVED5,
1049 	IB_QPT_RESERVED6,
1050 	IB_QPT_RESERVED7,
1051 	IB_QPT_RESERVED8,
1052 	IB_QPT_RESERVED9,
1053 	IB_QPT_RESERVED10,
1054 };
1055 
1056 enum ib_qp_create_flags {
1057 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1058 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1059 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1060 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1061 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1062 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1063 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1064 	/* FREE					= 1 << 7, */
1065 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1066 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1067 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1068 	/* reserve bits 26-31 for low level drivers' internal use */
1069 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1070 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1071 };
1072 
1073 /*
1074  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1075  * callback to destroy the passed in QP.
1076  */
1077 
1078 struct ib_qp_init_attr {
1079 	void                  (*event_handler)(struct ib_event *, void *);
1080 	void		       *qp_context;
1081 	struct ib_cq	       *send_cq;
1082 	struct ib_cq	       *recv_cq;
1083 	struct ib_srq	       *srq;
1084 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1085 	struct ib_qp_cap	cap;
1086 	enum ib_sig_type	sq_sig_type;
1087 	enum ib_qp_type		qp_type;
1088 	enum ib_qp_create_flags	create_flags;
1089 
1090 	/*
1091 	 * Only needed for special QP types, or when using the RW API.
1092 	 */
1093 	u8			port_num;
1094 	struct ib_rwq_ind_table *rwq_ind_tbl;
1095 	u32			source_qpn;
1096 };
1097 
1098 struct ib_qp_open_attr {
1099 	void                  (*event_handler)(struct ib_event *, void *);
1100 	void		       *qp_context;
1101 	u32			qp_num;
1102 	enum ib_qp_type		qp_type;
1103 };
1104 
1105 enum ib_rnr_timeout {
1106 	IB_RNR_TIMER_655_36 =  0,
1107 	IB_RNR_TIMER_000_01 =  1,
1108 	IB_RNR_TIMER_000_02 =  2,
1109 	IB_RNR_TIMER_000_03 =  3,
1110 	IB_RNR_TIMER_000_04 =  4,
1111 	IB_RNR_TIMER_000_06 =  5,
1112 	IB_RNR_TIMER_000_08 =  6,
1113 	IB_RNR_TIMER_000_12 =  7,
1114 	IB_RNR_TIMER_000_16 =  8,
1115 	IB_RNR_TIMER_000_24 =  9,
1116 	IB_RNR_TIMER_000_32 = 10,
1117 	IB_RNR_TIMER_000_48 = 11,
1118 	IB_RNR_TIMER_000_64 = 12,
1119 	IB_RNR_TIMER_000_96 = 13,
1120 	IB_RNR_TIMER_001_28 = 14,
1121 	IB_RNR_TIMER_001_92 = 15,
1122 	IB_RNR_TIMER_002_56 = 16,
1123 	IB_RNR_TIMER_003_84 = 17,
1124 	IB_RNR_TIMER_005_12 = 18,
1125 	IB_RNR_TIMER_007_68 = 19,
1126 	IB_RNR_TIMER_010_24 = 20,
1127 	IB_RNR_TIMER_015_36 = 21,
1128 	IB_RNR_TIMER_020_48 = 22,
1129 	IB_RNR_TIMER_030_72 = 23,
1130 	IB_RNR_TIMER_040_96 = 24,
1131 	IB_RNR_TIMER_061_44 = 25,
1132 	IB_RNR_TIMER_081_92 = 26,
1133 	IB_RNR_TIMER_122_88 = 27,
1134 	IB_RNR_TIMER_163_84 = 28,
1135 	IB_RNR_TIMER_245_76 = 29,
1136 	IB_RNR_TIMER_327_68 = 30,
1137 	IB_RNR_TIMER_491_52 = 31
1138 };
1139 
1140 enum ib_qp_attr_mask {
1141 	IB_QP_STATE			= 1,
1142 	IB_QP_CUR_STATE			= (1<<1),
1143 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1144 	IB_QP_ACCESS_FLAGS		= (1<<3),
1145 	IB_QP_PKEY_INDEX		= (1<<4),
1146 	IB_QP_PORT			= (1<<5),
1147 	IB_QP_QKEY			= (1<<6),
1148 	IB_QP_AV			= (1<<7),
1149 	IB_QP_PATH_MTU			= (1<<8),
1150 	IB_QP_TIMEOUT			= (1<<9),
1151 	IB_QP_RETRY_CNT			= (1<<10),
1152 	IB_QP_RNR_RETRY			= (1<<11),
1153 	IB_QP_RQ_PSN			= (1<<12),
1154 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1155 	IB_QP_ALT_PATH			= (1<<14),
1156 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1157 	IB_QP_SQ_PSN			= (1<<16),
1158 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1159 	IB_QP_PATH_MIG_STATE		= (1<<18),
1160 	IB_QP_CAP			= (1<<19),
1161 	IB_QP_DEST_QPN			= (1<<20),
1162 	IB_QP_RESERVED1			= (1<<21),
1163 	IB_QP_RESERVED2			= (1<<22),
1164 	IB_QP_RESERVED3			= (1<<23),
1165 	IB_QP_RESERVED4			= (1<<24),
1166 	IB_QP_RATE_LIMIT		= (1<<25),
1167 };
1168 
1169 enum ib_qp_state {
1170 	IB_QPS_RESET,
1171 	IB_QPS_INIT,
1172 	IB_QPS_RTR,
1173 	IB_QPS_RTS,
1174 	IB_QPS_SQD,
1175 	IB_QPS_SQE,
1176 	IB_QPS_ERR
1177 };
1178 
1179 enum ib_mig_state {
1180 	IB_MIG_MIGRATED,
1181 	IB_MIG_REARM,
1182 	IB_MIG_ARMED
1183 };
1184 
1185 enum ib_mw_type {
1186 	IB_MW_TYPE_1 = 1,
1187 	IB_MW_TYPE_2 = 2
1188 };
1189 
1190 struct ib_qp_attr {
1191 	enum ib_qp_state	qp_state;
1192 	enum ib_qp_state	cur_qp_state;
1193 	enum ib_mtu		path_mtu;
1194 	enum ib_mig_state	path_mig_state;
1195 	u32			qkey;
1196 	u32			rq_psn;
1197 	u32			sq_psn;
1198 	u32			dest_qp_num;
1199 	int			qp_access_flags;
1200 	struct ib_qp_cap	cap;
1201 	struct rdma_ah_attr	ah_attr;
1202 	struct rdma_ah_attr	alt_ah_attr;
1203 	u16			pkey_index;
1204 	u16			alt_pkey_index;
1205 	u8			en_sqd_async_notify;
1206 	u8			sq_draining;
1207 	u8			max_rd_atomic;
1208 	u8			max_dest_rd_atomic;
1209 	u8			min_rnr_timer;
1210 	u8			port_num;
1211 	u8			timeout;
1212 	u8			retry_cnt;
1213 	u8			rnr_retry;
1214 	u8			alt_port_num;
1215 	u8			alt_timeout;
1216 	u32			rate_limit;
1217 };
1218 
1219 enum ib_wr_opcode {
1220 	IB_WR_RDMA_WRITE,
1221 	IB_WR_RDMA_WRITE_WITH_IMM,
1222 	IB_WR_SEND,
1223 	IB_WR_SEND_WITH_IMM,
1224 	IB_WR_RDMA_READ,
1225 	IB_WR_ATOMIC_CMP_AND_SWP,
1226 	IB_WR_ATOMIC_FETCH_AND_ADD,
1227 	IB_WR_LSO,
1228 	IB_WR_SEND_WITH_INV,
1229 	IB_WR_RDMA_READ_WITH_INV,
1230 	IB_WR_LOCAL_INV,
1231 	IB_WR_REG_MR,
1232 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1233 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1234 	IB_WR_REG_SIG_MR,
1235 	/* reserve values for low level drivers' internal use.
1236 	 * These values will not be used at all in the ib core layer.
1237 	 */
1238 	IB_WR_RESERVED1 = 0xf0,
1239 	IB_WR_RESERVED2,
1240 	IB_WR_RESERVED3,
1241 	IB_WR_RESERVED4,
1242 	IB_WR_RESERVED5,
1243 	IB_WR_RESERVED6,
1244 	IB_WR_RESERVED7,
1245 	IB_WR_RESERVED8,
1246 	IB_WR_RESERVED9,
1247 	IB_WR_RESERVED10,
1248 };
1249 
1250 enum ib_send_flags {
1251 	IB_SEND_FENCE		= 1,
1252 	IB_SEND_SIGNALED	= (1<<1),
1253 	IB_SEND_SOLICITED	= (1<<2),
1254 	IB_SEND_INLINE		= (1<<3),
1255 	IB_SEND_IP_CSUM		= (1<<4),
1256 
1257 	/* reserve bits 26-31 for low level drivers' internal use */
1258 	IB_SEND_RESERVED_START	= (1 << 26),
1259 	IB_SEND_RESERVED_END	= (1 << 31),
1260 };
1261 
1262 struct ib_sge {
1263 	u64	addr;
1264 	u32	length;
1265 	u32	lkey;
1266 };
1267 
1268 struct ib_cqe {
1269 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1270 };
1271 
1272 struct ib_send_wr {
1273 	struct ib_send_wr      *next;
1274 	union {
1275 		u64		wr_id;
1276 		struct ib_cqe	*wr_cqe;
1277 	};
1278 	struct ib_sge	       *sg_list;
1279 	int			num_sge;
1280 	enum ib_wr_opcode	opcode;
1281 	int			send_flags;
1282 	union {
1283 		__be32		imm_data;
1284 		u32		invalidate_rkey;
1285 	} ex;
1286 };
1287 
1288 struct ib_rdma_wr {
1289 	struct ib_send_wr	wr;
1290 	u64			remote_addr;
1291 	u32			rkey;
1292 };
1293 
1294 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1295 {
1296 	return container_of(wr, struct ib_rdma_wr, wr);
1297 }
1298 
1299 struct ib_atomic_wr {
1300 	struct ib_send_wr	wr;
1301 	u64			remote_addr;
1302 	u64			compare_add;
1303 	u64			swap;
1304 	u64			compare_add_mask;
1305 	u64			swap_mask;
1306 	u32			rkey;
1307 };
1308 
1309 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1310 {
1311 	return container_of(wr, struct ib_atomic_wr, wr);
1312 }
1313 
1314 struct ib_ud_wr {
1315 	struct ib_send_wr	wr;
1316 	struct ib_ah		*ah;
1317 	void			*header;
1318 	int			hlen;
1319 	int			mss;
1320 	u32			remote_qpn;
1321 	u32			remote_qkey;
1322 	u16			pkey_index; /* valid for GSI only */
1323 	u8			port_num;   /* valid for DR SMPs on switch only */
1324 };
1325 
1326 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1327 {
1328 	return container_of(wr, struct ib_ud_wr, wr);
1329 }
1330 
1331 struct ib_reg_wr {
1332 	struct ib_send_wr	wr;
1333 	struct ib_mr		*mr;
1334 	u32			key;
1335 	int			access;
1336 };
1337 
1338 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1339 {
1340 	return container_of(wr, struct ib_reg_wr, wr);
1341 }
1342 
1343 struct ib_sig_handover_wr {
1344 	struct ib_send_wr	wr;
1345 	struct ib_sig_attrs    *sig_attrs;
1346 	struct ib_mr	       *sig_mr;
1347 	int			access_flags;
1348 	struct ib_sge	       *prot;
1349 };
1350 
1351 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1352 {
1353 	return container_of(wr, struct ib_sig_handover_wr, wr);
1354 }
1355 
1356 struct ib_recv_wr {
1357 	struct ib_recv_wr      *next;
1358 	union {
1359 		u64		wr_id;
1360 		struct ib_cqe	*wr_cqe;
1361 	};
1362 	struct ib_sge	       *sg_list;
1363 	int			num_sge;
1364 };
1365 
1366 enum ib_access_flags {
1367 	IB_ACCESS_LOCAL_WRITE	= 1,
1368 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1369 	IB_ACCESS_REMOTE_READ	= (1<<2),
1370 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1371 	IB_ACCESS_MW_BIND	= (1<<4),
1372 	IB_ZERO_BASED		= (1<<5),
1373 	IB_ACCESS_ON_DEMAND     = (1<<6),
1374 	IB_ACCESS_HUGETLB	= (1<<7),
1375 };
1376 
1377 /*
1378  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1379  * are hidden here instead of a uapi header!
