xref: /openbmc/linux/include/rdma/ib_verbs.h (revision aac5987a)
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 
59 #include <linux/if_link.h>
60 #include <linux/atomic.h>
61 #include <linux/mmu_notifier.h>
62 #include <linux/uaccess.h>
63 #include <linux/cgroup_rdma.h>
64 
65 extern struct workqueue_struct *ib_wq;
66 extern struct workqueue_struct *ib_comp_wq;
67 
68 union ib_gid {
69 	u8	raw[16];
70 	struct {
71 		__be64	subnet_prefix;
72 		__be64	interface_id;
73 	} global;
74 };
75 
76 extern union ib_gid zgid;
77 
78 enum ib_gid_type {
79 	/* If link layer is Ethernet, this is RoCE V1 */
80 	IB_GID_TYPE_IB        = 0,
81 	IB_GID_TYPE_ROCE      = 0,
82 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
83 	IB_GID_TYPE_SIZE
84 };
85 
86 #define ROCE_V2_UDP_DPORT      4791
87 struct ib_gid_attr {
88 	enum ib_gid_type	gid_type;
89 	struct net_device	*ndev;
90 };
91 
92 enum rdma_node_type {
93 	/* IB values map to NodeInfo:NodeType. */
94 	RDMA_NODE_IB_CA 	= 1,
95 	RDMA_NODE_IB_SWITCH,
96 	RDMA_NODE_IB_ROUTER,
97 	RDMA_NODE_RNIC,
98 	RDMA_NODE_USNIC,
99 	RDMA_NODE_USNIC_UDP,
100 };
101 
102 enum {
103 	/* set the local administered indication */
104 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
105 };
106 
107 enum rdma_transport_type {
108 	RDMA_TRANSPORT_IB,
109 	RDMA_TRANSPORT_IWARP,
110 	RDMA_TRANSPORT_USNIC,
111 	RDMA_TRANSPORT_USNIC_UDP
112 };
113 
114 enum rdma_protocol_type {
115 	RDMA_PROTOCOL_IB,
116 	RDMA_PROTOCOL_IBOE,
117 	RDMA_PROTOCOL_IWARP,
118 	RDMA_PROTOCOL_USNIC_UDP
119 };
120 
121 __attribute_const__ enum rdma_transport_type
122 rdma_node_get_transport(enum rdma_node_type node_type);
123 
124 enum rdma_network_type {
125 	RDMA_NETWORK_IB,
126 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
127 	RDMA_NETWORK_IPV4,
128 	RDMA_NETWORK_IPV6
129 };
130 
131 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
132 {
133 	if (network_type == RDMA_NETWORK_IPV4 ||
134 	    network_type == RDMA_NETWORK_IPV6)
135 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
136 
137 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
138 	return IB_GID_TYPE_IB;
139 }
140 
141 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
142 							    union ib_gid *gid)
143 {
144 	if (gid_type == IB_GID_TYPE_IB)
145 		return RDMA_NETWORK_IB;
146 
147 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
148 		return RDMA_NETWORK_IPV4;
149 	else
150 		return RDMA_NETWORK_IPV6;
151 }
152 
153 enum rdma_link_layer {
154 	IB_LINK_LAYER_UNSPECIFIED,
155 	IB_LINK_LAYER_INFINIBAND,
156 	IB_LINK_LAYER_ETHERNET,
157 };
158 
159 enum ib_device_cap_flags {
160 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
161 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
162 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
163 	IB_DEVICE_RAW_MULTI			= (1 << 3),
164 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
165 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
166 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
167 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
168 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
169 	IB_DEVICE_INIT_TYPE			= (1 << 9),
170 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
171 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
172 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
173 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
174 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
175 
176 	/*
177 	 * This device supports a per-device lkey or stag that can be
178 	 * used without performing a memory registration for the local
179 	 * memory.  Note that ULPs should never check this flag, but
180 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
181 	 * which will always contain a usable lkey.
182 	 */
183 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
184 	IB_DEVICE_RESERVED /* old SEND_W_INV */	= (1 << 16),
185 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
186 	/*
187 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
188 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
189 	 * messages and can verify the validity of checksum for
190 	 * incoming messages.  Setting this flag implies that the
191 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
192 	 */
193 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
194 	IB_DEVICE_UD_TSO			= (1 << 19),
195 	IB_DEVICE_XRC				= (1 << 20),
196 
197 	/*
198 	 * This device supports the IB "base memory management extension",
199 	 * which includes support for fast registrations (IB_WR_REG_MR,
200 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
201 	 * also be set by any iWarp device which must support FRs to comply
202 	 * to the iWarp verbs spec.  iWarp devices also support the
203 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
204 	 * stag.
205 	 */
206 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
207 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
208 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
209 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
210 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
211 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
212 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
213 	/*
214 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
215 	 * support execution of WQEs that involve synchronization
216 	 * of I/O operations with single completion queue managed
217 	 * by hardware.
218 	 */
219 	IB_DEVICE_CROSS_CHANNEL		= (1 << 27),
220 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
221 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
222 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
223 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
224 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
225 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
226 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
227 };
228 
229 enum ib_signature_prot_cap {
230 	IB_PROT_T10DIF_TYPE_1 = 1,
231 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
232 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
233 };
234 
235 enum ib_signature_guard_cap {
236 	IB_GUARD_T10DIF_CRC	= 1,
237 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
238 };
239 
240 enum ib_atomic_cap {
241 	IB_ATOMIC_NONE,
242 	IB_ATOMIC_HCA,
243 	IB_ATOMIC_GLOB
244 };
245 
246 enum ib_odp_general_cap_bits {
247 	IB_ODP_SUPPORT		= 1 << 0,
248 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
249 };
250 
251 enum ib_odp_transport_cap_bits {
252 	IB_ODP_SUPPORT_SEND	= 1 << 0,
253 	IB_ODP_SUPPORT_RECV	= 1 << 1,
254 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
255 	IB_ODP_SUPPORT_READ	= 1 << 3,
256 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
257 };
258 
259 struct ib_odp_caps {
260 	uint64_t general_caps;
261 	struct {
262 		uint32_t  rc_odp_caps;
263 		uint32_t  uc_odp_caps;
264 		uint32_t  ud_odp_caps;
265 	} per_transport_caps;
266 };
267 
268 struct ib_rss_caps {
269 	/* Corresponding bit will be set if qp type from
270 	 * 'enum ib_qp_type' is supported, e.g.
271 	 * supported_qpts |= 1 << IB_QPT_UD
272 	 */
273 	u32 supported_qpts;
274 	u32 max_rwq_indirection_tables;
275 	u32 max_rwq_indirection_table_size;
276 };
277 
278 enum ib_cq_creation_flags {
279 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
280 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
281 };
282 
283 struct ib_cq_init_attr {
284 	unsigned int	cqe;
285 	int		comp_vector;
286 	u32		flags;
287 };
288 
289 struct ib_device_attr {
290 	u64			fw_ver;
291 	__be64			sys_image_guid;
292 	u64			max_mr_size;
293 	u64			page_size_cap;
294 	u32			vendor_id;
295 	u32			vendor_part_id;
296 	u32			hw_ver;
297 	int			max_qp;
298 	int			max_qp_wr;
299 	u64			device_cap_flags;
300 	int			max_sge;
301 	int			max_sge_rd;
302 	int			max_cq;
303 	int			max_cqe;
304 	int			max_mr;
305 	int			max_pd;
306 	int			max_qp_rd_atom;
307 	int			max_ee_rd_atom;
308 	int			max_res_rd_atom;
309 	int			max_qp_init_rd_atom;
310 	int			max_ee_init_rd_atom;
311 	enum ib_atomic_cap	atomic_cap;
312 	enum ib_atomic_cap	masked_atomic_cap;
313 	int			max_ee;
314 	int			max_rdd;
315 	int			max_mw;
316 	int			max_raw_ipv6_qp;
317 	int			max_raw_ethy_qp;
318 	int			max_mcast_grp;
319 	int			max_mcast_qp_attach;
320 	int			max_total_mcast_qp_attach;
321 	int			max_ah;
322 	int			max_fmr;
323 	int			max_map_per_fmr;
324 	int			max_srq;
325 	int			max_srq_wr;
326 	int			max_srq_sge;
327 	unsigned int		max_fast_reg_page_list_len;
328 	u16			max_pkeys;
329 	u8			local_ca_ack_delay;
330 	int			sig_prot_cap;
331 	int			sig_guard_cap;
332 	struct ib_odp_caps	odp_caps;
333 	uint64_t		timestamp_mask;
334 	uint64_t		hca_core_clock; /* in KHZ */
335 	struct ib_rss_caps	rss_caps;
336 	u32			max_wq_type_rq;
337 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
338 };
339 
340 enum ib_mtu {
341 	IB_MTU_256  = 1,
342 	IB_MTU_512  = 2,
343 	IB_MTU_1024 = 3,
344 	IB_MTU_2048 = 4,
345 	IB_MTU_4096 = 5
346 };
347 
348 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
349 {
350 	switch (mtu) {
351 	case IB_MTU_256:  return  256;
352 	case IB_MTU_512:  return  512;
353 	case IB_MTU_1024: return 1024;
354 	case IB_MTU_2048: return 2048;
355 	case IB_MTU_4096: return 4096;
356 	default: 	  return -1;
357 	}
358 }
359 
360 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
361 {
362 	if (mtu >= 4096)
363 		return IB_MTU_4096;
364 	else if (mtu >= 2048)
365 		return IB_MTU_2048;
366 	else if (mtu >= 1024)
367 		return IB_MTU_1024;
368 	else if (mtu >= 512)
369 		return IB_MTU_512;
370 	else
371 		return IB_MTU_256;
372 }
373 
374 enum ib_port_state {
375 	IB_PORT_NOP		= 0,
376 	IB_PORT_DOWN		= 1,
377 	IB_PORT_INIT		= 2,
378 	IB_PORT_ARMED		= 3,
379 	IB_PORT_ACTIVE		= 4,
380 	IB_PORT_ACTIVE_DEFER	= 5
381 };
382 
383 enum ib_port_cap_flags {
384 	IB_PORT_SM				= 1 <<  1,
385 	IB_PORT_NOTICE_SUP			= 1 <<  2,
386 	IB_PORT_TRAP_SUP			= 1 <<  3,
387 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
388 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
389 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
390 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
391 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
392 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
393 	IB_PORT_SM_DISABLED			= 1 << 10,
394 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
395 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
396 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
397 	IB_PORT_CM_SUP				= 1 << 16,
398 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
399 	IB_PORT_REINIT_SUP			= 1 << 18,
400 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
401 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
402 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
403 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
404 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
405 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
406 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
407 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
408 };
409 
410 enum ib_port_width {
411 	IB_WIDTH_1X	= 1,
412 	IB_WIDTH_4X	= 2,
413 	IB_WIDTH_8X	= 4,
414 	IB_WIDTH_12X	= 8
415 };
416 
417 static inline int ib_width_enum_to_int(enum ib_port_width width)
418 {
419 	switch (width) {
420 	case IB_WIDTH_1X:  return  1;
421 	case IB_WIDTH_4X:  return  4;
422 	case IB_WIDTH_8X:  return  8;
423 	case IB_WIDTH_12X: return 12;
424 	default: 	  return -1;
425 	}
426 }
427 
428 enum ib_port_speed {
429 	IB_SPEED_SDR	= 1,
430 	IB_SPEED_DDR	= 2,
431 	IB_SPEED_QDR	= 4,
432 	IB_SPEED_FDR10	= 8,
433 	IB_SPEED_FDR	= 16,
434 	IB_SPEED_EDR	= 32
435 };
436 
437 /**
438  * struct rdma_hw_stats
439  * @timestamp - Used by the core code to track when the last update was
440  * @lifespan - Used by the core code to determine how old the counters
441  *   should be before being updated again.  Stored in jiffies, defaults
442  *   to 10 milliseconds, drivers can override the default be specifying
443  *   their own value during their allocation routine.
444  * @name - Array of pointers to static names used for the counters in
445  *   directory.
446  * @num_counters - How many hardware counters there are.  If name is
447  *   shorter than this number, a kernel oops will result.  Driver authors
448  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
449  *   in their code to prevent this.
450  * @value - Array of u64 counters that are accessed by the sysfs code and
451  *   filled in by the drivers get_stats routine
452  */
453 struct rdma_hw_stats {
454 	unsigned long	timestamp;
455 	unsigned long	lifespan;
456 	const char * const *names;
457 	int		num_counters;
458 	u64		value[];
459 };
460 
461 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
462 /**
463  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
464  *   for drivers.
465  * @names - Array of static const char *
466  * @num_counters - How many elements in array
467  * @lifespan - How many milliseconds between updates
468  */
469 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
470 		const char * const *names, int num_counters,
471 		unsigned long lifespan)
472 {
473 	struct rdma_hw_stats *stats;
474 
475 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
476 			GFP_KERNEL);
477 	if (!stats)
478 		return NULL;
479 	stats->names = names;
480 	stats->num_counters = num_counters;
481 	stats->lifespan = msecs_to_jiffies(lifespan);
482 
483 	return stats;
484 }
485 
486 
487 /* Define bits for the various functionality this port needs to be supported by
488  * the core.
