xref: /openbmc/linux/include/rdma/ib_verbs.h (revision 6dfcd296)
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 <asm/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 enum ib_port_state {
356 	IB_PORT_NOP		= 0,
357 	IB_PORT_DOWN		= 1,
358 	IB_PORT_INIT		= 2,
359 	IB_PORT_ARMED		= 3,
360 	IB_PORT_ACTIVE		= 4,
361 	IB_PORT_ACTIVE_DEFER	= 5
362 };
363 
364 enum ib_port_cap_flags {
365 	IB_PORT_SM				= 1 <<  1,
366 	IB_PORT_NOTICE_SUP			= 1 <<  2,
367 	IB_PORT_TRAP_SUP			= 1 <<  3,
368 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
369 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
370 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
371 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
372 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
373 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
374 	IB_PORT_SM_DISABLED			= 1 << 10,
375 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
376 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
377 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
378 	IB_PORT_CM_SUP				= 1 << 16,
379 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
380 	IB_PORT_REINIT_SUP			= 1 << 18,
381 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
382 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
383 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
384 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
385 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
386 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
387 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
388 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
389 };
390 
391 enum ib_port_width {
392 	IB_WIDTH_1X	= 1,
393 	IB_WIDTH_4X	= 2,
394 	IB_WIDTH_8X	= 4,
395 	IB_WIDTH_12X	= 8
396 };
397 
398 static inline int ib_width_enum_to_int(enum ib_port_width width)
399 {
400 	switch (width) {
401 	case IB_WIDTH_1X:  return  1;
402 	case IB_WIDTH_4X:  return  4;
403 	case IB_WIDTH_8X:  return  8;
404 	case IB_WIDTH_12X: return 12;
405 	default: 	  return -1;
406 	}
407 }
408 
409 enum ib_port_speed {
410 	IB_SPEED_SDR	= 1,
411 	IB_SPEED_DDR	= 2,
412 	IB_SPEED_QDR	= 4,
413 	IB_SPEED_FDR10	= 8,
414 	IB_SPEED_FDR	= 16,
415 	IB_SPEED_EDR	= 32
416 };
417 
418 /**
419  * struct rdma_hw_stats
420  * @timestamp - Used by the core code to track when the last update was
421  * @lifespan - Used by the core code to determine how old the counters
422  *   should be before being updated again.  Stored in jiffies, defaults
423  *   to 10 milliseconds, drivers can override the default be specifying
424  *   their own value during their allocation routine.
425  * @name - Array of pointers to static names used for the counters in
426  *   directory.
427  * @num_counters - How many hardware counters there are.  If name is
428  *   shorter than this number, a kernel oops will result.  Driver authors
429  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
430  *   in their code to prevent this.
431  * @value - Array of u64 counters that are accessed by the sysfs code and
432  *   filled in by the drivers get_stats routine
433  */
434 struct rdma_hw_stats {
435 	unsigned long	timestamp;
436 	unsigned long	lifespan;
437 	const char * const *names;
438 	int		num_counters;
439 	u64		value[];
440 };
441 
442 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
443 /**
444  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
445  *   for drivers.
446  * @names - Array of static const char *
447  * @num_counters - How many elements in array
448  * @lifespan - How many milliseconds between updates
449  */
450 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
451 		const char * const *names, int num_counters,
452 		unsigned long lifespan)
453 {
454 	struct rdma_hw_stats *stats;
455 
456 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
457 			GFP_KERNEL);
458 	if (!stats)
459 		return NULL;
460 	stats->names = names;
461 	stats->num_counters = num_counters;
462 	stats->lifespan = msecs_to_jiffies(lifespan);
463 
464 	return stats;
465 }
466 
467 
468 /* Define bits for the various functionality this port needs to be supported by
469  * the core.
470  */
471 /* Management                           0x00000FFF */
472 #define RDMA_CORE_CAP_IB_MAD            0x00000001
473 #define RDMA_CORE_CAP_IB_SMI            0x00000002
474 #define RDMA_CORE_CAP_IB_CM             0x00000004
475 #define RDMA_CORE_CAP_IW_CM             0x00000008
476 #define RDMA_CORE_CAP_IB_SA             0x00000010
477 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
478 
479 /* Address format                       0x000FF000 */
480 #define RDMA_CORE_CAP_AF_IB             0x00001000
481 #define RDMA_CORE_CAP_ETH_AH            0x00002000
482 
483 /* Protocol                             0xFFF00000 */
484 #define RDMA_CORE_CAP_PROT_IB           0x00100000
485 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
486 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
487 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
488 
489 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
490 					| RDMA_CORE_CAP_IB_MAD \
491 					| RDMA_CORE_CAP_IB_SMI \
492 					| RDMA_CORE_CAP_IB_CM  \
493 					| RDMA_CORE_CAP_IB_SA  \
494 					| RDMA_CORE_CAP_AF_IB)
495 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
496 					| RDMA_CORE_CAP_IB_MAD  \
497 					| RDMA_CORE_CAP_IB_CM   \
498 					| RDMA_CORE_CAP_AF_IB   \
499 					| RDMA_CORE_CAP_ETH_AH)
500 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
501 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
502 					| RDMA_CORE_CAP_IB_MAD  \
503 					| RDMA_CORE_CAP_IB_CM   \
504 					| RDMA_CORE_CAP_AF_IB   \
505 					| RDMA_CORE_CAP_ETH_AH)
506 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
507 					| RDMA_CORE_CAP_IW_CM)
508 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
509 					| RDMA_CORE_CAP_OPA_MAD)
510 
511 struct ib_port_attr {
512 	u64			subnet_prefix;
513 	enum ib_port_state	state;
514 	enum ib_mtu		max_mtu;
515 	enum ib_mtu		active_mtu;
516 	int			gid_tbl_len;
517 	u32			port_cap_flags;
518 	u32			max_msg_sz;
519 	u32			bad_pkey_cntr;
520 	u32			qkey_viol_cntr;
521 	u16			pkey_tbl_len;
522 	u16			lid;
523 	u16			sm_lid;
524 	u8			lmc;
525 	u8			max_vl_num;
526 	u8			sm_sl;
527 	u8			subnet_timeout;
528 	u8			init_type_reply;
529 	u8			active_width;
530 	u8			active_speed;
531 	u8                      phys_state;
532 	bool			grh_required;
533 };
534 
535 enum ib_device_modify_flags {
536 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
537 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
538 };
539 
540 #define IB_DEVICE_NODE_DESC_MAX 64
541 
542 struct ib_device_modify {
543 	u64	sys_image_guid;
544 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
545 };
546 
547 enum ib_port_modify_flags {
548 	IB_PORT_SHUTDOWN		= 1,
549 	IB_PORT_INIT_TYPE		= (1<<2),
550 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
551 };
552 
553 struct ib_port_modify {
554 	u32	set_port_cap_mask;
555 	u32	clr_port_cap_mask;
556 	u8	init_type;
557 };
558 
559 enum ib_event_type {
560 	IB_EVENT_CQ_ERR,
561 	IB_EVENT_QP_FATAL,
562 	IB_EVENT_QP_REQ_ERR,
563 	IB_EVENT_QP_ACCESS_ERR,
564 	IB_EVENT_COMM_EST,
565 	IB_EVENT_SQ_DRAINED,
566 	IB_EVENT_PATH_MIG,
567 	IB_EVENT_PATH_MIG_ERR,
568 	IB_EVENT_DEVICE_FATAL,
569 	IB_EVENT_PORT_ACTIVE,
570 	IB_EVENT_PORT_ERR,
571 	IB_EVENT_LID_CHANGE,
572 	IB_EVENT_PKEY_CHANGE,
573 	IB_EVENT_SM_CHANGE,
574 	IB_EVENT_SRQ_ERR,
575 	IB_EVENT_SRQ_LIMIT_REACHED,
576 	IB_EVENT_QP_LAST_WQE_REACHED,
577 	IB_EVENT_CLIENT_REREGISTER,
578 	IB_EVENT_GID_CHANGE,
579 	IB_EVENT_WQ_FATAL,
580 };
581 
582 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
583 
584 struct ib_event {
585 	struct ib_device	*device;
586 	union {
587 		struct ib_cq	*cq;
588 		struct ib_qp	*qp;
589 		struct ib_srq	*srq;
590 		struct ib_wq	*wq;
591 		u8		port_num;
592 	} element;
593 	enum ib_event_type	event;
594 };
595 
596 struct ib_event_handler {
597 	struct ib_device *device;
598 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
599 	struct list_head  list;
600 };
601 
602 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
603 	do {							\
604 		(_ptr)->device  = _device;			\
605 		(_ptr)->handler = _handler;			\
606 		INIT_LIST_HEAD(&(_ptr)->list);			\
607 	} while (0)
608 
609 struct ib_global_route {
610 	union ib_gid	dgid;
611 	u32		flow_label;
612 	u8		sgid_index;
613 	u8		hop_limit;
614 	u8		traffic_class;
615 };
616 
617 struct ib_grh {
618 	__be32		version_tclass_flow;
619 	__be16		paylen;
620 	u8		next_hdr;
621 	u8		hop_limit;
622 	union ib_gid	sgid;
623 	union ib_gid	dgid;
624 };
625 
626 union rdma_network_hdr {
627 	struct ib_grh ibgrh;
628 	struct {
629 		/* The IB spec states that if it's IPv4, the header
630 		 * is located in the last 20 bytes of the header.
