xref: /linux/include/rdma/ib_verbs.h (revision c1aac62f36c1e37ee81c9e09ee9ee733eef05dcb)
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 
64 extern struct workqueue_struct *ib_wq;
65 extern struct workqueue_struct *ib_comp_wq;
66 
67 union ib_gid {
68 	u8	raw[16];
69 	struct {
70 		__be64	subnet_prefix;
71 		__be64	interface_id;
72 	} global;
73 };
74 
75 extern union ib_gid zgid;
76 
77 enum ib_gid_type {
78 	/* If link layer is Ethernet, this is RoCE V1 */
79 	IB_GID_TYPE_IB        = 0,
80 	IB_GID_TYPE_ROCE      = 0,
81 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
82 	IB_GID_TYPE_SIZE
83 };
84 
85 #define ROCE_V2_UDP_DPORT      4791
86 struct ib_gid_attr {
87 	enum ib_gid_type	gid_type;
88 	struct net_device	*ndev;
89 };
90 
91 enum rdma_node_type {
92 	/* IB values map to NodeInfo:NodeType. */
93 	RDMA_NODE_IB_CA 	= 1,
94 	RDMA_NODE_IB_SWITCH,
95 	RDMA_NODE_IB_ROUTER,
96 	RDMA_NODE_RNIC,
97 	RDMA_NODE_USNIC,
98 	RDMA_NODE_USNIC_UDP,
99 };
100 
101 enum {
102 	/* set the local administered indication */
103 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
104 };
105 
106 enum rdma_transport_type {
107 	RDMA_TRANSPORT_IB,
108 	RDMA_TRANSPORT_IWARP,
109 	RDMA_TRANSPORT_USNIC,
110 	RDMA_TRANSPORT_USNIC_UDP
111 };
112 
113 enum rdma_protocol_type {
114 	RDMA_PROTOCOL_IB,
115 	RDMA_PROTOCOL_IBOE,
116 	RDMA_PROTOCOL_IWARP,
117 	RDMA_PROTOCOL_USNIC_UDP
118 };
119 
120 __attribute_const__ enum rdma_transport_type
121 rdma_node_get_transport(enum rdma_node_type node_type);
122 
123 enum rdma_network_type {
124 	RDMA_NETWORK_IB,
125 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
126 	RDMA_NETWORK_IPV4,
127 	RDMA_NETWORK_IPV6
128 };
129 
130 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
131 {
132 	if (network_type == RDMA_NETWORK_IPV4 ||
133 	    network_type == RDMA_NETWORK_IPV6)
134 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
135 
136 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
137 	return IB_GID_TYPE_IB;
138 }
139 
140 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
141 							    union ib_gid *gid)
142 {
143 	if (gid_type == IB_GID_TYPE_IB)
144 		return RDMA_NETWORK_IB;
145 
146 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
147 		return RDMA_NETWORK_IPV4;
148 	else
149 		return RDMA_NETWORK_IPV6;
150 }
151 
152 enum rdma_link_layer {
153 	IB_LINK_LAYER_UNSPECIFIED,
154 	IB_LINK_LAYER_INFINIBAND,
155 	IB_LINK_LAYER_ETHERNET,
156 };
157 
158 enum ib_device_cap_flags {
159 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
160 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
161 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
162 	IB_DEVICE_RAW_MULTI			= (1 << 3),
163 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
164 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
165 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
166 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
167 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
168 	IB_DEVICE_INIT_TYPE			= (1 << 9),
169 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
170 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
171 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
172 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
173 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
174 
175 	/*
176 	 * This device supports a per-device lkey or stag that can be
177 	 * used without performing a memory registration for the local
178 	 * memory.  Note that ULPs should never check this flag, but
179 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
180 	 * which will always contain a usable lkey.
181 	 */
182 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
183 	IB_DEVICE_RESERVED /* old SEND_W_INV */	= (1 << 16),
184 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
185 	/*
186 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
187 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
188 	 * messages and can verify the validity of checksum for
189 	 * incoming messages.  Setting this flag implies that the
190 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
191 	 */
192 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
193 	IB_DEVICE_UD_TSO			= (1 << 19),
194 	IB_DEVICE_XRC				= (1 << 20),
195 
196 	/*
197 	 * This device supports the IB "base memory management extension",
198 	 * which includes support for fast registrations (IB_WR_REG_MR,
199 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
200 	 * also be set by any iWarp device which must support FRs to comply
201 	 * to the iWarp verbs spec.  iWarp devices also support the
202 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
203 	 * stag.
204 	 */
205 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
206 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
207 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
208 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
209 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
210 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
211 	/*
212 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
213 	 * support execution of WQEs that involve synchronization
214 	 * of I/O operations with single completion queue managed
215 	 * by hardware.
216 	 */
217 	IB_DEVICE_CROSS_CHANNEL		= (1 << 27),
218 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
219 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
220 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
221 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
222 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
223 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
224 };
225 
226 enum ib_signature_prot_cap {
227 	IB_PROT_T10DIF_TYPE_1 = 1,
228 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
229 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
230 };
231 
232 enum ib_signature_guard_cap {
233 	IB_GUARD_T10DIF_CRC	= 1,
234 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
235 };
236 
237 enum ib_atomic_cap {
238 	IB_ATOMIC_NONE,
239 	IB_ATOMIC_HCA,
240 	IB_ATOMIC_GLOB
241 };
242 
243 enum ib_odp_general_cap_bits {
244 	IB_ODP_SUPPORT = 1 << 0,
245 };
246 
247 enum ib_odp_transport_cap_bits {
248 	IB_ODP_SUPPORT_SEND	= 1 << 0,
249 	IB_ODP_SUPPORT_RECV	= 1 << 1,
250 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
251 	IB_ODP_SUPPORT_READ	= 1 << 3,
252 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
253 };
254 
255 struct ib_odp_caps {
256 	uint64_t general_caps;
257 	struct {
258 		uint32_t  rc_odp_caps;
259 		uint32_t  uc_odp_caps;
260 		uint32_t  ud_odp_caps;
261 	} per_transport_caps;
262 };
263 
264 struct ib_rss_caps {
265 	/* Corresponding bit will be set if qp type from
266 	 * 'enum ib_qp_type' is supported, e.g.
267 	 * supported_qpts |= 1 << IB_QPT_UD
268 	 */
269 	u32 supported_qpts;
270 	u32 max_rwq_indirection_tables;
271 	u32 max_rwq_indirection_table_size;
272 };
273 
274 enum ib_cq_creation_flags {
275 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
276 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
277 };
278 
279 struct ib_cq_init_attr {
280 	unsigned int	cqe;
281 	int		comp_vector;
282 	u32		flags;
283 };
284 
285 struct ib_device_attr {
286 	u64			fw_ver;
287 	__be64			sys_image_guid;
288 	u64			max_mr_size;
289 	u64			page_size_cap;
290 	u32			vendor_id;
291 	u32			vendor_part_id;
292 	u32			hw_ver;
293 	int			max_qp;
294 	int			max_qp_wr;
295 	u64			device_cap_flags;
296 	int			max_sge;
297 	int			max_sge_rd;
298 	int			max_cq;
299 	int			max_cqe;
300 	int			max_mr;
301 	int			max_pd;
302 	int			max_qp_rd_atom;
303 	int			max_ee_rd_atom;
304 	int			max_res_rd_atom;
305 	int			max_qp_init_rd_atom;
306 	int			max_ee_init_rd_atom;
307 	enum ib_atomic_cap	atomic_cap;
308 	enum ib_atomic_cap	masked_atomic_cap;
309 	int			max_ee;
310 	int			max_rdd;
311 	int			max_mw;
312 	int			max_raw_ipv6_qp;
313 	int			max_raw_ethy_qp;
314 	int			max_mcast_grp;
315 	int			max_mcast_qp_attach;
316 	int			max_total_mcast_qp_attach;
317 	int			max_ah;
318 	int			max_fmr;
319 	int			max_map_per_fmr;
320 	int			max_srq;
321 	int			max_srq_wr;
322 	int			max_srq_sge;
323 	unsigned int		max_fast_reg_page_list_len;
324 	u16			max_pkeys;
325 	u8			local_ca_ack_delay;
326 	int			sig_prot_cap;
327 	int			sig_guard_cap;
328 	struct ib_odp_caps	odp_caps;
329 	uint64_t		timestamp_mask;
330 	uint64_t		hca_core_clock; /* in KHZ */
331 	struct ib_rss_caps	rss_caps;
332 	u32			max_wq_type_rq;
333 };
334 
335 enum ib_mtu {
336 	IB_MTU_256  = 1,
337 	IB_MTU_512  = 2,
338 	IB_MTU_1024 = 3,
339 	IB_MTU_2048 = 4,
340 	IB_MTU_4096 = 5
341 };
342 
343 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
344 {
345 	switch (mtu) {
346 	case IB_MTU_256:  return  256;
347 	case IB_MTU_512:  return  512;
348 	case IB_MTU_1024: return 1024;
349 	case IB_MTU_2048: return 2048;
350 	case IB_MTU_4096: return 4096;
351 	default: 	  return -1;
352 	}
353 }
354 
355 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
356 {
357 	if (mtu >= 4096)
358 		return IB_MTU_4096;
359 	else if (mtu >= 2048)
360 		return IB_MTU_2048;
361 	else if (mtu >= 1024)
362 		return IB_MTU_1024;
363 	else if (mtu >= 512)
364 		return IB_MTU_512;
365 	else
366 		return IB_MTU_256;
367 }
368 
369 enum ib_port_state {
370 	IB_PORT_NOP		= 0,
371 	IB_PORT_DOWN		= 1,
372 	IB_PORT_INIT		= 2,
373 	IB_PORT_ARMED		= 3,
374 	IB_PORT_ACTIVE		= 4,
375 	IB_PORT_ACTIVE_DEFER	= 5
376 };
377 
378 enum ib_port_cap_flags {
379 	IB_PORT_SM				= 1 <<  1,
380 	IB_PORT_NOTICE_SUP			= 1 <<  2,
381 	IB_PORT_TRAP_SUP			= 1 <<  3,
382 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
383 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
384 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
385 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
386 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
387 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
388 	IB_PORT_SM_DISABLED			= 1 << 10,
389 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
390 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
391 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
392 	IB_PORT_CM_SUP				= 1 << 16,
393 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
394 	IB_PORT_REINIT_SUP			= 1 << 18,
395 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
396 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
397 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
398 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
399 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
400 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
401 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
402 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
403 };
404 
405 enum ib_port_width {
406 	IB_WIDTH_1X	= 1,
407 	IB_WIDTH_4X	= 2,
408 	IB_WIDTH_8X	= 4,
409 	IB_WIDTH_12X	= 8
410 };
411 
412 static inline int ib_width_enum_to_int(enum ib_port_width width)
413 {
414 	switch (width) {
415 	case IB_WIDTH_1X:  return  1;
416 	case IB_WIDTH_4X:  return  4;
417 	case IB_WIDTH_8X:  return  8;
418 	case IB_WIDTH_12X: return 12;
419 	default: 	  return -1;
420 	}
421 }
422 
423 enum ib_port_speed {
424 	IB_SPEED_SDR	= 1,
425 	IB_SPEED_DDR	= 2,
426 	IB_SPEED_QDR	= 4,
427 	IB_SPEED_FDR10	= 8,
428 	IB_SPEED_FDR	= 16,
429 	IB_SPEED_EDR	= 32
430 };
431 
432 /**
433  * struct rdma_hw_stats
434  * @timestamp - Used by the core code to track when the last update was
435  * @lifespan - Used by the core code to determine how old the counters
436  *   should be before being updated again.  Stored in jiffies, defaults
437  *   to 10 milliseconds, drivers can override the default be specifying
438  *   their own value during their allocation routine.
439  * @name - Array of pointers to static names used for the counters in
440  *   directory.
441  * @num_counters - How many hardware counters there are.  If name is
442  *   shorter than this number, a kernel oops will result.  Driver authors
443  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
444  *   in their code to prevent this.
445  * @value - Array of u64 counters that are accessed by the sysfs code and
446  *   filled in by the drivers get_stats routine
447  */
448 struct rdma_hw_stats {
449 	unsigned long	timestamp;
450 	unsigned long	lifespan;
451 	const char * const *names;
452 	int		num_counters;
453 	u64		value[];
454 };
455 
456 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
457 /**
458  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
459  *   for drivers.
460  * @names - Array of static const char *
461  * @num_counters - How many elements in array
462  * @lifespan - How many milliseconds between updates
463  */
464 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
465 		const char * const *names, int num_counters,
466 		unsigned long lifespan)
467 {
468 	struct rdma_hw_stats *stats;
469 
470 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
471 			GFP_KERNEL);
472 	if (!stats)
473 		return NULL;
474 	stats->names = names;
475 	stats->num_counters = num_counters;
476 	stats->lifespan = msecs_to_jiffies(lifespan);
477 
478 	return stats;
479 }
480 
481 
482 /* Define bits for the various functionality this port needs to be supported by
483  * the core.