489  */
490 /* Management                           0x00000FFF */
491 #define RDMA_CORE_CAP_IB_MAD            0x00000001
492 #define RDMA_CORE_CAP_IB_SMI            0x00000002
493 #define RDMA_CORE_CAP_IB_CM             0x00000004
494 #define RDMA_CORE_CAP_IW_CM             0x00000008
495 #define RDMA_CORE_CAP_IB_SA             0x00000010
496 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
497 
498 /* Address format                       0x000FF000 */
499 #define RDMA_CORE_CAP_AF_IB             0x00001000
500 #define RDMA_CORE_CAP_ETH_AH            0x00002000
501 
502 /* Protocol                             0xFFF00000 */
503 #define RDMA_CORE_CAP_PROT_IB           0x00100000
504 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
505 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
506 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
507 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
508 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
509 
510 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
511 					| RDMA_CORE_CAP_IB_MAD \
512 					| RDMA_CORE_CAP_IB_SMI \
513 					| RDMA_CORE_CAP_IB_CM  \
514 					| RDMA_CORE_CAP_IB_SA  \
515 					| RDMA_CORE_CAP_AF_IB)
516 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
517 					| RDMA_CORE_CAP_IB_MAD  \
518 					| RDMA_CORE_CAP_IB_CM   \
519 					| RDMA_CORE_CAP_AF_IB   \
520 					| RDMA_CORE_CAP_ETH_AH)
521 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
522 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
523 					| RDMA_CORE_CAP_IB_MAD  \
524 					| RDMA_CORE_CAP_IB_CM   \
525 					| RDMA_CORE_CAP_AF_IB   \
526 					| RDMA_CORE_CAP_ETH_AH)
527 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
528 					| RDMA_CORE_CAP_IW_CM)
529 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
530 					| RDMA_CORE_CAP_OPA_MAD)
531 
532 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
533 
534 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
535 
536 struct ib_port_attr {
537 	u64			subnet_prefix;
538 	enum ib_port_state	state;
539 	enum ib_mtu		max_mtu;
540 	enum ib_mtu		active_mtu;
541 	int			gid_tbl_len;
542 	u32			port_cap_flags;
543 	u32			max_msg_sz;
544 	u32			bad_pkey_cntr;
545 	u32			qkey_viol_cntr;
546 	u16			pkey_tbl_len;
547 	u16			lid;
548 	u16			sm_lid;
549 	u8			lmc;
550 	u8			max_vl_num;
551 	u8			sm_sl;
552 	u8			subnet_timeout;
553 	u8			init_type_reply;
554 	u8			active_width;
555 	u8			active_speed;
556 	u8                      phys_state;
557 	bool			grh_required;
558 };
559 
560 enum ib_device_modify_flags {
561 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
562 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
563 };
564 
565 #define IB_DEVICE_NODE_DESC_MAX 64
566 
567 struct ib_device_modify {
568 	u64	sys_image_guid;
569 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
570 };
571 
572 enum ib_port_modify_flags {
573 	IB_PORT_SHUTDOWN		= 1,
574 	IB_PORT_INIT_TYPE		= (1<<2),
575 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
576 };
577 
578 struct ib_port_modify {
579 	u32	set_port_cap_mask;
580 	u32	clr_port_cap_mask;
581 	u8	init_type;
582 };
583 
584 enum ib_event_type {
585 	IB_EVENT_CQ_ERR,
586 	IB_EVENT_QP_FATAL,
587 	IB_EVENT_QP_REQ_ERR,
588 	IB_EVENT_QP_ACCESS_ERR,
589 	IB_EVENT_COMM_EST,
590 	IB_EVENT_SQ_DRAINED,
591 	IB_EVENT_PATH_MIG,
592 	IB_EVENT_PATH_MIG_ERR,
593 	IB_EVENT_DEVICE_FATAL,
594 	IB_EVENT_PORT_ACTIVE,
595 	IB_EVENT_PORT_ERR,
596 	IB_EVENT_LID_CHANGE,
597 	IB_EVENT_PKEY_CHANGE,
598 	IB_EVENT_SM_CHANGE,
599 	IB_EVENT_SRQ_ERR,
600 	IB_EVENT_SRQ_LIMIT_REACHED,
601 	IB_EVENT_QP_LAST_WQE_REACHED,
602 	IB_EVENT_CLIENT_REREGISTER,
603 	IB_EVENT_GID_CHANGE,
604 	IB_EVENT_WQ_FATAL,
605 };
606 
607 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
608 
609 struct ib_event {
610 	struct ib_device	*device;
611 	union {
612 		struct ib_cq	*cq;
613 		struct ib_qp	*qp;
614 		struct ib_srq	*srq;
615 		struct ib_wq	*wq;
616 		u8		port_num;
617 	} element;
618 	enum ib_event_type	event;
619 };
620 
621 struct ib_event_handler {
622 	struct ib_device *device;
623 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
624 	struct list_head  list;
625 };
626 
627 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
628 	do {							\
629 		(_ptr)->device  = _device;			\
630 		(_ptr)->handler = _handler;			\
631 		INIT_LIST_HEAD(&(_ptr)->list);			\
632 	} while (0)
633 
634 struct ib_global_route {
635 	union ib_gid	dgid;
636 	u32		flow_label;
637 	u8		sgid_index;
638 	u8		hop_limit;
639 	u8		traffic_class;
640 };
641 
642 struct ib_grh {
643 	__be32		version_tclass_flow;
644 	__be16		paylen;
645 	u8		next_hdr;
646 	u8		hop_limit;
647 	union ib_gid	sgid;
648 	union ib_gid	dgid;
649 };
650 
651 union rdma_network_hdr {
652 	struct ib_grh ibgrh;
653 	struct {
654 		/* The IB spec states that if it's IPv4, the header
655 		 * is located in the last 20 bytes of the header.
656 		 */
657 		u8		reserved[20];
658 		struct iphdr	roce4grh;
659 	};
660 };
661 
662 enum {
663 	IB_MULTICAST_QPN = 0xffffff
664 };
665 
666 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
667 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
668 
669 enum ib_ah_flags {
670 	IB_AH_GRH	= 1
671 };
672 
673 enum ib_rate {
674 	IB_RATE_PORT_CURRENT = 0,
675 	IB_RATE_2_5_GBPS = 2,
676 	IB_RATE_5_GBPS   = 5,
677 	IB_RATE_10_GBPS  = 3,
678 	IB_RATE_20_GBPS  = 6,
679 	IB_RATE_30_GBPS  = 4,
680 	IB_RATE_40_GBPS  = 7,
681 	IB_RATE_60_GBPS  = 8,
682 	IB_RATE_80_GBPS  = 9,
683 	IB_RATE_120_GBPS = 10,
684 	IB_RATE_14_GBPS  = 11,
685 	IB_RATE_56_GBPS  = 12,
686 	IB_RATE_112_GBPS = 13,
687 	IB_RATE_168_GBPS = 14,
688 	IB_RATE_25_GBPS  = 15,
689 	IB_RATE_100_GBPS = 16,
690 	IB_RATE_200_GBPS = 17,
691 	IB_RATE_300_GBPS = 18
692 };
693 
694 /**
695  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
696  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
697  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
698  * @rate: rate to convert.
699  */
700 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
701 
702 /**
703  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
704  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
705  * @rate: rate to convert.
706  */
707 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
708 
709 
710 /**
711  * enum ib_mr_type - memory region type
712  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
713  *                            normal registration
714  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
715  *                            signature operations (data-integrity
716  *                            capable regions)
717  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
718  *                            register any arbitrary sg lists (without
719  *                            the normal mr constraints - see
720  *                            ib_map_mr_sg)
721  */
722 enum ib_mr_type {
723 	IB_MR_TYPE_MEM_REG,
724 	IB_MR_TYPE_SIGNATURE,
725 	IB_MR_TYPE_SG_GAPS,
726 };
727 
728 /**
729  * Signature types
730  * IB_SIG_TYPE_NONE: Unprotected.
731  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
732  */
733 enum ib_signature_type {
734 	IB_SIG_TYPE_NONE,
735 	IB_SIG_TYPE_T10_DIF,
736 };
737 
738 /**
739  * Signature T10-DIF block-guard types
740  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
741  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
742  */
743 enum ib_t10_dif_bg_type {
744 	IB_T10DIF_CRC,
745 	IB_T10DIF_CSUM
746 };
747 
748 /**
749  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
750  *     domain.
751  * @bg_type: T10-DIF block guard type (CRC|CSUM)
752  * @pi_interval: protection information interval.
753  * @bg: seed of guard computation.
754  * @app_tag: application tag of guard block
755  * @ref_tag: initial guard block reference tag.
756  * @ref_remap: Indicate wethear the reftag increments each block
757  * @app_escape: Indicate to skip block check if apptag=0xffff
758  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
759  * @apptag_check_mask: check bitmask of application tag.
760  */
761 struct ib_t10_dif_domain {
762 	enum ib_t10_dif_bg_type bg_type;
763 	u16			pi_interval;
764 	u16			bg;
765 	u16			app_tag;
766 	u32			ref_tag;
767 	bool			ref_remap;
768 	bool			app_escape;
769 	bool			ref_escape;
770 	u16			apptag_check_mask;
771 };
772 
773 /**
774  * struct ib_sig_domain - Parameters for signature domain
775  * @sig_type: specific signauture type
776  * @sig: union of all signature domain attributes that may
777  *     be used to set domain layout.
778  */
779 struct ib_sig_domain {
780 	enum ib_signature_type sig_type;
781 	union {
782 		struct ib_t10_dif_domain dif;
783 	} sig;
784 };
785 
786 /**
787  * struct ib_sig_attrs - Parameters for signature handover operation
788  * @check_mask: bitmask for signature byte check (8 bytes)
789  * @mem: memory domain layout desciptor.
790  * @wire: wire domain layout desciptor.
791  */
792 struct ib_sig_attrs {
793 	u8			check_mask;
794 	struct ib_sig_domain	mem;
795 	struct ib_sig_domain	wire;
796 };
797 
798 enum ib_sig_err_type {
799 	IB_SIG_BAD_GUARD,
800 	IB_SIG_BAD_REFTAG,
801 	IB_SIG_BAD_APPTAG,
802 };
803 
804 /**
805  * struct ib_sig_err - signature error descriptor
806  */
807 struct ib_sig_err {
808 	enum ib_sig_err_type	err_type;
809 	u32			expected;
810 	u32			actual;
811 	u64			sig_err_offset;
812 	u32			key;
813 };
814 
815 enum ib_mr_status_check {
816 	IB_MR_CHECK_SIG_STATUS = 1,
817 };
818 
819 /**
820  * struct ib_mr_status - Memory region status container
821  *
822  * @fail_status: Bitmask of MR checks status. For each
823  *     failed check a corresponding status bit is set.
824  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
825  *     failure.
826  */
827 struct ib_mr_status {
828 	u32		    fail_status;
829 	struct ib_sig_err   sig_err;
830 };
831 
832 /**
833  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
834  * enum.
835  * @mult: multiple to convert.
836  */
837 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
838 
839 struct ib_ah_attr {
840 	struct ib_global_route	grh;
841 	u16			dlid;
842 	u8			sl;
843 	u8			src_path_bits;
844 	u8			static_rate;
845 	u8			ah_flags;
846 	u8			port_num;
847 	u8			dmac[ETH_ALEN];
848 };
849 
850 enum ib_wc_status {
851 	IB_WC_SUCCESS,
852 	IB_WC_LOC_LEN_ERR,
853 	IB_WC_LOC_QP_OP_ERR,
854 	IB_WC_LOC_EEC_OP_ERR,
855 	IB_WC_LOC_PROT_ERR,
856 	IB_WC_WR_FLUSH_ERR,
857 	IB_WC_MW_BIND_ERR,
858 	IB_WC_BAD_RESP_ERR,
859 	IB_WC_LOC_ACCESS_ERR,
860 	IB_WC_REM_INV_REQ_ERR,
861 	IB_WC_REM_ACCESS_ERR,
862 	IB_WC_REM_OP_ERR,
863 	IB_WC_RETRY_EXC_ERR,
864 	IB_WC_RNR_RETRY_EXC_ERR,
865 	IB_WC_LOC_RDD_VIOL_ERR,
866 	IB_WC_REM_INV_RD_REQ_ERR,
867 	IB_WC_REM_ABORT_ERR,
868 	IB_WC_INV_EECN_ERR,
869 	IB_WC_INV_EEC_STATE_ERR,
870 	IB_WC_FATAL_ERR,
871 	IB_WC_RESP_TIMEOUT_ERR,
872 	IB_WC_GENERAL_ERR
873 };
874 
875 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
876 
877 enum ib_wc_opcode {
878 	IB_WC_SEND,
879 	IB_WC_RDMA_WRITE,
880 	IB_WC_RDMA_READ,
881 	IB_WC_COMP_SWAP,
882 	IB_WC_FETCH_ADD,
883 	IB_WC_LSO,
884 	IB_WC_LOCAL_INV,
885 	IB_WC_REG_MR,
886 	IB_WC_MASKED_COMP_SWAP,
887 	IB_WC_MASKED_FETCH_ADD,
888 /*
889  * Set value of IB_WC_RECV so consumers can test if a completion is a
890  * receive by testing (opcode & IB_WC_RECV).