631 		 */
632 		u8		reserved[20];
633 		struct iphdr	roce4grh;
634 	};
635 };
636 
637 enum {
638 	IB_MULTICAST_QPN = 0xffffff
639 };
640 
641 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
642 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
643 
644 enum ib_ah_flags {
645 	IB_AH_GRH	= 1
646 };
647 
648 enum ib_rate {
649 	IB_RATE_PORT_CURRENT = 0,
650 	IB_RATE_2_5_GBPS = 2,
651 	IB_RATE_5_GBPS   = 5,
652 	IB_RATE_10_GBPS  = 3,
653 	IB_RATE_20_GBPS  = 6,
654 	IB_RATE_30_GBPS  = 4,
655 	IB_RATE_40_GBPS  = 7,
656 	IB_RATE_60_GBPS  = 8,
657 	IB_RATE_80_GBPS  = 9,
658 	IB_RATE_120_GBPS = 10,
659 	IB_RATE_14_GBPS  = 11,
660 	IB_RATE_56_GBPS  = 12,
661 	IB_RATE_112_GBPS = 13,
662 	IB_RATE_168_GBPS = 14,
663 	IB_RATE_25_GBPS  = 15,
664 	IB_RATE_100_GBPS = 16,
665 	IB_RATE_200_GBPS = 17,
666 	IB_RATE_300_GBPS = 18
667 };
668 
669 /**
670  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
671  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
672  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
673  * @rate: rate to convert.
674  */
675 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
676 
677 /**
678  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
679  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
680  * @rate: rate to convert.
681  */
682 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
683 
684 
685 /**
686  * enum ib_mr_type - memory region type
687  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
688  *                            normal registration
689  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
690  *                            signature operations (data-integrity
691  *                            capable regions)
692  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
693  *                            register any arbitrary sg lists (without
694  *                            the normal mr constraints - see
695  *                            ib_map_mr_sg)
696  */
697 enum ib_mr_type {
698 	IB_MR_TYPE_MEM_REG,
699 	IB_MR_TYPE_SIGNATURE,
700 	IB_MR_TYPE_SG_GAPS,
701 };
702 
703 /**
704  * Signature types
705  * IB_SIG_TYPE_NONE: Unprotected.
706  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
707  */
708 enum ib_signature_type {
709 	IB_SIG_TYPE_NONE,
710 	IB_SIG_TYPE_T10_DIF,
711 };
712 
713 /**
714  * Signature T10-DIF block-guard types
715  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
716  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
717  */
718 enum ib_t10_dif_bg_type {
719 	IB_T10DIF_CRC,
720 	IB_T10DIF_CSUM
721 };
722 
723 /**
724  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
725  *     domain.
726  * @bg_type: T10-DIF block guard type (CRC|CSUM)
727  * @pi_interval: protection information interval.
728  * @bg: seed of guard computation.
729  * @app_tag: application tag of guard block
730  * @ref_tag: initial guard block reference tag.
731  * @ref_remap: Indicate wethear the reftag increments each block
732  * @app_escape: Indicate to skip block check if apptag=0xffff
733  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
734  * @apptag_check_mask: check bitmask of application tag.
735  */
736 struct ib_t10_dif_domain {
737 	enum ib_t10_dif_bg_type bg_type;
738 	u16			pi_interval;
739 	u16			bg;
740 	u16			app_tag;
741 	u32			ref_tag;
742 	bool			ref_remap;
743 	bool			app_escape;
744 	bool			ref_escape;
745 	u16			apptag_check_mask;
746 };
747 
748 /**
749  * struct ib_sig_domain - Parameters for signature domain
750  * @sig_type: specific signauture type
751  * @sig: union of all signature domain attributes that may
752  *     be used to set domain layout.
753  */
754 struct ib_sig_domain {
755 	enum ib_signature_type sig_type;
756 	union {
757 		struct ib_t10_dif_domain dif;
758 	} sig;
759 };
760 
761 /**
762  * struct ib_sig_attrs - Parameters for signature handover operation
763  * @check_mask: bitmask for signature byte check (8 bytes)
764  * @mem: memory domain layout desciptor.
765  * @wire: wire domain layout desciptor.
766  */
767 struct ib_sig_attrs {
768 	u8			check_mask;
769 	struct ib_sig_domain	mem;
770 	struct ib_sig_domain	wire;
771 };
772 
773 enum ib_sig_err_type {
774 	IB_SIG_BAD_GUARD,
775 	IB_SIG_BAD_REFTAG,
776 	IB_SIG_BAD_APPTAG,
777 };
778 
779 /**
780  * struct ib_sig_err - signature error descriptor
781  */
782 struct ib_sig_err {
783 	enum ib_sig_err_type	err_type;
784 	u32			expected;
785 	u32			actual;
786 	u64			sig_err_offset;
787 	u32			key;
788 };
789 
790 enum ib_mr_status_check {
791 	IB_MR_CHECK_SIG_STATUS = 1,
792 };
793 
794 /**
795  * struct ib_mr_status - Memory region status container
796  *
797  * @fail_status: Bitmask of MR checks status. For each
798  *     failed check a corresponding status bit is set.
799  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
800  *     failure.
801  */
802 struct ib_mr_status {
803 	u32		    fail_status;
804 	struct ib_sig_err   sig_err;
805 };
806 
807 /**
808  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
809  * enum.
810  * @mult: multiple to convert.
811  */
812 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
813 
814 struct ib_ah_attr {
815 	struct ib_global_route	grh;
816 	u16			dlid;
817 	u8			sl;
818 	u8			src_path_bits;
819 	u8			static_rate;
820 	u8			ah_flags;
821 	u8			port_num;
822 	u8			dmac[ETH_ALEN];
823 };
824 
825 enum ib_wc_status {
826 	IB_WC_SUCCESS,
827 	IB_WC_LOC_LEN_ERR,
828 	IB_WC_LOC_QP_OP_ERR,
829 	IB_WC_LOC_EEC_OP_ERR,
830 	IB_WC_LOC_PROT_ERR,
831 	IB_WC_WR_FLUSH_ERR,
832 	IB_WC_MW_BIND_ERR,
833 	IB_WC_BAD_RESP_ERR,
834 	IB_WC_LOC_ACCESS_ERR,
835 	IB_WC_REM_INV_REQ_ERR,
836 	IB_WC_REM_ACCESS_ERR,
837 	IB_WC_REM_OP_ERR,
838 	IB_WC_RETRY_EXC_ERR,
839 	IB_WC_RNR_RETRY_EXC_ERR,
840 	IB_WC_LOC_RDD_VIOL_ERR,
841 	IB_WC_REM_INV_RD_REQ_ERR,
842 	IB_WC_REM_ABORT_ERR,
843 	IB_WC_INV_EECN_ERR,
844 	IB_WC_INV_EEC_STATE_ERR,
845 	IB_WC_FATAL_ERR,
846 	IB_WC_RESP_TIMEOUT_ERR,
847 	IB_WC_GENERAL_ERR
848 };
849 
850 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
851 
852 enum ib_wc_opcode {
853 	IB_WC_SEND,
854 	IB_WC_RDMA_WRITE,
855 	IB_WC_RDMA_READ,
856 	IB_WC_COMP_SWAP,
857 	IB_WC_FETCH_ADD,
858 	IB_WC_LSO,
859 	IB_WC_LOCAL_INV,
860 	IB_WC_REG_MR,
861 	IB_WC_MASKED_COMP_SWAP,
862 	IB_WC_MASKED_FETCH_ADD,
863 /*
864  * Set value of IB_WC_RECV so consumers can test if a completion is a
865  * receive by testing (opcode & IB_WC_RECV).
866  */
867 	IB_WC_RECV			= 1 << 7,
868 	IB_WC_RECV_RDMA_WITH_IMM
869 };
870 
871 enum ib_wc_flags {
872 	IB_WC_GRH		= 1,
873 	IB_WC_WITH_IMM		= (1<<1),
874 	IB_WC_WITH_INVALIDATE	= (1<<2),
875 	IB_WC_IP_CSUM_OK	= (1<<3),
876 	IB_WC_WITH_SMAC		= (1<<4),
877 	IB_WC_WITH_VLAN		= (1<<5),
878 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
879 };
880 
881 struct ib_wc {
882 	union {
883 		u64		wr_id;
884 		struct ib_cqe	*wr_cqe;
885 	};
886 	enum ib_wc_status	status;
887 	enum ib_wc_opcode	opcode;
888 	u32			vendor_err;
889 	u32			byte_len;
890 	struct ib_qp	       *qp;
891 	union {
892 		__be32		imm_data;
893 		u32		invalidate_rkey;
894 	} ex;
895 	u32			src_qp;
896 	int			wc_flags;
897 	u16			pkey_index;
898 	u16			slid;
899 	u8			sl;
900 	u8			dlid_path_bits;
901 	u8			port_num;	/* valid only for DR SMPs on switches */
902 	u8			smac[ETH_ALEN];
903 	u16			vlan_id;
904 	u8			network_hdr_type;
905 };
906 
907 enum ib_cq_notify_flags {
908 	IB_CQ_SOLICITED			= 1 << 0,
909 	IB_CQ_NEXT_COMP			= 1 << 1,
910 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
911 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
912 };
913 
914 enum ib_srq_type {
915 	IB_SRQT_BASIC,
916 	IB_SRQT_XRC
917 };
918 
919 enum ib_srq_attr_mask {
920 	IB_SRQ_MAX_WR	= 1 << 0,
921 	IB_SRQ_LIMIT	= 1 << 1,
922 };
923 
924 struct ib_srq_attr {
925 	u32	max_wr;
926 	u32	max_sge;
927 	u32	srq_limit;
928 };
929 
930 struct ib_srq_init_attr {
931 	void		      (*event_handler)(struct ib_event *, void *);
932 	void		       *srq_context;
933 	struct ib_srq_attr	attr;
934 	enum ib_srq_type	srq_type;
935 
936 	union {
937 		struct {
938 			struct ib_xrcd *xrcd;
939 			struct ib_cq   *cq;
940 		} xrc;
941 	} ext;
942 };
943 
944 struct ib_qp_cap {
945 	u32	max_send_wr;
946 	u32	max_recv_wr;
947 	u32	max_send_sge;
948 	u32	max_recv_sge;
949 	u32	max_inline_data;
950 
951 	/*
952 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
953 	 * ib_create_qp() will calculate the right amount of neededed WRs
954 	 * and MRs based on this.