1380  */
1381 enum ib_mr_rereg_flags {
1382 	IB_MR_REREG_TRANS	= 1,
1383 	IB_MR_REREG_PD		= (1<<1),
1384 	IB_MR_REREG_ACCESS	= (1<<2),
1385 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1386 };
1387 
1388 struct ib_fmr_attr {
1389 	int	max_pages;
1390 	int	max_maps;
1391 	u8	page_shift;
1392 };
1393 
1394 struct ib_umem;
1395 
1396 enum rdma_remove_reason {
1397 	/* Userspace requested uobject deletion. Call could fail */
1398 	RDMA_REMOVE_DESTROY,
1399 	/* Context deletion. This call should delete the actual object itself */
1400 	RDMA_REMOVE_CLOSE,
1401 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1402 	RDMA_REMOVE_DRIVER_REMOVE,
1403 	/* Context is being cleaned-up, but commit was just completed */
1404 	RDMA_REMOVE_DURING_CLEANUP,
1405 };
1406 
1407 struct ib_rdmacg_object {
1408 #ifdef CONFIG_CGROUP_RDMA
1409 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1410 #endif
1411 };
1412 
1413 struct ib_ucontext {
1414 	struct ib_device       *device;
1415 	struct ib_uverbs_file  *ufile;
1416 	int			closing;
1417 
1418 	/* locking the uobjects_list */
1419 	struct mutex		uobjects_lock;
1420 	struct list_head	uobjects;
1421 	/* protects cleanup process from other actions */
1422 	struct rw_semaphore	cleanup_rwsem;
1423 	enum rdma_remove_reason cleanup_reason;
1424 
1425 	struct pid             *tgid;
1426 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1427 	struct rb_root      umem_tree;
1428 	/*
1429 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1430 	 * mmu notifiers registration.
1431 	 */
1432 	struct rw_semaphore	umem_rwsem;
1433 	void (*invalidate_range)(struct ib_umem *umem,
1434 				 unsigned long start, unsigned long end);
1435 
1436 	struct mmu_notifier	mn;
1437 	atomic_t		notifier_count;
1438 	/* A list of umems that don't have private mmu notifier counters yet. */
1439 	struct list_head	no_private_counters;
1440 	int                     odp_mrs_count;
1441 #endif
1442 
1443 	struct ib_rdmacg_object	cg_obj;
1444 };
1445 
1446 struct ib_uobject {
1447 	u64			user_handle;	/* handle given to us by userspace */
1448 	struct ib_ucontext     *context;	/* associated user context */
1449 	void		       *object;		/* containing object */
1450 	struct list_head	list;		/* link to context's list */
1451 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1452 	int			id;		/* index into kernel idr */
1453 	struct kref		ref;
1454 	atomic_t		usecnt;		/* protects exclusive access */
1455 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1456 
1457 	const struct uverbs_obj_type *type;
1458 };
1459 
1460 struct ib_uobject_file {
1461 	struct ib_uobject	uobj;
1462 	/* ufile contains the lock between context release and file close */
1463 	struct ib_uverbs_file	*ufile;
1464 };
1465 
1466 struct ib_udata {
1467 	const void __user *inbuf;
1468 	void __user *outbuf;
1469 	size_t       inlen;
1470 	size_t       outlen;
1471 };
1472 
1473 struct ib_pd {
1474 	u32			local_dma_lkey;
1475 	u32			flags;
1476 	struct ib_device       *device;
1477 	struct ib_uobject      *uobject;
1478 	atomic_t          	usecnt; /* count all resources */
1479 
1480 	u32			unsafe_global_rkey;
1481 
1482 	/*
1483 	 * Implementation details of the RDMA core, don't use in drivers:
1484 	 */
1485 	struct ib_mr	       *__internal_mr;
1486 };
1487 
1488 struct ib_xrcd {
1489 	struct ib_device       *device;
1490 	atomic_t		usecnt; /* count all exposed resources */
1491 	struct inode	       *inode;
1492 
1493 	struct mutex		tgt_qp_mutex;
1494 	struct list_head	tgt_qp_list;
1495 };
1496 
1497 struct ib_ah {
1498 	struct ib_device	*device;
1499 	struct ib_pd		*pd;
1500 	struct ib_uobject	*uobject;
1501 	enum rdma_ah_attr_type	type;
1502 };
1503 
1504 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1505 
1506 enum ib_poll_context {
1507 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1508 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1509 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1510 };
1511 
1512 struct ib_cq {
1513 	struct ib_device       *device;
1514 	struct ib_uobject      *uobject;
1515 	ib_comp_handler   	comp_handler;
1516 	void                  (*event_handler)(struct ib_event *, void *);
1517 	void                   *cq_context;
1518 	int               	cqe;
1519 	atomic_t          	usecnt; /* count number of work queues */
1520 	enum ib_poll_context	poll_ctx;
1521 	struct ib_wc		*wc;
1522 	union {
1523 		struct irq_poll		iop;
1524 		struct work_struct	work;
1525 	};
1526 };
1527 
1528 struct ib_srq {
1529 	struct ib_device       *device;
1530 	struct ib_pd	       *pd;
1531 	struct ib_uobject      *uobject;
1532 	void		      (*event_handler)(struct ib_event *, void *);
1533 	void		       *srq_context;
1534 	enum ib_srq_type	srq_type;
1535 	atomic_t		usecnt;
1536 
1537 	union {
1538 		struct {
1539 			struct ib_xrcd *xrcd;
1540 			struct ib_cq   *cq;
1541 			u32		srq_num;
1542 		} xrc;
1543 	} ext;
1544 };
1545 
1546 enum ib_raw_packet_caps {
1547 	/* Strip cvlan from incoming packet and report it in the matching work
1548 	 * completion is supported.
1549 	 */
1550 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1551 	/* Scatter FCS field of an incoming packet to host memory is supported.
1552 	 */
1553 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1554 	/* Checksum offloads are supported (for both send and receive). */
1555 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1556 	/* When a packet is received for an RQ with no receive WQEs, the
1557 	 * packet processing is delayed.
1558 	 */
1559 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1560 };
1561 
1562 enum ib_wq_type {
1563 	IB_WQT_RQ
1564 };
1565 
1566 enum ib_wq_state {
1567 	IB_WQS_RESET,
1568 	IB_WQS_RDY,
1569 	IB_WQS_ERR
1570 };
1571 
1572 struct ib_wq {
1573 	struct ib_device       *device;
1574 	struct ib_uobject      *uobject;
1575 	void		    *wq_context;
1576 	void		    (*event_handler)(struct ib_event *, void *);
1577 	struct ib_pd	       *pd;
1578 	struct ib_cq	       *cq;
1579 	u32		wq_num;
1580 	enum ib_wq_state       state;
1581 	enum ib_wq_type	wq_type;
1582 	atomic_t		usecnt;
1583 };
1584 
1585 enum ib_wq_flags {
1586 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1587 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1588 	IB_WQ_FLAGS_DELAY_DROP		= 1 << 2,
1589 };
1590 
1591 struct ib_wq_init_attr {
1592 	void		       *wq_context;
1593 	enum ib_wq_type	wq_type;
1594 	u32		max_wr;
1595 	u32		max_sge;
1596 	struct	ib_cq	       *cq;
1597 	void		    (*event_handler)(struct ib_event *, void *);
1598 	u32		create_flags; /* Use enum ib_wq_flags */
1599 };
1600 
1601 enum ib_wq_attr_mask {
1602 	IB_WQ_STATE		= 1 << 0,
1603 	IB_WQ_CUR_STATE		= 1 << 1,
1604 	IB_WQ_FLAGS		= 1 << 2,
1605 };
1606 
1607 struct ib_wq_attr {
1608 	enum	ib_wq_state	wq_state;
1609 	enum	ib_wq_state	curr_wq_state;
1610 	u32			flags; /* Use enum ib_wq_flags */
1611 	u32			flags_mask; /* Use enum ib_wq_flags */
1612 };
1613 
1614 struct ib_rwq_ind_table {
1615 	struct ib_device	*device;
1616 	struct ib_uobject      *uobject;
1617 	atomic_t		usecnt;
1618 	u32		ind_tbl_num;
1619 	u32		log_ind_tbl_size;
1620 	struct ib_wq	**ind_tbl;
1621 };
1622 
1623 struct ib_rwq_ind_table_init_attr {
1624 	u32		log_ind_tbl_size;
1625 	/* Each entry is a pointer to Receive Work Queue */
1626 	struct ib_wq	**ind_tbl;
1627 };
1628 
1629 enum port_pkey_state {
1630 	IB_PORT_PKEY_NOT_VALID = 0,
1631 	IB_PORT_PKEY_VALID = 1,
1632 	IB_PORT_PKEY_LISTED = 2,
1633 };
1634 
1635 struct ib_qp_security;
1636 
1637 struct ib_port_pkey {
1638 	enum port_pkey_state	state;
1639 	u16			pkey_index;
1640 	u8			port_num;
1641 	struct list_head	qp_list;
1642 	struct list_head	to_error_list;
1643 	struct ib_qp_security  *sec;
1644 };
1645 
1646 struct ib_ports_pkeys {
1647 	struct ib_port_pkey	main;
1648 	struct ib_port_pkey	alt;
1649 };
1650 
1651 struct ib_qp_security {
1652 	struct ib_qp	       *qp;
1653 	struct ib_device       *dev;
1654 	/* Hold this mutex when changing port and pkey settings. */
1655 	struct mutex		mutex;
1656 	struct ib_ports_pkeys  *ports_pkeys;
1657 	/* A list of all open shared QP handles.  Required to enforce security
1658 	 * properly for all users of a shared QP.