891  */
892 	IB_WC_RECV			= 1 << 7,
893 	IB_WC_RECV_RDMA_WITH_IMM
894 };
895 
896 enum ib_wc_flags {
897 	IB_WC_GRH		= 1,
898 	IB_WC_WITH_IMM		= (1<<1),
899 	IB_WC_WITH_INVALIDATE	= (1<<2),
900 	IB_WC_IP_CSUM_OK	= (1<<3),
901 	IB_WC_WITH_SMAC		= (1<<4),
902 	IB_WC_WITH_VLAN		= (1<<5),
903 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
904 };
905 
906 struct ib_wc {
907 	union {
908 		u64		wr_id;
909 		struct ib_cqe	*wr_cqe;
910 	};
911 	enum ib_wc_status	status;
912 	enum ib_wc_opcode	opcode;
913 	u32			vendor_err;
914 	u32			byte_len;
915 	struct ib_qp	       *qp;
916 	union {
917 		__be32		imm_data;
918 		u32		invalidate_rkey;
919 	} ex;
920 	u32			src_qp;
921 	int			wc_flags;
922 	u16			pkey_index;
923 	u16			slid;
924 	u8			sl;
925 	u8			dlid_path_bits;
926 	u8			port_num;	/* valid only for DR SMPs on switches */
927 	u8			smac[ETH_ALEN];
928 	u16			vlan_id;
929 	u8			network_hdr_type;
930 };
931 
932 enum ib_cq_notify_flags {
933 	IB_CQ_SOLICITED			= 1 << 0,
934 	IB_CQ_NEXT_COMP			= 1 << 1,
935 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
936 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
937 };
938 
939 enum ib_srq_type {
940 	IB_SRQT_BASIC,
941 	IB_SRQT_XRC
942 };
943 
944 enum ib_srq_attr_mask {
945 	IB_SRQ_MAX_WR	= 1 << 0,
946 	IB_SRQ_LIMIT	= 1 << 1,
947 };
948 
949 struct ib_srq_attr {
950 	u32	max_wr;
951 	u32	max_sge;
952 	u32	srq_limit;
953 };
954 
955 struct ib_srq_init_attr {
956 	void		      (*event_handler)(struct ib_event *, void *);
957 	void		       *srq_context;
958 	struct ib_srq_attr	attr;
959 	enum ib_srq_type	srq_type;
960 
961 	union {
962 		struct {
963 			struct ib_xrcd *xrcd;
964 			struct ib_cq   *cq;
965 		} xrc;
966 	} ext;
967 };
968 
969 struct ib_qp_cap {
970 	u32	max_send_wr;
971 	u32	max_recv_wr;
972 	u32	max_send_sge;
973 	u32	max_recv_sge;
974 	u32	max_inline_data;
975 
976 	/*
977 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
978 	 * ib_create_qp() will calculate the right amount of neededed WRs
979 	 * and MRs based on this.
980 	 */
981 	u32	max_rdma_ctxs;
982 };
983 
984 enum ib_sig_type {
985 	IB_SIGNAL_ALL_WR,
986 	IB_SIGNAL_REQ_WR
987 };
988 
989 enum ib_qp_type {
990 	/*
991 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
992 	 * here (and in that order) since the MAD layer uses them as
993 	 * indices into a 2-entry table.
994 	 */
995 	IB_QPT_SMI,
996 	IB_QPT_GSI,
997 
998 	IB_QPT_RC,
999 	IB_QPT_UC,
1000 	IB_QPT_UD,
1001 	IB_QPT_RAW_IPV6,
1002 	IB_QPT_RAW_ETHERTYPE,
1003 	IB_QPT_RAW_PACKET = 8,
1004 	IB_QPT_XRC_INI = 9,
1005 	IB_QPT_XRC_TGT,
1006 	IB_QPT_MAX,
1007 	/* Reserve a range for qp types internal to the low level driver.
1008 	 * These qp types will not be visible at the IB core layer, so the
1009 	 * IB_QPT_MAX usages should not be affected in the core layer
1010 	 */
1011 	IB_QPT_RESERVED1 = 0x1000,
1012 	IB_QPT_RESERVED2,
1013 	IB_QPT_RESERVED3,
1014 	IB_QPT_RESERVED4,
1015 	IB_QPT_RESERVED5,
1016 	IB_QPT_RESERVED6,
1017 	IB_QPT_RESERVED7,
1018 	IB_QPT_RESERVED8,
1019 	IB_QPT_RESERVED9,
1020 	IB_QPT_RESERVED10,
1021 };
1022 
1023 enum ib_qp_create_flags {
1024 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1025 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1026 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1027 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1028 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1029 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1030 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1031 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
1032 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1033 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1034 	/* reserve bits 26-31 for low level drivers' internal use */
1035 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1036 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1037 };
1038 
1039 /*
1040  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1041  * callback to destroy the passed in QP.
1042  */
1043 
1044 struct ib_qp_init_attr {
1045 	void                  (*event_handler)(struct ib_event *, void *);
1046 	void		       *qp_context;
1047 	struct ib_cq	       *send_cq;
1048 	struct ib_cq	       *recv_cq;
1049 	struct ib_srq	       *srq;
1050 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1051 	struct ib_qp_cap	cap;
1052 	enum ib_sig_type	sq_sig_type;
1053 	enum ib_qp_type		qp_type;
1054 	enum ib_qp_create_flags	create_flags;
1055 
1056 	/*
1057 	 * Only needed for special QP types, or when using the RW API.
1058 	 */
1059 	u8			port_num;
1060 	struct ib_rwq_ind_table *rwq_ind_tbl;
1061 };
1062 
1063 struct ib_qp_open_attr {
1064 	void                  (*event_handler)(struct ib_event *, void *);
1065 	void		       *qp_context;
1066 	u32			qp_num;
1067 	enum ib_qp_type		qp_type;
1068 };
1069 
1070 enum ib_rnr_timeout {
1071 	IB_RNR_TIMER_655_36 =  0,
1072 	IB_RNR_TIMER_000_01 =  1,
1073 	IB_RNR_TIMER_000_02 =  2,
1074 	IB_RNR_TIMER_000_03 =  3,
1075 	IB_RNR_TIMER_000_04 =  4,
1076 	IB_RNR_TIMER_000_06 =  5,
1077 	IB_RNR_TIMER_000_08 =  6,
1078 	IB_RNR_TIMER_000_12 =  7,
1079 	IB_RNR_TIMER_000_16 =  8,
1080 	IB_RNR_TIMER_000_24 =  9,
1081 	IB_RNR_TIMER_000_32 = 10,
1082 	IB_RNR_TIMER_000_48 = 11,
1083 	IB_RNR_TIMER_000_64 = 12,
1084 	IB_RNR_TIMER_000_96 = 13,
1085 	IB_RNR_TIMER_001_28 = 14,
1086 	IB_RNR_TIMER_001_92 = 15,
1087 	IB_RNR_TIMER_002_56 = 16,
1088 	IB_RNR_TIMER_003_84 = 17,
1089 	IB_RNR_TIMER_005_12 = 18,
1090 	IB_RNR_TIMER_007_68 = 19,
1091 	IB_RNR_TIMER_010_24 = 20,
1092 	IB_RNR_TIMER_015_36 = 21,
1093 	IB_RNR_TIMER_020_48 = 22,
1094 	IB_RNR_TIMER_030_72 = 23,
1095 	IB_RNR_TIMER_040_96 = 24,
1096 	IB_RNR_TIMER_061_44 = 25,
1097 	IB_RNR_TIMER_081_92 = 26,
1098 	IB_RNR_TIMER_122_88 = 27,
1099 	IB_RNR_TIMER_163_84 = 28,
1100 	IB_RNR_TIMER_245_76 = 29,
1101 	IB_RNR_TIMER_327_68 = 30,
1102 	IB_RNR_TIMER_491_52 = 31
1103 };
1104 
1105 enum ib_qp_attr_mask {
1106 	IB_QP_STATE			= 1,
1107 	IB_QP_CUR_STATE			= (1<<1),
1108 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1109 	IB_QP_ACCESS_FLAGS		= (1<<3),
1110 	IB_QP_PKEY_INDEX		= (1<<4),
1111 	IB_QP_PORT			= (1<<5),
1112 	IB_QP_QKEY			= (1<<6),
1113 	IB_QP_AV			= (1<<7),
1114 	IB_QP_PATH_MTU			= (1<<8),
1115 	IB_QP_TIMEOUT			= (1<<9),
1116 	IB_QP_RETRY_CNT			= (1<<10),
1117 	IB_QP_RNR_RETRY			= (1<<11),
1118 	IB_QP_RQ_PSN			= (1<<12),
1119 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1120 	IB_QP_ALT_PATH			= (1<<14),
1121 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1122 	IB_QP_SQ_PSN			= (1<<16),
1123 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1124 	IB_QP_PATH_MIG_STATE		= (1<<18),
1125 	IB_QP_CAP			= (1<<19),
1126 	IB_QP_DEST_QPN			= (1<<20),
1127 	IB_QP_RESERVED1			= (1<<21),
1128 	IB_QP_RESERVED2			= (1<<22),
1129 	IB_QP_RESERVED3			= (1<<23),
1130 	IB_QP_RESERVED4			= (1<<24),
1131 	IB_QP_RATE_LIMIT		= (1<<25),
1132 };
1133 
1134 enum ib_qp_state {
1135 	IB_QPS_RESET,
1136 	IB_QPS_INIT,
1137 	IB_QPS_RTR,
1138 	IB_QPS_RTS,
1139 	IB_QPS_SQD,
1140 	IB_QPS_SQE,
1141 	IB_QPS_ERR
1142 };
1143 
1144 enum ib_mig_state {
1145 	IB_MIG_MIGRATED,
1146 	IB_MIG_REARM,
1147 	IB_MIG_ARMED
1148 };
1149 
1150 enum ib_mw_type {
1151 	IB_MW_TYPE_1 = 1,
1152 	IB_MW_TYPE_2 = 2
1153 };
1154 
1155 struct ib_qp_attr {
1156 	enum ib_qp_state	qp_state;
1157 	enum ib_qp_state	cur_qp_state;
1158 	enum ib_mtu		path_mtu;
1159 	enum ib_mig_state	path_mig_state;
1160 	u32			qkey;
1161 	u32			rq_psn;
1162 	u32			sq_psn;
1163 	u32			dest_qp_num;
1164 	int			qp_access_flags;
1165 	struct ib_qp_cap	cap;
1166 	struct ib_ah_attr	ah_attr;
1167 	struct ib_ah_attr	alt_ah_attr;
1168 	u16			pkey_index;
1169 	u16			alt_pkey_index;
1170 	u8			en_sqd_async_notify;
1171 	u8			sq_draining;
1172 	u8			max_rd_atomic;
1173 	u8			max_dest_rd_atomic;
1174 	u8			min_rnr_timer;
1175 	u8			port_num;
1176 	u8			timeout;
1177 	u8			retry_cnt;
1178 	u8			rnr_retry;
1179 	u8			alt_port_num;
1180 	u8			alt_timeout;
1181 	u32			rate_limit;
1182 };
1183 
1184 enum ib_wr_opcode {
1185 	IB_WR_RDMA_WRITE,
1186 	IB_WR_RDMA_WRITE_WITH_IMM,
1187 	IB_WR_SEND,
1188 	IB_WR_SEND_WITH_IMM,
1189 	IB_WR_RDMA_READ,
1190 	IB_WR_ATOMIC_CMP_AND_SWP,
1191 	IB_WR_ATOMIC_FETCH_AND_ADD,
1192 	IB_WR_LSO,
1193 	IB_WR_SEND_WITH_INV,
1194 	IB_WR_RDMA_READ_WITH_INV,
1195 	IB_WR_LOCAL_INV,
1196 	IB_WR_REG_MR,
1197 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1198 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1199 	IB_WR_REG_SIG_MR,
1200 	/* reserve values for low level drivers' internal use.
1201 	 * These values will not be used at all in the ib core layer.
1202 	 */
1203 	IB_WR_RESERVED1 = 0xf0,
1204 	IB_WR_RESERVED2,
1205 	IB_WR_RESERVED3,
1206 	IB_WR_RESERVED4,
1207 	IB_WR_RESERVED5,
1208 	IB_WR_RESERVED6,
1209 	IB_WR_RESERVED7,
1210 	IB_WR_RESERVED8,
1211 	IB_WR_RESERVED9,
1212 	IB_WR_RESERVED10,
1213 };
1214 
1215 enum ib_send_flags {
1216 	IB_SEND_FENCE		= 1,
1217 	IB_SEND_SIGNALED	= (1<<1),
1218 	IB_SEND_SOLICITED	= (1<<2),
1219 	IB_SEND_INLINE		= (1<<3),
1220 	IB_SEND_IP_CSUM		= (1<<4),
1221 
1222 	/* reserve bits 26-31 for low level drivers' internal use */
1223 	IB_SEND_RESERVED_START	= (1 << 26),
1224 	IB_SEND_RESERVED_END	= (1 << 31),
1225 };
1226 
1227 struct ib_sge {
1228 	u64	addr;
1229 	u32	length;
1230 	u32	lkey;
1231 };
1232 
1233 struct ib_cqe {
1234 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1235 };
1236 
1237 struct ib_send_wr {
1238 	struct ib_send_wr      *next;
1239 	union {
1240 		u64		wr_id;
1241 		struct ib_cqe	*wr_cqe;
1242 	};
1243 	struct ib_sge	       *sg_list;
1244 	int			num_sge;
1245 	enum ib_wr_opcode	opcode;
1246 	int			send_flags;
1247 	union {
1248 		__be32		imm_data;
1249 		u32		invalidate_rkey;
1250 	} ex;
1251 };
1252 
1253 struct ib_rdma_wr {
1254 	struct ib_send_wr	wr;
1255 	u64			remote_addr;
1256 	u32			rkey;
1257 };
1258 
1259 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1260 {
1261 	return container_of(wr, struct ib_rdma_wr, wr);
1262 }
1263 
1264 struct ib_atomic_wr {
1265 	struct ib_send_wr	wr;
1266 	u64			remote_addr;
1267 	u64			compare_add;
1268 	u64			swap;
1269 	u64			compare_add_mask;
1270 	u64			swap_mask;
1271 	u32			rkey;
1272 };
1273 
1274 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1275 {
1276 	return container_of(wr, struct ib_atomic_wr, wr);
1277 }
1278 
1279 struct ib_ud_wr {
1280 	struct ib_send_wr	wr;
1281 	struct ib_ah		*ah;
1282 	void			*header;
1283 	int			hlen;
1284 	int			mss;
1285 	u32			remote_qpn;
1286 	u32			remote_qkey;
1287 	u16			pkey_index; /* valid for GSI only */
1288 	u8			port_num;   /* valid for DR SMPs on switch only */
1289 };
1290 
1291 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1292 {
1293 	return container_of(wr, struct ib_ud_wr, wr);
1294 }
1295 
1296 struct ib_reg_wr {
1297 	struct ib_send_wr	wr;
1298 	struct ib_mr		*mr;
1299 	u32			key;
1300 	int			access;
1301 };
1302 
1303 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1304 {
1305 	return container_of(wr, struct ib_reg_wr, wr);
1306 }
1307 
1308 struct ib_sig_handover_wr {
1309 	struct ib_send_wr	wr;
1310 	struct ib_sig_attrs    *sig_attrs;
1311 	struct ib_mr	       *sig_mr;
1312 	int			access_flags;
1313 	struct ib_sge	       *prot;
1314 };
1315 
1316 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1317 {
1318 	return container_of(wr, struct ib_sig_handover_wr, wr);
1319 }
1320 
1321 struct ib_recv_wr {
1322 	struct ib_recv_wr      *next;
1323 	union {
1324 		u64		wr_id;
1325 		struct ib_cqe	*wr_cqe;
1326 	};
1327 	struct ib_sge	       *sg_list;
1328 	int			num_sge;
1329 };
1330 
1331 enum ib_access_flags {
1332 	IB_ACCESS_LOCAL_WRITE	= 1,
1333 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1334 	IB_ACCESS_REMOTE_READ	= (1<<2),
1335 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1336 	IB_ACCESS_MW_BIND	= (1<<4),
1337 	IB_ZERO_BASED		= (1<<5),
1338 	IB_ACCESS_ON_DEMAND     = (1<<6),
1339 };
1340 
1341 /*
1342  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1343  * are hidden here instead of a uapi header!