955 	 */
956 	u32	max_rdma_ctxs;
957 };
958 
959 enum ib_sig_type {
960 	IB_SIGNAL_ALL_WR,
961 	IB_SIGNAL_REQ_WR
962 };
963 
964 enum ib_qp_type {
965 	/*
966 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
967 	 * here (and in that order) since the MAD layer uses them as
968 	 * indices into a 2-entry table.
969 	 */
970 	IB_QPT_SMI,
971 	IB_QPT_GSI,
972 
973 	IB_QPT_RC,
974 	IB_QPT_UC,
975 	IB_QPT_UD,
976 	IB_QPT_RAW_IPV6,
977 	IB_QPT_RAW_ETHERTYPE,
978 	IB_QPT_RAW_PACKET = 8,
979 	IB_QPT_XRC_INI = 9,
980 	IB_QPT_XRC_TGT,
981 	IB_QPT_MAX,
982 	/* Reserve a range for qp types internal to the low level driver.
983 	 * These qp types will not be visible at the IB core layer, so the
984 	 * IB_QPT_MAX usages should not be affected in the core layer
985 	 */
986 	IB_QPT_RESERVED1 = 0x1000,
987 	IB_QPT_RESERVED2,
988 	IB_QPT_RESERVED3,
989 	IB_QPT_RESERVED4,
990 	IB_QPT_RESERVED5,
991 	IB_QPT_RESERVED6,
992 	IB_QPT_RESERVED7,
993 	IB_QPT_RESERVED8,
994 	IB_QPT_RESERVED9,
995 	IB_QPT_RESERVED10,
996 };
997 
998 enum ib_qp_create_flags {
999 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1000 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1001 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1002 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1003 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1004 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1005 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1006 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
1007 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1008 	/* reserve bits 26-31 for low level drivers' internal use */
1009 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1010 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1011 };
1012 
1013 /*
1014  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1015  * callback to destroy the passed in QP.
1016  */
1017 
1018 struct ib_qp_init_attr {
1019 	void                  (*event_handler)(struct ib_event *, void *);
1020 	void		       *qp_context;
1021 	struct ib_cq	       *send_cq;
1022 	struct ib_cq	       *recv_cq;
1023 	struct ib_srq	       *srq;
1024 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1025 	struct ib_qp_cap	cap;
1026 	enum ib_sig_type	sq_sig_type;
1027 	enum ib_qp_type		qp_type;
1028 	enum ib_qp_create_flags	create_flags;
1029 
1030 	/*
1031 	 * Only needed for special QP types, or when using the RW API.
1032 	 */
1033 	u8			port_num;
1034 	struct ib_rwq_ind_table *rwq_ind_tbl;
1035 };
1036 
1037 struct ib_qp_open_attr {
1038 	void                  (*event_handler)(struct ib_event *, void *);
1039 	void		       *qp_context;
1040 	u32			qp_num;
1041 	enum ib_qp_type		qp_type;
1042 };
1043 
1044 enum ib_rnr_timeout {
1045 	IB_RNR_TIMER_655_36 =  0,
1046 	IB_RNR_TIMER_000_01 =  1,
1047 	IB_RNR_TIMER_000_02 =  2,
1048 	IB_RNR_TIMER_000_03 =  3,
1049 	IB_RNR_TIMER_000_04 =  4,
1050 	IB_RNR_TIMER_000_06 =  5,
1051 	IB_RNR_TIMER_000_08 =  6,
1052 	IB_RNR_TIMER_000_12 =  7,
1053 	IB_RNR_TIMER_000_16 =  8,
1054 	IB_RNR_TIMER_000_24 =  9,
1055 	IB_RNR_TIMER_000_32 = 10,
1056 	IB_RNR_TIMER_000_48 = 11,
1057 	IB_RNR_TIMER_000_64 = 12,
1058 	IB_RNR_TIMER_000_96 = 13,
1059 	IB_RNR_TIMER_001_28 = 14,
1060 	IB_RNR_TIMER_001_92 = 15,
1061 	IB_RNR_TIMER_002_56 = 16,
1062 	IB_RNR_TIMER_003_84 = 17,
1063 	IB_RNR_TIMER_005_12 = 18,
1064 	IB_RNR_TIMER_007_68 = 19,
1065 	IB_RNR_TIMER_010_24 = 20,
1066 	IB_RNR_TIMER_015_36 = 21,
1067 	IB_RNR_TIMER_020_48 = 22,
1068 	IB_RNR_TIMER_030_72 = 23,
1069 	IB_RNR_TIMER_040_96 = 24,
1070 	IB_RNR_TIMER_061_44 = 25,
1071 	IB_RNR_TIMER_081_92 = 26,
1072 	IB_RNR_TIMER_122_88 = 27,
1073 	IB_RNR_TIMER_163_84 = 28,
1074 	IB_RNR_TIMER_245_76 = 29,
1075 	IB_RNR_TIMER_327_68 = 30,
1076 	IB_RNR_TIMER_491_52 = 31
1077 };
1078 
1079 enum ib_qp_attr_mask {
1080 	IB_QP_STATE			= 1,
1081 	IB_QP_CUR_STATE			= (1<<1),
1082 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1083 	IB_QP_ACCESS_FLAGS		= (1<<3),
1084 	IB_QP_PKEY_INDEX		= (1<<4),
1085 	IB_QP_PORT			= (1<<5),
1086 	IB_QP_QKEY			= (1<<6),
1087 	IB_QP_AV			= (1<<7),
1088 	IB_QP_PATH_MTU			= (1<<8),
1089 	IB_QP_TIMEOUT			= (1<<9),
1090 	IB_QP_RETRY_CNT			= (1<<10),
1091 	IB_QP_RNR_RETRY			= (1<<11),
1092 	IB_QP_RQ_PSN			= (1<<12),
1093 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1094 	IB_QP_ALT_PATH			= (1<<14),
1095 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1096 	IB_QP_SQ_PSN			= (1<<16),
1097 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1098 	IB_QP_PATH_MIG_STATE		= (1<<18),
1099 	IB_QP_CAP			= (1<<19),
1100 	IB_QP_DEST_QPN			= (1<<20),
1101 	IB_QP_RESERVED1			= (1<<21),
1102 	IB_QP_RESERVED2			= (1<<22),
1103 	IB_QP_RESERVED3			= (1<<23),
1104 	IB_QP_RESERVED4			= (1<<24),
1105 };
1106 
1107 enum ib_qp_state {
1108 	IB_QPS_RESET,
1109 	IB_QPS_INIT,
1110 	IB_QPS_RTR,
1111 	IB_QPS_RTS,
1112 	IB_QPS_SQD,
1113 	IB_QPS_SQE,
1114 	IB_QPS_ERR
1115 };
1116 
1117 enum ib_mig_state {
1118 	IB_MIG_MIGRATED,
1119 	IB_MIG_REARM,
1120 	IB_MIG_ARMED
1121 };
1122 
1123 enum ib_mw_type {
1124 	IB_MW_TYPE_1 = 1,
1125 	IB_MW_TYPE_2 = 2
1126 };
1127 
1128 struct ib_qp_attr {
1129 	enum ib_qp_state	qp_state;
1130 	enum ib_qp_state	cur_qp_state;
1131 	enum ib_mtu		path_mtu;
1132 	enum ib_mig_state	path_mig_state;
1133 	u32			qkey;
1134 	u32			rq_psn;
1135 	u32			sq_psn;
1136 	u32			dest_qp_num;
1137 	int			qp_access_flags;
1138 	struct ib_qp_cap	cap;
1139 	struct ib_ah_attr	ah_attr;
1140 	struct ib_ah_attr	alt_ah_attr;
1141 	u16			pkey_index;
1142 	u16			alt_pkey_index;
1143 	u8			en_sqd_async_notify;
1144 	u8			sq_draining;
1145 	u8			max_rd_atomic;
1146 	u8			max_dest_rd_atomic;
1147 	u8			min_rnr_timer;
1148 	u8			port_num;
1149 	u8			timeout;
1150 	u8			retry_cnt;
1151 	u8			rnr_retry;
1152 	u8			alt_port_num;
1153 	u8			alt_timeout;
1154 };
1155 
1156 enum ib_wr_opcode {
1157 	IB_WR_RDMA_WRITE,
1158 	IB_WR_RDMA_WRITE_WITH_IMM,
1159 	IB_WR_SEND,
1160 	IB_WR_SEND_WITH_IMM,
1161 	IB_WR_RDMA_READ,
1162 	IB_WR_ATOMIC_CMP_AND_SWP,
1163 	IB_WR_ATOMIC_FETCH_AND_ADD,
1164 	IB_WR_LSO,
1165 	IB_WR_SEND_WITH_INV,
1166 	IB_WR_RDMA_READ_WITH_INV,
1167 	IB_WR_LOCAL_INV,
1168 	IB_WR_REG_MR,
1169 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1170 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1171 	IB_WR_REG_SIG_MR,
1172 	/* reserve values for low level drivers' internal use.
1173 	 * These values will not be used at all in the ib core layer.