1659 	 */
1660 	struct list_head        shared_qp_list;
1661 	void                   *security;
1662 	bool			destroying;
1663 	atomic_t		error_list_count;
1664 	struct completion	error_complete;
1665 	int			error_comps_pending;
1666 };
1667 
1668 /*
1669  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1670  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1671  */
1672 struct ib_qp {
1673 	struct ib_device       *device;
1674 	struct ib_pd	       *pd;
1675 	struct ib_cq	       *send_cq;
1676 	struct ib_cq	       *recv_cq;
1677 	spinlock_t		mr_lock;
1678 	int			mrs_used;
1679 	struct list_head	rdma_mrs;
1680 	struct list_head	sig_mrs;
1681 	struct ib_srq	       *srq;
1682 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1683 	struct list_head	xrcd_list;
1684 
1685 	/* count times opened, mcast attaches, flow attaches */
1686 	atomic_t		usecnt;
1687 	struct list_head	open_list;
1688 	struct ib_qp           *real_qp;
1689 	struct ib_uobject      *uobject;
1690 	void                  (*event_handler)(struct ib_event *, void *);
1691 	void		       *qp_context;
1692 	u32			qp_num;
1693 	u32			max_write_sge;
1694 	u32			max_read_sge;
1695 	enum ib_qp_type		qp_type;
1696 	struct ib_rwq_ind_table *rwq_ind_tbl;
1697 	struct ib_qp_security  *qp_sec;
1698 };
1699 
1700 struct ib_mr {
1701 	struct ib_device  *device;
1702 	struct ib_pd	  *pd;
1703 	u32		   lkey;
1704 	u32		   rkey;
1705 	u64		   iova;
1706 	u32		   length;
1707 	unsigned int	   page_size;
1708 	bool		   need_inval;
1709 	union {
1710 		struct ib_uobject	*uobject;	/* user */
1711 		struct list_head	qp_entry;	/* FR */
1712 	};
1713 };
1714 
1715 struct ib_mw {
1716 	struct ib_device	*device;
1717 	struct ib_pd		*pd;
1718 	struct ib_uobject	*uobject;
1719 	u32			rkey;
1720 	enum ib_mw_type         type;
1721 };
1722 
1723 struct ib_fmr {
1724 	struct ib_device	*device;
1725 	struct ib_pd		*pd;
1726 	struct list_head	list;
1727 	u32			lkey;
1728 	u32			rkey;
1729 };
1730 
1731 /* Supported steering options */
1732 enum ib_flow_attr_type {
1733 	/* steering according to rule specifications */
1734 	IB_FLOW_ATTR_NORMAL		= 0x0,
1735 	/* default unicast and multicast rule -
1736 	 * receive all Eth traffic which isn't steered to any QP
1737 	 */
1738 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1739 	/* default multicast rule -
1740 	 * receive all Eth multicast traffic which isn't steered to any QP
1741 	 */
1742 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1743 	/* sniffer rule - receive all port traffic */
1744 	IB_FLOW_ATTR_SNIFFER		= 0x3
1745 };
1746 
1747 /* Supported steering header types */
1748 enum ib_flow_spec_type {
1749 	/* L2 headers*/
1750 	IB_FLOW_SPEC_ETH		= 0x20,
1751 	IB_FLOW_SPEC_IB			= 0x22,
1752 	/* L3 header*/
1753 	IB_FLOW_SPEC_IPV4		= 0x30,
1754 	IB_FLOW_SPEC_IPV6		= 0x31,
1755 	/* L4 headers*/
1756 	IB_FLOW_SPEC_TCP		= 0x40,
1757 	IB_FLOW_SPEC_UDP		= 0x41,
1758 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1759 	IB_FLOW_SPEC_INNER		= 0x100,
1760 	/* Actions */
1761 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1762 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1763 };
1764 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1765 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1766 
1767 /* Flow steering rule priority is set according to it's domain.
1768  * Lower domain value means higher priority.
1769  */
1770 enum ib_flow_domain {
1771 	IB_FLOW_DOMAIN_USER,
1772 	IB_FLOW_DOMAIN_ETHTOOL,
1773 	IB_FLOW_DOMAIN_RFS,
1774 	IB_FLOW_DOMAIN_NIC,
1775 	IB_FLOW_DOMAIN_NUM /* Must be last */
1776 };
1777 
1778 enum ib_flow_flags {
1779 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1780 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1781 };
1782 
1783 struct ib_flow_eth_filter {
1784 	u8	dst_mac[6];
1785 	u8	src_mac[6];
1786 	__be16	ether_type;
1787 	__be16	vlan_tag;
1788 	/* Must be last */
1789 	u8	real_sz[0];
1790 };
1791 
1792 struct ib_flow_spec_eth {
1793 	u32			  type;
1794 	u16			  size;
1795 	struct ib_flow_eth_filter val;
1796 	struct ib_flow_eth_filter mask;
1797 };
1798 
1799 struct ib_flow_ib_filter {
1800 	__be16 dlid;
1801 	__u8   sl;
1802 	/* Must be last */
1803 	u8	real_sz[0];
1804 };
1805 
1806 struct ib_flow_spec_ib {
1807 	u32			 type;
1808 	u16			 size;
1809 	struct ib_flow_ib_filter val;
1810 	struct ib_flow_ib_filter mask;
1811 };
1812 
1813 /* IPv4 header flags */
1814 enum ib_ipv4_flags {
1815 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1816 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1817 				    last have this flag set */
1818 };
1819 
1820 struct ib_flow_ipv4_filter {
1821 	__be32	src_ip;
1822 	__be32	dst_ip;
1823 	u8	proto;
1824 	u8	tos;
1825 	u8	ttl;
1826 	u8	flags;
1827 	/* Must be last */
1828 	u8	real_sz[0];
1829 };
1830 
1831 struct ib_flow_spec_ipv4 {
1832 	u32			   type;
1833 	u16			   size;
1834 	struct ib_flow_ipv4_filter val;
1835 	struct ib_flow_ipv4_filter mask;
1836 };
1837 
1838 struct ib_flow_ipv6_filter {
1839 	u8	src_ip[16];
1840 	u8	dst_ip[16];
1841 	__be32	flow_label;
1842 	u8	next_hdr;
1843 	u8	traffic_class;
1844 	u8	hop_limit;
1845 	/* Must be last */
1846 	u8	real_sz[0];
1847 };
1848 
1849 struct ib_flow_spec_ipv6 {
1850 	u32			   type;
1851 	u16			   size;
1852 	struct ib_flow_ipv6_filter val;
1853 	struct ib_flow_ipv6_filter mask;
1854 };
1855 
1856 struct ib_flow_tcp_udp_filter {
1857 	__be16	dst_port;
1858 	__be16	src_port;
1859 	/* Must be last */
1860 	u8	real_sz[0];
1861 };
1862 
1863 struct ib_flow_spec_tcp_udp {
1864 	u32			      type;
1865 	u16			      size;
1866 	struct ib_flow_tcp_udp_filter val;
1867 	struct ib_flow_tcp_udp_filter mask;
1868 };
1869 
1870 struct ib_flow_tunnel_filter {
1871 	__be32	tunnel_id;
1872 	u8	real_sz[0];
1873 };
1874 
1875 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1876  * the tunnel_id from val has the vni value
1877  */
1878 struct ib_flow_spec_tunnel {
1879 	u32			      type;
1880 	u16			      size;
1881 	struct ib_flow_tunnel_filter  val;
1882 	struct ib_flow_tunnel_filter  mask;
1883 };
1884 
1885 struct ib_flow_spec_action_tag {
1886 	enum ib_flow_spec_type	      type;
1887 	u16			      size;
1888 	u32                           tag_id;
1889 };
1890 
1891 struct ib_flow_spec_action_drop {
1892 	enum ib_flow_spec_type	      type;
1893 	u16			      size;
1894 };
1895 
1896 union ib_flow_spec {
1897 	struct {
1898 		u32			type;
1899 		u16			size;
1900 	};
1901 	struct ib_flow_spec_eth		eth;
1902 	struct ib_flow_spec_ib		ib;
1903 	struct ib_flow_spec_ipv4        ipv4;
1904 	struct ib_flow_spec_tcp_udp	tcp_udp;
1905 	struct ib_flow_spec_ipv6        ipv6;
1906 	struct ib_flow_spec_tunnel      tunnel;
1907 	struct ib_flow_spec_action_tag  flow_tag;
1908 	struct ib_flow_spec_action_drop drop;
1909 };
1910 
1911 struct ib_flow_attr {
1912 	enum ib_flow_attr_type type;
1913 	u16	     size;
1914 	u16	     priority;
1915 	u32	     flags;
1916 	u8	     num_of_specs;
1917 	u8	     port;
1918 	/* Following are the optional layers according to user request
1919 	 * struct ib_flow_spec_xxx
1920 	 * struct ib_flow_spec_yyy
1921 	 */
1922 };
1923 
1924 struct ib_flow {
1925 	struct ib_qp		*qp;
1926 	struct ib_uobject	*uobject;
1927 };
1928 
1929 struct ib_mad_hdr;
1930 struct ib_grh;
1931 
1932 enum ib_process_mad_flags {
1933 	IB_MAD_IGNORE_MKEY	= 1,
1934 	IB_MAD_IGNORE_BKEY	= 2,
1935 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1936 };
1937 
1938 enum ib_mad_result {
1939 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1940 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1941 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1942 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1943 };
1944 
1945 struct ib_port_cache {
1946 	u64		      subnet_prefix;
1947 	struct ib_pkey_cache  *pkey;
1948 	struct ib_gid_table   *gid;
1949 	u8                     lmc;
1950 	enum ib_port_state     port_state;
1951 };
1952 
1953 struct ib_cache {
1954 	rwlock_t                lock;
1955 	struct ib_event_handler event_handler;
1956 	struct ib_port_cache   *ports;
1957 };
1958 
1959 struct iw_cm_verbs;
1960 
1961 struct ib_port_immutable {
1962 	int                           pkey_tbl_len;
1963 	int                           gid_tbl_len;
1964 	u32                           core_cap_flags;
1965 	u32                           max_mad_size;
1966 };
1967 
1968 /* rdma netdev type - specifies protocol type */
1969 enum rdma_netdev_t {
1970 	RDMA_NETDEV_OPA_VNIC,
1971 	RDMA_NETDEV_IPOIB,
1972 };
1973 
1974 /**
1975  * struct rdma_netdev - rdma netdev
1976  * For cases where netstack interfacing is required.
1977  */
1978 struct rdma_netdev {
1979 	void              *clnt_priv;
1980 	struct ib_device  *hca;
1981 	u8                 port_num;
1982 
1983 	/* cleanup function must be specified */
1984 	void (*free_rdma_netdev)(struct net_device *netdev);
1985 
1986 	/* control functions */
1987 	void (*set_id)(struct net_device *netdev, int id);
1988 	/* send packet */
1989 	int (*send)(struct net_device *dev, struct sk_buff *skb,
1990 		    struct ib_ah *address, u32 dqpn);
1991 	/* multicast */
1992 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
1993 			    union ib_gid *gid, u16 mlid,
1994 			    int set_qkey, u32 qkey);
1995 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
1996 			    union ib_gid *gid, u16 mlid);
1997 };
1998 
1999 struct ib_port_pkey_list {
2000 	/* Lock to hold while modifying the list. */
2001 	spinlock_t                    list_lock;
2002 	struct list_head              pkey_list;
2003 };
2004 
2005 struct ib_device {
2006 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2007 	struct device                *dma_device;
2008 
2009 	char                          name[IB_DEVICE_NAME_MAX];
2010 
2011 	struct list_head              event_handler_list;
2012 	spinlock_t                    event_handler_lock;
2013 
2014 	spinlock_t                    client_data_lock;
2015 	struct list_head              core_list;
2016 	/* Access to the client_data_list is protected by the client_data_lock
2017 	 * spinlock and the lists_rwsem read-write semaphore */
2018 	struct list_head              client_data_list;
2019 
2020 	struct ib_cache               cache;
2021 	/**
2022 	 * port_immutable is indexed by port number
2023 	 */
2024 	struct ib_port_immutable     *port_immutable;
2025 
2026 	int			      num_comp_vectors;
2027 
2028 	struct ib_port_pkey_list     *port_pkey_list;
2029 
2030 	struct iw_cm_verbs	     *iwcm;
2031 
2032 	/**
2033 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2034 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2035 	 *   core when the device is removed.  A lifespan of -1 in the return
2036 	 *   struct tells the core to set a default lifespan.