1344  */
1345 enum ib_mr_rereg_flags {
1346 	IB_MR_REREG_TRANS	= 1,
1347 	IB_MR_REREG_PD		= (1<<1),
1348 	IB_MR_REREG_ACCESS	= (1<<2),
1349 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1350 };
1351 
1352 struct ib_fmr_attr {
1353 	int	max_pages;
1354 	int	max_maps;
1355 	u8	page_shift;
1356 };
1357 
1358 struct ib_umem;
1359 
1360 struct ib_rdmacg_object {
1361 #ifdef CONFIG_CGROUP_RDMA
1362 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1363 #endif
1364 };
1365 
1366 struct ib_ucontext {
1367 	struct ib_device       *device;
1368 	struct list_head	pd_list;
1369 	struct list_head	mr_list;
1370 	struct list_head	mw_list;
1371 	struct list_head	cq_list;
1372 	struct list_head	qp_list;
1373 	struct list_head	srq_list;
1374 	struct list_head	ah_list;
1375 	struct list_head	xrcd_list;
1376 	struct list_head	rule_list;
1377 	struct list_head	wq_list;
1378 	struct list_head	rwq_ind_tbl_list;
1379 	int			closing;
1380 
1381 	struct pid             *tgid;
1382 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1383 	struct rb_root      umem_tree;
1384 	/*
1385 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1386 	 * mmu notifiers registration.
1387 	 */
1388 	struct rw_semaphore	umem_rwsem;
1389 	void (*invalidate_range)(struct ib_umem *umem,
1390 				 unsigned long start, unsigned long end);
1391 
1392 	struct mmu_notifier	mn;
1393 	atomic_t		notifier_count;
1394 	/* A list of umems that don't have private mmu notifier counters yet. */
1395 	struct list_head	no_private_counters;
1396 	int                     odp_mrs_count;
1397 #endif
1398 
1399 	struct ib_rdmacg_object	cg_obj;
1400 };
1401 
1402 struct ib_uobject {
1403 	u64			user_handle;	/* handle given to us by userspace */
1404 	struct ib_ucontext     *context;	/* associated user context */
1405 	void		       *object;		/* containing object */
1406 	struct list_head	list;		/* link to context's list */
1407 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1408 	int			id;		/* index into kernel idr */
1409 	struct kref		ref;
1410 	struct rw_semaphore	mutex;		/* protects .live */
1411 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1412 	int			live;
1413 };
1414 
1415 struct ib_udata {
1416 	const void __user *inbuf;
1417 	void __user *outbuf;
1418 	size_t       inlen;
1419 	size_t       outlen;
1420 };
1421 
1422 struct ib_pd {
1423 	u32			local_dma_lkey;
1424 	u32			flags;
1425 	struct ib_device       *device;
1426 	struct ib_uobject      *uobject;
1427 	atomic_t          	usecnt; /* count all resources */
1428 
1429 	u32			unsafe_global_rkey;
1430 
1431 	/*
1432 	 * Implementation details of the RDMA core, don't use in drivers:
1433 	 */
1434 	struct ib_mr	       *__internal_mr;
1435 };
1436 
1437 struct ib_xrcd {
1438 	struct ib_device       *device;
1439 	atomic_t		usecnt; /* count all exposed resources */
1440 	struct inode	       *inode;
1441 
1442 	struct mutex		tgt_qp_mutex;
1443 	struct list_head	tgt_qp_list;
1444 };
1445 
1446 struct ib_ah {
1447 	struct ib_device	*device;
1448 	struct ib_pd		*pd;
1449 	struct ib_uobject	*uobject;
1450 };
1451 
1452 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1453 
1454 enum ib_poll_context {
1455 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1456 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1457 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1458 };
1459 
1460 struct ib_cq {
1461 	struct ib_device       *device;
1462 	struct ib_uobject      *uobject;
1463 	ib_comp_handler   	comp_handler;
1464 	void                  (*event_handler)(struct ib_event *, void *);
1465 	void                   *cq_context;
1466 	int               	cqe;
1467 	atomic_t          	usecnt; /* count number of work queues */
1468 	enum ib_poll_context	poll_ctx;
1469 	struct ib_wc		*wc;
1470 	union {
1471 		struct irq_poll		iop;
1472 		struct work_struct	work;
1473 	};
1474 };
1475 
1476 struct ib_srq {
1477 	struct ib_device       *device;
1478 	struct ib_pd	       *pd;
1479 	struct ib_uobject      *uobject;
1480 	void		      (*event_handler)(struct ib_event *, void *);
1481 	void		       *srq_context;
1482 	enum ib_srq_type	srq_type;
1483 	atomic_t		usecnt;
1484 
1485 	union {
1486 		struct {
1487 			struct ib_xrcd *xrcd;
1488 			struct ib_cq   *cq;
1489 			u32		srq_num;
1490 		} xrc;
1491 	} ext;
1492 };
1493 
1494 enum ib_raw_packet_caps {
1495 	/* Strip cvlan from incoming packet and report it in the matching work
1496 	 * completion is supported.
1497 	 */
1498 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1499 	/* Scatter FCS field of an incoming packet to host memory is supported.
1500 	 */
1501 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1502 	/* Checksum offloads are supported (for both send and receive). */
1503 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1504 };
1505 
1506 enum ib_wq_type {
1507 	IB_WQT_RQ
1508 };
1509 
1510 enum ib_wq_state {
1511 	IB_WQS_RESET,
1512 	IB_WQS_RDY,
1513 	IB_WQS_ERR
1514 };
1515 
1516 struct ib_wq {
1517 	struct ib_device       *device;
1518 	struct ib_uobject      *uobject;
1519 	void		    *wq_context;
1520 	void		    (*event_handler)(struct ib_event *, void *);
1521 	struct ib_pd	       *pd;
1522 	struct ib_cq	       *cq;
1523 	u32		wq_num;
1524 	enum ib_wq_state       state;
1525 	enum ib_wq_type	wq_type;
1526 	atomic_t		usecnt;
1527 };
1528 
1529 enum ib_wq_flags {
1530 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1531 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1532 };
1533 
1534 struct ib_wq_init_attr {
1535 	void		       *wq_context;
1536 	enum ib_wq_type	wq_type;
1537 	u32		max_wr;
1538 	u32		max_sge;
1539 	struct	ib_cq	       *cq;
1540 	void		    (*event_handler)(struct ib_event *, void *);
1541 	u32		create_flags; /* Use enum ib_wq_flags */
1542 };
1543 
1544 enum ib_wq_attr_mask {
1545 	IB_WQ_STATE		= 1 << 0,
1546 	IB_WQ_CUR_STATE		= 1 << 1,
1547 	IB_WQ_FLAGS		= 1 << 2,
1548 };
1549 
1550 struct ib_wq_attr {
1551 	enum	ib_wq_state	wq_state;
1552 	enum	ib_wq_state	curr_wq_state;
1553 	u32			flags; /* Use enum ib_wq_flags */
1554 	u32			flags_mask; /* Use enum ib_wq_flags */
1555 };
1556 
1557 struct ib_rwq_ind_table {
1558 	struct ib_device	*device;
1559 	struct ib_uobject      *uobject;
1560 	atomic_t		usecnt;
1561 	u32		ind_tbl_num;
1562 	u32		log_ind_tbl_size;
1563 	struct ib_wq	**ind_tbl;
1564 };
1565 
1566 struct ib_rwq_ind_table_init_attr {
1567 	u32		log_ind_tbl_size;
1568 	/* Each entry is a pointer to Receive Work Queue */
1569 	struct ib_wq	**ind_tbl;
1570 };
1571 
1572 /*
1573  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1574  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1575  */
1576 struct ib_qp {
1577 	struct ib_device       *device;
1578 	struct ib_pd	       *pd;
1579 	struct ib_cq	       *send_cq;
1580 	struct ib_cq	       *recv_cq;
1581 	spinlock_t		mr_lock;
1582 	int			mrs_used;
1583 	struct list_head	rdma_mrs;
1584 	struct list_head	sig_mrs;
1585 	struct ib_srq	       *srq;
1586 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1587 	struct list_head	xrcd_list;
1588 
1589 	/* count times opened, mcast attaches, flow attaches */
1590 	atomic_t		usecnt;
1591 	struct list_head	open_list;
1592 	struct ib_qp           *real_qp;
1593 	struct ib_uobject      *uobject;
1594 	void                  (*event_handler)(struct ib_event *, void *);
1595 	void		       *qp_context;
1596 	u32			qp_num;
1597 	u32			max_write_sge;
1598 	u32			max_read_sge;
1599 	enum ib_qp_type		qp_type;
1600 	struct ib_rwq_ind_table *rwq_ind_tbl;
1601 };
1602 
1603 struct ib_mr {
1604 	struct ib_device  *device;
1605 	struct ib_pd	  *pd;
1606 	u32		   lkey;
1607 	u32		   rkey;
1608 	u64		   iova;
1609 	u32		   length;
1610 	unsigned int	   page_size;
1611 	bool		   need_inval;
1612 	union {
1613 		struct ib_uobject	*uobject;	/* user */
1614 		struct list_head	qp_entry;	/* FR */
1615 	};
1616 };
1617 
1618 struct ib_mw {
1619 	struct ib_device	*device;
1620 	struct ib_pd		*pd;
1621 	struct ib_uobject	*uobject;
1622 	u32			rkey;
1623 	enum ib_mw_type         type;
1624 };
1625 
1626 struct ib_fmr {
1627 	struct ib_device	*device;
1628 	struct ib_pd		*pd;
1629 	struct list_head	list;
1630 	u32			lkey;
1631 	u32			rkey;
1632 };
1633 
1634 /* Supported steering options */
1635 enum ib_flow_attr_type {
1636 	/* steering according to rule specifications */
1637 	IB_FLOW_ATTR_NORMAL		= 0x0,
1638 	/* default unicast and multicast rule -
1639 	 * receive all Eth traffic which isn't steered to any QP
1640 	 */
1641 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1642 	/* default multicast rule -
1643 	 * receive all Eth multicast traffic which isn't steered to any QP
1644 	 */
1645 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1646 	/* sniffer rule - receive all port traffic */
1647 	IB_FLOW_ATTR_SNIFFER		= 0x3
1648 };
1649 
1650 /* Supported steering header types */
1651 enum ib_flow_spec_type {
1652 	/* L2 headers*/
1653 	IB_FLOW_SPEC_ETH		= 0x20,
1654 	IB_FLOW_SPEC_IB			= 0x22,
1655 	/* L3 header*/
1656 	IB_FLOW_SPEC_IPV4		= 0x30,
1657 	IB_FLOW_SPEC_IPV6		= 0x31,
1658 	/* L4 headers*/
1659 	IB_FLOW_SPEC_TCP		= 0x40,
1660 	IB_FLOW_SPEC_UDP		= 0x41,
1661 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1662 	IB_FLOW_SPEC_INNER		= 0x100,
1663 	/* Actions */
1664 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1665 };
1666 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1667 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1668 
1669 /* Flow steering rule priority is set according to it's domain.
1670  * Lower domain value means higher priority.