1174 	 */
1175 	IB_WR_RESERVED1 = 0xf0,
1176 	IB_WR_RESERVED2,
1177 	IB_WR_RESERVED3,
1178 	IB_WR_RESERVED4,
1179 	IB_WR_RESERVED5,
1180 	IB_WR_RESERVED6,
1181 	IB_WR_RESERVED7,
1182 	IB_WR_RESERVED8,
1183 	IB_WR_RESERVED9,
1184 	IB_WR_RESERVED10,
1185 };
1186 
1187 enum ib_send_flags {
1188 	IB_SEND_FENCE		= 1,
1189 	IB_SEND_SIGNALED	= (1<<1),
1190 	IB_SEND_SOLICITED	= (1<<2),
1191 	IB_SEND_INLINE		= (1<<3),
1192 	IB_SEND_IP_CSUM		= (1<<4),
1193 
1194 	/* reserve bits 26-31 for low level drivers' internal use */
1195 	IB_SEND_RESERVED_START	= (1 << 26),
1196 	IB_SEND_RESERVED_END	= (1 << 31),
1197 };
1198 
1199 struct ib_sge {
1200 	u64	addr;
1201 	u32	length;
1202 	u32	lkey;
1203 };
1204 
1205 struct ib_cqe {
1206 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1207 };
1208 
1209 struct ib_send_wr {
1210 	struct ib_send_wr      *next;
1211 	union {
1212 		u64		wr_id;
1213 		struct ib_cqe	*wr_cqe;
1214 	};
1215 	struct ib_sge	       *sg_list;
1216 	int			num_sge;
1217 	enum ib_wr_opcode	opcode;
1218 	int			send_flags;
1219 	union {
1220 		__be32		imm_data;
1221 		u32		invalidate_rkey;
1222 	} ex;
1223 };
1224 
1225 struct ib_rdma_wr {
1226 	struct ib_send_wr	wr;
1227 	u64			remote_addr;
1228 	u32			rkey;
1229 };
1230 
1231 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1232 {
1233 	return container_of(wr, struct ib_rdma_wr, wr);
1234 }
1235 
1236 struct ib_atomic_wr {
1237 	struct ib_send_wr	wr;
1238 	u64			remote_addr;
1239 	u64			compare_add;
1240 	u64			swap;
1241 	u64			compare_add_mask;
1242 	u64			swap_mask;
1243 	u32			rkey;
1244 };
1245 
1246 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1247 {
1248 	return container_of(wr, struct ib_atomic_wr, wr);
1249 }
1250 
1251 struct ib_ud_wr {
1252 	struct ib_send_wr	wr;
1253 	struct ib_ah		*ah;
1254 	void			*header;
1255 	int			hlen;
1256 	int			mss;
1257 	u32			remote_qpn;
1258 	u32			remote_qkey;
1259 	u16			pkey_index; /* valid for GSI only */
1260 	u8			port_num;   /* valid for DR SMPs on switch only */
1261 };
1262 
1263 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1264 {
1265 	return container_of(wr, struct ib_ud_wr, wr);
1266 }
1267 
1268 struct ib_reg_wr {
1269 	struct ib_send_wr	wr;
1270 	struct ib_mr		*mr;
1271 	u32			key;
1272 	int			access;
1273 };
1274 
1275 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1276 {
1277 	return container_of(wr, struct ib_reg_wr, wr);
1278 }
1279 
1280 struct ib_sig_handover_wr {
1281 	struct ib_send_wr	wr;
1282 	struct ib_sig_attrs    *sig_attrs;
1283 	struct ib_mr	       *sig_mr;
1284 	int			access_flags;
1285 	struct ib_sge	       *prot;
1286 };
1287 
1288 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1289 {
1290 	return container_of(wr, struct ib_sig_handover_wr, wr);
1291 }
1292 
1293 struct ib_recv_wr {
1294 	struct ib_recv_wr      *next;
1295 	union {
1296 		u64		wr_id;
1297 		struct ib_cqe	*wr_cqe;
1298 	};
1299 	struct ib_sge	       *sg_list;
1300 	int			num_sge;
1301 };
1302 
1303 enum ib_access_flags {
1304 	IB_ACCESS_LOCAL_WRITE	= 1,
1305 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1306 	IB_ACCESS_REMOTE_READ	= (1<<2),
1307 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1308 	IB_ACCESS_MW_BIND	= (1<<4),
1309 	IB_ZERO_BASED		= (1<<5),
1310 	IB_ACCESS_ON_DEMAND     = (1<<6),
1311 };
1312 
1313 /*
1314  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1315  * are hidden here instead of a uapi header!
1316  */
1317 enum ib_mr_rereg_flags {
1318 	IB_MR_REREG_TRANS	= 1,
1319 	IB_MR_REREG_PD		= (1<<1),
1320 	IB_MR_REREG_ACCESS	= (1<<2),
1321 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1322 };
1323 
1324 struct ib_fmr_attr {
1325 	int	max_pages;
1326 	int	max_maps;
1327 	u8	page_shift;
1328 };
1329 
1330 struct ib_umem;
1331 
1332 struct ib_ucontext {
1333 	struct ib_device       *device;
1334 	struct list_head	pd_list;
1335 	struct list_head	mr_list;
1336 	struct list_head	mw_list;
1337 	struct list_head	cq_list;
1338 	struct list_head	qp_list;
1339 	struct list_head	srq_list;
1340 	struct list_head	ah_list;
1341 	struct list_head	xrcd_list;
1342 	struct list_head	rule_list;
1343 	struct list_head	wq_list;
1344 	struct list_head	rwq_ind_tbl_list;
1345 	int			closing;
1346 
1347 	struct pid             *tgid;
1348 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1349 	struct rb_root      umem_tree;
1350 	/*
1351 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1352 	 * mmu notifiers registration.
1353 	 */
1354 	struct rw_semaphore	umem_rwsem;
1355 	void (*invalidate_range)(struct ib_umem *umem,
1356 				 unsigned long start, unsigned long end);
1357 
1358 	struct mmu_notifier	mn;
1359 	atomic_t		notifier_count;
1360 	/* A list of umems that don't have private mmu notifier counters yet. */
1361 	struct list_head	no_private_counters;
1362 	int                     odp_mrs_count;
1363 #endif
1364 };
1365 
1366 struct ib_uobject {
1367 	u64			user_handle;	/* handle given to us by userspace */
1368 	struct ib_ucontext     *context;	/* associated user context */
1369 	void		       *object;		/* containing object */
1370 	struct list_head	list;		/* link to context's list */
1371 	int			id;		/* index into kernel idr */
1372 	struct kref		ref;
1373 	struct rw_semaphore	mutex;		/* protects .live */
1374 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1375 	int			live;
1376 };
1377 
1378 struct ib_udata {
1379 	const void __user *inbuf;
1380 	void __user *outbuf;
1381 	size_t       inlen;
1382 	size_t       outlen;
1383 };
1384 
1385 struct ib_pd {
1386 	u32			local_dma_lkey;
1387 	u32			flags;
1388 	struct ib_device       *device;
1389 	struct ib_uobject      *uobject;
1390 	atomic_t          	usecnt; /* count all resources */
1391 
1392 	u32			unsafe_global_rkey;
1393 
1394 	/*
1395 	 * Implementation details of the RDMA core, don't use in drivers:
1396 	 */
1397 	struct ib_mr	       *__internal_mr;
1398 };
1399 
1400 struct ib_xrcd {
1401 	struct ib_device       *device;
1402 	atomic_t		usecnt; /* count all exposed resources */
1403 	struct inode	       *inode;
1404 
1405 	struct mutex		tgt_qp_mutex;
1406 	struct list_head	tgt_qp_list;
1407 };
1408 
1409 struct ib_ah {
1410 	struct ib_device	*device;
1411 	struct ib_pd		*pd;
1412 	struct ib_uobject	*uobject;
1413 };
1414 
1415 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1416 
1417 enum ib_poll_context {
1418 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1419 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1420 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1421 };
1422 
1423 struct ib_cq {
1424 	struct ib_device       *device;
1425 	struct ib_uobject      *uobject;
1426 	ib_comp_handler   	comp_handler;
1427 	void                  (*event_handler)(struct ib_event *, void *);
1428 	void                   *cq_context;
1429 	int               	cqe;
1430 	atomic_t          	usecnt; /* count number of work queues */
1431 	enum ib_poll_context	poll_ctx;
1432 	struct ib_wc		*wc;
1433 	union {
1434 		struct irq_poll		iop;
1435 		struct work_struct	work;
1436 	};
1437 };
1438 
1439 struct ib_srq {
1440 	struct ib_device       *device;
1441 	struct ib_pd	       *pd;
1442 	struct ib_uobject      *uobject;
1443 	void		      (*event_handler)(struct ib_event *, void *);
1444 	void		       *srq_context;
1445 	enum ib_srq_type	srq_type;
1446 	atomic_t		usecnt;
1447 
1448 	union {
1449 		struct {
1450 			struct ib_xrcd *xrcd;
1451 			struct ib_cq   *cq;
1452 			u32		srq_num;
1453 		} xrc;
1454 	} ext;
1455 };
1456 
1457 enum ib_wq_type {
1458 	IB_WQT_RQ
1459 };
1460 
1461 enum ib_wq_state {
1462 	IB_WQS_RESET,
1463 	IB_WQS_RDY,
1464 	IB_WQS_ERR
1465 };
1466 
1467 struct ib_wq {
1468 	struct ib_device       *device;
1469 	struct ib_uobject      *uobject;
1470 	void		    *wq_context;
1471 	void		    (*event_handler)(struct ib_event *, void *);
1472 	struct ib_pd	       *pd;
1473 	struct ib_cq	       *cq;
1474 	u32		wq_num;
1475 	enum ib_wq_state       state;
1476 	enum ib_wq_type	wq_type;
1477 	atomic_t		usecnt;
1478 };
1479 
1480 struct ib_wq_init_attr {
1481 	void		       *wq_context;
1482 	enum ib_wq_type	wq_type;
1483 	u32		max_wr;
1484 	u32		max_sge;
1485 	struct	ib_cq	       *cq;
1486 	void		    (*event_handler)(struct ib_event *, void *);
1487 };
1488 
1489 enum ib_wq_attr_mask {
1490 	IB_WQ_STATE	= 1 << 0,
1491 	IB_WQ_CUR_STATE	= 1 << 1,
1492 };
1493 
1494 struct ib_wq_attr {
1495 	enum	ib_wq_state	wq_state;
1496 	enum	ib_wq_state	curr_wq_state;
1497 };
1498 
1499 struct ib_rwq_ind_table {
1500 	struct ib_device	*device;
1501 	struct ib_uobject      *uobject;
1502 	atomic_t		usecnt;
1503 	u32		ind_tbl_num;
1504 	u32		log_ind_tbl_size;
1505 	struct ib_wq	**ind_tbl;
1506 };
1507 
1508 struct ib_rwq_ind_table_init_attr {
1509 	u32		log_ind_tbl_size;
1510 	/* Each entry is a pointer to Receive Work Queue */
1511 	struct ib_wq	**ind_tbl;
1512 };
1513 
1514 /*
1515  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1516  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1517  */
1518 struct ib_qp {
1519 	struct ib_device       *device;
1520 	struct ib_pd	       *pd;
1521 	struct ib_cq	       *send_cq;
1522 	struct ib_cq	       *recv_cq;
1523 	spinlock_t		mr_lock;
1524 	int			mrs_used;
1525 	struct list_head	rdma_mrs;
1526 	struct list_head	sig_mrs;
1527 	struct ib_srq	       *srq;
1528 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1529 	struct list_head	xrcd_list;