2037 	 */
2038 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
2039 						     u8 port_num);
2040 	/**
2041 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2042 	 * @index - The index in the value array we wish to have updated, or
2043 	 *   num_counters if we want all stats updated
2044 	 * Return codes -
2045 	 *   < 0 - Error, no counters updated
2046 	 *   index - Updated the single counter pointed to by index
2047 	 *   num_counters - Updated all counters (will reset the timestamp
2048 	 *     and prevent further calls for lifespan milliseconds)
2049 	 * Drivers are allowed to update all counters in leiu of just the
2050 	 *   one given in index at their option
2051 	 */
2052 	int		           (*get_hw_stats)(struct ib_device *device,
2053 						   struct rdma_hw_stats *stats,
2054 						   u8 port, int index);
2055 	int		           (*query_device)(struct ib_device *device,
2056 						   struct ib_device_attr *device_attr,
2057 						   struct ib_udata *udata);
2058 	int		           (*query_port)(struct ib_device *device,
2059 						 u8 port_num,
2060 						 struct ib_port_attr *port_attr);
2061 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
2062 						     u8 port_num);
2063 	/* When calling get_netdev, the HW vendor's driver should return the
2064 	 * net device of device @device at port @port_num or NULL if such
2065 	 * a net device doesn't exist. The vendor driver should call dev_hold
2066 	 * on this net device. The HW vendor's device driver must guarantee
2067 	 * that this function returns NULL before the net device reaches
2068 	 * NETDEV_UNREGISTER_FINAL state.
2069 	 */
2070 	struct net_device	  *(*get_netdev)(struct ib_device *device,
2071 						 u8 port_num);
2072 	int		           (*query_gid)(struct ib_device *device,
2073 						u8 port_num, int index,
2074 						union ib_gid *gid);
2075 	/* When calling add_gid, the HW vendor's driver should
2076 	 * add the gid of device @device at gid index @index of
2077 	 * port @port_num to be @gid. Meta-info of that gid (for example,
2078 	 * the network device related to this gid is available
2079 	 * at @attr. @context allows the HW vendor driver to store extra
2080 	 * information together with a GID entry. The HW vendor may allocate
2081 	 * memory to contain this information and store it in @context when a
2082 	 * new GID entry is written to. Params are consistent until the next
2083 	 * call of add_gid or delete_gid. The function should return 0 on
2084 	 * success or error otherwise. The function could be called
2085 	 * concurrently for different ports. This function is only called
2086 	 * when roce_gid_table is used.
2087 	 */
2088 	int		           (*add_gid)(struct ib_device *device,
2089 					      u8 port_num,
2090 					      unsigned int index,
2091 					      const union ib_gid *gid,
2092 					      const struct ib_gid_attr *attr,
2093 					      void **context);
2094 	/* When calling del_gid, the HW vendor's driver should delete the
2095 	 * gid of device @device at gid index @index of port @port_num.
2096 	 * Upon the deletion of a GID entry, the HW vendor must free any
2097 	 * allocated memory. The caller will clear @context afterwards.
2098 	 * This function is only called when roce_gid_table is used.
2099 	 */
2100 	int		           (*del_gid)(struct ib_device *device,
2101 					      u8 port_num,
2102 					      unsigned int index,
2103 					      void **context);
2104 	int		           (*query_pkey)(struct ib_device *device,
2105 						 u8 port_num, u16 index, u16 *pkey);
2106 	int		           (*modify_device)(struct ib_device *device,
2107 						    int device_modify_mask,
2108 						    struct ib_device_modify *device_modify);
2109 	int		           (*modify_port)(struct ib_device *device,
2110 						  u8 port_num, int port_modify_mask,
2111 						  struct ib_port_modify *port_modify);
2112 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
2113 						     struct ib_udata *udata);
2114 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
2115 	int                        (*mmap)(struct ib_ucontext *context,
2116 					   struct vm_area_struct *vma);
2117 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
2118 					       struct ib_ucontext *context,
2119 					       struct ib_udata *udata);
2120 	int                        (*dealloc_pd)(struct ib_pd *pd);
2121 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
2122 						struct rdma_ah_attr *ah_attr,
2123 						struct ib_udata *udata);
2124 	int                        (*modify_ah)(struct ib_ah *ah,
2125 						struct rdma_ah_attr *ah_attr);
2126 	int                        (*query_ah)(struct ib_ah *ah,
2127 					       struct rdma_ah_attr *ah_attr);
2128 	int                        (*destroy_ah)(struct ib_ah *ah);
2129 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
2130 						 struct ib_srq_init_attr *srq_init_attr,
2131 						 struct ib_udata *udata);
2132 	int                        (*modify_srq)(struct ib_srq *srq,
2133 						 struct ib_srq_attr *srq_attr,
2134 						 enum ib_srq_attr_mask srq_attr_mask,
2135 						 struct ib_udata *udata);
2136 	int                        (*query_srq)(struct ib_srq *srq,
2137 						struct ib_srq_attr *srq_attr);
2138 	int                        (*destroy_srq)(struct ib_srq *srq);
2139 	int                        (*post_srq_recv)(struct ib_srq *srq,
2140 						    struct ib_recv_wr *recv_wr,
2141 						    struct ib_recv_wr **bad_recv_wr);
2142 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
2143 						struct ib_qp_init_attr *qp_init_attr,
2144 						struct ib_udata *udata);
2145 	int                        (*modify_qp)(struct ib_qp *qp,
2146 						struct ib_qp_attr *qp_attr,
2147 						int qp_attr_mask,
2148 						struct ib_udata *udata);
2149 	int                        (*query_qp)(struct ib_qp *qp,
2150 					       struct ib_qp_attr *qp_attr,
2151 					       int qp_attr_mask,
2152 					       struct ib_qp_init_attr *qp_init_attr);
2153 	int                        (*destroy_qp)(struct ib_qp *qp);
2154 	int                        (*post_send)(struct ib_qp *qp,
2155 						struct ib_send_wr *send_wr,
2156 						struct ib_send_wr **bad_send_wr);
2157 	int                        (*post_recv)(struct ib_qp *qp,
2158 						struct ib_recv_wr *recv_wr,
2159 						struct ib_recv_wr **bad_recv_wr);
2160 	struct ib_cq *             (*create_cq)(struct ib_device *device,
2161 						const struct ib_cq_init_attr *attr,
2162 						struct ib_ucontext *context,
2163 						struct ib_udata *udata);
2164 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2165 						u16 cq_period);
2166 	int                        (*destroy_cq)(struct ib_cq *cq);
2167 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
2168 						struct ib_udata *udata);
2169 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
2170 					      struct ib_wc *wc);
2171 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2172 	int                        (*req_notify_cq)(struct ib_cq *cq,
2173 						    enum ib_cq_notify_flags flags);
2174 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
2175 						      int wc_cnt);
2176 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
2177 						 int mr_access_flags);
2178 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
2179 						  u64 start, u64 length,
2180 						  u64 virt_addr,
2181 						  int mr_access_flags,
2182 						  struct ib_udata *udata);
2183 	int			   (*rereg_user_mr)(struct ib_mr *mr,
2184 						    int flags,
2185 						    u64 start, u64 length,
2186 						    u64 virt_addr,
2187 						    int mr_access_flags,
2188 						    struct ib_pd *pd,
2189 						    struct ib_udata *udata);
2190 	int                        (*dereg_mr)(struct ib_mr *mr);
2191 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2192 					       enum ib_mr_type mr_type,
2193 					       u32 max_num_sg);
2194 	int                        (*map_mr_sg)(struct ib_mr *mr,
2195 						struct scatterlist *sg,
2196 						int sg_nents,
2197 						unsigned int *sg_offset);
2198 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2199 					       enum ib_mw_type type,
2200 					       struct ib_udata *udata);
2201 	int                        (*dealloc_mw)(struct ib_mw *mw);
2202 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2203 						int mr_access_flags,
2204 						struct ib_fmr_attr *fmr_attr);
2205 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2206 						   u64 *page_list, int list_len,
2207 						   u64 iova);
2208 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2209 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2210 	int                        (*attach_mcast)(struct ib_qp *qp,
2211 						   union ib_gid *gid,
2212 						   u16 lid);
2213 	int                        (*detach_mcast)(struct ib_qp *qp,
2214 						   union ib_gid *gid,
2215 						   u16 lid);
2216 	int                        (*process_mad)(struct ib_device *device,
2217 						  int process_mad_flags,
2218 						  u8 port_num,
2219 						  const struct ib_wc *in_wc,
2220 						  const struct ib_grh *in_grh,
2221 						  const struct ib_mad_hdr *in_mad,
2222 						  size_t in_mad_size,
2223 						  struct ib_mad_hdr *out_mad,
2224 						  size_t *out_mad_size,
2225 						  u16 *out_mad_pkey_index);
2226 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2227 						 struct ib_ucontext *ucontext,
2228 						 struct ib_udata *udata);
2229 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2230 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2231 						  struct ib_flow_attr
2232 						  *flow_attr,
2233 						  int domain);
2234 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2235 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2236 						      struct ib_mr_status *mr_status);
2237 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2238 	void			   (*drain_rq)(struct ib_qp *qp);
2239 	void			   (*drain_sq)(struct ib_qp *qp);
2240 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2241 							int state);
2242 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2243 						   struct ifla_vf_info *ivf);
2244 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2245 						   struct ifla_vf_stats *stats);
2246 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2247 						  int type);
2248 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2249 						struct ib_wq_init_attr *init_attr,
2250 						struct ib_udata *udata);
2251 	int			   (*destroy_wq)(struct ib_wq *wq);
2252 	int			   (*modify_wq)(struct ib_wq *wq,
2253 						struct ib_wq_attr *attr,
2254 						u32 wq_attr_mask,
2255 						struct ib_udata *udata);
2256 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2257 							   struct ib_rwq_ind_table_init_attr *init_attr,
2258 							   struct ib_udata *udata);
2259 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2260 	/**
2261 	 * rdma netdev operation
2262 	 *
2263 	 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2264 	 * doesn't support the specified rdma netdev type.