1671  */
1672 enum ib_flow_domain {
1673 	IB_FLOW_DOMAIN_USER,
1674 	IB_FLOW_DOMAIN_ETHTOOL,
1675 	IB_FLOW_DOMAIN_RFS,
1676 	IB_FLOW_DOMAIN_NIC,
1677 	IB_FLOW_DOMAIN_NUM /* Must be last */
1678 };
1679 
1680 enum ib_flow_flags {
1681 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1682 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1683 };
1684 
1685 struct ib_flow_eth_filter {
1686 	u8	dst_mac[6];
1687 	u8	src_mac[6];
1688 	__be16	ether_type;
1689 	__be16	vlan_tag;
1690 	/* Must be last */
1691 	u8	real_sz[0];
1692 };
1693 
1694 struct ib_flow_spec_eth {
1695 	u32			  type;
1696 	u16			  size;
1697 	struct ib_flow_eth_filter val;
1698 	struct ib_flow_eth_filter mask;
1699 };
1700 
1701 struct ib_flow_ib_filter {
1702 	__be16 dlid;
1703 	__u8   sl;
1704 	/* Must be last */
1705 	u8	real_sz[0];
1706 };
1707 
1708 struct ib_flow_spec_ib {
1709 	u32			 type;
1710 	u16			 size;
1711 	struct ib_flow_ib_filter val;
1712 	struct ib_flow_ib_filter mask;
1713 };
1714 
1715 /* IPv4 header flags */
1716 enum ib_ipv4_flags {
1717 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1718 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1719 				    last have this flag set */
1720 };
1721 
1722 struct ib_flow_ipv4_filter {
1723 	__be32	src_ip;
1724 	__be32	dst_ip;
1725 	u8	proto;
1726 	u8	tos;
1727 	u8	ttl;
1728 	u8	flags;
1729 	/* Must be last */
1730 	u8	real_sz[0];
1731 };
1732 
1733 struct ib_flow_spec_ipv4 {
1734 	u32			   type;
1735 	u16			   size;
1736 	struct ib_flow_ipv4_filter val;
1737 	struct ib_flow_ipv4_filter mask;
1738 };
1739 
1740 struct ib_flow_ipv6_filter {
1741 	u8	src_ip[16];
1742 	u8	dst_ip[16];
1743 	__be32	flow_label;
1744 	u8	next_hdr;
1745 	u8	traffic_class;
1746 	u8	hop_limit;
1747 	/* Must be last */
1748 	u8	real_sz[0];
1749 };
1750 
1751 struct ib_flow_spec_ipv6 {
1752 	u32			   type;
1753 	u16			   size;
1754 	struct ib_flow_ipv6_filter val;
1755 	struct ib_flow_ipv6_filter mask;
1756 };
1757 
1758 struct ib_flow_tcp_udp_filter {
1759 	__be16	dst_port;
1760 	__be16	src_port;
1761 	/* Must be last */
1762 	u8	real_sz[0];
1763 };
1764 
1765 struct ib_flow_spec_tcp_udp {
1766 	u32			      type;
1767 	u16			      size;
1768 	struct ib_flow_tcp_udp_filter val;
1769 	struct ib_flow_tcp_udp_filter mask;
1770 };
1771 
1772 struct ib_flow_tunnel_filter {
1773 	__be32	tunnel_id;
1774 	u8	real_sz[0];
1775 };
1776 
1777 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1778  * the tunnel_id from val has the vni value
1779  */
1780 struct ib_flow_spec_tunnel {
1781 	u32			      type;
1782 	u16			      size;
1783 	struct ib_flow_tunnel_filter  val;
1784 	struct ib_flow_tunnel_filter  mask;
1785 };
1786 
1787 struct ib_flow_spec_action_tag {
1788 	enum ib_flow_spec_type	      type;
1789 	u16			      size;
1790 	u32                           tag_id;
1791 };
1792 
1793 union ib_flow_spec {
1794 	struct {
1795 		u32			type;
1796 		u16			size;
1797 	};
1798 	struct ib_flow_spec_eth		eth;
1799 	struct ib_flow_spec_ib		ib;
1800 	struct ib_flow_spec_ipv4        ipv4;
1801 	struct ib_flow_spec_tcp_udp	tcp_udp;
1802 	struct ib_flow_spec_ipv6        ipv6;
1803 	struct ib_flow_spec_tunnel      tunnel;
1804 	struct ib_flow_spec_action_tag  flow_tag;
1805 };
1806 
1807 struct ib_flow_attr {
1808 	enum ib_flow_attr_type type;
1809 	u16	     size;
1810 	u16	     priority;
1811 	u32	     flags;
1812 	u8	     num_of_specs;
1813 	u8	     port;
1814 	/* Following are the optional layers according to user request
1815 	 * struct ib_flow_spec_xxx
1816 	 * struct ib_flow_spec_yyy
1817 	 */
1818 };
1819 
1820 struct ib_flow {
1821 	struct ib_qp		*qp;
1822 	struct ib_uobject	*uobject;
1823 };
1824 
1825 struct ib_mad_hdr;
1826 struct ib_grh;
1827 
1828 enum ib_process_mad_flags {
1829 	IB_MAD_IGNORE_MKEY	= 1,
1830 	IB_MAD_IGNORE_BKEY	= 2,
1831 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1832 };
1833 
1834 enum ib_mad_result {
1835 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1836 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1837 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1838 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1839 };
1840 
1841 #define IB_DEVICE_NAME_MAX 64
1842 
1843 struct ib_port_cache {
1844 	struct ib_pkey_cache  *pkey;
1845 	struct ib_gid_table   *gid;
1846 	u8                     lmc;
1847 	enum ib_port_state     port_state;
1848 };
1849 
1850 struct ib_cache {
1851 	rwlock_t                lock;
1852 	struct ib_event_handler event_handler;
1853 	struct ib_port_cache   *ports;
1854 };
1855 
1856 struct iw_cm_verbs;
1857 
1858 struct ib_port_immutable {
1859 	int                           pkey_tbl_len;
1860 	int                           gid_tbl_len;
1861 	u32                           core_cap_flags;
1862 	u32                           max_mad_size;
1863 };
1864 
1865 struct ib_device {
1866 	char                          name[IB_DEVICE_NAME_MAX];
1867 
1868 	struct list_head              event_handler_list;
1869 	spinlock_t                    event_handler_lock;
1870 
1871 	spinlock_t                    client_data_lock;
1872 	struct list_head              core_list;
1873 	/* Access to the client_data_list is protected by the client_data_lock
1874 	 * spinlock and the lists_rwsem read-write semaphore */
1875 	struct list_head              client_data_list;
1876 
1877 	struct ib_cache               cache;
1878 	/**
1879 	 * port_immutable is indexed by port number
1880 	 */
1881 	struct ib_port_immutable     *port_immutable;
1882 
1883 	int			      num_comp_vectors;
1884 
1885 	struct iw_cm_verbs	     *iwcm;
1886 
1887 	/**
1888 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1889 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
1890 	 *   core when the device is removed.  A lifespan of -1 in the return
1891 	 *   struct tells the core to set a default lifespan.
1892 	 */
1893 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
1894 						     u8 port_num);
1895 	/**
1896 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1897 	 * @index - The index in the value array we wish to have updated, or
1898 	 *   num_counters if we want all stats updated
1899 	 * Return codes -
1900 	 *   < 0 - Error, no counters updated
1901 	 *   index - Updated the single counter pointed to by index
1902 	 *   num_counters - Updated all counters (will reset the timestamp
1903 	 *     and prevent further calls for lifespan milliseconds)
1904 	 * Drivers are allowed to update all counters in leiu of just the
1905 	 *   one given in index at their option
1906 	 */
1907 	int		           (*get_hw_stats)(struct ib_device *device,
1908 						   struct rdma_hw_stats *stats,
1909 						   u8 port, int index);
1910 	int		           (*query_device)(struct ib_device *device,
1911 						   struct ib_device_attr *device_attr,
1912 						   struct ib_udata *udata);
1913 	int		           (*query_port)(struct ib_device *device,
1914 						 u8 port_num,
1915 						 struct ib_port_attr *port_attr);
1916 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1917 						     u8 port_num);
1918 	/* When calling get_netdev, the HW vendor's driver should return the
1919 	 * net device of device @device at port @port_num or NULL if such
1920 	 * a net device doesn't exist. The vendor driver should call dev_hold
1921 	 * on this net device. The HW vendor's device driver must guarantee
1922 	 * that this function returns NULL before the net device reaches
1923 	 * NETDEV_UNREGISTER_FINAL state.
1924 	 */
1925 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1926 						 u8 port_num);
1927 	int		           (*query_gid)(struct ib_device *device,
1928 						u8 port_num, int index,
1929 						union ib_gid *gid);
1930 	/* When calling add_gid, the HW vendor's driver should
1931 	 * add the gid of device @device at gid index @index of
1932 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1933 	 * the network device related to this gid is available
1934 	 * at @attr. @context allows the HW vendor driver to store extra
1935 	 * information together with a GID entry. The HW vendor may allocate
1936 	 * memory to contain this information and store it in @context when a
1937 	 * new GID entry is written to. Params are consistent until the next
1938 	 * call of add_gid or delete_gid. The function should return 0 on
1939 	 * success or error otherwise. The function could be called
1940 	 * concurrently for different ports. This function is only called
1941 	 * when roce_gid_table is used.
1942 	 */
1943 	int		           (*add_gid)(struct ib_device *device,
1944 					      u8 port_num,
1945 					      unsigned int index,
1946 					      const union ib_gid *gid,
1947 					      const struct ib_gid_attr *attr,
1948 					      void **context);
1949 	/* When calling del_gid, the HW vendor's driver should delete the
1950 	 * gid of device @device at gid index @index of port @port_num.
1951 	 * Upon the deletion of a GID entry, the HW vendor must free any
1952 	 * allocated memory. The caller will clear @context afterwards.
1953 	 * This function is only called when roce_gid_table is used.
1954 	 */
1955 	int		           (*del_gid)(struct ib_device *device,
1956 					      u8 port_num,
1957 					      unsigned int index,
1958 					      void **context);
1959 	int		           (*query_pkey)(struct ib_device *device,
1960 						 u8 port_num, u16 index, u16 *pkey);
1961 	int		           (*modify_device)(struct ib_device *device,
1962 						    int device_modify_mask,
1963 						    struct ib_device_modify *device_modify);
1964 	int		           (*modify_port)(struct ib_device *device,
1965 						  u8 port_num, int port_modify_mask,
1966 						  struct ib_port_modify *port_modify);
1967 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1968 						     struct ib_udata *udata);
1969 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1970 	int                        (*mmap)(struct ib_ucontext *context,
1971 					   struct vm_area_struct *vma);
1972 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1973 					       struct ib_ucontext *context,
1974 					       struct ib_udata *udata);
1975 	int                        (*dealloc_pd)(struct ib_pd *pd);
1976 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1977 						struct ib_ah_attr *ah_attr,
1978 						struct ib_udata *udata);
1979 	int                        (*modify_ah)(struct ib_ah *ah,
1980 						struct ib_ah_attr *ah_attr);
1981 	int                        (*query_ah)(struct ib_ah *ah,
1982 					       struct ib_ah_attr *ah_attr);
1983 	int                        (*destroy_ah)(struct ib_ah *ah);
1984 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1985 						 struct ib_srq_init_attr *srq_init_attr,
1986 						 struct ib_udata *udata);
1987 	int                        (*modify_srq)(struct ib_srq *srq,
1988 						 struct ib_srq_attr *srq_attr,
1989 						 enum ib_srq_attr_mask srq_attr_mask,
1990 						 struct ib_udata *udata);
1991 	int                        (*query_srq)(struct ib_srq *srq,
1992 						struct ib_srq_attr *srq_attr);
1993 	int                        (*destroy_srq)(struct ib_srq *srq);
1994 	int                        (*post_srq_recv)(struct ib_srq *srq,
1995 						    struct ib_recv_wr *recv_wr,
1996 						    struct ib_recv_wr **bad_recv_wr);
1997 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1998 						struct ib_qp_init_attr *qp_init_attr,
1999 						struct ib_udata *udata);
2000 	int                        (*modify_qp)(struct ib_qp *qp,
2001 						struct ib_qp_attr *qp_attr,
2002 						int qp_attr_mask,
2003 						struct ib_udata *udata);
2004 	int                        (*query_qp)(struct ib_qp *qp,
2005 					       struct ib_qp_attr *qp_attr,
2006 					       int qp_attr_mask,
2007 					       struct ib_qp_init_attr *qp_init_attr);
2008 	int                        (*destroy_qp)(struct ib_qp *qp);
2009 	int                        (*post_send)(struct ib_qp *qp,
2010 						struct ib_send_wr *send_wr,
2011 						struct ib_send_wr **bad_send_wr);
2012 	int                        (*post_recv)(struct ib_qp *qp,
2013 						struct ib_recv_wr *recv_wr,
2014 						struct ib_recv_wr **bad_recv_wr);
2015 	struct ib_cq *             (*create_cq)(struct ib_device *device,
2016 						const struct ib_cq_init_attr *attr,
2017 						struct ib_ucontext *context,
2018 						struct ib_udata *udata);
2019 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2020 						u16 cq_period);
2021 	int                        (*destroy_cq)(struct ib_cq *cq);
2022 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
2023 						struct ib_udata *udata);
2024 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
2025 					      struct ib_wc *wc);
2026 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2027 	int                        (*req_notify_cq)(struct ib_cq *cq,
2028 						    enum ib_cq_notify_flags flags);
2029 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
2030 						      int wc_cnt);
2031 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
2032 						 int mr_access_flags);
2033 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
2034 						  u64 start, u64 length,
2035 						  u64 virt_addr,
2036 						  int mr_access_flags,
2037 						  struct ib_udata *udata);
2038 	int			   (*rereg_user_mr)(struct ib_mr *mr,
2039 						    int flags,
2040 						    u64 start, u64 length,
2041 						    u64 virt_addr,
2042 						    int mr_access_flags,
2043 						    struct ib_pd *pd,
2044 						    struct ib_udata *udata);
2045 	int                        (*dereg_mr)(struct ib_mr *mr);
2046 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2047 					       enum ib_mr_type mr_type,
2048 					       u32 max_num_sg);
2049 	int                        (*map_mr_sg)(struct ib_mr *mr,
2050 						struct scatterlist *sg,
2051 						int sg_nents,
2052 						unsigned int *sg_offset);
2053 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2054 					       enum ib_mw_type type,
2055 					       struct ib_udata *udata);
2056 	int                        (*dealloc_mw)(struct ib_mw *mw);
2057 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2058 						int mr_access_flags,
2059 						struct ib_fmr_attr *fmr_attr);
2060 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2061 						   u64 *page_list, int list_len,
2062 						   u64 iova);
2063 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2064 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2065 	int                        (*attach_mcast)(struct ib_qp *qp,
2066 						   union ib_gid *gid,
2067 						   u16 lid);
2068 	int                        (*detach_mcast)(struct ib_qp *qp,
2069 						   union ib_gid *gid,
2070 						   u16 lid);
2071 	int                        (*process_mad)(struct ib_device *device,
2072 						  int process_mad_flags,
2073 						  u8 port_num,
2074 						  const struct ib_wc *in_wc,
2075 						  const struct ib_grh *in_grh,
2076 						  const struct ib_mad_hdr *in_mad,
2077 						  size_t in_mad_size,
2078 						  struct ib_mad_hdr *out_mad,
2079 						  size_t *out_mad_size,
2080 						  u16 *out_mad_pkey_index);
2081 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2082 						 struct ib_ucontext *ucontext,
2083 						 struct ib_udata *udata);
2084 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2085 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2086 						  struct ib_flow_attr
2087 						  *flow_attr,
2088 						  int domain);
2089 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2090 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2091 						      struct ib_mr_status *mr_status);
2092 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2093 	void			   (*drain_rq)(struct ib_qp *qp);
2094 	void			   (*drain_sq)(struct ib_qp *qp);
2095 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2096 							int state);
2097 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2098 						   struct ifla_vf_info *ivf);
2099 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2100 						   struct ifla_vf_stats *stats);
2101 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2102 						  int type);
2103 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2104 						struct ib_wq_init_attr *init_attr,
2105 						struct ib_udata *udata);
2106 	int			   (*destroy_wq)(struct ib_wq *wq);
2107 	int			   (*modify_wq)(struct ib_wq *wq,
2108 						struct ib_wq_attr *attr,
2109 						u32 wq_attr_mask,
2110 						struct ib_udata *udata);
2111 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2112 							   struct ib_rwq_ind_table_init_attr *init_attr,
2113 							   struct ib_udata *udata);
2114 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2115 
2116 	struct module               *owner;
2117 	struct device                dev;
2118 	struct kobject               *ports_parent;
2119 	struct list_head             port_list;
2120 
2121 	enum {
2122 		IB_DEV_UNINITIALIZED,
2123 		IB_DEV_REGISTERED,
2124 		IB_DEV_UNREGISTERED
2125 	}                            reg_state;
2126 
2127 	int			     uverbs_abi_ver;
2128 	u64			     uverbs_cmd_mask;
2129 	u64			     uverbs_ex_cmd_mask;
2130 
2131 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2132 	__be64			     node_guid;
2133 	u32			     local_dma_lkey;
2134 	u16                          is_switch:1;
2135 	u8                           node_type;
2136 	u8                           phys_port_cnt;
2137 	struct ib_device_attr        attrs;
2138 	struct attribute_group	     *hw_stats_ag;
2139 	struct rdma_hw_stats         *hw_stats;
2140 
2141 #ifdef CONFIG_CGROUP_RDMA
2142 	struct rdmacg_device         cg_device;
2143 #endif
2144 
2145 	/**
2146 	 * The following mandatory functions are used only at device
2147 	 * registration.  Keep functions such as these at the end of this
2148 	 * structure to avoid cache line misses when accessing struct ib_device
2149 	 * in fast paths.