1530 
1531 	/* count times opened, mcast attaches, flow attaches */
1532 	atomic_t		usecnt;
1533 	struct list_head	open_list;
1534 	struct ib_qp           *real_qp;
1535 	struct ib_uobject      *uobject;
1536 	void                  (*event_handler)(struct ib_event *, void *);
1537 	void		       *qp_context;
1538 	u32			qp_num;
1539 	u32			max_write_sge;
1540 	u32			max_read_sge;
1541 	enum ib_qp_type		qp_type;
1542 	struct ib_rwq_ind_table *rwq_ind_tbl;
1543 };
1544 
1545 struct ib_mr {
1546 	struct ib_device  *device;
1547 	struct ib_pd	  *pd;
1548 	u32		   lkey;
1549 	u32		   rkey;
1550 	u64		   iova;
1551 	u32		   length;
1552 	unsigned int	   page_size;
1553 	bool		   need_inval;
1554 	union {
1555 		struct ib_uobject	*uobject;	/* user */
1556 		struct list_head	qp_entry;	/* FR */
1557 	};
1558 };
1559 
1560 struct ib_mw {
1561 	struct ib_device	*device;
1562 	struct ib_pd		*pd;
1563 	struct ib_uobject	*uobject;
1564 	u32			rkey;
1565 	enum ib_mw_type         type;
1566 };
1567 
1568 struct ib_fmr {
1569 	struct ib_device	*device;
1570 	struct ib_pd		*pd;
1571 	struct list_head	list;
1572 	u32			lkey;
1573 	u32			rkey;
1574 };
1575 
1576 /* Supported steering options */
1577 enum ib_flow_attr_type {
1578 	/* steering according to rule specifications */
1579 	IB_FLOW_ATTR_NORMAL		= 0x0,
1580 	/* default unicast and multicast rule -
1581 	 * receive all Eth traffic which isn't steered to any QP
1582 	 */
1583 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1584 	/* default multicast rule -
1585 	 * receive all Eth multicast traffic which isn't steered to any QP
1586 	 */
1587 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1588 	/* sniffer rule - receive all port traffic */
1589 	IB_FLOW_ATTR_SNIFFER		= 0x3
1590 };
1591 
1592 /* Supported steering header types */
1593 enum ib_flow_spec_type {
1594 	/* L2 headers*/
1595 	IB_FLOW_SPEC_ETH	= 0x20,
1596 	IB_FLOW_SPEC_IB		= 0x22,
1597 	/* L3 header*/
1598 	IB_FLOW_SPEC_IPV4	= 0x30,
1599 	IB_FLOW_SPEC_IPV6	= 0x31,
1600 	/* L4 headers*/
1601 	IB_FLOW_SPEC_TCP	= 0x40,
1602 	IB_FLOW_SPEC_UDP	= 0x41
1603 };
1604 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1605 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1606 
1607 /* Flow steering rule priority is set according to it's domain.
1608  * Lower domain value means higher priority.
1609  */
1610 enum ib_flow_domain {
1611 	IB_FLOW_DOMAIN_USER,
1612 	IB_FLOW_DOMAIN_ETHTOOL,
1613 	IB_FLOW_DOMAIN_RFS,
1614 	IB_FLOW_DOMAIN_NIC,
1615 	IB_FLOW_DOMAIN_NUM /* Must be last */
1616 };
1617 
1618 enum ib_flow_flags {
1619 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1620 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1621 };
1622 
1623 struct ib_flow_eth_filter {
1624 	u8	dst_mac[6];
1625 	u8	src_mac[6];
1626 	__be16	ether_type;
1627 	__be16	vlan_tag;
1628 	/* Must be last */
1629 	u8	real_sz[0];
1630 };
1631 
1632 struct ib_flow_spec_eth {
1633 	enum ib_flow_spec_type	  type;
1634 	u16			  size;
1635 	struct ib_flow_eth_filter val;
1636 	struct ib_flow_eth_filter mask;
1637 };
1638 
1639 struct ib_flow_ib_filter {
1640 	__be16 dlid;
1641 	__u8   sl;
1642 	/* Must be last */
1643 	u8	real_sz[0];
1644 };
1645 
1646 struct ib_flow_spec_ib {
1647 	enum ib_flow_spec_type	 type;
1648 	u16			 size;
1649 	struct ib_flow_ib_filter val;
1650 	struct ib_flow_ib_filter mask;
1651 };
1652 
1653 /* IPv4 header flags */
1654 enum ib_ipv4_flags {
1655 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1656 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1657 				    last have this flag set */
1658 };
1659 
1660 struct ib_flow_ipv4_filter {
1661 	__be32	src_ip;
1662 	__be32	dst_ip;
1663 	u8	proto;
1664 	u8	tos;
1665 	u8	ttl;
1666 	u8	flags;
1667 	/* Must be last */
1668 	u8	real_sz[0];
1669 };
1670 
1671 struct ib_flow_spec_ipv4 {
1672 	enum ib_flow_spec_type	   type;
1673 	u16			   size;
1674 	struct ib_flow_ipv4_filter val;
1675 	struct ib_flow_ipv4_filter mask;
1676 };
1677 
1678 struct ib_flow_ipv6_filter {
1679 	u8	src_ip[16];
1680 	u8	dst_ip[16];
1681 	__be32	flow_label;
1682 	u8	next_hdr;
1683 	u8	traffic_class;
1684 	u8	hop_limit;
1685 	/* Must be last */
1686 	u8	real_sz[0];
1687 };
1688 
1689 struct ib_flow_spec_ipv6 {
1690 	enum ib_flow_spec_type	   type;
1691 	u16			   size;
1692 	struct ib_flow_ipv6_filter val;
1693 	struct ib_flow_ipv6_filter mask;
1694 };
1695 
1696 struct ib_flow_tcp_udp_filter {
1697 	__be16	dst_port;
1698 	__be16	src_port;
1699 	/* Must be last */
1700 	u8	real_sz[0];
1701 };
1702 
1703 struct ib_flow_spec_tcp_udp {
1704 	enum ib_flow_spec_type	      type;
1705 	u16			      size;
1706 	struct ib_flow_tcp_udp_filter val;
1707 	struct ib_flow_tcp_udp_filter mask;
1708 };
1709 
1710 union ib_flow_spec {
1711 	struct {
1712 		enum ib_flow_spec_type	type;
1713 		u16			size;
1714 	};
1715 	struct ib_flow_spec_eth		eth;
1716 	struct ib_flow_spec_ib		ib;
1717 	struct ib_flow_spec_ipv4        ipv4;
1718 	struct ib_flow_spec_tcp_udp	tcp_udp;
1719 	struct ib_flow_spec_ipv6        ipv6;
1720 };
1721 
1722 struct ib_flow_attr {
1723 	enum ib_flow_attr_type type;
1724 	u16	     size;
1725 	u16	     priority;
1726 	u32	     flags;
1727 	u8	     num_of_specs;
1728 	u8	     port;
1729 	/* Following are the optional layers according to user request
1730 	 * struct ib_flow_spec_xxx
1731 	 * struct ib_flow_spec_yyy
1732 	 */
1733 };
1734 
1735 struct ib_flow {
1736 	struct ib_qp		*qp;
1737 	struct ib_uobject	*uobject;
1738 };
1739 
1740 struct ib_mad_hdr;
1741 struct ib_grh;
1742 
1743 enum ib_process_mad_flags {
1744 	IB_MAD_IGNORE_MKEY	= 1,
1745 	IB_MAD_IGNORE_BKEY	= 2,
1746 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1747 };
1748 
1749 enum ib_mad_result {
1750 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1751 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1752 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1753 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1754 };
1755 
1756 #define IB_DEVICE_NAME_MAX 64
1757 
1758 struct ib_cache {
1759 	rwlock_t                lock;
1760 	struct ib_event_handler event_handler;
1761 	struct ib_pkey_cache  **pkey_cache;
1762 	struct ib_gid_table   **gid_cache;
1763 	u8                     *lmc_cache;
1764 };
1765 
1766 struct ib_dma_mapping_ops {
1767 	int		(*mapping_error)(struct ib_device *dev,
1768 					 u64 dma_addr);
1769 	u64		(*map_single)(struct ib_device *dev,
1770 				      void *ptr, size_t size,
1771 				      enum dma_data_direction direction);
1772 	void		(*unmap_single)(struct ib_device *dev,
1773 					u64 addr, size_t size,
1774 					enum dma_data_direction direction);
1775 	u64		(*map_page)(struct ib_device *dev,
1776 				    struct page *page, unsigned long offset,
1777 				    size_t size,
1778 				    enum dma_data_direction direction);
1779 	void		(*unmap_page)(struct ib_device *dev,
1780 				      u64 addr, size_t size,
1781 				      enum dma_data_direction direction);
1782 	int		(*map_sg)(struct ib_device *dev,
1783 				  struct scatterlist *sg, int nents,
1784 				  enum dma_data_direction direction);
1785 	void		(*unmap_sg)(struct ib_device *dev,
1786 				    struct scatterlist *sg, int nents,
1787 				    enum dma_data_direction direction);
1788 	int		(*map_sg_attrs)(struct ib_device *dev,
1789 					struct scatterlist *sg, int nents,
1790 					enum dma_data_direction direction,
1791 					unsigned long attrs);
1792 	void		(*unmap_sg_attrs)(struct ib_device *dev,
1793 					  struct scatterlist *sg, int nents,
1794 					  enum dma_data_direction direction,
1795 					  unsigned long attrs);
1796 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1797 					       u64 dma_handle,
1798 					       size_t size,
1799 					       enum dma_data_direction dir);
1800 	void		(*sync_single_for_device)(struct ib_device *dev,
1801 						  u64 dma_handle,
1802 						  size_t size,
1803 						  enum dma_data_direction dir);
1804 	void		*(*alloc_coherent)(struct ib_device *dev,
1805 					   size_t size,
1806 					   u64 *dma_handle,
1807 					   gfp_t flag);
1808 	void		(*free_coherent)(struct ib_device *dev,
1809 					 size_t size, void *cpu_addr,
1810 					 u64 dma_handle);
1811 };
1812 
1813 struct iw_cm_verbs;
1814 
1815 struct ib_port_immutable {
1816 	int                           pkey_tbl_len;
1817 	int                           gid_tbl_len;
1818 	u32                           core_cap_flags;
1819 	u32                           max_mad_size;
1820 };
1821 
1822 struct ib_device {
1823 	struct device                *dma_device;
1824 
1825 	char                          name[IB_DEVICE_NAME_MAX];
1826 
1827 	struct list_head              event_handler_list;
1828 	spinlock_t                    event_handler_lock;
1829 
1830 	spinlock_t                    client_data_lock;
1831 	struct list_head              core_list;
1832 	/* Access to the client_data_list is protected by the client_data_lock
1833 	 * spinlock and the lists_rwsem read-write semaphore */
1834 	struct list_head              client_data_list;
1835 
1836 	struct ib_cache               cache;
1837 	/**
1838 	 * port_immutable is indexed by port number
1839 	 */
1840 	struct ib_port_immutable     *port_immutable;
1841 
1842 	int			      num_comp_vectors;
1843 
1844 	struct iw_cm_verbs	     *iwcm;
1845 
1846 	/**
1847 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1848 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