484  */
485 /* Management                           0x00000FFF */
486 #define RDMA_CORE_CAP_IB_MAD            0x00000001
487 #define RDMA_CORE_CAP_IB_SMI            0x00000002
488 #define RDMA_CORE_CAP_IB_CM             0x00000004
489 #define RDMA_CORE_CAP_IW_CM             0x00000008
490 #define RDMA_CORE_CAP_IB_SA             0x00000010
491 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
492 
493 /* Address format                       0x000FF000 */
494 #define RDMA_CORE_CAP_AF_IB             0x00001000
495 #define RDMA_CORE_CAP_ETH_AH            0x00002000
496 
497 /* Protocol                             0xFFF00000 */
498 #define RDMA_CORE_CAP_PROT_IB           0x00100000
499 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
500 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
501 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
502 
503 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
504 					| RDMA_CORE_CAP_IB_MAD \
505 					| RDMA_CORE_CAP_IB_SMI \
506 					| RDMA_CORE_CAP_IB_CM  \
507 					| RDMA_CORE_CAP_IB_SA  \
508 					| RDMA_CORE_CAP_AF_IB)
509 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
510 					| RDMA_CORE_CAP_IB_MAD  \
511 					| RDMA_CORE_CAP_IB_CM   \
512 					| RDMA_CORE_CAP_AF_IB   \
513 					| RDMA_CORE_CAP_ETH_AH)
514 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
515 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
516 					| RDMA_CORE_CAP_IB_MAD  \
517 					| RDMA_CORE_CAP_IB_CM   \
518 					| RDMA_CORE_CAP_AF_IB   \
519 					| RDMA_CORE_CAP_ETH_AH)
520 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
521 					| RDMA_CORE_CAP_IW_CM)
522 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
523 					| RDMA_CORE_CAP_OPA_MAD)
524 
525 struct ib_port_attr {
526 	u64			subnet_prefix;
527 	enum ib_port_state	state;
528 	enum ib_mtu		max_mtu;
529 	enum ib_mtu		active_mtu;
530 	int			gid_tbl_len;
531 	u32			port_cap_flags;
532 	u32			max_msg_sz;
533 	u32			bad_pkey_cntr;
534 	u32			qkey_viol_cntr;
535 	u16			pkey_tbl_len;
536 	u16			lid;
537 	u16			sm_lid;
538 	u8			lmc;
539 	u8			max_vl_num;
540 	u8			sm_sl;
541 	u8			subnet_timeout;
542 	u8			init_type_reply;
543 	u8			active_width;
544 	u8			active_speed;
545 	u8                      phys_state;
546 	bool			grh_required;
547 };
548 
549 enum ib_device_modify_flags {
550 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
551 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
552 };
553 
554 #define IB_DEVICE_NODE_DESC_MAX 64
555 
556 struct ib_device_modify {
557 	u64	sys_image_guid;
558 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
559 };
560 
561 enum ib_port_modify_flags {
562 	IB_PORT_SHUTDOWN		= 1,
563 	IB_PORT_INIT_TYPE		= (1<<2),
564 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
565 };
566 
567 struct ib_port_modify {
568 	u32	set_port_cap_mask;
569 	u32	clr_port_cap_mask;
570 	u8	init_type;
571 };
572 
573 enum ib_event_type {
574 	IB_EVENT_CQ_ERR,
575 	IB_EVENT_QP_FATAL,
576 	IB_EVENT_QP_REQ_ERR,
577 	IB_EVENT_QP_ACCESS_ERR,
578 	IB_EVENT_COMM_EST,
579 	IB_EVENT_SQ_DRAINED,
580 	IB_EVENT_PATH_MIG,
581 	IB_EVENT_PATH_MIG_ERR,
582 	IB_EVENT_DEVICE_FATAL,
583 	IB_EVENT_PORT_ACTIVE,
584 	IB_EVENT_PORT_ERR,
585 	IB_EVENT_LID_CHANGE,
586 	IB_EVENT_PKEY_CHANGE,
587 	IB_EVENT_SM_CHANGE,
588 	IB_EVENT_SRQ_ERR,
589 	IB_EVENT_SRQ_LIMIT_REACHED,
590 	IB_EVENT_QP_LAST_WQE_REACHED,
591 	IB_EVENT_CLIENT_REREGISTER,
592 	IB_EVENT_GID_CHANGE,
593 	IB_EVENT_WQ_FATAL,
594 };
595 
596 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
597 
598 struct ib_event {
599 	struct ib_device	*device;
600 	union {
601 		struct ib_cq	*cq;
602 		struct ib_qp	*qp;
603 		struct ib_srq	*srq;
604 		struct ib_wq	*wq;
605 		u8		port_num;
606 	} element;
607 	enum ib_event_type	event;
608 };
609 
610 struct ib_event_handler {
611 	struct ib_device *device;
612 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
613 	struct list_head  list;
614 };
615 
616 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
617 	do {							\
618 		(_ptr)->device  = _device;			\
619 		(_ptr)->handler = _handler;			\
620 		INIT_LIST_HEAD(&(_ptr)->list);			\
621 	} while (0)
622 
623 struct ib_global_route {
624 	union ib_gid	dgid;
625 	u32		flow_label;
626 	u8		sgid_index;
627 	u8		hop_limit;
628 	u8		traffic_class;
629 };
630 
631 struct ib_grh {
632 	__be32		version_tclass_flow;
633 	__be16		paylen;
634 	u8		next_hdr;
635 	u8		hop_limit;
636 	union ib_gid	sgid;
637 	union ib_gid	dgid;
638 };
639 
640 union rdma_network_hdr {
641 	struct ib_grh ibgrh;
642 	struct {
643 		/* The IB spec states that if it's IPv4, the header
644 		 * is located in the last 20 bytes of the header.
645 		 */
646 		u8		reserved[20];
647 		struct iphdr	roce4grh;
648 	};
649 };
650 
651 enum {
652 	IB_MULTICAST_QPN = 0xffffff
653 };
654 
655 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
656 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
657 
658 enum ib_ah_flags {
659 	IB_AH_GRH	= 1
660 };
661 
662 enum ib_rate {
663 	IB_RATE_PORT_CURRENT = 0,
664 	IB_RATE_2_5_GBPS = 2,
665 	IB_RATE_5_GBPS   = 5,
666 	IB_RATE_10_GBPS  = 3,
667 	IB_RATE_20_GBPS  = 6,
668 	IB_RATE_30_GBPS  = 4,
669 	IB_RATE_40_GBPS  = 7,
670 	IB_RATE_60_GBPS  = 8,
671 	IB_RATE_80_GBPS  = 9,
672 	IB_RATE_120_GBPS = 10,
673 	IB_RATE_14_GBPS  = 11,
674 	IB_RATE_56_GBPS  = 12,
675 	IB_RATE_112_GBPS = 13,
676 	IB_RATE_168_GBPS = 14,
677 	IB_RATE_25_GBPS  = 15,
678 	IB_RATE_100_GBPS = 16,
679 	IB_RATE_200_GBPS = 17,
680 	IB_RATE_300_GBPS = 18
681 };
682 
683 /**
684  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
685  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
686  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
687  * @rate: rate to convert.
688  */
689 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
690 
691 /**
692  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
693  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
694  * @rate: rate to convert.
695  */
696 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
697 
698 
699 /**
700  * enum ib_mr_type - memory region type
701  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
702  *                            normal registration
703  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
704  *                            signature operations (data-integrity
705  *                            capable regions)
706  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
707  *                            register any arbitrary sg lists (without
708  *                            the normal mr constraints - see
709  *                            ib_map_mr_sg)
710  */
711 enum ib_mr_type {
712 	IB_MR_TYPE_MEM_REG,
713 	IB_MR_TYPE_SIGNATURE,
714 	IB_MR_TYPE_SG_GAPS,
715 };
716 
717 /**
718  * Signature types
719  * IB_SIG_TYPE_NONE: Unprotected.
720  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
721  */
722 enum ib_signature_type {
723 	IB_SIG_TYPE_NONE,
724 	IB_SIG_TYPE_T10_DIF,
725 };
726 
727 /**
728  * Signature T10-DIF block-guard types
729  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
730  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
731  */
732 enum ib_t10_dif_bg_type {
733 	IB_T10DIF_CRC,
734 	IB_T10DIF_CSUM
735 };
736 
737 /**
738  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
739  *     domain.
740  * @bg_type: T10-DIF block guard type (CRC|CSUM)
741  * @pi_interval: protection information interval.
742  * @bg: seed of guard computation.
743  * @app_tag: application tag of guard block
744  * @ref_tag: initial guard block reference tag.
745  * @ref_remap: Indicate wethear the reftag increments each block
746  * @app_escape: Indicate to skip block check if apptag=0xffff
747  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
748  * @apptag_check_mask: check bitmask of application tag.
749  */
750 struct ib_t10_dif_domain {
751 	enum ib_t10_dif_bg_type bg_type;
752 	u16			pi_interval;
753 	u16			bg;
754 	u16			app_tag;
755 	u32			ref_tag;
756 	bool			ref_remap;
757 	bool			app_escape;
758 	bool			ref_escape;
759 	u16			apptag_check_mask;
760 };
761 
762 /**
763  * struct ib_sig_domain - Parameters for signature domain
764  * @sig_type: specific signauture type
765  * @sig: union of all signature domain attributes that may
766  *     be used to set domain layout.
767  */
768 struct ib_sig_domain {
769 	enum ib_signature_type sig_type;
770 	union {
771 		struct ib_t10_dif_domain dif;
772 	} sig;
773 };
774 
775 /**
776  * struct ib_sig_attrs - Parameters for signature handover operation
777  * @check_mask: bitmask for signature byte check (8 bytes)
778  * @mem: memory domain layout desciptor.
779  * @wire: wire domain layout desciptor.
780  */
781 struct ib_sig_attrs {
782 	u8			check_mask;
783 	struct ib_sig_domain	mem;
784 	struct ib_sig_domain	wire;
785 };
786 
787 enum ib_sig_err_type {
788 	IB_SIG_BAD_GUARD,
789 	IB_SIG_BAD_REFTAG,
790 	IB_SIG_BAD_APPTAG,
791 };
792 
793 /**
794  * struct ib_sig_err - signature error descriptor
795  */
796 struct ib_sig_err {
797 	enum ib_sig_err_type	err_type;
798 	u32			expected;
799 	u32			actual;
800 	u64			sig_err_offset;
801 	u32			key;
802 };
803 
804 enum ib_mr_status_check {
805 	IB_MR_CHECK_SIG_STATUS = 1,
806 };
807 
808 /**
809  * struct ib_mr_status - Memory region status container
810  *
811  * @fail_status: Bitmask of MR checks status. For each
812  *     failed check a corresponding status bit is set.
813  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
814  *     failure.
815  */
816 struct ib_mr_status {
817 	u32		    fail_status;
818 	struct ib_sig_err   sig_err;
819 };
820 
821 /**
822  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
823  * enum.
824  * @mult: multiple to convert.
825  */
826 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
827 
828 struct ib_ah_attr {
829 	struct ib_global_route	grh;
830 	u16			dlid;
831 	u8			sl;
832 	u8			src_path_bits;
833 	u8			static_rate;
834 	u8			ah_flags;
835 	u8			port_num;
836 	u8			dmac[ETH_ALEN];
837 };
838 
839 enum ib_wc_status {
840 	IB_WC_SUCCESS,
841 	IB_WC_LOC_LEN_ERR,
842 	IB_WC_LOC_QP_OP_ERR,
843 	IB_WC_LOC_EEC_OP_ERR,
844 	IB_WC_LOC_PROT_ERR,
845 	IB_WC_WR_FLUSH_ERR,
846 	IB_WC_MW_BIND_ERR,
847 	IB_WC_BAD_RESP_ERR,
848 	IB_WC_LOC_ACCESS_ERR,
849 	IB_WC_REM_INV_REQ_ERR,
850 	IB_WC_REM_ACCESS_ERR,
851 	IB_WC_REM_OP_ERR,
852 	IB_WC_RETRY_EXC_ERR,
853 	IB_WC_RNR_RETRY_EXC_ERR,
854 	IB_WC_LOC_RDD_VIOL_ERR,
855 	IB_WC_REM_INV_RD_REQ_ERR,
856 	IB_WC_REM_ABORT_ERR,
857 	IB_WC_INV_EECN_ERR,
858 	IB_WC_INV_EEC_STATE_ERR,
859 	IB_WC_FATAL_ERR,
860 	IB_WC_RESP_TIMEOUT_ERR,
861 	IB_WC_GENERAL_ERR
862 };
863 
864 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
865 
866 enum ib_wc_opcode {
867 	IB_WC_SEND,
868 	IB_WC_RDMA_WRITE,
869 	IB_WC_RDMA_READ,
870 	IB_WC_COMP_SWAP,
871 	IB_WC_FETCH_ADD,
872 	IB_WC_LSO,
873 	IB_WC_LOCAL_INV,
874 	IB_WC_REG_MR,
875 	IB_WC_MASKED_COMP_SWAP,
876 	IB_WC_MASKED_FETCH_ADD,
877 /*
878  * Set value of IB_WC_RECV so consumers can test if a completion is a
879  * receive by testing (opcode & IB_WC_RECV).
880  */
881 	IB_WC_RECV			= 1 << 7,
882 	IB_WC_RECV_RDMA_WITH_IMM
883 };
884 
885 enum ib_wc_flags {
886 	IB_WC_GRH		= 1,
887 	IB_WC_WITH_IMM		= (1<<1),
888 	IB_WC_WITH_INVALIDATE	= (1<<2),
889 	IB_WC_IP_CSUM_OK	= (1<<3),
890 	IB_WC_WITH_SMAC		= (1<<4),
891 	IB_WC_WITH_VLAN		= (1<<5),
892 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
893 };
894 
895 struct ib_wc {
896 	union {
897 		u64		wr_id;
898 		struct ib_cqe	*wr_cqe;
899 	};
900 	enum ib_wc_status	status;
901 	enum ib_wc_opcode	opcode;
902 	u32			vendor_err;
903 	u32			byte_len;
904 	struct ib_qp	       *qp;
905 	union {
906 		__be32		imm_data;
907 		u32		invalidate_rkey;
908 	} ex;
909 	u32			src_qp;
910 	int			wc_flags;
911 	u16			pkey_index;
912 	u16			slid;
913 	u8			sl;
914 	u8			dlid_path_bits;
915 	u8			port_num;	/* valid only for DR SMPs on switches */
916 	u8			smac[ETH_ALEN];
917 	u16			vlan_id;
918 	u8			network_hdr_type;
919 };
920 
921 enum ib_cq_notify_flags {
922 	IB_CQ_SOLICITED			= 1 << 0,
923 	IB_CQ_NEXT_COMP			= 1 << 1,
924 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
925 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
926 };
927 
928 enum ib_srq_type {
929 	IB_SRQT_BASIC,
930 	IB_SRQT_XRC
931 };
932 
933 enum ib_srq_attr_mask {
934 	IB_SRQ_MAX_WR	= 1 << 0,
935 	IB_SRQ_LIMIT	= 1 << 1,
936 };
937 
938 struct ib_srq_attr {
939 	u32	max_wr;
940 	u32	max_sge;
941 	u32	srq_limit;
942 };
943 
944 struct ib_srq_init_attr {
945 	void		      (*event_handler)(struct ib_event *, void *);
946 	void		       *srq_context;
947 	struct ib_srq_attr	attr;
948 	enum ib_srq_type	srq_type;
949 
950 	union {
951 		struct {
952 			struct ib_xrcd *xrcd;
953 			struct ib_cq   *cq;
954 		} xrc;
955 	} ext;
956 };
957 
958 struct ib_qp_cap {
959 	u32	max_send_wr;
960 	u32	max_recv_wr;
961 	u32	max_send_sge;
962 	u32	max_recv_sge;
963 	u32	max_inline_data;
964 
965 	/*
966 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
967 	 * ib_create_qp() will calculate the right amount of neededed WRs
968 	 * and MRs based on this.