2265 	 */
2266 	struct net_device *(*alloc_rdma_netdev)(
2267 					struct ib_device *device,
2268 					u8 port_num,
2269 					enum rdma_netdev_t type,
2270 					const char *name,
2271 					unsigned char name_assign_type,
2272 					void (*setup)(struct net_device *));
2273 
2274 	struct module               *owner;
2275 	struct device                dev;
2276 	struct kobject               *ports_parent;
2277 	struct list_head             port_list;
2278 
2279 	enum {
2280 		IB_DEV_UNINITIALIZED,
2281 		IB_DEV_REGISTERED,
2282 		IB_DEV_UNREGISTERED
2283 	}                            reg_state;
2284 
2285 	int			     uverbs_abi_ver;
2286 	u64			     uverbs_cmd_mask;
2287 	u64			     uverbs_ex_cmd_mask;
2288 
2289 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2290 	__be64			     node_guid;
2291 	u32			     local_dma_lkey;
2292 	u16                          is_switch:1;
2293 	u8                           node_type;
2294 	u8                           phys_port_cnt;
2295 	struct ib_device_attr        attrs;
2296 	struct attribute_group	     *hw_stats_ag;
2297 	struct rdma_hw_stats         *hw_stats;
2298 
2299 #ifdef CONFIG_CGROUP_RDMA
2300 	struct rdmacg_device         cg_device;
2301 #endif
2302 
2303 	u32                          index;
2304 
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 *, u8, struct ib_port_immutable *);
2312 	void (*get_dev_fw_str)(struct ib_device *, char *str);
2313 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2314 						     int comp_vector);
2315 };
2316 
2317 struct ib_client {
2318 	char  *name;
2319 	void (*add)   (struct ib_device *);
2320 	void (*remove)(struct ib_device *, void *client_data);
2321 
2322 	/* Returns the net_dev belonging to this ib_client and matching the
2323 	 * given parameters.
2324 	 * @dev:	 An RDMA device that the net_dev use for communication.
2325 	 * @port:	 A physical port number on the RDMA device.
2326 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2327 	 * @gid:	 A GID that the net_dev uses to communicate.
2328 	 * @addr:	 An IP address the net_dev is configured with.
2329 	 * @client_data: The device's client data set by ib_set_client_data().
2330 	 *
2331 	 * An ib_client that implements a net_dev on top of RDMA devices
2332 	 * (such as IP over IB) should implement this callback, allowing the
2333 	 * rdma_cm module to find the right net_dev for a given request.
2334 	 *
2335 	 * The caller is responsible for calling dev_put on the returned
2336 	 * netdev. */
2337 	struct net_device *(*get_net_dev_by_params)(
2338 			struct ib_device *dev,
2339 			u8 port,
2340 			u16 pkey,
2341 			const union ib_gid *gid,
2342 			const struct sockaddr *addr,
2343 			void *client_data);
2344 	struct list_head list;
2345 };
2346 
2347 struct ib_device *ib_alloc_device(size_t size);
2348 void ib_dealloc_device(struct ib_device *device);
2349 
2350 void ib_get_device_fw_str(struct ib_device *device, char *str);
2351 
2352 int ib_register_device(struct ib_device *device,
2353 		       int (*port_callback)(struct ib_device *,
2354 					    u8, struct kobject *));
2355 void ib_unregister_device(struct ib_device *device);
2356 
2357 int ib_register_client   (struct ib_client *client);
2358 void ib_unregister_client(struct ib_client *client);
2359 
2360 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2361 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2362 			 void *data);
2363 
2364 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2365 {
2366 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2367 }
2368 
2369 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2370 {
2371 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2372 }
2373 
2374 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2375 				       size_t offset,
2376 				       size_t len)
2377 {
2378 	const void __user *p = udata->inbuf + offset;
2379 	bool ret;
2380 	u8 *buf;
2381 
2382 	if (len > USHRT_MAX)
2383 		return false;
2384 
2385 	buf = memdup_user(p, len);
2386 	if (IS_ERR(buf))
2387 		return false;
2388 
2389 	ret = !memchr_inv(buf, 0, len);
2390 	kfree(buf);
2391 	return ret;
2392 }
2393 
2394 /**
2395  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2396  * contains all required attributes and no attributes not allowed for
2397  * the given QP state transition.
2398  * @cur_state: Current QP state
2399  * @next_state: Next QP state
2400  * @type: QP type
2401  * @mask: Mask of supplied QP attributes
2402  * @ll : link layer of port
2403  *
2404  * This function is a helper function that a low-level driver's
2405  * modify_qp method can use to validate the consumer's input.  It
2406  * checks that cur_state and next_state are valid QP states, that a
2407  * transition from cur_state to next_state is allowed by the IB spec,
2408  * and that the attribute mask supplied is allowed for the transition.
2409  */
2410 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2411 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2412 		       enum rdma_link_layer ll);
2413 
2414 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2415 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2416 void ib_dispatch_event(struct ib_event *event);
2417 
2418 int ib_query_port(struct ib_device *device,
2419 		  u8 port_num, struct ib_port_attr *port_attr);
2420 
2421 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2422 					       u8 port_num);
2423 
2424 /**
2425  * rdma_cap_ib_switch - Check if the device is IB switch
2426  * @device: Device to check
2427  *
2428  * Device driver is responsible for setting is_switch bit on
2429  * in ib_device structure at init time.
2430  *
2431  * Return: true if the device is IB switch.
2432  */
2433 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2434 {
2435 	return device->is_switch;
2436 }
2437 
2438 /**
2439  * rdma_start_port - Return the first valid port number for the device
2440  * specified
2441  *
2442  * @device: Device to be checked
2443  *
2444  * Return start port number
2445  */
2446 static inline u8 rdma_start_port(const struct ib_device *device)
2447 {
2448 	return rdma_cap_ib_switch(device) ? 0 : 1;
2449 }
2450 
2451 /**
2452  * rdma_end_port - Return the last valid port number for the device
2453  * specified
2454  *
2455  * @device: Device to be checked
2456  *
2457  * Return last port number
2458  */
2459 static inline u8 rdma_end_port(const struct ib_device *device)
2460 {
2461 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2462 }
2463 
2464 static inline int rdma_is_port_valid(const struct ib_device *device,
2465 				     unsigned int port)
2466 {
2467 	return (port >= rdma_start_port(device) &&
2468 		port <= rdma_end_port(device));
2469 }
2470 
2471 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2472 {
2473 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2474 }
2475 
2476 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2477 {
2478 	return device->port_immutable[port_num].core_cap_flags &
2479 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2480 }
2481 
2482 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2483 {
2484 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2485 }
2486 
2487 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2488 {
2489 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2490 }
2491 
2492 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2493 {
2494 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2495 }
2496 
2497 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2498 {
2499 	return rdma_protocol_ib(device, port_num) ||
2500 		rdma_protocol_roce(device, port_num);
2501 }
2502 
2503 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2504 {
2505 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2506 }
2507 
2508 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2509 {
2510 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2511 }
2512 
2513 /**
2514  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2515  * Management Datagrams.
2516  * @device: Device to check
2517  * @port_num: Port number to check
2518  *
2519  * Management Datagrams (MAD) are a required part of the InfiniBand
2520  * specification and are supported on all InfiniBand devices.  A slightly
2521  * extended version are also supported on OPA interfaces.
2522  *
2523  * Return: true if the port supports sending/receiving of MAD packets.
2524  */
2525 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2526 {
2527 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2528 }
2529 
2530 /**
2531  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2532  * Management Datagrams.
2533  * @device: Device to check
2534  * @port_num: Port number to check
2535  *
2536  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2537  * datagrams with their own versions.  These OPA MADs share many but not all of
2538  * the characteristics of InfiniBand MADs.
2539  *
2540  * OPA MADs differ in the following ways:
2541  *
2542  *    1) MADs are variable size up to 2K
2543  *       IBTA defined MADs remain fixed at 256 bytes
2544  *    2) OPA SMPs must carry valid PKeys
2545  *    3) OPA SMP packets are a different format
2546  *
2547  * Return: true if the port supports OPA MAD packet formats.
2548  */
2549 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2550 {
2551 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2552 		== RDMA_CORE_CAP_OPA_MAD;
2553 }
2554 
2555 /**
2556  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2557  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2558  * @device: Device to check
2559  * @port_num: Port number to check
2560  *
2561  * Each InfiniBand node is required to provide a Subnet Management Agent
2562  * that the subnet manager can access.  Prior to the fabric being fully
2563  * configured by the subnet manager, the SMA is accessed via a well known
2564  * interface called the Subnet Management Interface (SMI).  This interface
2565  * uses directed route packets to communicate with the SM to get around the
2566  * chicken and egg problem of the SM needing to know what's on the fabric
2567  * in order to configure the fabric, and needing to configure the fabric in
2568  * order to send packets to the devices on the fabric.  These directed
2569  * route packets do not need the fabric fully configured in order to reach
2570  * their destination.  The SMI is the only method allowed to send
2571  * directed route packets on an InfiniBand fabric.
2572  *
2573  * Return: true if the port provides an SMI.
2574  */
2575 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2576 {
2577 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2578 }
2579 
2580 /**
2581  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2582  * Communication Manager.
2583  * @device: Device to check
2584  * @port_num: Port number to check
2585  *
2586  * The InfiniBand Communication Manager is one of many pre-defined General
2587  * Service Agents (GSA) that are accessed via the General Service
2588  * Interface (GSI).  It's role is to facilitate establishment of connections
2589  * between nodes as well as other management related tasks for established
2590  * connections.
2591  *
2592  * Return: true if the port supports an IB CM (this does not guarantee that
2593  * a CM is actually running however).
2594  */
2595 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2596 {
2597 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2598 }
2599 
2600 /**
2601  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2602  * Communication Manager.
2603  * @device: Device to check
2604  * @port_num: Port number to check
2605  *
2606  * Similar to above, but specific to iWARP connections which have a different
2607  * managment protocol than InfiniBand.
2608  *
2609  * Return: true if the port supports an iWARP CM (this does not guarantee that
2610  * a CM is actually running however).
2611  */
2612 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2613 {
2614 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2615 }
2616 
2617 /**
2618  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2619  * Subnet Administration.
2620  * @device: Device to check
2621  * @port_num: Port number to check
2622  *
2623  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2624  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2625  * fabrics, devices should resolve routes to other hosts by contacting the
2626  * SA to query the proper route.
2627  *
2628  * Return: true if the port should act as a client to the fabric Subnet
2629  * Administration interface.  This does not imply that the SA service is
2630  * running locally.
2631  */
2632 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2633 {
2634 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2635 }
2636 
2637 /**
2638  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2639  * Multicast.
2640  * @device: Device to check
2641  * @port_num: Port number to check
2642  *
2643  * InfiniBand multicast registration is more complex than normal IPv4 or
2644  * IPv6 multicast registration.  Each Host Channel Adapter must register
2645  * with the Subnet Manager when it wishes to join a multicast group.  It
2646  * should do so only once regardless of how many queue pairs it subscribes
2647  * to this group.  And it should leave the group only after all queue pairs
2648  * attached to the group have been detached.
2649  *
2650  * Return: true if the port must undertake the additional adminstrative
2651  * overhead of registering/unregistering with the SM and tracking of the
2652  * total number of queue pairs attached to the multicast group.
2653  */
2654 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2655 {
2656 	return rdma_cap_ib_sa(device, port_num);
2657 }
2658 
2659 /**
2660  * rdma_cap_af_ib - Check if the port of device has the capability
2661  * Native Infiniband Address.
2662  * @device: Device to check
2663  * @port_num: Port number to check
2664  *
2665  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2666  * GID.  RoCE uses a different mechanism, but still generates a GID via
2667  * a prescribed mechanism and port specific data.
2668  *
2669  * Return: true if the port uses a GID address to identify devices on the
2670  * network.
2671  */
2672 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2673 {
2674 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2675 }
2676 
2677 /**
2678  * rdma_cap_eth_ah - Check if the port of device has the capability
2679  * Ethernet Address Handle.
2680  * @device: Device to check
2681  * @port_num: Port number to check
2682  *
2683  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2684  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2685  * port.  Normally, packet headers are generated by the sending host
2686  * adapter, but when sending connectionless datagrams, we must manually
2687  * inject the proper headers for the fabric we are communicating over.