1849 	 *   core when the device is removed.  A lifespan of -1 in the return
1850 	 *   struct tells the core to set a default lifespan.
1851 	 */
1852 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
1853 						     u8 port_num);
1854 	/**
1855 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1856 	 * @index - The index in the value array we wish to have updated, or
1857 	 *   num_counters if we want all stats updated
1858 	 * Return codes -
1859 	 *   < 0 - Error, no counters updated
1860 	 *   index - Updated the single counter pointed to by index
1861 	 *   num_counters - Updated all counters (will reset the timestamp
1862 	 *     and prevent further calls for lifespan milliseconds)
1863 	 * Drivers are allowed to update all counters in leiu of just the
1864 	 *   one given in index at their option
1865 	 */
1866 	int		           (*get_hw_stats)(struct ib_device *device,
1867 						   struct rdma_hw_stats *stats,
1868 						   u8 port, int index);
1869 	int		           (*query_device)(struct ib_device *device,
1870 						   struct ib_device_attr *device_attr,
1871 						   struct ib_udata *udata);
1872 	int		           (*query_port)(struct ib_device *device,
1873 						 u8 port_num,
1874 						 struct ib_port_attr *port_attr);
1875 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1876 						     u8 port_num);
1877 	/* When calling get_netdev, the HW vendor's driver should return the
1878 	 * net device of device @device at port @port_num or NULL if such
1879 	 * a net device doesn't exist. The vendor driver should call dev_hold
1880 	 * on this net device. The HW vendor's device driver must guarantee
1881 	 * that this function returns NULL before the net device reaches
1882 	 * NETDEV_UNREGISTER_FINAL state.
1883 	 */
1884 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1885 						 u8 port_num);
1886 	int		           (*query_gid)(struct ib_device *device,
1887 						u8 port_num, int index,
1888 						union ib_gid *gid);
1889 	/* When calling add_gid, the HW vendor's driver should
1890 	 * add the gid of device @device at gid index @index of
1891 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1892 	 * the network device related to this gid is available
1893 	 * at @attr. @context allows the HW vendor driver to store extra
1894 	 * information together with a GID entry. The HW vendor may allocate
1895 	 * memory to contain this information and store it in @context when a
1896 	 * new GID entry is written to. Params are consistent until the next
1897 	 * call of add_gid or delete_gid. The function should return 0 on
1898 	 * success or error otherwise. The function could be called
1899 	 * concurrently for different ports. This function is only called
1900 	 * when roce_gid_table is used.
1901 	 */
1902 	int		           (*add_gid)(struct ib_device *device,
1903 					      u8 port_num,
1904 					      unsigned int index,
1905 					      const union ib_gid *gid,
1906 					      const struct ib_gid_attr *attr,
1907 					      void **context);
1908 	/* When calling del_gid, the HW vendor's driver should delete the
1909 	 * gid of device @device at gid index @index of port @port_num.
1910 	 * Upon the deletion of a GID entry, the HW vendor must free any
1911 	 * allocated memory. The caller will clear @context afterwards.
1912 	 * This function is only called when roce_gid_table is used.
1913 	 */
1914 	int		           (*del_gid)(struct ib_device *device,
1915 					      u8 port_num,
1916 					      unsigned int index,
1917 					      void **context);
1918 	int		           (*query_pkey)(struct ib_device *device,
1919 						 u8 port_num, u16 index, u16 *pkey);
1920 	int		           (*modify_device)(struct ib_device *device,
1921 						    int device_modify_mask,
1922 						    struct ib_device_modify *device_modify);
1923 	int		           (*modify_port)(struct ib_device *device,
1924 						  u8 port_num, int port_modify_mask,
1925 						  struct ib_port_modify *port_modify);
1926 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1927 						     struct ib_udata *udata);
1928 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1929 	int                        (*mmap)(struct ib_ucontext *context,
1930 					   struct vm_area_struct *vma);
1931 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1932 					       struct ib_ucontext *context,
1933 					       struct ib_udata *udata);
1934 	int                        (*dealloc_pd)(struct ib_pd *pd);
1935 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1936 						struct ib_ah_attr *ah_attr);
1937 	int                        (*modify_ah)(struct ib_ah *ah,
1938 						struct ib_ah_attr *ah_attr);
1939 	int                        (*query_ah)(struct ib_ah *ah,
1940 					       struct ib_ah_attr *ah_attr);
1941 	int                        (*destroy_ah)(struct ib_ah *ah);
1942 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1943 						 struct ib_srq_init_attr *srq_init_attr,
1944 						 struct ib_udata *udata);
1945 	int                        (*modify_srq)(struct ib_srq *srq,
1946 						 struct ib_srq_attr *srq_attr,
1947 						 enum ib_srq_attr_mask srq_attr_mask,
1948 						 struct ib_udata *udata);
1949 	int                        (*query_srq)(struct ib_srq *srq,
1950 						struct ib_srq_attr *srq_attr);
1951 	int                        (*destroy_srq)(struct ib_srq *srq);
1952 	int                        (*post_srq_recv)(struct ib_srq *srq,
1953 						    struct ib_recv_wr *recv_wr,
1954 						    struct ib_recv_wr **bad_recv_wr);
1955 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1956 						struct ib_qp_init_attr *qp_init_attr,
1957 						struct ib_udata *udata);
1958 	int                        (*modify_qp)(struct ib_qp *qp,
1959 						struct ib_qp_attr *qp_attr,
1960 						int qp_attr_mask,
1961 						struct ib_udata *udata);
1962 	int                        (*query_qp)(struct ib_qp *qp,
1963 					       struct ib_qp_attr *qp_attr,
1964 					       int qp_attr_mask,
1965 					       struct ib_qp_init_attr *qp_init_attr);
1966 	int                        (*destroy_qp)(struct ib_qp *qp);
1967 	int                        (*post_send)(struct ib_qp *qp,
1968 						struct ib_send_wr *send_wr,
1969 						struct ib_send_wr **bad_send_wr);
1970 	int                        (*post_recv)(struct ib_qp *qp,
1971 						struct ib_recv_wr *recv_wr,
1972 						struct ib_recv_wr **bad_recv_wr);
1973 	struct ib_cq *             (*create_cq)(struct ib_device *device,
1974 						const struct ib_cq_init_attr *attr,
1975 						struct ib_ucontext *context,
1976 						struct ib_udata *udata);
1977 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1978 						u16 cq_period);
1979 	int                        (*destroy_cq)(struct ib_cq *cq);
1980 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1981 						struct ib_udata *udata);
1982 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1983 					      struct ib_wc *wc);
1984 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1985 	int                        (*req_notify_cq)(struct ib_cq *cq,
1986 						    enum ib_cq_notify_flags flags);
1987 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1988 						      int wc_cnt);
1989 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1990 						 int mr_access_flags);
1991 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1992 						  u64 start, u64 length,
1993 						  u64 virt_addr,
1994 						  int mr_access_flags,
1995 						  struct ib_udata *udata);
1996 	int			   (*rereg_user_mr)(struct ib_mr *mr,
1997 						    int flags,
1998 						    u64 start, u64 length,
1999 						    u64 virt_addr,
2000 						    int mr_access_flags,
2001 						    struct ib_pd *pd,
2002 						    struct ib_udata *udata);
2003 	int                        (*dereg_mr)(struct ib_mr *mr);
2004 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2005 					       enum ib_mr_type mr_type,
2006 					       u32 max_num_sg);
2007 	int                        (*map_mr_sg)(struct ib_mr *mr,
2008 						struct scatterlist *sg,
2009 						int sg_nents,
2010 						unsigned int *sg_offset);
2011 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2012 					       enum ib_mw_type type,
2013 					       struct ib_udata *udata);
2014 	int                        (*dealloc_mw)(struct ib_mw *mw);
2015 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2016 						int mr_access_flags,
2017 						struct ib_fmr_attr *fmr_attr);
2018 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2019 						   u64 *page_list, int list_len,
2020 						   u64 iova);
2021 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2022 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2023 	int                        (*attach_mcast)(struct ib_qp *qp,
2024 						   union ib_gid *gid,
2025 						   u16 lid);
2026 	int                        (*detach_mcast)(struct ib_qp *qp,
2027 						   union ib_gid *gid,
2028 						   u16 lid);
2029 	int                        (*process_mad)(struct ib_device *device,
2030 						  int process_mad_flags,
2031 						  u8 port_num,
2032 						  const struct ib_wc *in_wc,
2033 						  const struct ib_grh *in_grh,
2034 						  const struct ib_mad_hdr *in_mad,
2035 						  size_t in_mad_size,
2036 						  struct ib_mad_hdr *out_mad,
2037 						  size_t *out_mad_size,
2038 						  u16 *out_mad_pkey_index);
2039 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2040 						 struct ib_ucontext *ucontext,