969 	 */
970 	u32	max_rdma_ctxs;
971 };
972 
973 enum ib_sig_type {
974 	IB_SIGNAL_ALL_WR,
975 	IB_SIGNAL_REQ_WR
976 };
977 
978 enum ib_qp_type {
979 	/*
980 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
981 	 * here (and in that order) since the MAD layer uses them as
982 	 * indices into a 2-entry table.
983 	 */
984 	IB_QPT_SMI,
985 	IB_QPT_GSI,
986 
987 	IB_QPT_RC,
988 	IB_QPT_UC,
989 	IB_QPT_UD,
990 	IB_QPT_RAW_IPV6,
991 	IB_QPT_RAW_ETHERTYPE,
992 	IB_QPT_RAW_PACKET = 8,
993 	IB_QPT_XRC_INI = 9,
994 	IB_QPT_XRC_TGT,
995 	IB_QPT_MAX,
996 	/* Reserve a range for qp types internal to the low level driver.
997 	 * These qp types will not be visible at the IB core layer, so the
998 	 * IB_QPT_MAX usages should not be affected in the core layer
999 	 */
1000 	IB_QPT_RESERVED1 = 0x1000,
1001 	IB_QPT_RESERVED2,
1002 	IB_QPT_RESERVED3,
1003 	IB_QPT_RESERVED4,
1004 	IB_QPT_RESERVED5,
1005 	IB_QPT_RESERVED6,
1006 	IB_QPT_RESERVED7,
1007 	IB_QPT_RESERVED8,
1008 	IB_QPT_RESERVED9,
1009 	IB_QPT_RESERVED10,
1010 };
1011 
1012 enum ib_qp_create_flags {
1013 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1014 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1015 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1016 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1017 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1018 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1019 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1020 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
1021 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1022 	/* reserve bits 26-31 for low level drivers' internal use */
1023 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1024 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1025 };
1026 
1027 /*
1028  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1029  * callback to destroy the passed in QP.
1030  */
1031 
1032 struct ib_qp_init_attr {
1033 	void                  (*event_handler)(struct ib_event *, void *);
1034 	void		       *qp_context;
1035 	struct ib_cq	       *send_cq;
1036 	struct ib_cq	       *recv_cq;
1037 	struct ib_srq	       *srq;
1038 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1039 	struct ib_qp_cap	cap;
1040 	enum ib_sig_type	sq_sig_type;
1041 	enum ib_qp_type		qp_type;
1042 	enum ib_qp_create_flags	create_flags;
1043 
1044 	/*
1045 	 * Only needed for special QP types, or when using the RW API.
1046 	 */
1047 	u8			port_num;
1048 	struct ib_rwq_ind_table *rwq_ind_tbl;
1049 };
1050 
1051 struct ib_qp_open_attr {
1052 	void                  (*event_handler)(struct ib_event *, void *);
1053 	void		       *qp_context;
1054 	u32			qp_num;
1055 	enum ib_qp_type		qp_type;
1056 };
1057 
1058 enum ib_rnr_timeout {
1059 	IB_RNR_TIMER_655_36 =  0,
1060 	IB_RNR_TIMER_000_01 =  1,
1061 	IB_RNR_TIMER_000_02 =  2,
1062 	IB_RNR_TIMER_000_03 =  3,
1063 	IB_RNR_TIMER_000_04 =  4,
1064 	IB_RNR_TIMER_000_06 =  5,
1065 	IB_RNR_TIMER_000_08 =  6,
1066 	IB_RNR_TIMER_000_12 =  7,
1067 	IB_RNR_TIMER_000_16 =  8,
1068 	IB_RNR_TIMER_000_24 =  9,
1069 	IB_RNR_TIMER_000_32 = 10,
1070 	IB_RNR_TIMER_000_48 = 11,
1071 	IB_RNR_TIMER_000_64 = 12,
1072 	IB_RNR_TIMER_000_96 = 13,
1073 	IB_RNR_TIMER_001_28 = 14,
1074 	IB_RNR_TIMER_001_92 = 15,
1075 	IB_RNR_TIMER_002_56 = 16,
1076 	IB_RNR_TIMER_003_84 = 17,
1077 	IB_RNR_TIMER_005_12 = 18,
1078 	IB_RNR_TIMER_007_68 = 19,
1079 	IB_RNR_TIMER_010_24 = 20,
1080 	IB_RNR_TIMER_015_36 = 21,
1081 	IB_RNR_TIMER_020_48 = 22,
1082 	IB_RNR_TIMER_030_72 = 23,
1083 	IB_RNR_TIMER_040_96 = 24,
1084 	IB_RNR_TIMER_061_44 = 25,
1085 	IB_RNR_TIMER_081_92 = 26,
1086 	IB_RNR_TIMER_122_88 = 27,
1087 	IB_RNR_TIMER_163_84 = 28,
1088 	IB_RNR_TIMER_245_76 = 29,
1089 	IB_RNR_TIMER_327_68 = 30,
1090 	IB_RNR_TIMER_491_52 = 31
1091 };
1092 
1093 enum ib_qp_attr_mask {
1094 	IB_QP_STATE			= 1,
1095 	IB_QP_CUR_STATE			= (1<<1),
1096 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1097 	IB_QP_ACCESS_FLAGS		= (1<<3),
1098 	IB_QP_PKEY_INDEX		= (1<<4),
1099 	IB_QP_PORT			= (1<<5),
1100 	IB_QP_QKEY			= (1<<6),
1101 	IB_QP_AV			= (1<<7),
1102 	IB_QP_PATH_MTU			= (1<<8),
1103 	IB_QP_TIMEOUT			= (1<<9),
1104 	IB_QP_RETRY_CNT			= (1<<10),
1105 	IB_QP_RNR_RETRY			= (1<<11),
1106 	IB_QP_RQ_PSN			= (1<<12),
1107 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1108 	IB_QP_ALT_PATH			= (1<<14),
1109 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1110 	IB_QP_SQ_PSN			= (1<<16),
1111 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1112 	IB_QP_PATH_MIG_STATE		= (1<<18),
1113 	IB_QP_CAP			= (1<<19),
1114 	IB_QP_DEST_QPN			= (1<<20),
1115 	IB_QP_RESERVED1			= (1<<21),
1116 	IB_QP_RESERVED2			= (1<<22),
1117 	IB_QP_RESERVED3			= (1<<23),
1118 	IB_QP_RESERVED4			= (1<<24),
1119 	IB_QP_RATE_LIMIT		= (1<<25),
1120 };
1121 
1122 enum ib_qp_state {
1123 	IB_QPS_RESET,
1124 	IB_QPS_INIT,
1125 	IB_QPS_RTR,
1126 	IB_QPS_RTS,
1127 	IB_QPS_SQD,
1128 	IB_QPS_SQE,
1129 	IB_QPS_ERR
1130 };
1131 
1132 enum ib_mig_state {
1133 	IB_MIG_MIGRATED,
1134 	IB_MIG_REARM,
1135 	IB_MIG_ARMED
1136 };
1137 
1138 enum ib_mw_type {
1139 	IB_MW_TYPE_1 = 1,
1140 	IB_MW_TYPE_2 = 2
1141 };
1142 
1143 struct ib_qp_attr {
1144 	enum ib_qp_state	qp_state;
1145 	enum ib_qp_state	cur_qp_state;
1146 	enum ib_mtu		path_mtu;
1147 	enum ib_mig_state	path_mig_state;
1148 	u32			qkey;
1149 	u32			rq_psn;
1150 	u32			sq_psn;
1151 	u32			dest_qp_num;
1152 	int			qp_access_flags;
1153 	struct ib_qp_cap	cap;
1154 	struct ib_ah_attr	ah_attr;
1155 	struct ib_ah_attr	alt_ah_attr;
1156 	u16			pkey_index;
1157 	u16			alt_pkey_index;
1158 	u8			en_sqd_async_notify;
1159 	u8			sq_draining;
1160 	u8			max_rd_atomic;
1161 	u8			max_dest_rd_atomic;
1162 	u8			min_rnr_timer;
1163 	u8			port_num;
1164 	u8			timeout;
1165 	u8			retry_cnt;
1166 	u8			rnr_retry;
1167 	u8			alt_port_num;
1168 	u8			alt_timeout;
1169 	u32			rate_limit;
1170 };
1171 
1172 enum ib_wr_opcode {
1173 	IB_WR_RDMA_WRITE,
1174 	IB_WR_RDMA_WRITE_WITH_IMM,
1175 	IB_WR_SEND,
1176 	IB_WR_SEND_WITH_IMM,
1177 	IB_WR_RDMA_READ,
1178 	IB_WR_ATOMIC_CMP_AND_SWP,
1179 	IB_WR_ATOMIC_FETCH_AND_ADD,
1180 	IB_WR_LSO,
1181 	IB_WR_SEND_WITH_INV,
1182 	IB_WR_RDMA_READ_WITH_INV,
1183 	IB_WR_LOCAL_INV,
1184 	IB_WR_REG_MR,
1185 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1186 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1187 	IB_WR_REG_SIG_MR,
1188 	/* reserve values for low level drivers' internal use.
1189 	 * These values will not be used at all in the ib core layer.
1190 	 */
1191 	IB_WR_RESERVED1 = 0xf0,
1192 	IB_WR_RESERVED2,
1193 	IB_WR_RESERVED3,
1194 	IB_WR_RESERVED4,
1195 	IB_WR_RESERVED5,
1196 	IB_WR_RESERVED6,
1197 	IB_WR_RESERVED7,
1198 	IB_WR_RESERVED8,
1199 	IB_WR_RESERVED9,
1200 	IB_WR_RESERVED10,
1201 };
1202 
1203 enum ib_send_flags {
1204 	IB_SEND_FENCE		= 1,
1205 	IB_SEND_SIGNALED	= (1<<1),
1206 	IB_SEND_SOLICITED	= (1<<2),
1207 	IB_SEND_INLINE		= (1<<3),
1208 	IB_SEND_IP_CSUM		= (1<<4),
1209 
1210 	/* reserve bits 26-31 for low level drivers' internal use */
1211 	IB_SEND_RESERVED_START	= (1 << 26),
1212 	IB_SEND_RESERVED_END	= (1 << 31),
1213 };
1214 
1215 struct ib_sge {
1216 	u64	addr;
1217 	u32	length;
1218 	u32	lkey;
1219 };
1220 
1221 struct ib_cqe {
1222 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1223 };
1224 
1225 struct ib_send_wr {
1226 	struct ib_send_wr      *next;
1227 	union {
1228 		u64		wr_id;
1229 		struct ib_cqe	*wr_cqe;
1230 	};
1231 	struct ib_sge	       *sg_list;
1232 	int			num_sge;
1233 	enum ib_wr_opcode	opcode;
1234 	int			send_flags;
1235 	union {
1236 		__be32		imm_data;
1237 		u32		invalidate_rkey;
1238 	} ex;
1239 };
1240 
1241 struct ib_rdma_wr {
1242 	struct ib_send_wr	wr;
1243 	u64			remote_addr;
1244 	u32			rkey;
1245 };
1246 
1247 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1248 {
1249 	return container_of(wr, struct ib_rdma_wr, wr);
1250 }
1251 
1252 struct ib_atomic_wr {
1253 	struct ib_send_wr	wr;
1254 	u64			remote_addr;
1255 	u64			compare_add;
1256 	u64			swap;
1257 	u64			compare_add_mask;
1258 	u64			swap_mask;
1259 	u32			rkey;
1260 };
1261 
1262 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1263 {
1264 	return container_of(wr, struct ib_atomic_wr, wr);
1265 }
1266 
1267 struct ib_ud_wr {
1268 	struct ib_send_wr	wr;
1269 	struct ib_ah		*ah;
1270 	void			*header;
1271 	int			hlen;
1272 	int			mss;
1273 	u32			remote_qpn;
1274 	u32			remote_qkey;
1275 	u16			pkey_index; /* valid for GSI only */
1276 	u8			port_num;   /* valid for DR SMPs on switch only */
1277 };
1278 
1279 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1280 {
1281 	return container_of(wr, struct ib_ud_wr, wr);
1282 }
1283 
1284 struct ib_reg_wr {
1285 	struct ib_send_wr	wr;
1286 	struct ib_mr		*mr;
1287 	u32			key;
1288 	int			access;
1289 };
1290 
1291 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1292 {
1293 	return container_of(wr, struct ib_reg_wr, wr);
1294 }
1295 
1296 struct ib_sig_handover_wr {
1297 	struct ib_send_wr	wr;
1298 	struct ib_sig_attrs    *sig_attrs;
1299 	struct ib_mr	       *sig_mr;
1300 	int			access_flags;
1301 	struct ib_sge	       *prot;
1302 };
1303 
1304 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1305 {
1306 	return container_of(wr, struct ib_sig_handover_wr, wr);
1307 }
1308 
1309 struct ib_recv_wr {
1310 	struct ib_recv_wr      *next;
1311 	union {
1312 		u64		wr_id;
1313 		struct ib_cqe	*wr_cqe;
1314 	};
1315 	struct ib_sge	       *sg_list;
1316 	int			num_sge;
1317 };
1318 
1319 enum ib_access_flags {
1320 	IB_ACCESS_LOCAL_WRITE	= 1,
1321 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1322 	IB_ACCESS_REMOTE_READ	= (1<<2),
1323 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1324 	IB_ACCESS_MW_BIND	= (1<<4),
1325 	IB_ZERO_BASED		= (1<<5),
1326 	IB_ACCESS_ON_DEMAND     = (1<<6),
1327 };
1328 
1329 /*
1330  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1331  * are hidden here instead of a uapi header!