2688  *
2689  * Return: true if we are running as a RoCE port and must force the
2690  * addition of a Global Route Header built from our Ethernet Address
2691  * Handle into our header list for connectionless packets.
2692  */
2693 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2694 {
2695 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2696 }
2697 
2698 /**
2699  * rdma_cap_opa_ah - Check if the port of device supports
2700  * OPA Address handles
2701  * @device: Device to check
2702  * @port_num: Port number to check
2703  *
2704  * Return: true if we are running on an OPA device which supports
2705  * the extended OPA addressing.
2706  */
2707 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
2708 {
2709 	return (device->port_immutable[port_num].core_cap_flags &
2710 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
2711 }
2712 
2713 /**
2714  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2715  *
2716  * @device: Device
2717  * @port_num: Port number
2718  *
2719  * This MAD size includes the MAD headers and MAD payload.  No other headers
2720  * are included.
2721  *
2722  * Return the max MAD size required by the Port.  Will return 0 if the port
2723  * does not support MADs
2724  */
2725 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2726 {
2727 	return device->port_immutable[port_num].max_mad_size;
2728 }
2729 
2730 /**
2731  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2732  * @device: Device to check
2733  * @port_num: Port number to check
2734  *
2735  * RoCE GID table mechanism manages the various GIDs for a device.
2736  *
2737  * NOTE: if allocating the port's GID table has failed, this call will still
2738  * return true, but any RoCE GID table API will fail.
2739  *
2740  * Return: true if the port uses RoCE GID table mechanism in order to manage
2741  * its GIDs.
2742  */
2743 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2744 					   u8 port_num)
2745 {
2746 	return rdma_protocol_roce(device, port_num) &&
2747 		device->add_gid && device->del_gid;
2748 }
2749 
2750 /*
2751  * Check if the device supports READ W/ INVALIDATE.
2752  */
2753 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2754 {
2755 	/*
2756 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2757 	 * has support for it yet.
2758 	 */
2759 	return rdma_protocol_iwarp(dev, port_num);
2760 }
2761 
2762 int ib_query_gid(struct ib_device *device,
2763 		 u8 port_num, int index, union ib_gid *gid,
2764 		 struct ib_gid_attr *attr);
2765 
2766 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2767 			 int state);
2768 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2769 		     struct ifla_vf_info *info);
2770 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2771 		    struct ifla_vf_stats *stats);
2772 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2773 		   int type);
2774 
2775 int ib_query_pkey(struct ib_device *device,
2776 		  u8 port_num, u16 index, u16 *pkey);
2777 
2778 int ib_modify_device(struct ib_device *device,
2779 		     int device_modify_mask,
2780 		     struct ib_device_modify *device_modify);
2781 
2782 int ib_modify_port(struct ib_device *device,
2783 		   u8 port_num, int port_modify_mask,
2784 		   struct ib_port_modify *port_modify);
2785 
2786 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2787 		enum ib_gid_type gid_type, struct net_device *ndev,
2788 		u8 *port_num, u16 *index);
2789 
2790 int ib_find_pkey(struct ib_device *device,
2791 		 u8 port_num, u16 pkey, u16 *index);
2792 
2793 enum ib_pd_flags {
2794 	/*
2795 	 * Create a memory registration for all memory in the system and place
2796 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2797 	 * ULPs to avoid the overhead of dynamic MRs.
2798 	 *
2799 	 * This flag is generally considered unsafe and must only be used in
2800 	 * extremly trusted environments.  Every use of it will log a warning
2801 	 * in the kernel log.
2802 	 */
2803 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2804 };
2805 
2806 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2807 		const char *caller);
2808 #define ib_alloc_pd(device, flags) \
2809 	__ib_alloc_pd((device), (flags), __func__)
2810 void ib_dealloc_pd(struct ib_pd *pd);
2811 
2812 /**
2813  * rdma_create_ah - Creates an address handle for the given address vector.
2814  * @pd: The protection domain associated with the address handle.
2815  * @ah_attr: The attributes of the address vector.
2816  *
2817  * The address handle is used to reference a local or global destination
2818  * in all UD QP post sends.
2819  */
2820 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
2821 
2822 /**
2823  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2824  *   work completion.
2825  * @hdr: the L3 header to parse
2826  * @net_type: type of header to parse
2827  * @sgid: place to store source gid
2828  * @dgid: place to store destination gid
2829  */
2830 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2831 			      enum rdma_network_type net_type,
2832 			      union ib_gid *sgid, union ib_gid *dgid);
2833 
2834 /**
2835  * ib_get_rdma_header_version - Get the header version
2836  * @hdr: the L3 header to parse
2837  */
2838 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2839 
2840 /**
2841  * ib_init_ah_from_wc - Initializes address handle attributes from a
2842  *   work completion.
2843  * @device: Device on which the received message arrived.
2844  * @port_num: Port on which the received message arrived.
2845  * @wc: Work completion associated with the received message.
2846  * @grh: References the received global route header.  This parameter is
2847  *   ignored unless the work completion indicates that the GRH is valid.
2848  * @ah_attr: Returned attributes that can be used when creating an address
2849  *   handle for replying to the message.
2850  */
2851 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2852 		       const struct ib_wc *wc, const struct ib_grh *grh,
2853 		       struct rdma_ah_attr *ah_attr);
2854 
2855 /**
2856  * ib_create_ah_from_wc - Creates an address handle associated with the
2857  *   sender of the specified work completion.
2858  * @pd: The protection domain associated with the address handle.
2859  * @wc: Work completion information associated with a received message.
2860  * @grh: References the received global route header.  This parameter is
2861  *   ignored unless the work completion indicates that the GRH is valid.
2862  * @port_num: The outbound port number to associate with the address.
2863  *
2864  * The address handle is used to reference a local or global destination
2865  * in all UD QP post sends.
2866  */
2867 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2868 				   const struct ib_grh *grh, u8 port_num);
2869 
2870 /**
2871  * rdma_modify_ah - Modifies the address vector associated with an address
2872  *   handle.
2873  * @ah: The address handle to modify.
2874  * @ah_attr: The new address vector attributes to associate with the
2875  *   address handle.
2876  */
2877 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2878 
2879 /**
2880  * rdma_query_ah - Queries the address vector associated with an address
2881  *   handle.
2882  * @ah: The address handle to query.
2883  * @ah_attr: The address vector attributes associated with the address
2884  *   handle.
2885  */
2886 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2887 
2888 /**
2889  * rdma_destroy_ah - Destroys an address handle.
2890  * @ah: The address handle to destroy.
2891  */
2892 int rdma_destroy_ah(struct ib_ah *ah);
2893 
2894 /**
2895  * ib_create_srq - Creates a SRQ associated with the specified protection
2896  *   domain.
2897  * @pd: The protection domain associated with the SRQ.
2898  * @srq_init_attr: A list of initial attributes required to create the
2899  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2900  *   the actual capabilities of the created SRQ.
2901  *
2902  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2903  * requested size of the SRQ, and set to the actual values allocated
2904  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2905  * will always be at least as large as the requested values.
2906  */
2907 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2908 			     struct ib_srq_init_attr *srq_init_attr);
2909 
2910 /**
2911  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2912  * @srq: The SRQ to modify.
2913  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2914  *   the current values of selected SRQ attributes are returned.
2915  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2916  *   are being modified.
2917  *
2918  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2919  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2920  * the number of receives queued drops below the limit.
2921  */
2922 int ib_modify_srq(struct ib_srq *srq,
2923 		  struct ib_srq_attr *srq_attr,
2924 		  enum ib_srq_attr_mask srq_attr_mask);
2925 
2926 /**
2927  * ib_query_srq - Returns the attribute list and current values for the
2928  *   specified SRQ.
2929  * @srq: The SRQ to query.
2930  * @srq_attr: The attributes of the specified SRQ.
2931  */
2932 int ib_query_srq(struct ib_srq *srq,
2933 		 struct ib_srq_attr *srq_attr);
2934 
2935 /**
2936  * ib_destroy_srq - Destroys the specified SRQ.
2937  * @srq: The SRQ to destroy.
2938  */
2939 int ib_destroy_srq(struct ib_srq *srq);
2940 
2941 /**
2942  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2943  * @srq: The SRQ to post the work request on.
2944  * @recv_wr: A list of work requests to post on the receive queue.
2945  * @bad_recv_wr: On an immediate failure, this parameter will reference
2946  *   the work request that failed to be posted on the QP.
2947  */
2948 static inline int ib_post_srq_recv(struct ib_srq *srq,
2949 				   struct ib_recv_wr *recv_wr,
2950 				   struct ib_recv_wr **bad_recv_wr)
2951 {
2952 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2953 }
2954 
2955 /**
2956  * ib_create_qp - Creates a QP associated with the specified protection
2957  *   domain.
2958  * @pd: The protection domain associated with the QP.
2959  * @qp_init_attr: A list of initial attributes required to create the
2960  *   QP.  If QP creation succeeds, then the attributes are updated to
2961  *   the actual capabilities of the created QP.
2962  */
2963 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2964 			   struct ib_qp_init_attr *qp_init_attr);
2965 
2966 /**
2967  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
2968  * @qp: The QP to modify.
2969  * @attr: On input, specifies the QP attributes to modify.  On output,
2970  *   the current values of selected QP attributes are returned.
2971  * @attr_mask: A bit-mask used to specify which attributes of the QP
2972  *   are being modified.
2973  * @udata: pointer to user's input output buffer information
2974  *   are being modified.
2975  * It returns 0 on success and returns appropriate error code on error.
2976  */
2977 int ib_modify_qp_with_udata(struct ib_qp *qp,
2978 			    struct ib_qp_attr *attr,
2979 			    int attr_mask,
2980 			    struct ib_udata *udata);
2981 
2982 /**
2983  * ib_modify_qp - Modifies the attributes for the specified QP and then
2984  *   transitions the QP to the given state.
2985  * @qp: The QP to modify.
2986  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2987  *   the current values of selected QP attributes are returned.
2988  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2989  *   are being modified.
2990  */
2991 int ib_modify_qp(struct ib_qp *qp,
2992 		 struct ib_qp_attr *qp_attr,
2993 		 int qp_attr_mask);
2994 
2995 /**
2996  * ib_query_qp - Returns the attribute list and current values for the
2997  *   specified QP.
2998  * @qp: The QP to query.
2999  * @qp_attr: The attributes of the specified QP.
3000  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3001  * @qp_init_attr: Additional attributes of the selected QP.
3002  *
3003  * The qp_attr_mask may be used to limit the query to gathering only the
3004  * selected attributes.
3005  */
3006 int ib_query_qp(struct ib_qp *qp,
3007 		struct ib_qp_attr *qp_attr,
3008 		int qp_attr_mask,
3009 		struct ib_qp_init_attr *qp_init_attr);
3010 
3011 /**
3012  * ib_destroy_qp - Destroys the specified QP.
3013  * @qp: The QP to destroy.
3014  */
3015 int ib_destroy_qp(struct ib_qp *qp);
3016 
3017 /**
3018  * ib_open_qp - Obtain a reference to an existing sharable QP.
3019  * @xrcd - XRC domain
3020  * @qp_open_attr: Attributes identifying the QP to open.
3021  *
3022  * Returns a reference to a sharable QP.
3023  */
3024 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3025 			 struct ib_qp_open_attr *qp_open_attr);
3026 
3027 /**
3028  * ib_close_qp - Release an external reference to a QP.