2150 	 */
2151 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2152 	void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2153 };
2154 
2155 struct ib_client {
2156 	char  *name;
2157 	void (*add)   (struct ib_device *);
2158 	void (*remove)(struct ib_device *, void *client_data);
2159 
2160 	/* Returns the net_dev belonging to this ib_client and matching the
2161 	 * given parameters.
2162 	 * @dev:	 An RDMA device that the net_dev use for communication.
2163 	 * @port:	 A physical port number on the RDMA device.
2164 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2165 	 * @gid:	 A GID that the net_dev uses to communicate.
2166 	 * @addr:	 An IP address the net_dev is configured with.
2167 	 * @client_data: The device's client data set by ib_set_client_data().
2168 	 *
2169 	 * An ib_client that implements a net_dev on top of RDMA devices
2170 	 * (such as IP over IB) should implement this callback, allowing the
2171 	 * rdma_cm module to find the right net_dev for a given request.
2172 	 *
2173 	 * The caller is responsible for calling dev_put on the returned
2174 	 * netdev. */
2175 	struct net_device *(*get_net_dev_by_params)(
2176 			struct ib_device *dev,
2177 			u8 port,
2178 			u16 pkey,
2179 			const union ib_gid *gid,
2180 			const struct sockaddr *addr,
2181 			void *client_data);
2182 	struct list_head list;
2183 };
2184 
2185 struct ib_device *ib_alloc_device(size_t size);
2186 void ib_dealloc_device(struct ib_device *device);
2187 
2188 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2189 
2190 int ib_register_device(struct ib_device *device,
2191 		       int (*port_callback)(struct ib_device *,
2192 					    u8, struct kobject *));
2193 void ib_unregister_device(struct ib_device *device);
2194 
2195 int ib_register_client   (struct ib_client *client);
2196 void ib_unregister_client(struct ib_client *client);
2197 
2198 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2199 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2200 			 void *data);
2201 
2202 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2203 {
2204 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2205 }
2206 
2207 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2208 {
2209 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2210 }
2211 
2212 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2213 				       size_t offset,
2214 				       size_t len)
2215 {
2216 	const void __user *p = udata->inbuf + offset;
2217 	bool ret;
2218 	u8 *buf;
2219 
2220 	if (len > USHRT_MAX)
2221 		return false;
2222 
2223 	buf = memdup_user(p, len);
2224 	if (IS_ERR(buf))
2225 		return false;
2226 
2227 	ret = !memchr_inv(buf, 0, len);
2228 	kfree(buf);
2229 	return ret;
2230 }
2231 
2232 /**
2233  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2234  * contains all required attributes and no attributes not allowed for
2235  * the given QP state transition.
2236  * @cur_state: Current QP state
2237  * @next_state: Next QP state
2238  * @type: QP type
2239  * @mask: Mask of supplied QP attributes
2240  * @ll : link layer of port
2241  *
2242  * This function is a helper function that a low-level driver's
2243  * modify_qp method can use to validate the consumer's input.  It
2244  * checks that cur_state and next_state are valid QP states, that a
2245  * transition from cur_state to next_state is allowed by the IB spec,
2246  * and that the attribute mask supplied is allowed for the transition.
2247  */
2248 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2249 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2250 		       enum rdma_link_layer ll);
2251 
2252 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2253 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2254 void ib_dispatch_event(struct ib_event *event);
2255 
2256 int ib_query_port(struct ib_device *device,
2257 		  u8 port_num, struct ib_port_attr *port_attr);
2258 
2259 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2260 					       u8 port_num);
2261 
2262 /**
2263  * rdma_cap_ib_switch - Check if the device is IB switch
2264  * @device: Device to check
2265  *
2266  * Device driver is responsible for setting is_switch bit on
2267  * in ib_device structure at init time.
2268  *
2269  * Return: true if the device is IB switch.
2270  */
2271 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2272 {
2273 	return device->is_switch;
2274 }
2275 
2276 /**
2277  * rdma_start_port - Return the first valid port number for the device
2278  * specified
2279  *
2280  * @device: Device to be checked
2281  *
2282  * Return start port number
2283  */
2284 static inline u8 rdma_start_port(const struct ib_device *device)
2285 {
2286 	return rdma_cap_ib_switch(device) ? 0 : 1;
2287 }
2288 
2289 /**
2290  * rdma_end_port - Return the last valid port number for the device
2291  * specified
2292  *
2293  * @device: Device to be checked
2294  *
2295  * Return last port number
2296  */
2297 static inline u8 rdma_end_port(const struct ib_device *device)
2298 {
2299 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2300 }
2301 
2302 static inline int rdma_is_port_valid(const struct ib_device *device,
2303 				     unsigned int port)
2304 {
2305 	return (port >= rdma_start_port(device) &&
2306 		port <= rdma_end_port(device));
2307 }
2308 
2309 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2310 {
2311 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2312 }
2313 
2314 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2315 {
2316 	return device->port_immutable[port_num].core_cap_flags &
2317 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2318 }
2319 
2320 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2321 {
2322 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2323 }
2324 
2325 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2326 {
2327 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2328 }
2329 
2330 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2331 {
2332 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2333 }
2334 
2335 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2336 {
2337 	return rdma_protocol_ib(device, port_num) ||
2338 		rdma_protocol_roce(device, port_num);
2339 }
2340 
2341 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2342 {
2343 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2344 }
2345 
2346 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2347 {
2348 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2349 }
2350 
2351 /**
2352  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2353  * Management Datagrams.
2354  * @device: Device to check
2355  * @port_num: Port number to check
2356  *
2357  * Management Datagrams (MAD) are a required part of the InfiniBand
2358  * specification and are supported on all InfiniBand devices.  A slightly
2359  * extended version are also supported on OPA interfaces.
2360  *
2361  * Return: true if the port supports sending/receiving of MAD packets.
2362  */
2363 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2364 {
2365 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2366 }
2367 
2368 /**
2369  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2370  * Management Datagrams.
2371  * @device: Device to check
2372  * @port_num: Port number to check
2373  *
2374  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2375  * datagrams with their own versions.  These OPA MADs share many but not all of
2376  * the characteristics of InfiniBand MADs.
2377  *
2378  * OPA MADs differ in the following ways:
2379  *
2380  *    1) MADs are variable size up to 2K
2381  *       IBTA defined MADs remain fixed at 256 bytes
2382  *    2) OPA SMPs must carry valid PKeys
2383  *    3) OPA SMP packets are a different format
2384  *
2385  * Return: true if the port supports OPA MAD packet formats.
2386  */
2387 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2388 {
2389 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2390 		== RDMA_CORE_CAP_OPA_MAD;
2391 }
2392 
2393 /**
2394  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2395  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2396  * @device: Device to check
2397  * @port_num: Port number to check
2398  *
2399  * Each InfiniBand node is required to provide a Subnet Management Agent
2400  * that the subnet manager can access.  Prior to the fabric being fully
2401  * configured by the subnet manager, the SMA is accessed via a well known
2402  * interface called the Subnet Management Interface (SMI).  This interface
2403  * uses directed route packets to communicate with the SM to get around the
2404  * chicken and egg problem of the SM needing to know what's on the fabric
2405  * in order to configure the fabric, and needing to configure the fabric in
2406  * order to send packets to the devices on the fabric.  These directed
2407  * route packets do not need the fabric fully configured in order to reach
2408  * their destination.  The SMI is the only method allowed to send
2409  * directed route packets on an InfiniBand fabric.
2410  *
2411  * Return: true if the port provides an SMI.
2412  */
2413 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2414 {
2415 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2416 }
2417 
2418 /**
2419  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2420  * Communication Manager.
2421  * @device: Device to check
2422  * @port_num: Port number to check
2423  *
2424  * The InfiniBand Communication Manager is one of many pre-defined General
2425  * Service Agents (GSA) that are accessed via the General Service
2426  * Interface (GSI).  It's role is to facilitate establishment of connections
2427  * between nodes as well as other management related tasks for established
2428  * connections.
2429  *
2430  * Return: true if the port supports an IB CM (this does not guarantee that
2431  * a CM is actually running however).
2432  */
2433 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2434 {
2435 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2436 }
2437 
2438 /**
2439  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2440  * Communication Manager.
2441  * @device: Device to check
2442  * @port_num: Port number to check
2443  *
2444  * Similar to above, but specific to iWARP connections which have a different
2445  * managment protocol than InfiniBand.
2446  *
2447  * Return: true if the port supports an iWARP CM (this does not guarantee that
2448  * a CM is actually running however).
2449  */
2450 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2451 {
2452 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2453 }
2454 
2455 /**
2456  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2457  * Subnet Administration.
2458  * @device: Device to check
2459  * @port_num: Port number to check
2460  *
2461  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2462  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2463  * fabrics, devices should resolve routes to other hosts by contacting the
2464  * SA to query the proper route.
2465  *
2466  * Return: true if the port should act as a client to the fabric Subnet
2467  * Administration interface.  This does not imply that the SA service is
2468  * running locally.
2469  */
2470 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2471 {
2472 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2473 }
2474 
2475 /**
2476  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2477  * Multicast.
2478  * @device: Device to check
2479  * @port_num: Port number to check
2480  *
2481  * InfiniBand multicast registration is more complex than normal IPv4 or
2482  * IPv6 multicast registration.  Each Host Channel Adapter must register
2483  * with the Subnet Manager when it wishes to join a multicast group.  It
2484  * should do so only once regardless of how many queue pairs it subscribes
2485  * to this group.  And it should leave the group only after all queue pairs
2486  * attached to the group have been detached.
2487  *
2488  * Return: true if the port must undertake the additional adminstrative
2489  * overhead of registering/unregistering with the SM and tracking of the
2490  * total number of queue pairs attached to the multicast group.
2491  */
2492 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2493 {
2494 	return rdma_cap_ib_sa(device, port_num);
2495 }
2496 
2497 /**
2498  * rdma_cap_af_ib - Check if the port of device has the capability
2499  * Native Infiniband Address.
2500  * @device: Device to check
2501  * @port_num: Port number to check
2502  *
2503  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2504  * GID.  RoCE uses a different mechanism, but still generates a GID via
2505  * a prescribed mechanism and port specific data.
2506  *
2507  * Return: true if the port uses a GID address to identify devices on the
2508  * network.
2509  */
2510 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2511 {
2512 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2513 }
2514 
2515 /**
2516  * rdma_cap_eth_ah - Check if the port of device has the capability
2517  * Ethernet Address Handle.
2518  * @device: Device to check
2519  * @port_num: Port number to check
2520  *
2521  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2522  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2523  * port.  Normally, packet headers are generated by the sending host
2524  * adapter, but when sending connectionless datagrams, we must manually
2525  * inject the proper headers for the fabric we are communicating over.
2526  *
2527  * Return: true if we are running as a RoCE port and must force the
2528  * addition of a Global Route Header built from our Ethernet Address
2529  * Handle into our header list for connectionless packets.
2530  */
2531 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2532 {
2533 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2534 }
2535 
2536 /**
2537  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2538  *
2539  * @device: Device
2540  * @port_num: Port number
2541  *
2542  * This MAD size includes the MAD headers and MAD payload.  No other headers
2543  * are included.
2544  *
2545  * Return the max MAD size required by the Port.  Will return 0 if the port
2546  * does not support MADs
2547  */
2548 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2549 {
2550 	return device->port_immutable[port_num].max_mad_size;
2551 }
2552 
2553 /**
2554  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2555  * @device: Device to check
2556  * @port_num: Port number to check
2557  *
2558  * RoCE GID table mechanism manages the various GIDs for a device.