1332  */
1333 enum ib_mr_rereg_flags {
1334 	IB_MR_REREG_TRANS	= 1,
1335 	IB_MR_REREG_PD		= (1<<1),
1336 	IB_MR_REREG_ACCESS	= (1<<2),
1337 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1338 };
1339 
1340 struct ib_fmr_attr {
1341 	int	max_pages;
1342 	int	max_maps;
1343 	u8	page_shift;
1344 };
1345 
1346 struct ib_umem;
1347 
1348 struct ib_ucontext {
1349 	struct ib_device       *device;
1350 	struct list_head	pd_list;
1351 	struct list_head	mr_list;
1352 	struct list_head	mw_list;
1353 	struct list_head	cq_list;
1354 	struct list_head	qp_list;
1355 	struct list_head	srq_list;
1356 	struct list_head	ah_list;
1357 	struct list_head	xrcd_list;
1358 	struct list_head	rule_list;
1359 	struct list_head	wq_list;
1360 	struct list_head	rwq_ind_tbl_list;
1361 	int			closing;
1362 
1363 	struct pid             *tgid;
1364 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1365 	struct rb_root      umem_tree;
1366 	/*
1367 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1368 	 * mmu notifiers registration.
1369 	 */
1370 	struct rw_semaphore	umem_rwsem;
1371 	void (*invalidate_range)(struct ib_umem *umem,
1372 				 unsigned long start, unsigned long end);
1373 
1374 	struct mmu_notifier	mn;
1375 	atomic_t		notifier_count;
1376 	/* A list of umems that don't have private mmu notifier counters yet. */
1377 	struct list_head	no_private_counters;
1378 	int                     odp_mrs_count;
1379 #endif
1380 };
1381 
1382 struct ib_uobject {
1383 	u64			user_handle;	/* handle given to us by userspace */
1384 	struct ib_ucontext     *context;	/* associated user context */
1385 	void		       *object;		/* containing object */
1386 	struct list_head	list;		/* link to context's list */
1387 	int			id;		/* index into kernel idr */
1388 	struct kref		ref;
1389 	struct rw_semaphore	mutex;		/* protects .live */
1390 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1391 	int			live;
1392 };
1393 
1394 struct ib_udata {
1395 	const void __user *inbuf;
1396 	void __user *outbuf;
1397 	size_t       inlen;
1398 	size_t       outlen;
1399 };
1400 
1401 struct ib_pd {
1402 	u32			local_dma_lkey;
1403 	u32			flags;
1404 	struct ib_device       *device;
1405 	struct ib_uobject      *uobject;
1406 	atomic_t          	usecnt; /* count all resources */
1407 
1408 	u32			unsafe_global_rkey;
1409 
1410 	/*
1411 	 * Implementation details of the RDMA core, don't use in drivers:
1412 	 */
1413 	struct ib_mr	       *__internal_mr;
1414 };
1415 
1416 struct ib_xrcd {
1417 	struct ib_device       *device;
1418 	atomic_t		usecnt; /* count all exposed resources */
1419 	struct inode	       *inode;
1420 
1421 	struct mutex		tgt_qp_mutex;
1422 	struct list_head	tgt_qp_list;
1423 };
1424 
1425 struct ib_ah {
1426 	struct ib_device	*device;
1427 	struct ib_pd		*pd;
1428 	struct ib_uobject	*uobject;
1429 };
1430 
1431 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1432 
1433 enum ib_poll_context {
1434 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1435 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1436 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1437 };
1438 
1439 struct ib_cq {
1440 	struct ib_device       *device;
1441 	struct ib_uobject      *uobject;
1442 	ib_comp_handler   	comp_handler;
1443 	void                  (*event_handler)(struct ib_event *, void *);
1444 	void                   *cq_context;
1445 	int               	cqe;
1446 	atomic_t          	usecnt; /* count number of work queues */
1447 	enum ib_poll_context	poll_ctx;
1448 	struct ib_wc		*wc;
1449 	union {
1450 		struct irq_poll		iop;
1451 		struct work_struct	work;
1452 	};
1453 };
1454 
1455 struct ib_srq {
1456 	struct ib_device       *device;
1457 	struct ib_pd	       *pd;
1458 	struct ib_uobject      *uobject;
1459 	void		      (*event_handler)(struct ib_event *, void *);
1460 	void		       *srq_context;
1461 	enum ib_srq_type	srq_type;
1462 	atomic_t		usecnt;
1463 
1464 	union {
1465 		struct {
1466 			struct ib_xrcd *xrcd;
1467 			struct ib_cq   *cq;
1468 			u32		srq_num;
1469 		} xrc;
1470 	} ext;
1471 };
1472 
1473 enum ib_wq_type {
1474 	IB_WQT_RQ
1475 };
1476 
1477 enum ib_wq_state {
1478 	IB_WQS_RESET,
1479 	IB_WQS_RDY,
1480 	IB_WQS_ERR
1481 };
1482 
1483 struct ib_wq {
1484 	struct ib_device       *device;
1485 	struct ib_uobject      *uobject;
1486 	void		    *wq_context;
1487 	void		    (*event_handler)(struct ib_event *, void *);
1488 	struct ib_pd	       *pd;
1489 	struct ib_cq	       *cq;
1490 	u32		wq_num;
1491 	enum ib_wq_state       state;
1492 	enum ib_wq_type	wq_type;
1493 	atomic_t		usecnt;
1494 };
1495 
1496 struct ib_wq_init_attr {
1497 	void		       *wq_context;
1498 	enum ib_wq_type	wq_type;
1499 	u32		max_wr;
1500 	u32		max_sge;
1501 	struct	ib_cq	       *cq;
1502 	void		    (*event_handler)(struct ib_event *, void *);
1503 };
1504 
1505 enum ib_wq_attr_mask {
1506 	IB_WQ_STATE	= 1 << 0,
1507 	IB_WQ_CUR_STATE	= 1 << 1,
1508 };
1509 
1510 struct ib_wq_attr {
1511 	enum	ib_wq_state	wq_state;
1512 	enum	ib_wq_state	curr_wq_state;
1513 };
1514 
1515 struct ib_rwq_ind_table {
1516 	struct ib_device	*device;
1517 	struct ib_uobject      *uobject;
1518 	atomic_t		usecnt;
1519 	u32		ind_tbl_num;
1520 	u32		log_ind_tbl_size;
1521 	struct ib_wq	**ind_tbl;
1522 };
1523 
1524 struct ib_rwq_ind_table_init_attr {
1525 	u32		log_ind_tbl_size;
1526 	/* Each entry is a pointer to Receive Work Queue */
1527 	struct ib_wq	**ind_tbl;
1528 };
1529 
1530 /*
1531  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1532  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1533  */
1534 struct ib_qp {
1535 	struct ib_device       *device;
1536 	struct ib_pd	       *pd;
1537 	struct ib_cq	       *send_cq;
1538 	struct ib_cq	       *recv_cq;
1539 	spinlock_t		mr_lock;
1540 	int			mrs_used;
1541 	struct list_head	rdma_mrs;
1542 	struct list_head	sig_mrs;
1543 	struct ib_srq	       *srq;
1544 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1545 	struct list_head	xrcd_list;
1546 
1547 	/* count times opened, mcast attaches, flow attaches */
1548 	atomic_t		usecnt;
1549 	struct list_head	open_list;
1550 	struct ib_qp           *real_qp;
1551 	struct ib_uobject      *uobject;
1552 	void                  (*event_handler)(struct ib_event *, void *);
1553 	void		       *qp_context;
1554 	u32			qp_num;
1555 	u32			max_write_sge;
1556 	u32			max_read_sge;
1557 	enum ib_qp_type		qp_type;
1558 	struct ib_rwq_ind_table *rwq_ind_tbl;
1559 };
1560 
1561 struct ib_mr {
1562 	struct ib_device  *device;
1563 	struct ib_pd	  *pd;
1564 	u32		   lkey;
1565 	u32		   rkey;
1566 	u64		   iova;
1567 	u32		   length;
1568 	unsigned int	   page_size;
1569 	bool		   need_inval;
1570 	union {
1571 		struct ib_uobject	*uobject;	/* user */
1572 		struct list_head	qp_entry;	/* FR */
1573 	};
1574 };
1575 
1576 struct ib_mw {
1577 	struct ib_device	*device;
1578 	struct ib_pd		*pd;
1579 	struct ib_uobject	*uobject;
1580 	u32			rkey;
1581 	enum ib_mw_type         type;
1582 };
1583 
1584 struct ib_fmr {
1585 	struct ib_device	*device;
1586 	struct ib_pd		*pd;
1587 	struct list_head	list;
1588 	u32			lkey;
1589 	u32			rkey;
1590 };
1591 
1592 /* Supported steering options */
1593 enum ib_flow_attr_type {
1594 	/* steering according to rule specifications */
1595 	IB_FLOW_ATTR_NORMAL		= 0x0,
1596 	/* default unicast and multicast rule -
1597 	 * receive all Eth traffic which isn't steered to any QP
1598 	 */
1599 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1600 	/* default multicast rule -
1601 	 * receive all Eth multicast traffic which isn't steered to any QP
1602 	 */
1603 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1604 	/* sniffer rule - receive all port traffic */
1605 	IB_FLOW_ATTR_SNIFFER		= 0x3
1606 };
1607 
1608 /* Supported steering header types */
1609 enum ib_flow_spec_type {
1610 	/* L2 headers*/
1611 	IB_FLOW_SPEC_ETH		= 0x20,
1612 	IB_FLOW_SPEC_IB			= 0x22,
1613 	/* L3 header*/
1614 	IB_FLOW_SPEC_IPV4		= 0x30,
1615 	IB_FLOW_SPEC_IPV6		= 0x31,
1616 	/* L4 headers*/
1617 	IB_FLOW_SPEC_TCP		= 0x40,
1618 	IB_FLOW_SPEC_UDP		= 0x41,
1619 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1620 	IB_FLOW_SPEC_INNER		= 0x100,
1621 };
1622 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1623 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1624 
1625 /* Flow steering rule priority is set according to it's domain.
1626  * Lower domain value means higher priority.