3029  * @qp: The QP handle to release
3030  *
3031  * The opened QP handle is released by the caller.  The underlying
3032  * shared QP is not destroyed until all internal references are released.
3033  */
3034 int ib_close_qp(struct ib_qp *qp);
3035 
3036 /**
3037  * ib_post_send - Posts a list of work requests to the send queue of
3038  *   the specified QP.
3039  * @qp: The QP to post the work request on.
3040  * @send_wr: A list of work requests to post on the send queue.
3041  * @bad_send_wr: On an immediate failure, this parameter will reference
3042  *   the work request that failed to be posted on the QP.
3043  *
3044  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3045  * error is returned, the QP state shall not be affected,
3046  * ib_post_send() will return an immediate error after queueing any
3047  * earlier work requests in the list.
3048  */
3049 static inline int ib_post_send(struct ib_qp *qp,
3050 			       struct ib_send_wr *send_wr,
3051 			       struct ib_send_wr **bad_send_wr)
3052 {
3053 	return qp->device->post_send(qp, send_wr, bad_send_wr);
3054 }
3055 
3056 /**
3057  * ib_post_recv - Posts a list of work requests to the receive queue of
3058  *   the specified QP.
3059  * @qp: The QP to post the work request on.
3060  * @recv_wr: A list of work requests to post on the receive queue.
3061  * @bad_recv_wr: On an immediate failure, this parameter will reference
3062  *   the work request that failed to be posted on the QP.
3063  */
3064 static inline int ib_post_recv(struct ib_qp *qp,
3065 			       struct ib_recv_wr *recv_wr,
3066 			       struct ib_recv_wr **bad_recv_wr)
3067 {
3068 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
3069 }
3070 
3071 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3072 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
3073 void ib_free_cq(struct ib_cq *cq);
3074 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3075 
3076 /**
3077  * ib_create_cq - Creates a CQ on the specified device.
3078  * @device: The device on which to create the CQ.
3079  * @comp_handler: A user-specified callback that is invoked when a
3080  *   completion event occurs on the CQ.
3081  * @event_handler: A user-specified callback that is invoked when an
3082  *   asynchronous event not associated with a completion occurs on the CQ.
3083  * @cq_context: Context associated with the CQ returned to the user via
3084  *   the associated completion and event handlers.
3085  * @cq_attr: The attributes the CQ should be created upon.
3086  *
3087  * Users can examine the cq structure to determine the actual CQ size.
3088  */
3089 struct ib_cq *ib_create_cq(struct ib_device *device,
3090 			   ib_comp_handler comp_handler,
3091 			   void (*event_handler)(struct ib_event *, void *),
3092 			   void *cq_context,
3093 			   const struct ib_cq_init_attr *cq_attr);
3094 
3095 /**
3096  * ib_resize_cq - Modifies the capacity of the CQ.
3097  * @cq: The CQ to resize.
3098  * @cqe: The minimum size of the CQ.
3099  *
3100  * Users can examine the cq structure to determine the actual CQ size.
3101  */
3102 int ib_resize_cq(struct ib_cq *cq, int cqe);
3103 
3104 /**
3105  * ib_modify_cq - Modifies moderation params of the CQ
3106  * @cq: The CQ to modify.
3107  * @cq_count: number of CQEs that will trigger an event
3108  * @cq_period: max period of time in usec before triggering an event
3109  *
3110  */
3111 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3112 
3113 /**
3114  * ib_destroy_cq - Destroys the specified CQ.
3115  * @cq: The CQ to destroy.
3116  */
3117 int ib_destroy_cq(struct ib_cq *cq);
3118 
3119 /**
3120  * ib_poll_cq - poll a CQ for completion(s)
3121  * @cq:the CQ being polled
3122  * @num_entries:maximum number of completions to return
3123  * @wc:array of at least @num_entries &struct ib_wc where completions
3124  *   will be returned
3125  *
3126  * Poll a CQ for (possibly multiple) completions.  If the return value
3127  * is < 0, an error occurred.  If the return value is >= 0, it is the
3128  * number of completions returned.  If the return value is
3129  * non-negative and < num_entries, then the CQ was emptied.
3130  */
3131 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3132 			     struct ib_wc *wc)
3133 {
3134 	return cq->device->poll_cq(cq, num_entries, wc);
3135 }
3136 
3137 /**
3138  * ib_peek_cq - Returns the number of unreaped completions currently
3139  *   on the specified CQ.
3140  * @cq: The CQ to peek.
3141  * @wc_cnt: A minimum number of unreaped completions to check for.
3142  *
3143  * If the number of unreaped completions is greater than or equal to wc_cnt,
3144  * this function returns wc_cnt, otherwise, it returns the actual number of
3145  * unreaped completions.
3146  */
3147 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
3148 
3149 /**
3150  * ib_req_notify_cq - Request completion notification on a CQ.
3151  * @cq: The CQ to generate an event for.
3152  * @flags:
3153  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3154  *   to request an event on the next solicited event or next work
3155  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3156  *   may also be |ed in to request a hint about missed events, as
3157  *   described below.
3158  *
3159  * Return Value:
3160  *    < 0 means an error occurred while requesting notification
3161  *   == 0 means notification was requested successfully, and if
3162  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3163  *        were missed and it is safe to wait for another event.  In
3164  *        this case is it guaranteed that any work completions added
3165  *        to the CQ since the last CQ poll will trigger a completion
3166  *        notification event.
3167  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3168  *        in.  It means that the consumer must poll the CQ again to
3169  *        make sure it is empty to avoid missing an event because of a
3170  *        race between requesting notification and an entry being
3171  *        added to the CQ.  This return value means it is possible
3172  *        (but not guaranteed) that a work completion has been added
3173  *        to the CQ since the last poll without triggering a
3174  *        completion notification event.
3175  */
3176 static inline int ib_req_notify_cq(struct ib_cq *cq,
3177 				   enum ib_cq_notify_flags flags)
3178 {
3179 	return cq->device->req_notify_cq(cq, flags);
3180 }
3181 
3182 /**
3183  * ib_req_ncomp_notif - Request completion notification when there are
3184  *   at least the specified number of unreaped completions on the CQ.
3185  * @cq: The CQ to generate an event for.
3186  * @wc_cnt: The number of unreaped completions that should be on the
3187  *   CQ before an event is generated.
3188  */
3189 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3190 {
3191 	return cq->device->req_ncomp_notif ?
3192 		cq->device->req_ncomp_notif(cq, wc_cnt) :
3193 		-ENOSYS;
3194 }
3195 
3196 /**
3197  * ib_dma_mapping_error - check a DMA addr for error
3198  * @dev: The device for which the dma_addr was created
3199  * @dma_addr: The DMA address to check
3200  */
3201 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3202 {
3203 	return dma_mapping_error(dev->dma_device, dma_addr);
3204 }
3205 
3206 /**
3207  * ib_dma_map_single - Map a kernel virtual address to DMA address
3208  * @dev: The device for which the dma_addr is to be created
3209  * @cpu_addr: The kernel virtual address
3210  * @size: The size of the region in bytes
3211  * @direction: The direction of the DMA
3212  */
3213 static inline u64 ib_dma_map_single(struct ib_device *dev,
3214 				    void *cpu_addr, size_t size,
3215 				    enum dma_data_direction direction)
3216 {
3217 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3218 }
3219 
3220 /**
3221  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3222  * @dev: The device for which the DMA address was created
3223  * @addr: The DMA address
3224  * @size: The size of the region in bytes
3225  * @direction: The direction of the DMA
3226  */
3227 static inline void ib_dma_unmap_single(struct ib_device *dev,
3228 				       u64 addr, size_t size,
3229 				       enum dma_data_direction direction)
3230 {
3231 	dma_unmap_single(dev->dma_device, addr, size, direction);
3232 }
3233 
3234 /**
3235  * ib_dma_map_page - Map a physical page to DMA address
3236  * @dev: The device for which the dma_addr is to be created
3237  * @page: The page to be mapped
3238  * @offset: The offset within the page
3239  * @size: The size of the region in bytes
3240  * @direction: The direction of the DMA
3241  */
3242 static inline u64 ib_dma_map_page(struct ib_device *dev,
3243 				  struct page *page,
3244 				  unsigned long offset,
3245 				  size_t size,
3246 					 enum dma_data_direction direction)
3247 {
3248 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3249 }
3250 
3251 /**
3252  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3253  * @dev: The device for which the DMA address was created
3254  * @addr: The DMA address
3255  * @size: The size of the region in bytes
3256  * @direction: The direction of the DMA
3257  */
3258 static inline void ib_dma_unmap_page(struct ib_device *dev,
3259 				     u64 addr, size_t size,
3260 				     enum dma_data_direction direction)
3261 {
3262 	dma_unmap_page(dev->dma_device, addr, size, direction);
3263 }
3264 
3265 /**
3266  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3267  * @dev: The device for which the DMA addresses are to be created
3268  * @sg: The array of scatter/gather entries
3269  * @nents: The number of scatter/gather entries
3270  * @direction: The direction of the DMA
3271  */
3272 static inline int ib_dma_map_sg(struct ib_device *dev,
3273 				struct scatterlist *sg, int nents,
3274 				enum dma_data_direction direction)
3275 {
3276 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3277 }
3278 
3279 /**
3280  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3281  * @dev: The device for which the DMA addresses were created
3282  * @sg: The array of scatter/gather entries
3283  * @nents: The number of scatter/gather entries
3284  * @direction: The direction of the DMA
3285  */
3286 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3287 				   struct scatterlist *sg, int nents,
3288 				   enum dma_data_direction direction)
3289 {
3290 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
3291 }
3292 
3293 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3294 				      struct scatterlist *sg, int nents,
3295 				      enum dma_data_direction direction,
3296 				      unsigned long dma_attrs)
3297 {
3298 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3299 				dma_attrs);
3300 }
3301 
3302 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3303 					 struct scatterlist *sg, int nents,
3304 					 enum dma_data_direction direction,
3305 					 unsigned long dma_attrs)
3306 {
3307 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3308 }
3309 /**
3310  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3311  * @dev: The device for which the DMA addresses were created
3312  * @sg: The scatter/gather entry
3313  *
3314  * Note: this function is obsolete. To do: change all occurrences of
3315  * ib_sg_dma_address() into sg_dma_address().
3316  */
3317 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3318 				    struct scatterlist *sg)
3319 {
3320 	return sg_dma_address(sg);
3321 }
3322 
3323 /**
3324  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3325  * @dev: The device for which the DMA addresses were created
3326  * @sg: The scatter/gather entry
3327  *
3328  * Note: this function is obsolete. To do: change all occurrences of
3329  * ib_sg_dma_len() into sg_dma_len().