2559  *
2560  * NOTE: if allocating the port's GID table has failed, this call will still
2561  * return true, but any RoCE GID table API will fail.
2562  *
2563  * Return: true if the port uses RoCE GID table mechanism in order to manage
2564  * its GIDs.
2565  */
2566 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2567 					   u8 port_num)
2568 {
2569 	return rdma_protocol_roce(device, port_num) &&
2570 		device->add_gid && device->del_gid;
2571 }
2572 
2573 /*
2574  * Check if the device supports READ W/ INVALIDATE.
2575  */
2576 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2577 {
2578 	/*
2579 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2580 	 * has support for it yet.
2581 	 */
2582 	return rdma_protocol_iwarp(dev, port_num);
2583 }
2584 
2585 int ib_query_gid(struct ib_device *device,
2586 		 u8 port_num, int index, union ib_gid *gid,
2587 		 struct ib_gid_attr *attr);
2588 
2589 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2590 			 int state);
2591 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2592 		     struct ifla_vf_info *info);
2593 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2594 		    struct ifla_vf_stats *stats);
2595 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2596 		   int type);
2597 
2598 int ib_query_pkey(struct ib_device *device,
2599 		  u8 port_num, u16 index, u16 *pkey);
2600 
2601 int ib_modify_device(struct ib_device *device,
2602 		     int device_modify_mask,
2603 		     struct ib_device_modify *device_modify);
2604 
2605 int ib_modify_port(struct ib_device *device,
2606 		   u8 port_num, int port_modify_mask,
2607 		   struct ib_port_modify *port_modify);
2608 
2609 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2610 		enum ib_gid_type gid_type, struct net_device *ndev,
2611 		u8 *port_num, u16 *index);
2612 
2613 int ib_find_pkey(struct ib_device *device,
2614 		 u8 port_num, u16 pkey, u16 *index);
2615 
2616 enum ib_pd_flags {
2617 	/*
2618 	 * Create a memory registration for all memory in the system and place
2619 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2620 	 * ULPs to avoid the overhead of dynamic MRs.
2621 	 *
2622 	 * This flag is generally considered unsafe and must only be used in
2623 	 * extremly trusted environments.  Every use of it will log a warning
2624 	 * in the kernel log.
2625 	 */
2626 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2627 };
2628 
2629 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2630 		const char *caller);
2631 #define ib_alloc_pd(device, flags) \
2632 	__ib_alloc_pd((device), (flags), __func__)
2633 void ib_dealloc_pd(struct ib_pd *pd);
2634 
2635 /**
2636  * ib_create_ah - Creates an address handle for the given address vector.
2637  * @pd: The protection domain associated with the address handle.
2638  * @ah_attr: The attributes of the address vector.
2639  *
2640  * The address handle is used to reference a local or global destination
2641  * in all UD QP post sends.
2642  */
2643 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2644 
2645 /**
2646  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2647  *   work completion.
2648  * @hdr: the L3 header to parse
2649  * @net_type: type of header to parse
2650  * @sgid: place to store source gid
2651  * @dgid: place to store destination gid
2652  */
2653 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2654 			      enum rdma_network_type net_type,
2655 			      union ib_gid *sgid, union ib_gid *dgid);
2656 
2657 /**
2658  * ib_get_rdma_header_version - Get the header version
2659  * @hdr: the L3 header to parse
2660  */
2661 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2662 
2663 /**
2664  * ib_init_ah_from_wc - Initializes address handle attributes from a
2665  *   work completion.
2666  * @device: Device on which the received message arrived.
2667  * @port_num: Port on which the received message arrived.
2668  * @wc: Work completion associated with the received message.
2669  * @grh: References the received global route header.  This parameter is
2670  *   ignored unless the work completion indicates that the GRH is valid.
2671  * @ah_attr: Returned attributes that can be used when creating an address
2672  *   handle for replying to the message.
2673  */
2674 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2675 		       const struct ib_wc *wc, const struct ib_grh *grh,
2676 		       struct ib_ah_attr *ah_attr);
2677 
2678 /**
2679  * ib_create_ah_from_wc - Creates an address handle associated with the
2680  *   sender of the specified work completion.
2681  * @pd: The protection domain associated with the address handle.
2682  * @wc: Work completion information associated with a received message.
2683  * @grh: References the received global route header.  This parameter is
2684  *   ignored unless the work completion indicates that the GRH is valid.
2685  * @port_num: The outbound port number to associate with the address.
2686  *
2687  * The address handle is used to reference a local or global destination
2688  * in all UD QP post sends.
2689  */
2690 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2691 				   const struct ib_grh *grh, u8 port_num);
2692 
2693 /**
2694  * ib_modify_ah - Modifies the address vector associated with an address
2695  *   handle.
2696  * @ah: The address handle to modify.
2697  * @ah_attr: The new address vector attributes to associate with the
2698  *   address handle.
2699  */
2700 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2701 
2702 /**
2703  * ib_query_ah - Queries the address vector associated with an address
2704  *   handle.
2705  * @ah: The address handle to query.
2706  * @ah_attr: The address vector attributes associated with the address
2707  *   handle.
2708  */
2709 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2710 
2711 /**
2712  * ib_destroy_ah - Destroys an address handle.
2713  * @ah: The address handle to destroy.
2714  */
2715 int ib_destroy_ah(struct ib_ah *ah);
2716 
2717 /**
2718  * ib_create_srq - Creates a SRQ associated with the specified protection
2719  *   domain.
2720  * @pd: The protection domain associated with the SRQ.
2721  * @srq_init_attr: A list of initial attributes required to create the
2722  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2723  *   the actual capabilities of the created SRQ.
2724  *
2725  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2726  * requested size of the SRQ, and set to the actual values allocated
2727  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2728  * will always be at least as large as the requested values.
2729  */
2730 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2731 			     struct ib_srq_init_attr *srq_init_attr);
2732 
2733 /**
2734  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2735  * @srq: The SRQ to modify.
2736  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2737  *   the current values of selected SRQ attributes are returned.
2738  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2739  *   are being modified.
2740  *
2741  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2742  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2743  * the number of receives queued drops below the limit.
2744  */
2745 int ib_modify_srq(struct ib_srq *srq,
2746 		  struct ib_srq_attr *srq_attr,
2747 		  enum ib_srq_attr_mask srq_attr_mask);
2748 
2749 /**
2750  * ib_query_srq - Returns the attribute list and current values for the
2751  *   specified SRQ.
2752  * @srq: The SRQ to query.
2753  * @srq_attr: The attributes of the specified SRQ.
2754  */
2755 int ib_query_srq(struct ib_srq *srq,
2756 		 struct ib_srq_attr *srq_attr);
2757 
2758 /**
2759  * ib_destroy_srq - Destroys the specified SRQ.
2760  * @srq: The SRQ to destroy.
2761  */
2762 int ib_destroy_srq(struct ib_srq *srq);
2763 
2764 /**
2765  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2766  * @srq: The SRQ to post the work request on.
2767  * @recv_wr: A list of work requests to post on the receive queue.
2768  * @bad_recv_wr: On an immediate failure, this parameter will reference
2769  *   the work request that failed to be posted on the QP.
2770  */
2771 static inline int ib_post_srq_recv(struct ib_srq *srq,
2772 				   struct ib_recv_wr *recv_wr,
2773 				   struct ib_recv_wr **bad_recv_wr)
2774 {
2775 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2776 }
2777 
2778 /**
2779  * ib_create_qp - Creates a QP associated with the specified protection
2780  *   domain.
2781  * @pd: The protection domain associated with the QP.
2782  * @qp_init_attr: A list of initial attributes required to create the
2783  *   QP.  If QP creation succeeds, then the attributes are updated to
2784  *   the actual capabilities of the created QP.
2785  */
2786 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2787 			   struct ib_qp_init_attr *qp_init_attr);
2788 
2789 /**
2790  * ib_modify_qp - Modifies the attributes for the specified QP and then
2791  *   transitions the QP to the given state.
2792  * @qp: The QP to modify.
2793  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2794  *   the current values of selected QP attributes are returned.
2795  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2796  *   are being modified.
2797  */
2798 int ib_modify_qp(struct ib_qp *qp,
2799 		 struct ib_qp_attr *qp_attr,
2800 		 int qp_attr_mask);
2801 
2802 /**
2803  * ib_query_qp - Returns the attribute list and current values for the
2804  *   specified QP.
2805  * @qp: The QP to query.
2806  * @qp_attr: The attributes of the specified QP.
2807  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2808  * @qp_init_attr: Additional attributes of the selected QP.
2809  *
2810  * The qp_attr_mask may be used to limit the query to gathering only the
2811  * selected attributes.
2812  */
2813 int ib_query_qp(struct ib_qp *qp,
2814 		struct ib_qp_attr *qp_attr,
2815 		int qp_attr_mask,
2816 		struct ib_qp_init_attr *qp_init_attr);
2817 
2818 /**
2819  * ib_destroy_qp - Destroys the specified QP.
2820  * @qp: The QP to destroy.
2821  */
2822 int ib_destroy_qp(struct ib_qp *qp);
2823 
2824 /**
2825  * ib_open_qp - Obtain a reference to an existing sharable QP.
2826  * @xrcd - XRC domain
2827  * @qp_open_attr: Attributes identifying the QP to open.
2828  *
2829  * Returns a reference to a sharable QP.
2830  */
2831 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2832 			 struct ib_qp_open_attr *qp_open_attr);
2833 
2834 /**
2835  * ib_close_qp - Release an external reference to a QP.
2836  * @qp: The QP handle to release
2837  *
2838  * The opened QP handle is released by the caller.  The underlying
2839  * shared QP is not destroyed until all internal references are released.
2840  */
2841 int ib_close_qp(struct ib_qp *qp);
2842 
2843 /**
2844  * ib_post_send - Posts a list of work requests to the send queue of
2845  *   the specified QP.
2846  * @qp: The QP to post the work request on.
2847  * @send_wr: A list of work requests to post on the send queue.
2848  * @bad_send_wr: On an immediate failure, this parameter will reference
2849  *   the work request that failed to be posted on the QP.
2850  *
2851  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2852  * error is returned, the QP state shall not be affected,
2853  * ib_post_send() will return an immediate error after queueing any
2854  * earlier work requests in the list.
2855  */
2856 static inline int ib_post_send(struct ib_qp *qp,
2857 			       struct ib_send_wr *send_wr,
2858 			       struct ib_send_wr **bad_send_wr)
2859 {
2860 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2861 }
2862 
2863 /**
2864  * ib_post_recv - Posts a list of work requests to the receive queue of
2865  *   the specified QP.
2866  * @qp: The QP to post the work request on.
2867  * @recv_wr: A list of work requests to post on the receive queue.
2868  * @bad_recv_wr: On an immediate failure, this parameter will reference
2869  *   the work request that failed to be posted on the QP.
2870  */
2871 static inline int ib_post_recv(struct ib_qp *qp,
2872 			       struct ib_recv_wr *recv_wr,
2873 			       struct ib_recv_wr **bad_recv_wr)
2874 {
2875 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2876 }
2877 
2878 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2879 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2880 void ib_free_cq(struct ib_cq *cq);
2881 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2882 
2883 /**
2884  * ib_create_cq - Creates a CQ on the specified device.
2885  * @device: The device on which to create the CQ.
2886  * @comp_handler: A user-specified callback that is invoked when a
2887  *   completion event occurs on the CQ.
2888  * @event_handler: A user-specified callback that is invoked when an
2889  *   asynchronous event not associated with a completion occurs on the CQ.
2890  * @cq_context: Context associated with the CQ returned to the user via
2891  *   the associated completion and event handlers.
2892  * @cq_attr: The attributes the CQ should be created upon.
2893  *
2894  * Users can examine the cq structure to determine the actual CQ size.
2895  */
2896 struct ib_cq *ib_create_cq(struct ib_device *device,
2897 			   ib_comp_handler comp_handler,
2898 			   void (*event_handler)(struct ib_event *, void *),
2899 			   void *cq_context,
2900 			   const struct ib_cq_init_attr *cq_attr);
2901 
2902 /**
2903  * ib_resize_cq - Modifies the capacity of the CQ.
2904  * @cq: The CQ to resize.
2905  * @cqe: The minimum size of the CQ.
2906  *
2907  * Users can examine the cq structure to determine the actual CQ size.
2908  */
2909 int ib_resize_cq(struct ib_cq *cq, int cqe);
2910 
2911 /**
2912  * ib_modify_cq - Modifies moderation params of the CQ
2913  * @cq: The CQ to modify.
2914  * @cq_count: number of CQEs that will trigger an event
2915  * @cq_period: max period of time in usec before triggering an event
2916  *
2917  */
2918 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2919 
2920 /**
2921  * ib_destroy_cq - Destroys the specified CQ.
2922  * @cq: The CQ to destroy.
2923  */
2924 int ib_destroy_cq(struct ib_cq *cq);
2925 
2926 /**
2927  * ib_poll_cq - poll a CQ for completion(s)
2928  * @cq:the CQ being polled
2929  * @num_entries:maximum number of completions to return
2930  * @wc:array of at least @num_entries &struct ib_wc where completions
2931  *   will be returned
2932  *
2933  * Poll a CQ for (possibly multiple) completions.  If the return value
2934  * is < 0, an error occurred.  If the return value is >= 0, it is the
2935  * number of completions returned.  If the return value is
2936  * non-negative and < num_entries, then the CQ was emptied.
2937  */
2938 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2939 			     struct ib_wc *wc)
2940 {
2941 	return cq->device->poll_cq(cq, num_entries, wc);
2942 }
2943 
2944 /**
2945  * ib_peek_cq - Returns the number of unreaped completions currently
2946  *   on the specified CQ.
2947  * @cq: The CQ to peek.
2948  * @wc_cnt: A minimum number of unreaped completions to check for.
2949  *
2950  * If the number of unreaped completions is greater than or equal to wc_cnt,
2951  * this function returns wc_cnt, otherwise, it returns the actual number of
2952  * unreaped completions.