1627  */
1628 enum ib_flow_domain {
1629 	IB_FLOW_DOMAIN_USER,
1630 	IB_FLOW_DOMAIN_ETHTOOL,
1631 	IB_FLOW_DOMAIN_RFS,
1632 	IB_FLOW_DOMAIN_NIC,
1633 	IB_FLOW_DOMAIN_NUM /* Must be last */
1634 };
1635 
1636 enum ib_flow_flags {
1637 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1638 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1639 };
1640 
1641 struct ib_flow_eth_filter {
1642 	u8	dst_mac[6];
1643 	u8	src_mac[6];
1644 	__be16	ether_type;
1645 	__be16	vlan_tag;
1646 	/* Must be last */
1647 	u8	real_sz[0];
1648 };
1649 
1650 struct ib_flow_spec_eth {
1651 	u32			  type;
1652 	u16			  size;
1653 	struct ib_flow_eth_filter val;
1654 	struct ib_flow_eth_filter mask;
1655 };
1656 
1657 struct ib_flow_ib_filter {
1658 	__be16 dlid;
1659 	__u8   sl;
1660 	/* Must be last */
1661 	u8	real_sz[0];
1662 };
1663 
1664 struct ib_flow_spec_ib {
1665 	u32			 type;
1666 	u16			 size;
1667 	struct ib_flow_ib_filter val;
1668 	struct ib_flow_ib_filter mask;
1669 };
1670 
1671 /* IPv4 header flags */
1672 enum ib_ipv4_flags {
1673 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1674 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1675 				    last have this flag set */
1676 };
1677 
1678 struct ib_flow_ipv4_filter {
1679 	__be32	src_ip;
1680 	__be32	dst_ip;
1681 	u8	proto;
1682 	u8	tos;
1683 	u8	ttl;
1684 	u8	flags;
1685 	/* Must be last */
1686 	u8	real_sz[0];
1687 };
1688 
1689 struct ib_flow_spec_ipv4 {
1690 	u32			   type;
1691 	u16			   size;
1692 	struct ib_flow_ipv4_filter val;
1693 	struct ib_flow_ipv4_filter mask;
1694 };
1695 
1696 struct ib_flow_ipv6_filter {
1697 	u8	src_ip[16];
1698 	u8	dst_ip[16];
1699 	__be32	flow_label;
1700 	u8	next_hdr;
1701 	u8	traffic_class;
1702 	u8	hop_limit;
1703 	/* Must be last */
1704 	u8	real_sz[0];
1705 };
1706 
1707 struct ib_flow_spec_ipv6 {
1708 	u32			   type;
1709 	u16			   size;
1710 	struct ib_flow_ipv6_filter val;
1711 	struct ib_flow_ipv6_filter mask;
1712 };
1713 
1714 struct ib_flow_tcp_udp_filter {
1715 	__be16	dst_port;
1716 	__be16	src_port;
1717 	/* Must be last */
1718 	u8	real_sz[0];
1719 };
1720 
1721 struct ib_flow_spec_tcp_udp {
1722 	u32			      type;
1723 	u16			      size;
1724 	struct ib_flow_tcp_udp_filter val;
1725 	struct ib_flow_tcp_udp_filter mask;
1726 };
1727 
1728 struct ib_flow_tunnel_filter {
1729 	__be32	tunnel_id;
1730 	u8	real_sz[0];
1731 };
1732 
1733 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1734  * the tunnel_id from val has the vni value
1735  */
1736 struct ib_flow_spec_tunnel {
1737 	u32			      type;
1738 	u16			      size;
1739 	struct ib_flow_tunnel_filter  val;
1740 	struct ib_flow_tunnel_filter  mask;
1741 };
1742 
1743 union ib_flow_spec {
1744 	struct {
1745 		u32			type;
1746 		u16			size;
1747 	};
1748 	struct ib_flow_spec_eth		eth;
1749 	struct ib_flow_spec_ib		ib;
1750 	struct ib_flow_spec_ipv4        ipv4;
1751 	struct ib_flow_spec_tcp_udp	tcp_udp;
1752 	struct ib_flow_spec_ipv6        ipv6;
1753 	struct ib_flow_spec_tunnel      tunnel;
1754 };
1755 
1756 struct ib_flow_attr {
1757 	enum ib_flow_attr_type type;
1758 	u16	     size;
1759 	u16	     priority;
1760 	u32	     flags;
1761 	u8	     num_of_specs;
1762 	u8	     port;
1763 	/* Following are the optional layers according to user request
1764 	 * struct ib_flow_spec_xxx
1765 	 * struct ib_flow_spec_yyy
1766 	 */
1767 };
1768 
1769 struct ib_flow {
1770 	struct ib_qp		*qp;
1771 	struct ib_uobject	*uobject;
1772 };
1773 
1774 struct ib_mad_hdr;
1775 struct ib_grh;
1776 
1777 enum ib_process_mad_flags {
1778 	IB_MAD_IGNORE_MKEY	= 1,
1779 	IB_MAD_IGNORE_BKEY	= 2,
1780 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1781 };
1782 
1783 enum ib_mad_result {
1784 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1785 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1786 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1787 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1788 };
1789 
1790 #define IB_DEVICE_NAME_MAX 64
1791 
1792 struct ib_cache {
1793 	rwlock_t                lock;
1794 	struct ib_event_handler event_handler;
1795 	struct ib_pkey_cache  **pkey_cache;
1796 	struct ib_gid_table   **gid_cache;
1797 	u8                     *lmc_cache;
1798 };
1799 
1800 struct ib_dma_mapping_ops {
1801 	int		(*mapping_error)(struct ib_device *dev,
1802 					 u64 dma_addr);
1803 	u64		(*map_single)(struct ib_device *dev,
1804 				      void *ptr, size_t size,
1805 				      enum dma_data_direction direction);
1806 	void		(*unmap_single)(struct ib_device *dev,
1807 					u64 addr, size_t size,
1808 					enum dma_data_direction direction);
1809 	u64		(*map_page)(struct ib_device *dev,
1810 				    struct page *page, unsigned long offset,
1811 				    size_t size,
1812 				    enum dma_data_direction direction);
1813 	void		(*unmap_page)(struct ib_device *dev,
1814 				      u64 addr, size_t size,
1815 				      enum dma_data_direction direction);
1816 	int		(*map_sg)(struct ib_device *dev,
1817 				  struct scatterlist *sg, int nents,
1818 				  enum dma_data_direction direction);
1819 	void		(*unmap_sg)(struct ib_device *dev,
1820 				    struct scatterlist *sg, int nents,
1821 				    enum dma_data_direction direction);
1822 	int		(*map_sg_attrs)(struct ib_device *dev,
1823 					struct scatterlist *sg, int nents,
1824 					enum dma_data_direction direction,
1825 					unsigned long attrs);
1826 	void		(*unmap_sg_attrs)(struct ib_device *dev,
1827 					  struct scatterlist *sg, int nents,
1828 					  enum dma_data_direction direction,
1829 					  unsigned long attrs);
1830 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1831 					       u64 dma_handle,
1832 					       size_t size,
1833 					       enum dma_data_direction dir);
1834 	void		(*sync_single_for_device)(struct ib_device *dev,
1835 						  u64 dma_handle,
1836 						  size_t size,
1837 						  enum dma_data_direction dir);
1838 	void		*(*alloc_coherent)(struct ib_device *dev,
1839 					   size_t size,
1840 					   u64 *dma_handle,
1841 					   gfp_t flag);
1842 	void		(*free_coherent)(struct ib_device *dev,
1843 					 size_t size, void *cpu_addr,
1844 					 u64 dma_handle);
1845 };
1846 
1847 struct iw_cm_verbs;
1848 
1849 struct ib_port_immutable {
1850 	int                           pkey_tbl_len;
1851 	int                           gid_tbl_len;
1852 	u32                           core_cap_flags;
1853 	u32                           max_mad_size;
1854 };
1855 
1856 struct ib_device {
1857 	struct device                *dma_device;
1858 
1859 	char                          name[IB_DEVICE_NAME_MAX];
1860 
1861 	struct list_head              event_handler_list;
1862 	spinlock_t                    event_handler_lock;
1863 
1864 	spinlock_t                    client_data_lock;
1865 	struct list_head              core_list;
1866 	/* Access to the client_data_list is protected by the client_data_lock
1867 	 * spinlock and the lists_rwsem read-write semaphore */
1868 	struct list_head              client_data_list;
1869 
1870 	struct ib_cache               cache;
1871 	/**
1872 	 * port_immutable is indexed by port number
1873 	 */
1874 	struct ib_port_immutable     *port_immutable;
1875 
1876 	int			      num_comp_vectors;
1877 
1878 	struct iw_cm_verbs	     *iwcm;
1879 
1880 	/**
1881 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1882 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
1883 	 *   core when the device is removed.  A lifespan of -1 in the return
1884 	 *   struct tells the core to set a default lifespan.
1885 	 */
1886 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
1887 						     u8 port_num);
1888 	/**
1889 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1890 	 * @index - The index in the value array we wish to have updated, or
1891 	 *   num_counters if we want all stats updated
1892 	 * Return codes -
1893 	 *   < 0 - Error, no counters updated
1894 	 *   index - Updated the single counter pointed to by index
1895 	 *   num_counters - Updated all counters (will reset the timestamp
1896 	 *     and prevent further calls for lifespan milliseconds)
1897 	 * Drivers are allowed to update all counters in leiu of just the
1898 	 *   one given in index at their option
1899 	 */
1900 	int		           (*get_hw_stats)(struct ib_device *device,
1901 						   struct rdma_hw_stats *stats,
1902 						   u8 port, int index);
1903 	int		           (*query_device)(struct ib_device *device,
1904 						   struct ib_device_attr *device_attr,
1905 						   struct ib_udata *udata);
1906 	int		           (*query_port)(struct ib_device *device,
1907 						 u8 port_num,
1908 						 struct ib_port_attr *port_attr);
1909 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1910 						     u8 port_num);
1911 	/* When calling get_netdev, the HW vendor's driver should return the
1912 	 * net device of device @device at port @port_num or NULL if such
1913 	 * a net device doesn't exist. The vendor driver should call dev_hold
1914 	 * on this net device. The HW vendor's device driver must guarantee
1915 	 * that this function returns NULL before the net device reaches
1916 	 * NETDEV_UNREGISTER_FINAL state.
1917 	 */
1918 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1919 						 u8 port_num);
1920 	int		           (*query_gid)(struct ib_device *device,
1921 						u8 port_num, int index,
1922 						union ib_gid *gid);
1923 	/* When calling add_gid, the HW vendor's driver should
1924 	 * add the gid of device @device at gid index @index of
1925 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1926 	 * the network device related to this gid is available
1927 	 * at @attr. @context allows the HW vendor driver to store extra
1928 	 * information together with a GID entry. The HW vendor may allocate
1929 	 * memory to contain this information and store it in @context when a
1930 	 * new GID entry is written to. Params are consistent until the next
1931 	 * call of add_gid or delete_gid. The function should return 0 on
1932 	 * success or error otherwise. The function could be called
1933 	 * concurrently for different ports. This function is only called
1934 	 * when roce_gid_table is used.
1935 	 */
1936 	int		           (*add_gid)(struct ib_device *device,
1937 					      u8 port_num,
1938 					      unsigned int index,
1939 					      const union ib_gid *gid,
1940 					      const struct ib_gid_attr *attr,
1941 					      void **context);
1942 	/* When calling del_gid, the HW vendor's driver should delete the
1943 	 * gid of device @device at gid index @index of port @port_num.
1944 	 * Upon the deletion of a GID entry, the HW vendor must free any
1945 	 * allocated memory. The caller will clear @context afterwards.
1946 	 * This function is only called when roce_gid_table is used.
1947 	 */
1948 	int		           (*del_gid)(struct ib_device *device,
1949 					      u8 port_num,
1950 					      unsigned int index,
1951 					      void **context);
1952 	int		           (*query_pkey)(struct ib_device *device,
1953 						 u8 port_num, u16 index, u16 *pkey);
1954 	int		           (*modify_device)(struct ib_device *device,
1955 						    int device_modify_mask,
1956 						    struct ib_device_modify *device_modify);
1957 	int		           (*modify_port)(struct ib_device *device,
1958 						  u8 port_num, int port_modify_mask,
1959 						  struct ib_port_modify *port_modify);
1960 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1961 						     struct ib_udata *udata);
1962 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1963 	int                        (*mmap)(struct ib_ucontext *context,
1964 					   struct vm_area_struct *vma);
1965 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1966 					       struct ib_ucontext *context,
1967 					       struct ib_udata *udata);
1968 	int                        (*dealloc_pd)(struct ib_pd *pd);
1969 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1970 						struct ib_ah_attr *ah_attr,
1971 						struct ib_udata *udata);
1972 	int                        (*modify_ah)(struct ib_ah *ah,
1973 						struct ib_ah_attr *ah_attr);
1974 	int                        (*query_ah)(struct ib_ah *ah,
1975 					       struct ib_ah_attr *ah_attr);
1976 	int                        (*destroy_ah)(struct ib_ah *ah);
1977 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1978 						 struct ib_srq_init_attr *srq_init_attr,
1979 						 struct ib_udata *udata);
1980 	int                        (*modify_srq)(struct ib_srq *srq,
1981 						 struct ib_srq_attr *srq_attr,
1982 						 enum ib_srq_attr_mask srq_attr_mask,
1983 						 struct ib_udata *udata);
1984 	int                        (*query_srq)(struct ib_srq *srq,
1985 						struct ib_srq_attr *srq_attr);
1986 	int                        (*destroy_srq)(struct ib_srq *srq);
1987 	int                        (*post_srq_recv)(struct ib_srq *srq,
1988 						    struct ib_recv_wr *recv_wr,
1989 						    struct ib_recv_wr **bad_recv_wr);
1990 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1991 						struct ib_qp_init_attr *qp_init_attr,
1992 						struct ib_udata *udata);
1993 	int                        (*modify_qp)(struct ib_qp *qp,
1994 						struct ib_qp_attr *qp_attr,
1995 						int qp_attr_mask,
1996 						struct ib_udata *udata);
1997 	int                        (*query_qp)(struct ib_qp *qp,
1998 					       struct ib_qp_attr *qp_attr,
1999 					       int qp_attr_mask,
2000 					       struct ib_qp_init_attr *qp_init_attr);
2001 	int                        (*destroy_qp)(struct ib_qp *qp);
2002 	int                        (*post_send)(struct ib_qp *qp,
2003 						struct ib_send_wr *send_wr,
2004 						struct ib_send_wr **bad_send_wr);
2005 	int                        (*post_recv)(struct ib_qp *qp,
2006 						struct ib_recv_wr *recv_wr,
2007 						struct ib_recv_wr **bad_recv_wr);
2008 	struct ib_cq *             (*create_cq)(struct ib_device *device,
2009 						const struct ib_cq_init_attr *attr,
2010 						struct ib_ucontext *context,
2011 						struct ib_udata *udata);
2012 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2013 						u16 cq_period);
2014 	int                        (*destroy_cq)(struct ib_cq *cq);
2015 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
2016 						struct ib_udata *udata);
2017 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
2018 					      struct ib_wc *wc);
2019 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2020 	int                        (*req_notify_cq)(struct ib_cq *cq,
2021 						    enum ib_cq_notify_flags flags);
2022 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
2023 						      int wc_cnt);
2024 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
2025 						 int mr_access_flags);
2026 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
2027 						  u64 start, u64 length,
2028 						  u64 virt_addr,
2029 						  int mr_access_flags,
2030 						  struct ib_udata *udata);
2031 	int			   (*rereg_user_mr)(struct ib_mr *mr,
2032 						    int flags,
2033 						    u64 start, u64 length,
2034 						    u64 virt_addr,
2035 						    int mr_access_flags,
2036 						    struct ib_pd *pd,
2037 						    struct ib_udata *udata);
2038 	int                        (*dereg_mr)(struct ib_mr *mr);
2039 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2040 					       enum ib_mr_type mr_type,
2041 					       u32 max_num_sg);
2042 	int                        (*map_mr_sg)(struct ib_mr *mr,
2043 						struct scatterlist *sg,
2044 						int sg_nents,
2045 						unsigned int *sg_offset);
2046 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2047 					       enum ib_mw_type type,
2048 					       struct ib_udata *udata);
2049 	int                        (*dealloc_mw)(struct ib_mw *mw);
2050 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2051 						int mr_access_flags,
2052 						struct ib_fmr_attr *fmr_attr);
2053 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2054 						   u64 *page_list, int list_len,
2055 						   u64 iova);
2056 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2057 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2058 	int                        (*attach_mcast)(struct ib_qp *qp,
2059 						   union ib_gid *gid,
2060 						   u16 lid);
2061 	int                        (*detach_mcast)(struct ib_qp *qp,
2062 						   union ib_gid *gid,
2063 						   u16 lid);
2064 	int                        (*process_mad)(struct ib_device *device,
2065 						  int process_mad_flags,
2066 						  u8 port_num,
2067 						  const struct ib_wc *in_wc,
2068 						  const struct ib_grh *in_grh,
2069 						  const struct ib_mad_hdr *in_mad,
2070 						  size_t in_mad_size,
2071 						  struct ib_mad_hdr *out_mad,
2072 						  size_t *out_mad_size,
2073 						  u16 *out_mad_pkey_index);
2074 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2075 						 struct ib_ucontext *ucontext,
2076 						 struct ib_udata *udata);
2077 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2078 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2079 						  struct ib_flow_attr
2080 						  *flow_attr,
2081 						  int domain);
2082 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2083 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2084 						      struct ib_mr_status *mr_status);
2085 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2086 	void			   (*drain_rq)(struct ib_qp *qp);
2087 	void			   (*drain_sq)(struct ib_qp *qp);
2088 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2089 							int state);
2090 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2091 						   struct ifla_vf_info *ivf);
2092 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2093 						   struct ifla_vf_stats *stats);
2094 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2095 						  int type);
2096 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2097 						struct ib_wq_init_attr *init_attr,
2098 						struct ib_udata *udata);
2099 	int			   (*destroy_wq)(struct ib_wq *wq);
2100 	int			   (*modify_wq)(struct ib_wq *wq,
2101 						struct ib_wq_attr *attr,
2102 						u32 wq_attr_mask,
2103 						struct ib_udata *udata);
2104 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2105 							   struct ib_rwq_ind_table_init_attr *init_attr,
2106 							   struct ib_udata *udata);
2107 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2108 	struct ib_dma_mapping_ops   *dma_ops;
2109 
2110 	struct module               *owner;
2111 	struct device                dev;
2112 	struct kobject               *ports_parent;
2113 	struct list_head             port_list;
2114 
2115 	enum {
2116 		IB_DEV_UNINITIALIZED,
2117 		IB_DEV_REGISTERED,
2118 		IB_DEV_UNREGISTERED
2119 	}                            reg_state;
2120 
2121 	int			     uverbs_abi_ver;
2122 	u64			     uverbs_cmd_mask;
2123 	u64			     uverbs_ex_cmd_mask;
2124 
2125 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2126 	__be64			     node_guid;
2127 	u32			     local_dma_lkey;
2128 	u16                          is_switch:1;
2129 	u8                           node_type;
2130 	u8                           phys_port_cnt;
2131 	struct ib_device_attr        attrs;
2132 	struct attribute_group	     *hw_stats_ag;
2133 	struct rdma_hw_stats         *hw_stats;
2134 
2135 	/**
2136 	 * The following mandatory functions are used only at device
2137 	 * registration.  Keep functions such as these at the end of this
2138 	 * structure to avoid cache line misses when accessing struct ib_device
2139 	 * in fast paths.