3330  */
3331 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3332 					 struct scatterlist *sg)
3333 {
3334 	return sg_dma_len(sg);
3335 }
3336 
3337 /**
3338  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3339  * @dev: The device for which the DMA address was created
3340  * @addr: The DMA address
3341  * @size: The size of the region in bytes
3342  * @dir: The direction of the DMA
3343  */
3344 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3345 					      u64 addr,
3346 					      size_t size,
3347 					      enum dma_data_direction dir)
3348 {
3349 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3350 }
3351 
3352 /**
3353  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3354  * @dev: The device for which the DMA address was created
3355  * @addr: The DMA address
3356  * @size: The size of the region in bytes
3357  * @dir: The direction of the DMA
3358  */
3359 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3360 						 u64 addr,
3361 						 size_t size,
3362 						 enum dma_data_direction dir)
3363 {
3364 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3365 }
3366 
3367 /**
3368  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3369  * @dev: The device for which the DMA address is requested
3370  * @size: The size of the region to allocate in bytes
3371  * @dma_handle: A pointer for returning the DMA address of the region
3372  * @flag: memory allocator flags
3373  */
3374 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3375 					   size_t size,
3376 					   dma_addr_t *dma_handle,
3377 					   gfp_t flag)
3378 {
3379 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3380 }
3381 
3382 /**
3383  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3384  * @dev: The device for which the DMA addresses were allocated
3385  * @size: The size of the region
3386  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3387  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3388  */
3389 static inline void ib_dma_free_coherent(struct ib_device *dev,
3390 					size_t size, void *cpu_addr,
3391 					dma_addr_t dma_handle)
3392 {
3393 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3394 }
3395 
3396 /**
3397  * ib_dereg_mr - Deregisters a memory region and removes it from the
3398  *   HCA translation table.
3399  * @mr: The memory region to deregister.
3400  *
3401  * This function can fail, if the memory region has memory windows bound to it.
3402  */
3403 int ib_dereg_mr(struct ib_mr *mr);
3404 
3405 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3406 			  enum ib_mr_type mr_type,
3407 			  u32 max_num_sg);
3408 
3409 /**
3410  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3411  *   R_Key and L_Key.
3412  * @mr - struct ib_mr pointer to be updated.
3413  * @newkey - new key to be used.
3414  */
3415 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3416 {
3417 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3418 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3419 }
3420 
3421 /**
3422  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3423  * for calculating a new rkey for type 2 memory windows.
3424  * @rkey - the rkey to increment.
3425  */
3426 static inline u32 ib_inc_rkey(u32 rkey)
3427 {
3428 	const u32 mask = 0x000000ff;
3429 	return ((rkey + 1) & mask) | (rkey & ~mask);
3430 }
3431 
3432 /**
3433  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3434  * @pd: The protection domain associated with the unmapped region.
3435  * @mr_access_flags: Specifies the memory access rights.
3436  * @fmr_attr: Attributes of the unmapped region.
3437  *
3438  * A fast memory region must be mapped before it can be used as part of
3439  * a work request.
3440  */
3441 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3442 			    int mr_access_flags,
3443 			    struct ib_fmr_attr *fmr_attr);
3444 
3445 /**
3446  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3447  * @fmr: The fast memory region to associate with the pages.
3448  * @page_list: An array of physical pages to map to the fast memory region.
3449  * @list_len: The number of pages in page_list.
3450  * @iova: The I/O virtual address to use with the mapped region.
3451  */
3452 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3453 				  u64 *page_list, int list_len,
3454 				  u64 iova)
3455 {
3456 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3457 }
3458 
3459 /**
3460  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3461  * @fmr_list: A linked list of fast memory regions to unmap.
3462  */
3463 int ib_unmap_fmr(struct list_head *fmr_list);
3464 
3465 /**
3466  * ib_dealloc_fmr - Deallocates a fast memory region.
3467  * @fmr: The fast memory region to deallocate.
3468  */
3469 int ib_dealloc_fmr(struct ib_fmr *fmr);
3470 
3471 /**
3472  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3473  * @qp: QP to attach to the multicast group.  The QP must be type
3474  *   IB_QPT_UD.
3475  * @gid: Multicast group GID.
3476  * @lid: Multicast group LID in host byte order.
3477  *
3478  * In order to send and receive multicast packets, subnet
3479  * administration must have created the multicast group and configured
3480  * the fabric appropriately.  The port associated with the specified
3481  * QP must also be a member of the multicast group.
3482  */
3483 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3484 
3485 /**
3486  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3487  * @qp: QP to detach from the multicast group.
3488  * @gid: Multicast group GID.
3489  * @lid: Multicast group LID in host byte order.
3490  */
3491 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3492 
3493 /**
3494  * ib_alloc_xrcd - Allocates an XRC domain.
3495  * @device: The device on which to allocate the XRC domain.
3496  */
3497 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3498 
3499 /**
3500  * ib_dealloc_xrcd - Deallocates an XRC domain.
3501  * @xrcd: The XRC domain to deallocate.
3502  */
3503 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3504 
3505 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3506 			       struct ib_flow_attr *flow_attr, int domain);
3507 int ib_destroy_flow(struct ib_flow *flow_id);
3508 
3509 static inline int ib_check_mr_access(int flags)
3510 {
3511 	/*
3512 	 * Local write permission is required if remote write or
3513 	 * remote atomic permission is also requested.
3514 	 */
3515 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3516 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3517 		return -EINVAL;
3518 
3519 	return 0;
3520 }
3521 
3522 /**
3523  * ib_check_mr_status: lightweight check of MR status.
3524  *     This routine may provide status checks on a selected
3525  *     ib_mr. first use is for signature status check.
3526  *
3527  * @mr: A memory region.
3528  * @check_mask: Bitmask of which checks to perform from
3529  *     ib_mr_status_check enumeration.
3530  * @mr_status: The container of relevant status checks.
3531  *     failed checks will be indicated in the status bitmask
3532  *     and the relevant info shall be in the error item.
3533  */
3534 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3535 		       struct ib_mr_status *mr_status);
3536 
3537 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3538 					    u16 pkey, const union ib_gid *gid,
3539 					    const struct sockaddr *addr);
3540 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3541 			   struct ib_wq_init_attr *init_attr);
3542 int ib_destroy_wq(struct ib_wq *wq);
3543 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3544 		 u32 wq_attr_mask);
3545 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3546 						 struct ib_rwq_ind_table_init_attr*
3547 						 wq_ind_table_init_attr);
3548 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3549 
3550 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3551 		 unsigned int *sg_offset, unsigned int page_size);
3552 
3553 static inline int
3554 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3555 		  unsigned int *sg_offset, unsigned int page_size)
3556 {
3557 	int n;
3558 
3559 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3560 	mr->iova = 0;
3561 
3562 	return n;
3563 }
3564 
3565 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3566 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3567 
3568 void ib_drain_rq(struct ib_qp *qp);
3569 void ib_drain_sq(struct ib_qp *qp);
3570 void ib_drain_qp(struct ib_qp *qp);
3571 
3572 int ib_resolve_eth_dmac(struct ib_device *device,
3573 			struct rdma_ah_attr *ah_attr);
3574 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3575 
3576 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3577 {
3578 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3579 		return attr->roce.dmac;
3580 	return NULL;
3581 }
3582 
3583 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3584 {
3585 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3586 		attr->ib.dlid = (u16)dlid;
3587 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3588 		attr->opa.dlid = dlid;
3589 }
3590 
3591 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3592 {
3593 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3594 		return attr->ib.dlid;
3595 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3596 		return attr->opa.dlid;
3597 	return 0;
3598 }
3599 
3600 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3601 {
3602 	attr->sl = sl;
3603 }
3604 
3605 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3606 {
3607 	return attr->sl;
3608 }
3609 
3610 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3611 					 u8 src_path_bits)
3612 {
3613 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3614 		attr->ib.src_path_bits = src_path_bits;
3615 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3616 		attr->opa.src_path_bits = src_path_bits;
3617 }
3618 
3619 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3620 {
3621 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3622 		return attr->ib.src_path_bits;
3623 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3624 		return attr->opa.src_path_bits;
3625 	return 0;
3626 }
3627 
3628 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3629 {
3630 	attr->port_num = port_num;
3631 }
3632 
3633 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3634 {
3635 	return attr->port_num;
3636 }
3637 
3638 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3639 					   u8 static_rate)
3640 {
3641 	attr->static_rate = static_rate;
3642 }
3643 
3644 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3645 {
3646 	return attr->static_rate;
3647 }
3648 
3649 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3650 					enum ib_ah_flags flag)
3651 {
3652 	attr->ah_flags = flag;
3653 }
3654 
3655 static inline enum ib_ah_flags
3656 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3657 {
3658 	return attr->ah_flags;
3659 }
3660 
3661 static inline const struct ib_global_route
3662 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
3663 {
3664 	return &attr->grh;
3665 }
3666 
3667 /*To retrieve and modify the grh */
3668 static inline struct ib_global_route
3669 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
3670 {
3671 	return &attr->grh;
3672 }
3673 
3674 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
3675 {
3676 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3677 
3678 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
3679 }
3680 
3681 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
3682 					     __be64 prefix)
3683 {
3684 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3685 
3686 	grh->dgid.global.subnet_prefix = prefix;
3687 }
3688 
3689 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
3690 					    __be64 if_id)
3691 {
3692 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3693 
3694 	grh->dgid.global.interface_id = if_id;
3695 }
3696 
3697 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
3698 				   union ib_gid *dgid, u32 flow_label,
3699 				   u8 sgid_index, u8 hop_limit,
3700 				   u8 traffic_class)
3701 {
3702 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3703 
3704 	attr->ah_flags = IB_AH_GRH;
3705 	if (dgid)
3706 		grh->dgid = *dgid;
3707 	grh->flow_label = flow_label;
3708 	grh->sgid_index = sgid_index;
3709 	grh->hop_limit = hop_limit;
3710 	grh->traffic_class = traffic_class;
3711 }
3712 
3713 /*Get AH type */
3714 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
3715 						       u32 port_num)
3716 {
3717 	if ((rdma_protocol_roce(dev, port_num)) ||
3718 	    (rdma_protocol_iwarp(dev, port_num)))
3719 		return RDMA_AH_ATTR_TYPE_ROCE;
3720 	else if ((rdma_protocol_ib(dev, port_num)) &&
3721 		 (rdma_cap_opa_ah(dev, port_num)))
3722 		return RDMA_AH_ATTR_TYPE_OPA;
3723 	else
3724 		return RDMA_AH_ATTR_TYPE_IB;
3725 }
3726 
3727 /**
3728  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
3729  *     In the current implementation the only way to get
3730  *     get the 32bit lid is from other sources for OPA.
3731  *     For IB, lids will always be 16bits so cast the
3732  *     value accordingly.
3733  *
3734  * @lid: A 32bit LID
3735  */
3736 static inline u16 ib_lid_cpu16(u32 lid)
3737 {
3738 	WARN_ON_ONCE(lid & 0xFFFF0000);
3739 	return (u16)lid;
3740 }
3741 
3742 /**
3743  * ib_lid_be16 - Return lid in 16bit BE encoding.
3744  *
3745  * @lid: A 32bit LID
3746  */
3747 static inline __be16 ib_lid_be16(u32 lid)
3748 {
3749 	WARN_ON_ONCE(lid & 0xFFFF0000);
3750 	return cpu_to_be16((u16)lid);
3751 }
3752 
3753 /**
3754  * ib_get_vector_affinity - Get the affinity mappings of a given completion
3755  *   vector
3756  * @device:         the rdma device
3757  * @comp_vector:    index of completion vector
3758  *
3759  * Returns NULL on failure, otherwise a corresponding cpu map of the
3760  * completion vector (returns all-cpus map if the device driver doesn't
3761  * implement get_vector_affinity).
3762  */
3763 static inline const struct cpumask *
3764 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
3765 {
3766 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
3767 	    !device->get_vector_affinity)
3768 		return NULL;
3769 
3770 	return device->get_vector_affinity(device, comp_vector);
3771 
3772 }
3773 
3774 #endif /* IB_VERBS_H */
3775