2041 						 struct ib_udata *udata);
2042 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2043 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2044 						  struct ib_flow_attr
2045 						  *flow_attr,
2046 						  int domain);
2047 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2048 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2049 						      struct ib_mr_status *mr_status);
2050 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2051 	void			   (*drain_rq)(struct ib_qp *qp);
2052 	void			   (*drain_sq)(struct ib_qp *qp);
2053 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2054 							int state);
2055 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2056 						   struct ifla_vf_info *ivf);
2057 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2058 						   struct ifla_vf_stats *stats);
2059 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2060 						  int type);
2061 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2062 						struct ib_wq_init_attr *init_attr,
2063 						struct ib_udata *udata);
2064 	int			   (*destroy_wq)(struct ib_wq *wq);
2065 	int			   (*modify_wq)(struct ib_wq *wq,
2066 						struct ib_wq_attr *attr,
2067 						u32 wq_attr_mask,
2068 						struct ib_udata *udata);
2069 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2070 							   struct ib_rwq_ind_table_init_attr *init_attr,
2071 							   struct ib_udata *udata);
2072 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2073 	struct ib_dma_mapping_ops   *dma_ops;
2074 
2075 	struct module               *owner;
2076 	struct device                dev;
2077 	struct kobject               *ports_parent;
2078 	struct list_head             port_list;
2079 
2080 	enum {
2081 		IB_DEV_UNINITIALIZED,
2082 		IB_DEV_REGISTERED,
2083 		IB_DEV_UNREGISTERED
2084 	}                            reg_state;
2085 
2086 	int			     uverbs_abi_ver;
2087 	u64			     uverbs_cmd_mask;
2088 	u64			     uverbs_ex_cmd_mask;
2089 
2090 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2091 	__be64			     node_guid;
2092 	u32			     local_dma_lkey;
2093 	u16                          is_switch:1;
2094 	u8                           node_type;
2095 	u8                           phys_port_cnt;
2096 	struct ib_device_attr        attrs;
2097 	struct attribute_group	     *hw_stats_ag;
2098 	struct rdma_hw_stats         *hw_stats;
2099 
2100 	/**
2101 	 * The following mandatory functions are used only at device
2102 	 * registration.  Keep functions such as these at the end of this
2103 	 * structure to avoid cache line misses when accessing struct ib_device
2104 	 * in fast paths.
2105 	 */
2106 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2107 	void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2108 };
2109 
2110 struct ib_client {
2111 	char  *name;
2112 	void (*add)   (struct ib_device *);
2113 	void (*remove)(struct ib_device *, void *client_data);
2114 
2115 	/* Returns the net_dev belonging to this ib_client and matching the
2116 	 * given parameters.
2117 	 * @dev:	 An RDMA device that the net_dev use for communication.
2118 	 * @port:	 A physical port number on the RDMA device.
2119 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2120 	 * @gid:	 A GID that the net_dev uses to communicate.
2121 	 * @addr:	 An IP address the net_dev is configured with.
2122 	 * @client_data: The device's client data set by ib_set_client_data().
2123 	 *
2124 	 * An ib_client that implements a net_dev on top of RDMA devices
2125 	 * (such as IP over IB) should implement this callback, allowing the
2126 	 * rdma_cm module to find the right net_dev for a given request.
2127 	 *
2128 	 * The caller is responsible for calling dev_put on the returned
2129 	 * netdev. */
2130 	struct net_device *(*get_net_dev_by_params)(
2131 			struct ib_device *dev,
2132 			u8 port,
2133 			u16 pkey,
2134 			const union ib_gid *gid,
2135 			const struct sockaddr *addr,
2136 			void *client_data);
2137 	struct list_head list;
2138 };
2139 
2140 struct ib_device *ib_alloc_device(size_t size);
2141 void ib_dealloc_device(struct ib_device *device);
2142 
2143 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2144 
2145 int ib_register_device(struct ib_device *device,
2146 		       int (*port_callback)(struct ib_device *,
2147 					    u8, struct kobject *));
2148 void ib_unregister_device(struct ib_device *device);
2149 
2150 int ib_register_client   (struct ib_client *client);
2151 void ib_unregister_client(struct ib_client *client);
2152 
2153 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2154 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2155 			 void *data);
2156 
2157 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2158 {
2159 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2160 }
2161 
2162 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2163 {
2164 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2165 }
2166 
2167 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2168 				       size_t offset,
2169 				       size_t len)
2170 {
2171 	const void __user *p = udata->inbuf + offset;
2172 	bool ret;
2173 	u8 *buf;
2174 
2175 	if (len > USHRT_MAX)
2176 		return false;
2177 
2178 	buf = memdup_user(p, len);
2179 	if (IS_ERR(buf))
2180 		return false;
2181 
2182 	ret = !memchr_inv(buf, 0, len);
2183 	kfree(buf);
2184 	return ret;
2185 }
2186 
2187 /**
2188  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2189  * contains all required attributes and no attributes not allowed for
2190  * the given QP state transition.
2191  * @cur_state: Current QP state
2192  * @next_state: Next QP state
2193  * @type: QP type
2194  * @mask: Mask of supplied QP attributes
2195  * @ll : link layer of port
2196  *
2197  * This function is a helper function that a low-level driver's
2198  * modify_qp method can use to validate the consumer's input.  It
2199  * checks that cur_state and next_state are valid QP states, that a
2200  * transition from cur_state to next_state is allowed by the IB spec,
2201  * and that the attribute mask supplied is allowed for the transition.
2202  */
2203 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2204 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2205 		       enum rdma_link_layer ll);
2206 
2207 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2208 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2209 void ib_dispatch_event(struct ib_event *event);
2210 
2211 int ib_query_port(struct ib_device *device,
2212 		  u8 port_num, struct ib_port_attr *port_attr);
2213 
2214 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2215 					       u8 port_num);
2216 
2217 /**
2218  * rdma_cap_ib_switch - Check if the device is IB switch
2219  * @device: Device to check
2220  *
2221  * Device driver is responsible for setting is_switch bit on
2222  * in ib_device structure at init time.
2223  *
2224  * Return: true if the device is IB switch.
2225  */
2226 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2227 {
2228 	return device->is_switch;
2229 }
2230 
2231 /**
2232  * rdma_start_port - Return the first valid port number for the device
2233  * specified
2234  *
2235  * @device: Device to be checked
2236  *
2237  * Return start port number
2238  */
2239 static inline u8 rdma_start_port(const struct ib_device *device)
2240 {
2241 	return rdma_cap_ib_switch(device) ? 0 : 1;
2242 }
2243 
2244 /**
2245  * rdma_end_port - Return the last valid port number for the device
2246  * specified
2247  *
2248  * @device: Device to be checked
2249  *
2250  * Return last port number
2251  */
2252 static inline u8 rdma_end_port(const struct ib_device *device)
2253 {
2254 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2255 }
2256 
2257 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2258 {
2259 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2260 }
2261 
2262 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2263 {
2264 	return device->port_immutable[port_num].core_cap_flags &
2265 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2266 }
2267 
2268 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2269 {
2270 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2271 }
2272 
2273 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2274 {
2275 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2276 }
2277 
2278 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2279 {
2280 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2281 }
2282 
2283 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2284 {
2285 	return rdma_protocol_ib(device, port_num) ||
2286 		rdma_protocol_roce(device, port_num);
2287 }
2288 
2289 /**
2290  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2291  * Management Datagrams.
2292  * @device: Device to check
2293  * @port_num: Port number to check
2294  *
2295  * Management Datagrams (MAD) are a required part of the InfiniBand
2296  * specification and are supported on all InfiniBand devices.  A slightly
2297  * extended version are also supported on OPA interfaces.
2298  *
2299  * Return: true if the port supports sending/receiving of MAD packets.
2300  */
2301 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2302 {
2303 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2304 }
2305 
2306 /**
2307  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2308  * Management Datagrams.