2140 	 */
2141 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2142 	void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2143 };
2144 
2145 struct ib_client {
2146 	char  *name;
2147 	void (*add)   (struct ib_device *);
2148 	void (*remove)(struct ib_device *, void *client_data);
2149 
2150 	/* Returns the net_dev belonging to this ib_client and matching the
2151 	 * given parameters.
2152 	 * @dev:	 An RDMA device that the net_dev use for communication.
2153 	 * @port:	 A physical port number on the RDMA device.
2154 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2155 	 * @gid:	 A GID that the net_dev uses to communicate.
2156 	 * @addr:	 An IP address the net_dev is configured with.
2157 	 * @client_data: The device's client data set by ib_set_client_data().
2158 	 *
2159 	 * An ib_client that implements a net_dev on top of RDMA devices
2160 	 * (such as IP over IB) should implement this callback, allowing the
2161 	 * rdma_cm module to find the right net_dev for a given request.
2162 	 *
2163 	 * The caller is responsible for calling dev_put on the returned
2164 	 * netdev. */
2165 	struct net_device *(*get_net_dev_by_params)(
2166 			struct ib_device *dev,
2167 			u8 port,
2168 			u16 pkey,
2169 			const union ib_gid *gid,
2170 			const struct sockaddr *addr,
2171 			void *client_data);
2172 	struct list_head list;
2173 };
2174 
2175 struct ib_device *ib_alloc_device(size_t size);
2176 void ib_dealloc_device(struct ib_device *device);
2177 
2178 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2179 
2180 int ib_register_device(struct ib_device *device,
2181 		       int (*port_callback)(struct ib_device *,
2182 					    u8, struct kobject *));
2183 void ib_unregister_device(struct ib_device *device);
2184 
2185 int ib_register_client   (struct ib_client *client);
2186 void ib_unregister_client(struct ib_client *client);
2187 
2188 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2189 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2190 			 void *data);
2191 
2192 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2193 {
2194 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2195 }
2196 
2197 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2198 {
2199 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2200 }
2201 
2202 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2203 				       size_t offset,
2204 				       size_t len)
2205 {
2206 	const void __user *p = udata->inbuf + offset;
2207 	bool ret;
2208 	u8 *buf;
2209 
2210 	if (len > USHRT_MAX)
2211 		return false;
2212 
2213 	buf = memdup_user(p, len);
2214 	if (IS_ERR(buf))
2215 		return false;
2216 
2217 	ret = !memchr_inv(buf, 0, len);
2218 	kfree(buf);
2219 	return ret;
2220 }
2221 
2222 /**
2223  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2224  * contains all required attributes and no attributes not allowed for
2225  * the given QP state transition.
2226  * @cur_state: Current QP state
2227  * @next_state: Next QP state
2228  * @type: QP type
2229  * @mask: Mask of supplied QP attributes
2230  * @ll : link layer of port
2231  *
2232  * This function is a helper function that a low-level driver's
2233  * modify_qp method can use to validate the consumer's input.  It
2234  * checks that cur_state and next_state are valid QP states, that a
2235  * transition from cur_state to next_state is allowed by the IB spec,
2236  * and that the attribute mask supplied is allowed for the transition.
2237  */
2238 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2239 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2240 		       enum rdma_link_layer ll);
2241 
2242 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2243 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2244 void ib_dispatch_event(struct ib_event *event);
2245 
2246 int ib_query_port(struct ib_device *device,
2247 		  u8 port_num, struct ib_port_attr *port_attr);
2248 
2249 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2250 					       u8 port_num);
2251 
2252 /**
2253  * rdma_cap_ib_switch - Check if the device is IB switch
2254  * @device: Device to check
2255  *
2256  * Device driver is responsible for setting is_switch bit on
2257  * in ib_device structure at init time.
2258  *
2259  * Return: true if the device is IB switch.
2260  */
2261 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2262 {
2263 	return device->is_switch;
2264 }
2265 
2266 /**
2267  * rdma_start_port - Return the first valid port number for the device
2268  * specified
2269  *
2270  * @device: Device to be checked
2271  *
2272  * Return start port number
2273  */
2274 static inline u8 rdma_start_port(const struct ib_device *device)
2275 {
2276 	return rdma_cap_ib_switch(device) ? 0 : 1;
2277 }
2278 
2279 /**
2280  * rdma_end_port - Return the last valid port number for the device
2281  * specified
2282  *
2283  * @device: Device to be checked
2284  *
2285  * Return last port number
2286  */
2287 static inline u8 rdma_end_port(const struct ib_device *device)
2288 {
2289 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2290 }
2291 
2292 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2293 {
2294 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2295 }
2296 
2297 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2298 {
2299 	return device->port_immutable[port_num].core_cap_flags &
2300 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2301 }
2302 
2303 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2304 {
2305 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2306 }
2307 
2308 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2309 {
2310 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2311 }
2312 
2313 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2314 {
2315 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2316 }
2317 
2318 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2319 {
2320 	return rdma_protocol_ib(device, port_num) ||
2321 		rdma_protocol_roce(device, port_num);
2322 }
2323 
2324 /**
2325  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2326  * Management Datagrams.
2327  * @device: Device to check
2328  * @port_num: Port number to check
2329  *
2330  * Management Datagrams (MAD) are a required part of the InfiniBand
2331  * specification and are supported on all InfiniBand devices.  A slightly
2332  * extended version are also supported on OPA interfaces.
2333  *
2334  * Return: true if the port supports sending/receiving of MAD packets.
2335  */
2336 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2337 {
2338 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2339 }
2340 
2341 /**
2342  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2343  * Management Datagrams.
2344  * @device: Device to check
2345  * @port_num: Port number to check
2346  *
2347  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2348  * datagrams with their own versions.  These OPA MADs share many but not all of
2349  * the characteristics of InfiniBand MADs.
2350  *
2351  * OPA MADs differ in the following ways:
2352  *
2353  *    1) MADs are variable size up to 2K
2354  *       IBTA defined MADs remain fixed at 256 bytes
2355  *    2) OPA SMPs must carry valid PKeys
2356  *    3) OPA SMP packets are a different format
2357  *
2358  * Return: true if the port supports OPA MAD packet formats.
2359  */
2360 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2361 {
2362 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2363 		== RDMA_CORE_CAP_OPA_MAD;
2364 }
2365 
2366 /**
2367  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2368  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2369  * @device: Device to check
2370  * @port_num: Port number to check
2371  *
2372  * Each InfiniBand node is required to provide a Subnet Management Agent
2373  * that the subnet manager can access.  Prior to the fabric being fully
2374  * configured by the subnet manager, the SMA is accessed via a well known
2375  * interface called the Subnet Management Interface (SMI).  This interface
2376  * uses directed route packets to communicate with the SM to get around the
2377  * chicken and egg problem of the SM needing to know what's on the fabric
2378  * in order to configure the fabric, and needing to configure the fabric in
2379  * order to send packets to the devices on the fabric.  These directed
2380  * route packets do not need the fabric fully configured in order to reach
2381  * their destination.  The SMI is the only method allowed to send
2382  * directed route packets on an InfiniBand fabric.
2383  *
2384  * Return: true if the port provides an SMI.
2385  */
2386 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2387 {
2388 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2389 }
2390 
2391 /**
2392  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2393  * Communication Manager.
2394  * @device: Device to check
2395  * @port_num: Port number to check
2396  *
2397  * The InfiniBand Communication Manager is one of many pre-defined General
2398  * Service Agents (GSA) that are accessed via the General Service
2399  * Interface (GSI).  It's role is to facilitate establishment of connections
2400  * between nodes as well as other management related tasks for established
2401  * connections.
2402  *
2403  * Return: true if the port supports an IB CM (this does not guarantee that
2404  * a CM is actually running however).
2405  */
2406 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2407 {
2408 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2409 }
2410 
2411 /**
2412  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2413  * Communication Manager.
2414  * @device: Device to check
2415  * @port_num: Port number to check
2416  *
2417  * Similar to above, but specific to iWARP connections which have a different
2418  * managment protocol than InfiniBand.
2419  *
2420  * Return: true if the port supports an iWARP CM (this does not guarantee that
2421  * a CM is actually running however).
2422  */
2423 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2424 {
2425 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2426 }
2427 
2428 /**
2429  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2430  * Subnet Administration.
2431  * @device: Device to check
2432  * @port_num: Port number to check
2433  *
2434  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2435  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2436  * fabrics, devices should resolve routes to other hosts by contacting the
2437  * SA to query the proper route.
2438  *
2439  * Return: true if the port should act as a client to the fabric Subnet
2440  * Administration interface.  This does not imply that the SA service is
2441  * running locally.
2442  */
2443 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2444 {
2445 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2446 }
2447 
2448 /**
2449  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2450  * Multicast.
2451  * @device: Device to check
2452  * @port_num: Port number to check
2453  *
2454  * InfiniBand multicast registration is more complex than normal IPv4 or
2455  * IPv6 multicast registration.  Each Host Channel Adapter must register
2456  * with the Subnet Manager when it wishes to join a multicast group.  It
2457  * should do so only once regardless of how many queue pairs it subscribes
2458  * to this group.  And it should leave the group only after all queue pairs
2459  * attached to the group have been detached.
2460  *
2461  * Return: true if the port must undertake the additional adminstrative
2462  * overhead of registering/unregistering with the SM and tracking of the
2463  * total number of queue pairs attached to the multicast group.
2464  */
2465 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2466 {
2467 	return rdma_cap_ib_sa(device, port_num);
2468 }
2469 
2470 /**
2471  * rdma_cap_af_ib - Check if the port of device has the capability
2472  * Native Infiniband Address.
2473  * @device: Device to check
2474  * @port_num: Port number to check
2475  *
2476  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2477  * GID.  RoCE uses a different mechanism, but still generates a GID via
2478  * a prescribed mechanism and port specific data.
2479  *
2480  * Return: true if the port uses a GID address to identify devices on the
2481  * network.
2482  */
2483 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2484 {
2485 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2486 }
2487 
2488 /**
2489  * rdma_cap_eth_ah - Check if the port of device has the capability
2490  * Ethernet Address Handle.
2491  * @device: Device to check
2492  * @port_num: Port number to check
2493  *
2494  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2495  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2496  * port.  Normally, packet headers are generated by the sending host
2497  * adapter, but when sending connectionless datagrams, we must manually
2498  * inject the proper headers for the fabric we are communicating over.
2499  *
2500  * Return: true if we are running as a RoCE port and must force the
2501  * addition of a Global Route Header built from our Ethernet Address
2502  * Handle into our header list for connectionless packets.
2503  */
2504 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2505 {
2506 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2507 }
2508 
2509 /**
2510  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2511  *
2512  * @device: Device
2513  * @port_num: Port number
2514  *
2515  * This MAD size includes the MAD headers and MAD payload.  No other headers
2516  * are included.
2517  *
2518  * Return the max MAD size required by the Port.  Will return 0 if the port
2519  * does not support MADs
2520  */
2521 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2522 {
2523 	return device->port_immutable[port_num].max_mad_size;
2524 }
2525 
2526 /**
2527  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2528  * @device: Device to check
2529  * @port_num: Port number to check
2530  *
2531  * RoCE GID table mechanism manages the various GIDs for a device.
2532  *
2533  * NOTE: if allocating the port's GID table has failed, this call will still
2534  * return true, but any RoCE GID table API will fail.
2535  *
2536  * Return: true if the port uses RoCE GID table mechanism in order to manage
2537  * its GIDs.
2538  */
2539 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2540 					   u8 port_num)
2541 {
2542 	return rdma_protocol_roce(device, port_num) &&
2543 		device->add_gid && device->del_gid;
2544 }
2545 
2546 /*
2547  * Check if the device supports READ W/ INVALIDATE.