2953  */
2954 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2955 
2956 /**
2957  * ib_req_notify_cq - Request completion notification on a CQ.
2958  * @cq: The CQ to generate an event for.
2959  * @flags:
2960  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2961  *   to request an event on the next solicited event or next work
2962  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2963  *   may also be |ed in to request a hint about missed events, as
2964  *   described below.
2965  *
2966  * Return Value:
2967  *    < 0 means an error occurred while requesting notification
2968  *   == 0 means notification was requested successfully, and if
2969  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2970  *        were missed and it is safe to wait for another event.  In
2971  *        this case is it guaranteed that any work completions added
2972  *        to the CQ since the last CQ poll will trigger a completion
2973  *        notification event.
2974  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2975  *        in.  It means that the consumer must poll the CQ again to
2976  *        make sure it is empty to avoid missing an event because of a
2977  *        race between requesting notification and an entry being
2978  *        added to the CQ.  This return value means it is possible
2979  *        (but not guaranteed) that a work completion has been added
2980  *        to the CQ since the last poll without triggering a
2981  *        completion notification event.
2982  */
2983 static inline int ib_req_notify_cq(struct ib_cq *cq,
2984 				   enum ib_cq_notify_flags flags)
2985 {
2986 	return cq->device->req_notify_cq(cq, flags);
2987 }
2988 
2989 /**
2990  * ib_req_ncomp_notif - Request completion notification when there are
2991  *   at least the specified number of unreaped completions on the CQ.
2992  * @cq: The CQ to generate an event for.
2993  * @wc_cnt: The number of unreaped completions that should be on the
2994  *   CQ before an event is generated.
2995  */
2996 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2997 {
2998 	return cq->device->req_ncomp_notif ?
2999 		cq->device->req_ncomp_notif(cq, wc_cnt) :
3000 		-ENOSYS;
3001 }
3002 
3003 /**
3004  * ib_dma_mapping_error - check a DMA addr for error
3005  * @dev: The device for which the dma_addr was created
3006  * @dma_addr: The DMA address to check
3007  */
3008 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3009 {
3010 	return dma_mapping_error(&dev->dev, dma_addr);
3011 }
3012 
3013 /**
3014  * ib_dma_map_single - Map a kernel virtual address to DMA address
3015  * @dev: The device for which the dma_addr is to be created
3016  * @cpu_addr: The kernel virtual address
3017  * @size: The size of the region in bytes
3018  * @direction: The direction of the DMA
3019  */
3020 static inline u64 ib_dma_map_single(struct ib_device *dev,
3021 				    void *cpu_addr, size_t size,
3022 				    enum dma_data_direction direction)
3023 {
3024 	return dma_map_single(&dev->dev, cpu_addr, size, direction);
3025 }
3026 
3027 /**
3028  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3029  * @dev: The device for which the DMA address was created
3030  * @addr: The DMA address
3031  * @size: The size of the region in bytes
3032  * @direction: The direction of the DMA
3033  */
3034 static inline void ib_dma_unmap_single(struct ib_device *dev,
3035 				       u64 addr, size_t size,
3036 				       enum dma_data_direction direction)
3037 {
3038 	dma_unmap_single(&dev->dev, addr, size, direction);
3039 }
3040 
3041 /**
3042  * ib_dma_map_page - Map a physical page to DMA address
3043  * @dev: The device for which the dma_addr is to be created
3044  * @page: The page to be mapped
3045  * @offset: The offset within the page
3046  * @size: The size of the region in bytes
3047  * @direction: The direction of the DMA
3048  */
3049 static inline u64 ib_dma_map_page(struct ib_device *dev,
3050 				  struct page *page,
3051 				  unsigned long offset,
3052 				  size_t size,
3053 					 enum dma_data_direction direction)
3054 {
3055 	return dma_map_page(&dev->dev, page, offset, size, direction);
3056 }
3057 
3058 /**
3059  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3060  * @dev: The device for which the DMA address was created
3061  * @addr: The DMA address
3062  * @size: The size of the region in bytes
3063  * @direction: The direction of the DMA
3064  */
3065 static inline void ib_dma_unmap_page(struct ib_device *dev,
3066 				     u64 addr, size_t size,
3067 				     enum dma_data_direction direction)
3068 {
3069 	dma_unmap_page(&dev->dev, addr, size, direction);
3070 }
3071 
3072 /**
3073  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3074  * @dev: The device for which the DMA addresses are to be created
3075  * @sg: The array of scatter/gather entries
3076  * @nents: The number of scatter/gather entries
3077  * @direction: The direction of the DMA
3078  */
3079 static inline int ib_dma_map_sg(struct ib_device *dev,
3080 				struct scatterlist *sg, int nents,
3081 				enum dma_data_direction direction)
3082 {
3083 	return dma_map_sg(&dev->dev, sg, nents, direction);
3084 }
3085 
3086 /**
3087  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3088  * @dev: The device for which the DMA addresses were created
3089  * @sg: The array of scatter/gather entries
3090  * @nents: The number of scatter/gather entries
3091  * @direction: The direction of the DMA
3092  */
3093 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3094 				   struct scatterlist *sg, int nents,
3095 				   enum dma_data_direction direction)
3096 {
3097 	dma_unmap_sg(&dev->dev, sg, nents, direction);
3098 }
3099 
3100 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3101 				      struct scatterlist *sg, int nents,
3102 				      enum dma_data_direction direction,
3103 				      unsigned long dma_attrs)
3104 {
3105 	return dma_map_sg_attrs(&dev->dev, sg, nents, direction, dma_attrs);
3106 }
3107 
3108 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3109 					 struct scatterlist *sg, int nents,
3110 					 enum dma_data_direction direction,
3111 					 unsigned long dma_attrs)
3112 {
3113 	dma_unmap_sg_attrs(&dev->dev, sg, nents, direction, dma_attrs);
3114 }
3115 /**
3116  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3117  * @dev: The device for which the DMA addresses were created
3118  * @sg: The scatter/gather entry
3119  *
3120  * Note: this function is obsolete. To do: change all occurrences of
3121  * ib_sg_dma_address() into sg_dma_address().
3122  */
3123 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3124 				    struct scatterlist *sg)
3125 {
3126 	return sg_dma_address(sg);
3127 }
3128 
3129 /**
3130  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3131  * @dev: The device for which the DMA addresses were created
3132  * @sg: The scatter/gather entry
3133  *
3134  * Note: this function is obsolete. To do: change all occurrences of
3135  * ib_sg_dma_len() into sg_dma_len().
3136  */
3137 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3138 					 struct scatterlist *sg)
3139 {
3140 	return sg_dma_len(sg);
3141 }
3142 
3143 /**
3144  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3145  * @dev: The device for which the DMA address was created
3146  * @addr: The DMA address
3147  * @size: The size of the region in bytes
3148  * @dir: The direction of the DMA
3149  */
3150 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3151 					      u64 addr,
3152 					      size_t size,
3153 					      enum dma_data_direction dir)
3154 {
3155 	dma_sync_single_for_cpu(&dev->dev, addr, size, dir);
3156 }
3157 
3158 /**
3159  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3160  * @dev: The device for which the DMA address was created
3161  * @addr: The DMA address
3162  * @size: The size of the region in bytes
3163  * @dir: The direction of the DMA
3164  */
3165 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3166 						 u64 addr,
3167 						 size_t size,
3168 						 enum dma_data_direction dir)
3169 {
3170 	dma_sync_single_for_device(&dev->dev, addr, size, dir);
3171 }
3172 
3173 /**
3174  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3175  * @dev: The device for which the DMA address is requested
3176  * @size: The size of the region to allocate in bytes
3177  * @dma_handle: A pointer for returning the DMA address of the region
3178  * @flag: memory allocator flags
3179  */
3180 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3181 					   size_t size,
3182 					   dma_addr_t *dma_handle,
3183 					   gfp_t flag)
3184 {
3185 	return dma_alloc_coherent(&dev->dev, size, dma_handle, flag);
3186 }
3187 
3188 /**
3189  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3190  * @dev: The device for which the DMA addresses were allocated
3191  * @size: The size of the region
3192  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3193  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3194  */
3195 static inline void ib_dma_free_coherent(struct ib_device *dev,
3196 					size_t size, void *cpu_addr,
3197 					dma_addr_t dma_handle)
3198 {
3199 	dma_free_coherent(&dev->dev, size, cpu_addr, dma_handle);
3200 }
3201 
3202 /**
3203  * ib_dereg_mr - Deregisters a memory region and removes it from the
3204  *   HCA translation table.
3205  * @mr: The memory region to deregister.
3206  *
3207  * This function can fail, if the memory region has memory windows bound to it.
3208  */
3209 int ib_dereg_mr(struct ib_mr *mr);
3210 
3211 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3212 			  enum ib_mr_type mr_type,
3213 			  u32 max_num_sg);
3214 
3215 /**
3216  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3217  *   R_Key and L_Key.
3218  * @mr - struct ib_mr pointer to be updated.
3219  * @newkey - new key to be used.
3220  */
3221 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3222 {
3223 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3224 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3225 }
3226 
3227 /**
3228  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3229  * for calculating a new rkey for type 2 memory windows.
3230  * @rkey - the rkey to increment.
3231  */
3232 static inline u32 ib_inc_rkey(u32 rkey)
3233 {
3234 	const u32 mask = 0x000000ff;
3235 	return ((rkey + 1) & mask) | (rkey & ~mask);
3236 }
3237 
3238 /**
3239  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3240  * @pd: The protection domain associated with the unmapped region.
3241  * @mr_access_flags: Specifies the memory access rights.
3242  * @fmr_attr: Attributes of the unmapped region.
3243  *
3244  * A fast memory region must be mapped before it can be used as part of
3245  * a work request.
3246  */
3247 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3248 			    int mr_access_flags,
3249 			    struct ib_fmr_attr *fmr_attr);
3250 
3251 /**
3252  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3253  * @fmr: The fast memory region to associate with the pages.
3254  * @page_list: An array of physical pages to map to the fast memory region.
3255  * @list_len: The number of pages in page_list.
3256  * @iova: The I/O virtual address to use with the mapped region.
3257  */
3258 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3259 				  u64 *page_list, int list_len,
3260 				  u64 iova)
3261 {
3262 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3263 }
3264 
3265 /**
3266  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3267  * @fmr_list: A linked list of fast memory regions to unmap.
3268  */
3269 int ib_unmap_fmr(struct list_head *fmr_list);
3270 
3271 /**
3272  * ib_dealloc_fmr - Deallocates a fast memory region.
3273  * @fmr: The fast memory region to deallocate.
3274  */
3275 int ib_dealloc_fmr(struct ib_fmr *fmr);
3276 
3277 /**
3278  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3279  * @qp: QP to attach to the multicast group.  The QP must be type
3280  *   IB_QPT_UD.
3281  * @gid: Multicast group GID.
3282  * @lid: Multicast group LID in host byte order.
3283  *
3284  * In order to send and receive multicast packets, subnet
3285  * administration must have created the multicast group and configured
3286  * the fabric appropriately.  The port associated with the specified
3287  * QP must also be a member of the multicast group.
3288  */
3289 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3290 
3291 /**
3292  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3293  * @qp: QP to detach from the multicast group.
3294  * @gid: Multicast group GID.
3295  * @lid: Multicast group LID in host byte order.
3296  */
3297 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3298 
3299 /**
3300  * ib_alloc_xrcd - Allocates an XRC domain.
3301  * @device: The device on which to allocate the XRC domain.
3302  */
3303 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3304 
3305 /**
3306  * ib_dealloc_xrcd - Deallocates an XRC domain.
3307  * @xrcd: The XRC domain to deallocate.
3308  */
3309 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3310 
3311 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3312 			       struct ib_flow_attr *flow_attr, int domain);
3313 int ib_destroy_flow(struct ib_flow *flow_id);
3314 
3315 static inline int ib_check_mr_access(int flags)
3316 {
3317 	/*
3318 	 * Local write permission is required if remote write or
3319 	 * remote atomic permission is also requested.
3320 	 */
3321 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3322 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3323 		return -EINVAL;
3324 
3325 	return 0;
3326 }
3327 
3328 /**
3329  * ib_check_mr_status: lightweight check of MR status.
3330  *     This routine may provide status checks on a selected
3331  *     ib_mr. first use is for signature status check.
3332  *
3333  * @mr: A memory region.
3334  * @check_mask: Bitmask of which checks to perform from
3335  *     ib_mr_status_check enumeration.
3336  * @mr_status: The container of relevant status checks.
3337  *     failed checks will be indicated in the status bitmask
3338  *     and the relevant info shall be in the error item.
3339  */
3340 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3341 		       struct ib_mr_status *mr_status);
3342 
3343 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3344 					    u16 pkey, const union ib_gid *gid,
3345 					    const struct sockaddr *addr);
3346 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3347 			   struct ib_wq_init_attr *init_attr);
3348 int ib_destroy_wq(struct ib_wq *wq);
3349 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3350 		 u32 wq_attr_mask);
3351 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3352 						 struct ib_rwq_ind_table_init_attr*
3353 						 wq_ind_table_init_attr);
3354 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3355 
3356 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3357 		 unsigned int *sg_offset, unsigned int page_size);
3358 
3359 static inline int
3360 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3361 		  unsigned int *sg_offset, unsigned int page_size)
3362 {
3363 	int n;
3364 
3365 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3366 	mr->iova = 0;
3367 
3368 	return n;
3369 }
3370 
3371 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3372 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3373 
3374 void ib_drain_rq(struct ib_qp *qp);
3375 void ib_drain_sq(struct ib_qp *qp);
3376 void ib_drain_qp(struct ib_qp *qp);
3377 
3378 int ib_resolve_eth_dmac(struct ib_device *device,
3379 			struct ib_ah_attr *ah_attr);
3380 #endif /* IB_VERBS_H */
3381