2548  */
2549 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2550 {
2551 	/*
2552 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2553 	 * has support for it yet.
2554 	 */
2555 	return rdma_protocol_iwarp(dev, port_num);
2556 }
2557 
2558 int ib_query_gid(struct ib_device *device,
2559 		 u8 port_num, int index, union ib_gid *gid,
2560 		 struct ib_gid_attr *attr);
2561 
2562 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2563 			 int state);
2564 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2565 		     struct ifla_vf_info *info);
2566 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2567 		    struct ifla_vf_stats *stats);
2568 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2569 		   int type);
2570 
2571 int ib_query_pkey(struct ib_device *device,
2572 		  u8 port_num, u16 index, u16 *pkey);
2573 
2574 int ib_modify_device(struct ib_device *device,
2575 		     int device_modify_mask,
2576 		     struct ib_device_modify *device_modify);
2577 
2578 int ib_modify_port(struct ib_device *device,
2579 		   u8 port_num, int port_modify_mask,
2580 		   struct ib_port_modify *port_modify);
2581 
2582 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2583 		enum ib_gid_type gid_type, struct net_device *ndev,
2584 		u8 *port_num, u16 *index);
2585 
2586 int ib_find_pkey(struct ib_device *device,
2587 		 u8 port_num, u16 pkey, u16 *index);
2588 
2589 enum ib_pd_flags {
2590 	/*
2591 	 * Create a memory registration for all memory in the system and place
2592 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2593 	 * ULPs to avoid the overhead of dynamic MRs.
2594 	 *
2595 	 * This flag is generally considered unsafe and must only be used in
2596 	 * extremly trusted environments.  Every use of it will log a warning
2597 	 * in the kernel log.
2598 	 */
2599 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2600 };
2601 
2602 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2603 		const char *caller);
2604 #define ib_alloc_pd(device, flags) \
2605 	__ib_alloc_pd((device), (flags), __func__)
2606 void ib_dealloc_pd(struct ib_pd *pd);
2607 
2608 /**
2609  * ib_create_ah - Creates an address handle for the given address vector.
2610  * @pd: The protection domain associated with the address handle.
2611  * @ah_attr: The attributes of the address vector.
2612  *
2613  * The address handle is used to reference a local or global destination
2614  * in all UD QP post sends.
2615  */
2616 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2617 
2618 /**
2619  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2620  *   work completion.
2621  * @hdr: the L3 header to parse
2622  * @net_type: type of header to parse
2623  * @sgid: place to store source gid
2624  * @dgid: place to store destination gid
2625  */
2626 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2627 			      enum rdma_network_type net_type,
2628 			      union ib_gid *sgid, union ib_gid *dgid);
2629 
2630 /**
2631  * ib_get_rdma_header_version - Get the header version
2632  * @hdr: the L3 header to parse
2633  */
2634 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2635 
2636 /**
2637  * ib_init_ah_from_wc - Initializes address handle attributes from a
2638  *   work completion.
2639  * @device: Device on which the received message arrived.
2640  * @port_num: Port on which the received message arrived.
2641  * @wc: Work completion associated with the received message.
2642  * @grh: References the received global route header.  This parameter is
2643  *   ignored unless the work completion indicates that the GRH is valid.
2644  * @ah_attr: Returned attributes that can be used when creating an address
2645  *   handle for replying to the message.
2646  */
2647 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2648 		       const struct ib_wc *wc, const struct ib_grh *grh,
2649 		       struct ib_ah_attr *ah_attr);
2650 
2651 /**
2652  * ib_create_ah_from_wc - Creates an address handle associated with the
2653  *   sender of the specified work completion.
2654  * @pd: The protection domain associated with the address handle.
2655  * @wc: Work completion information associated with a received message.
2656  * @grh: References the received global route header.  This parameter is
2657  *   ignored unless the work completion indicates that the GRH is valid.
2658  * @port_num: The outbound port number to associate with the address.
2659  *
2660  * The address handle is used to reference a local or global destination
2661  * in all UD QP post sends.
2662  */
2663 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2664 				   const struct ib_grh *grh, u8 port_num);
2665 
2666 /**
2667  * ib_modify_ah - Modifies the address vector associated with an address
2668  *   handle.
2669  * @ah: The address handle to modify.
2670  * @ah_attr: The new address vector attributes to associate with the
2671  *   address handle.
2672  */
2673 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2674 
2675 /**
2676  * ib_query_ah - Queries the address vector associated with an address
2677  *   handle.
2678  * @ah: The address handle to query.
2679  * @ah_attr: The address vector attributes associated with the address
2680  *   handle.
2681  */
2682 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2683 
2684 /**
2685  * ib_destroy_ah - Destroys an address handle.
2686  * @ah: The address handle to destroy.
2687  */
2688 int ib_destroy_ah(struct ib_ah *ah);
2689 
2690 /**
2691  * ib_create_srq - Creates a SRQ associated with the specified protection
2692  *   domain.
2693  * @pd: The protection domain associated with the SRQ.
2694  * @srq_init_attr: A list of initial attributes required to create the
2695  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2696  *   the actual capabilities of the created SRQ.
2697  *
2698  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2699  * requested size of the SRQ, and set to the actual values allocated
2700  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2701  * will always be at least as large as the requested values.
2702  */
2703 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2704 			     struct ib_srq_init_attr *srq_init_attr);
2705 
2706 /**
2707  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2708  * @srq: The SRQ to modify.
2709  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2710  *   the current values of selected SRQ attributes are returned.
2711  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2712  *   are being modified.
2713  *
2714  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2715  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2716  * the number of receives queued drops below the limit.
2717  */
2718 int ib_modify_srq(struct ib_srq *srq,
2719 		  struct ib_srq_attr *srq_attr,
2720 		  enum ib_srq_attr_mask srq_attr_mask);
2721 
2722 /**
2723  * ib_query_srq - Returns the attribute list and current values for the
2724  *   specified SRQ.
2725  * @srq: The SRQ to query.
2726  * @srq_attr: The attributes of the specified SRQ.
2727  */
2728 int ib_query_srq(struct ib_srq *srq,
2729 		 struct ib_srq_attr *srq_attr);
2730 
2731 /**
2732  * ib_destroy_srq - Destroys the specified SRQ.
2733  * @srq: The SRQ to destroy.
2734  */
2735 int ib_destroy_srq(struct ib_srq *srq);
2736 
2737 /**
2738  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2739  * @srq: The SRQ to post the work request on.
2740  * @recv_wr: A list of work requests to post on the receive queue.
2741  * @bad_recv_wr: On an immediate failure, this parameter will reference
2742  *   the work request that failed to be posted on the QP.
2743  */
2744 static inline int ib_post_srq_recv(struct ib_srq *srq,
2745 				   struct ib_recv_wr *recv_wr,
2746 				   struct ib_recv_wr **bad_recv_wr)
2747 {
2748 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2749 }
2750 
2751 /**
2752  * ib_create_qp - Creates a QP associated with the specified protection
2753  *   domain.
2754  * @pd: The protection domain associated with the QP.
2755  * @qp_init_attr: A list of initial attributes required to create the
2756  *   QP.  If QP creation succeeds, then the attributes are updated to
2757  *   the actual capabilities of the created QP.
2758  */
2759 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2760 			   struct ib_qp_init_attr *qp_init_attr);
2761 
2762 /**
2763  * ib_modify_qp - Modifies the attributes for the specified QP and then
2764  *   transitions the QP to the given state.
2765  * @qp: The QP to modify.
2766  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2767  *   the current values of selected QP attributes are returned.
2768  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2769  *   are being modified.
2770  */
2771 int ib_modify_qp(struct ib_qp *qp,
2772 		 struct ib_qp_attr *qp_attr,
2773 		 int qp_attr_mask);
2774 
2775 /**
2776  * ib_query_qp - Returns the attribute list and current values for the
2777  *   specified QP.
2778  * @qp: The QP to query.
2779  * @qp_attr: The attributes of the specified QP.
2780  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2781  * @qp_init_attr: Additional attributes of the selected QP.
2782  *
2783  * The qp_attr_mask may be used to limit the query to gathering only the
2784  * selected attributes.
2785  */
2786 int ib_query_qp(struct ib_qp *qp,
2787 		struct ib_qp_attr *qp_attr,
2788 		int qp_attr_mask,
2789 		struct ib_qp_init_attr *qp_init_attr);
2790 
2791 /**
2792  * ib_destroy_qp - Destroys the specified QP.
2793  * @qp: The QP to destroy.
2794  */
2795 int ib_destroy_qp(struct ib_qp *qp);
2796 
2797 /**
2798  * ib_open_qp - Obtain a reference to an existing sharable QP.
2799  * @xrcd - XRC domain
2800  * @qp_open_attr: Attributes identifying the QP to open.
2801  *
2802  * Returns a reference to a sharable QP.
2803  */
2804 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2805 			 struct ib_qp_open_attr *qp_open_attr);
2806 
2807 /**
2808  * ib_close_qp - Release an external reference to a QP.
2809  * @qp: The QP handle to release
2810  *
2811  * The opened QP handle is released by the caller.  The underlying
2812  * shared QP is not destroyed until all internal references are released.
2813  */
2814 int ib_close_qp(struct ib_qp *qp);
2815 
2816 /**
2817  * ib_post_send - Posts a list of work requests to the send queue of
2818  *   the specified QP.
2819  * @qp: The QP to post the work request on.
2820  * @send_wr: A list of work requests to post on the send queue.
2821  * @bad_send_wr: On an immediate failure, this parameter will reference
2822  *   the work request that failed to be posted on the QP.
2823  *
2824  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2825  * error is returned, the QP state shall not be affected,
2826  * ib_post_send() will return an immediate error after queueing any
2827  * earlier work requests in the list.
2828  */
2829 static inline int ib_post_send(struct ib_qp *qp,
2830 			       struct ib_send_wr *send_wr,
2831 			       struct ib_send_wr **bad_send_wr)
2832 {
2833 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2834 }
2835 
2836 /**
2837  * ib_post_recv - Posts a list of work requests to the receive queue of
2838  *   the specified QP.
2839  * @qp: The QP to post the work request on.
2840  * @recv_wr: A list of work requests to post on the receive queue.
2841  * @bad_recv_wr: On an immediate failure, this parameter will reference
2842  *   the work request that failed to be posted on the QP.
2843  */
2844 static inline int ib_post_recv(struct ib_qp *qp,
2845 			       struct ib_recv_wr *recv_wr,
2846 			       struct ib_recv_wr **bad_recv_wr)
2847 {
2848 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2849 }
2850 
2851 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2852 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2853 void ib_free_cq(struct ib_cq *cq);
2854 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2855 
2856 /**
2857  * ib_create_cq - Creates a CQ on the specified device.
2858  * @device: The device on which to create the CQ.
2859  * @comp_handler: A user-specified callback that is invoked when a
2860  *   completion event occurs on the CQ.
2861  * @event_handler: A user-specified callback that is invoked when an
2862  *   asynchronous event not associated with a completion occurs on the CQ.
2863  * @cq_context: Context associated with the CQ returned to the user via
2864  *   the associated completion and event handlers.
2865  * @cq_attr: The attributes the CQ should be created upon.
2866  *
2867  * Users can examine the cq structure to determine the actual CQ size.
2868  */
2869 struct ib_cq *ib_create_cq(struct ib_device *device,
2870 			   ib_comp_handler comp_handler,
2871 			   void (*event_handler)(struct ib_event *, void *),
2872 			   void *cq_context,
2873 			   const struct ib_cq_init_attr *cq_attr);
2874 
2875 /**
2876  * ib_resize_cq - Modifies the capacity of the CQ.
2877  * @cq: The CQ to resize.
2878  * @cqe: The minimum size of the CQ.
2879  *
2880  * Users can examine the cq structure to determine the actual CQ size.
2881  */
2882 int ib_resize_cq(struct ib_cq *cq, int cqe);
2883 
2884 /**
2885  * ib_modify_cq - Modifies moderation params of the CQ
2886  * @cq: The CQ to modify.
2887  * @cq_count: number of CQEs that will trigger an event
2888  * @cq_period: max period of time in usec before triggering an event
2889  *
2890  */
2891 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2892 
2893 /**
2894  * ib_destroy_cq - Destroys the specified CQ.
2895  * @cq: The CQ to destroy.
2896  */
2897 int ib_destroy_cq(struct ib_cq *cq);
2898 
2899 /**
2900  * ib_poll_cq - poll a CQ for completion(s)
2901  * @cq:the CQ being polled
2902  * @num_entries:maximum number of completions to return
2903  * @wc:array of at least @num_entries &struct ib_wc where completions
2904  *   will be returned
2905  *
2906  * Poll a CQ for (possibly multiple) completions.  If the return value
2907  * is < 0, an error occurred.  If the return value is >= 0, it is the
2908  * number of completions returned.  If the return value is
2909  * non-negative and < num_entries, then the CQ was emptied.
2910  */
2911 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2912 			     struct ib_wc *wc)
2913 {
2914 	return cq->device->poll_cq(cq, num_entries, wc);
2915 }
2916 
2917 /**
2918  * ib_peek_cq - Returns the number of unreaped completions currently
2919  *   on the specified CQ.
2920  * @cq: The CQ to peek.
2921  * @wc_cnt: A minimum number of unreaped completions to check for.
2922  *
2923  * If the number of unreaped completions is greater than or equal to wc_cnt,
2924  * this function returns wc_cnt, otherwise, it returns the actual number of
2925  * unreaped completions.
2926  */
2927 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2928 
2929 /**
2930  * ib_req_notify_cq - Request completion notification on a CQ.
2931  * @cq: The CQ to generate an event for.
2932  * @flags:
2933  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2934  *   to request an event on the next solicited event or next work
2935  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2936  *   may also be |ed in to request a hint about missed events, as
2937  *   described below.
2938  *
2939  * Return Value:
2940  *    < 0 means an error occurred while requesting notification
2941  *   == 0 means notification was requested successfully, and if
2942  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2943  *        were missed and it is safe to wait for another event.  In
2944  *        this case is it guaranteed that any work completions added
2945  *        to the CQ since the last CQ poll will trigger a completion
2946  *        notification event.
2947  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2948  *        in.  It means that the consumer must poll the CQ again to
2949  *        make sure it is empty to avoid missing an event because of a
2950  *        race between requesting notification and an entry being
2951  *        added to the CQ.  This return value means it is possible
2952  *        (but not guaranteed) that a work completion has been added
2953  *        to the CQ since the last poll without triggering a
2954  *        completion notification event.
2955  */
2956 static inline int ib_req_notify_cq(struct ib_cq *cq,
2957 				   enum ib_cq_notify_flags flags)
2958 {
2959 	return cq->device->req_notify_cq(cq, flags);
2960 }
2961 
2962 /**
2963  * ib_req_ncomp_notif - Request completion notification when there are
2964  *   at least the specified number of unreaped completions on the CQ.
2965  * @cq: The CQ to generate an event for.
2966  * @wc_cnt: The number of unreaped completions that should be on the
2967  *   CQ before an event is generated.
2968  */
2969 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2970 {
2971 	return cq->device->req_ncomp_notif ?
2972 		cq->device->req_ncomp_notif(cq, wc_cnt) :
2973 		-ENOSYS;
2974 }
2975 
2976 /**
2977  * ib_dma_mapping_error - check a DMA addr for error
2978  * @dev: The device for which the dma_addr was created
2979  * @dma_addr: The DMA address to check
2980  */
2981 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2982 {
2983 	if (dev->dma_ops)
2984 		return dev->dma_ops->mapping_error(dev, dma_addr);
2985 	return dma_mapping_error(dev->dma_device, dma_addr);
2986 }
2987 
2988 /**
2989  * ib_dma_map_single - Map a kernel virtual address to DMA address
2990  * @dev: The device for which the dma_addr is to be created
2991  * @cpu_addr: The kernel virtual address
2992  * @size: The size of the region in bytes
2993  * @direction: The direction of the DMA
2994  */
2995 static inline u64 ib_dma_map_single(struct ib_device *dev,
2996 				    void *cpu_addr, size_t size,
2997 				    enum dma_data_direction direction)
2998 {
2999 	if (dev->dma_ops)
3000 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
3001 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3002 }
3003 
3004 /**
3005  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3006  * @dev: The device for which the DMA address was created
3007  * @addr: The DMA address
3008  * @size: The size of the region in bytes
3009  * @direction: The direction of the DMA
3010  */
3011 static inline void ib_dma_unmap_single(struct ib_device *dev,
3012 				       u64 addr, size_t size,
3013 				       enum dma_data_direction direction)
3014 {
3015 	if (dev->dma_ops)
3016 		dev->dma_ops->unmap_single(dev, addr, size, direction);
3017 	else
3018 		dma_unmap_single(dev->dma_device, addr, size, direction);
3019 }
3020 
3021 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
3022 					  void *cpu_addr, size_t size,
3023 					  enum dma_data_direction direction,
3024 					  unsigned long dma_attrs)
3025 {
3026 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
3027 				    direction, dma_attrs);
3028 }
3029 
3030 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
3031 					     u64 addr, size_t size,
3032 					     enum dma_data_direction direction,
3033 					     unsigned long dma_attrs)
3034 {
3035 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
3036 				      direction, dma_attrs);
3037 }
3038 
3039 /**
3040  * ib_dma_map_page - Map a physical page to DMA address
3041  * @dev: The device for which the dma_addr is to be created
3042  * @page: The page to be mapped
3043  * @offset: The offset within the page
3044  * @size: The size of the region in bytes
3045  * @direction: The direction of the DMA
3046  */
3047 static inline u64 ib_dma_map_page(struct ib_device *dev,
3048 				  struct page *page,
3049 				  unsigned long offset,
3050 				  size_t size,
3051 					 enum dma_data_direction direction)
3052 {
3053 	if (dev->dma_ops)
3054 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
3055 	return dma_map_page(dev->dma_device, 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 	if (dev->dma_ops)
3070 		dev->dma_ops->unmap_page(dev, addr, size, direction);
3071 	else
3072 		dma_unmap_page(dev->dma_device, addr, size, direction);
3073 }
3074 
3075 /**
3076  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3077  * @dev: The device for which the DMA addresses are to be created
3078  * @sg: The array of scatter/gather entries
3079  * @nents: The number of scatter/gather entries
3080  * @direction: The direction of the DMA
3081  */
3082 static inline int ib_dma_map_sg(struct ib_device *dev,
3083 				struct scatterlist *sg, int nents,
3084 				enum dma_data_direction direction)
3085 {
3086 	if (dev->dma_ops)
3087 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
3088 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3089 }
3090 
3091 /**
3092  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3093  * @dev: The device for which the DMA addresses were created
3094  * @sg: The array of scatter/gather entries
3095  * @nents: The number of scatter/gather entries
3096  * @direction: The direction of the DMA
3097  */
3098 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3099 				   struct scatterlist *sg, int nents,
3100 				   enum dma_data_direction direction)
3101 {
3102 	if (dev->dma_ops)
3103 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
3104 	else
3105 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
3106 }
3107 
3108 static inline int ib_dma_map_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 	if (dev->dma_ops)
3114 		return dev->dma_ops->map_sg_attrs(dev, sg, nents, direction,
3115 						  dma_attrs);
3116 	else
3117 		return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3118 					dma_attrs);
3119 }
3120 
3121 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3122 					 struct scatterlist *sg, int nents,
3123 					 enum dma_data_direction direction,
3124 					 unsigned long dma_attrs)
3125 {
3126 	if (dev->dma_ops)
3127 		return dev->dma_ops->unmap_sg_attrs(dev, sg, nents, direction,
3128 						  dma_attrs);
3129 	else
3130 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
3131 				   dma_attrs);
3132 }
3133 /**
3134  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3135  * @dev: The device for which the DMA addresses were created
3136  * @sg: The scatter/gather entry
3137  *
3138  * Note: this function is obsolete. To do: change all occurrences of
3139  * ib_sg_dma_address() into sg_dma_address().
3140  */
3141 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3142 				    struct scatterlist *sg)
3143 {
3144 	return sg_dma_address(sg);
3145 }
3146 
3147 /**
3148  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3149  * @dev: The device for which the DMA addresses were created
3150  * @sg: The scatter/gather entry
3151  *
3152  * Note: this function is obsolete. To do: change all occurrences of
3153  * ib_sg_dma_len() into sg_dma_len().
3154  */
3155 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3156 					 struct scatterlist *sg)
3157 {
3158 	return sg_dma_len(sg);
3159 }
3160 
3161 /**
3162  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3163  * @dev: The device for which the DMA address was created
3164  * @addr: The DMA address
3165  * @size: The size of the region in bytes
3166  * @dir: The direction of the DMA
3167  */
3168 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3169 					      u64 addr,
3170 					      size_t size,
3171 					      enum dma_data_direction dir)
3172 {
3173 	if (dev->dma_ops)
3174 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
3175 	else
3176 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3177 }
3178 
3179 /**
3180  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3181  * @dev: The device for which the DMA address was created
3182  * @addr: The DMA address
3183  * @size: The size of the region in bytes
3184  * @dir: The direction of the DMA
3185  */
3186 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3187 						 u64 addr,
3188 						 size_t size,
3189 						 enum dma_data_direction dir)
3190 {
3191 	if (dev->dma_ops)
3192 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
3193 	else
3194 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3195 }
3196 
3197 /**
3198  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3199  * @dev: The device for which the DMA address is requested
3200  * @size: The size of the region to allocate in bytes
3201  * @dma_handle: A pointer for returning the DMA address of the region
3202  * @flag: memory allocator flags
3203  */
3204 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3205 					   size_t size,
3206 					   u64 *dma_handle,
3207 					   gfp_t flag)
3208 {
3209 	if (dev->dma_ops)
3210 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
3211 	else {
3212 		dma_addr_t handle;
3213 		void *ret;
3214 
3215 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3216 		*dma_handle = handle;
3217 		return ret;
3218 	}
3219 }
3220 
3221 /**
3222  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3223  * @dev: The device for which the DMA addresses were allocated
3224  * @size: The size of the region
3225  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3226  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3227  */
3228 static inline void ib_dma_free_coherent(struct ib_device *dev,
3229 					size_t size, void *cpu_addr,
3230 					u64 dma_handle)
3231 {
3232 	if (dev->dma_ops)
3233 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3234 	else
3235 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3236 }
3237 
3238 /**
3239  * ib_dereg_mr - Deregisters a memory region and removes it from the
3240  *   HCA translation table.
3241  * @mr: The memory region to deregister.
3242  *
3243  * This function can fail, if the memory region has memory windows bound to it.
3244  */
3245 int ib_dereg_mr(struct ib_mr *mr);
3246 
3247 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3248 			  enum ib_mr_type mr_type,
3249 			  u32 max_num_sg);
3250 
3251 /**
3252  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3253  *   R_Key and L_Key.
3254  * @mr - struct ib_mr pointer to be updated.
3255  * @newkey - new key to be used.
3256  */
3257 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3258 {
3259 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3260 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3261 }
3262 
3263 /**
3264  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3265  * for calculating a new rkey for type 2 memory windows.
3266  * @rkey - the rkey to increment.
3267  */
3268 static inline u32 ib_inc_rkey(u32 rkey)
3269 {
3270 	const u32 mask = 0x000000ff;
3271 	return ((rkey + 1) & mask) | (rkey & ~mask);
3272 }
3273 
3274 /**
3275  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3276  * @pd: The protection domain associated with the unmapped region.
3277  * @mr_access_flags: Specifies the memory access rights.
3278  * @fmr_attr: Attributes of the unmapped region.
3279  *
3280  * A fast memory region must be mapped before it can be used as part of
3281  * a work request.
3282  */
3283 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3284 			    int mr_access_flags,
3285 			    struct ib_fmr_attr *fmr_attr);
3286 
3287 /**
3288  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3289  * @fmr: The fast memory region to associate with the pages.
3290  * @page_list: An array of physical pages to map to the fast memory region.
3291  * @list_len: The number of pages in page_list.
3292  * @iova: The I/O virtual address to use with the mapped region.
3293  */
3294 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3295 				  u64 *page_list, int list_len,
3296 				  u64 iova)
3297 {
3298 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3299 }
3300 
3301 /**
3302  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3303  * @fmr_list: A linked list of fast memory regions to unmap.
3304  */
3305 int ib_unmap_fmr(struct list_head *fmr_list);
3306 
3307 /**
3308  * ib_dealloc_fmr - Deallocates a fast memory region.
3309  * @fmr: The fast memory region to deallocate.
3310  */
3311 int ib_dealloc_fmr(struct ib_fmr *fmr);
3312 
3313 /**
3314  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3315  * @qp: QP to attach to the multicast group.  The QP must be type
3316  *   IB_QPT_UD.
3317  * @gid: Multicast group GID.
3318  * @lid: Multicast group LID in host byte order.
3319  *
3320  * In order to send and receive multicast packets, subnet
3321  * administration must have created the multicast group and configured
3322  * the fabric appropriately.  The port associated with the specified
3323  * QP must also be a member of the multicast group.
3324  */
3325 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3326 
3327 /**
3328  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3329  * @qp: QP to detach from the multicast group.
3330  * @gid: Multicast group GID.
3331  * @lid: Multicast group LID in host byte order.
3332  */
3333 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3334 
3335 /**
3336  * ib_alloc_xrcd - Allocates an XRC domain.
3337  * @device: The device on which to allocate the XRC domain.
3338  */
3339 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3340 
3341 /**
3342  * ib_dealloc_xrcd - Deallocates an XRC domain.
3343  * @xrcd: The XRC domain to deallocate.
3344  */
3345 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3346 
3347 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3348 			       struct ib_flow_attr *flow_attr, int domain);
3349 int ib_destroy_flow(struct ib_flow *flow_id);
3350 
3351 static inline int ib_check_mr_access(int flags)
3352 {
3353 	/*
3354 	 * Local write permission is required if remote write or
3355 	 * remote atomic permission is also requested.
3356 	 */
3357 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3358 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3359 		return -EINVAL;
3360 
3361 	return 0;
3362 }
3363 
3364 /**
3365  * ib_check_mr_status: lightweight check of MR status.
3366  *     This routine may provide status checks on a selected
3367  *     ib_mr. first use is for signature status check.
3368  *
3369  * @mr: A memory region.
3370  * @check_mask: Bitmask of which checks to perform from
3371  *     ib_mr_status_check enumeration.
3372  * @mr_status: The container of relevant status checks.
3373  *     failed checks will be indicated in the status bitmask
3374  *     and the relevant info shall be in the error item.
3375  */
3376 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3377 		       struct ib_mr_status *mr_status);
3378 
3379 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3380 					    u16 pkey, const union ib_gid *gid,
3381 					    const struct sockaddr *addr);
3382 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3383 			   struct ib_wq_init_attr *init_attr);
3384 int ib_destroy_wq(struct ib_wq *wq);
3385 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3386 		 u32 wq_attr_mask);
3387 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3388 						 struct ib_rwq_ind_table_init_attr*
3389 						 wq_ind_table_init_attr);
3390 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3391 
3392 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3393 		 unsigned int *sg_offset, unsigned int page_size);
3394 
3395 static inline int
3396 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3397 		  unsigned int *sg_offset, unsigned int page_size)
3398 {
3399 	int n;
3400 
3401 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3402 	mr->iova = 0;
3403 
3404 	return n;
3405 }
3406 
3407 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3408 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3409 
3410 void ib_drain_rq(struct ib_qp *qp);
3411 void ib_drain_sq(struct ib_qp *qp);
3412 void ib_drain_qp(struct ib_qp *qp);
3413 
3414 int ib_resolve_eth_dmac(struct ib_device *device,
3415 			struct ib_ah_attr *ah_attr);
3416 #endif /* IB_VERBS_H */
3417