2309  * @device: Device to check
2310  * @port_num: Port number to check
2311  *
2312  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2313  * datagrams with their own versions.  These OPA MADs share many but not all of
2314  * the characteristics of InfiniBand MADs.
2315  *
2316  * OPA MADs differ in the following ways:
2317  *
2318  *    1) MADs are variable size up to 2K
2319  *       IBTA defined MADs remain fixed at 256 bytes
2320  *    2) OPA SMPs must carry valid PKeys
2321  *    3) OPA SMP packets are a different format
2322  *
2323  * Return: true if the port supports OPA MAD packet formats.
2324  */
2325 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2326 {
2327 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2328 		== RDMA_CORE_CAP_OPA_MAD;
2329 }
2330 
2331 /**
2332  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2333  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2334  * @device: Device to check
2335  * @port_num: Port number to check
2336  *
2337  * Each InfiniBand node is required to provide a Subnet Management Agent
2338  * that the subnet manager can access.  Prior to the fabric being fully
2339  * configured by the subnet manager, the SMA is accessed via a well known
2340  * interface called the Subnet Management Interface (SMI).  This interface
2341  * uses directed route packets to communicate with the SM to get around the
2342  * chicken and egg problem of the SM needing to know what's on the fabric
2343  * in order to configure the fabric, and needing to configure the fabric in
2344  * order to send packets to the devices on the fabric.  These directed
2345  * route packets do not need the fabric fully configured in order to reach
2346  * their destination.  The SMI is the only method allowed to send
2347  * directed route packets on an InfiniBand fabric.
2348  *
2349  * Return: true if the port provides an SMI.
2350  */
2351 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2352 {
2353 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2354 }
2355 
2356 /**
2357  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2358  * Communication Manager.
2359  * @device: Device to check
2360  * @port_num: Port number to check
2361  *
2362  * The InfiniBand Communication Manager is one of many pre-defined General
2363  * Service Agents (GSA) that are accessed via the General Service
2364  * Interface (GSI).  It's role is to facilitate establishment of connections
2365  * between nodes as well as other management related tasks for established
2366  * connections.
2367  *
2368  * Return: true if the port supports an IB CM (this does not guarantee that
2369  * a CM is actually running however).
2370  */
2371 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2372 {
2373 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2374 }
2375 
2376 /**
2377  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2378  * Communication Manager.
2379  * @device: Device to check
2380  * @port_num: Port number to check
2381  *
2382  * Similar to above, but specific to iWARP connections which have a different
2383  * managment protocol than InfiniBand.
2384  *
2385  * Return: true if the port supports an iWARP CM (this does not guarantee that
2386  * a CM is actually running however).
2387  */
2388 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2389 {
2390 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2391 }
2392 
2393 /**
2394  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2395  * Subnet Administration.
2396  * @device: Device to check
2397  * @port_num: Port number to check
2398  *
2399  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2400  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2401  * fabrics, devices should resolve routes to other hosts by contacting the
2402  * SA to query the proper route.
2403  *
2404  * Return: true if the port should act as a client to the fabric Subnet
2405  * Administration interface.  This does not imply that the SA service is
2406  * running locally.
2407  */
2408 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2409 {
2410 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2411 }
2412 
2413 /**
2414  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2415  * Multicast.
2416  * @device: Device to check
2417  * @port_num: Port number to check
2418  *
2419  * InfiniBand multicast registration is more complex than normal IPv4 or
2420  * IPv6 multicast registration.  Each Host Channel Adapter must register
2421  * with the Subnet Manager when it wishes to join a multicast group.  It
2422  * should do so only once regardless of how many queue pairs it subscribes
2423  * to this group.  And it should leave the group only after all queue pairs
2424  * attached to the group have been detached.
2425  *
2426  * Return: true if the port must undertake the additional adminstrative
2427  * overhead of registering/unregistering with the SM and tracking of the
2428  * total number of queue pairs attached to the multicast group.
2429  */
2430 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2431 {
2432 	return rdma_cap_ib_sa(device, port_num);
2433 }
2434 
2435 /**
2436  * rdma_cap_af_ib - Check if the port of device has the capability
2437  * Native Infiniband Address.
2438  * @device: Device to check
2439  * @port_num: Port number to check
2440  *
2441  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2442  * GID.  RoCE uses a different mechanism, but still generates a GID via
2443  * a prescribed mechanism and port specific data.
2444  *
2445  * Return: true if the port uses a GID address to identify devices on the
2446  * network.
2447  */
2448 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2449 {
2450 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2451 }
2452 
2453 /**
2454  * rdma_cap_eth_ah - Check if the port of device has the capability
2455  * Ethernet Address Handle.
2456  * @device: Device to check
2457  * @port_num: Port number to check
2458  *
2459  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2460  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2461  * port.  Normally, packet headers are generated by the sending host
2462  * adapter, but when sending connectionless datagrams, we must manually
2463  * inject the proper headers for the fabric we are communicating over.
2464  *
2465  * Return: true if we are running as a RoCE port and must force the
2466  * addition of a Global Route Header built from our Ethernet Address
2467  * Handle into our header list for connectionless packets.
2468  */
2469 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2470 {
2471 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2472 }
2473 
2474 /**
2475  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2476  *
2477  * @device: Device
2478  * @port_num: Port number
2479  *
2480  * This MAD size includes the MAD headers and MAD payload.  No other headers
2481  * are included.
2482  *
2483  * Return the max MAD size required by the Port.  Will return 0 if the port
2484  * does not support MADs
2485  */
2486 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2487 {
2488 	return device->port_immutable[port_num].max_mad_size;
2489 }
2490 
2491 /**
2492  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2493  * @device: Device to check
2494  * @port_num: Port number to check
2495  *
2496  * RoCE GID table mechanism manages the various GIDs for a device.
2497  *
2498  * NOTE: if allocating the port's GID table has failed, this call will still
2499  * return true, but any RoCE GID table API will fail.
2500  *
2501  * Return: true if the port uses RoCE GID table mechanism in order to manage
2502  * its GIDs.
2503  */
2504 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2505 					   u8 port_num)
2506 {
2507 	return rdma_protocol_roce(device, port_num) &&
2508 		device->add_gid && device->del_gid;
2509 }
2510 
2511 /*
2512  * Check if the device supports READ W/ INVALIDATE.
2513  */
2514 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2515 {
2516 	/*
2517 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2518 	 * has support for it yet.
2519 	 */
2520 	return rdma_protocol_iwarp(dev, port_num);
2521 }
2522 
2523 int ib_query_gid(struct ib_device *device,
2524 		 u8 port_num, int index, union ib_gid *gid,
2525 		 struct ib_gid_attr *attr);
2526 
2527 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2528 			 int state);
2529 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2530 		     struct ifla_vf_info *info);
2531 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2532 		    struct ifla_vf_stats *stats);
2533 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2534 		   int type);
2535 
2536 int ib_query_pkey(struct ib_device *device,
2537 		  u8 port_num, u16 index, u16 *pkey);
2538 
2539 int ib_modify_device(struct ib_device *device,
2540 		     int device_modify_mask,
2541 		     struct ib_device_modify *device_modify);
2542 
2543 int ib_modify_port(struct ib_device *device,
2544 		   u8 port_num, int port_modify_mask,
2545 		   struct ib_port_modify *port_modify);
2546 
2547 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2548 		enum ib_gid_type gid_type, struct net_device *ndev,
2549 		u8 *port_num, u16 *index);
2550 
2551 int ib_find_pkey(struct ib_device *device,
2552 		 u8 port_num, u16 pkey, u16 *index);
2553 
2554 enum ib_pd_flags {
2555 	/*
2556 	 * Create a memory registration for all memory in the system and place
2557 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2558 	 * ULPs to avoid the overhead of dynamic MRs.
2559 	 *
2560 	 * This flag is generally considered unsafe and must only be used in
2561 	 * extremly trusted environments.  Every use of it will log a warning
2562 	 * in the kernel log.
2563 	 */
2564 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2565 };
2566 
2567 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2568 		const char *caller);
2569 #define ib_alloc_pd(device, flags) \
2570 	__ib_alloc_pd((device), (flags), __func__)
2571 void ib_dealloc_pd(struct ib_pd *pd);
2572 
2573 /**
2574  * ib_create_ah - Creates an address handle for the given address vector.
2575  * @pd: The protection domain associated with the address handle.
2576  * @ah_attr: The attributes of the address vector.
2577  *
2578  * The address handle is used to reference a local or global destination
2579  * in all UD QP post sends.
2580  */
2581 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2582 
2583 /**
2584  * ib_init_ah_from_wc - Initializes address handle attributes from a
2585  *   work completion.
2586  * @device: Device on which the received message arrived.
2587  * @port_num: Port on which the received message arrived.
2588  * @wc: Work completion associated with the received message.
2589  * @grh: References the received global route header.  This parameter is
2590  *   ignored unless the work completion indicates that the GRH is valid.
2591  * @ah_attr: Returned attributes that can be used when creating an address
2592  *   handle for replying to the message.
2593  */
2594 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2595 		       const struct ib_wc *wc, const struct ib_grh *grh,
2596 		       struct ib_ah_attr *ah_attr);
2597 
2598 /**
2599  * ib_create_ah_from_wc - Creates an address handle associated with the
2600  *   sender of the specified work completion.
2601  * @pd: The protection domain associated with the address handle.
2602  * @wc: Work completion information associated with a received message.
2603  * @grh: References the received global route header.  This parameter is
2604  *   ignored unless the work completion indicates that the GRH is valid.
2605  * @port_num: The outbound port number to associate with the address.
2606  *
2607  * The address handle is used to reference a local or global destination
2608  * in all UD QP post sends.
2609  */
2610 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2611 				   const struct ib_grh *grh, u8 port_num);
2612 
2613 /**
2614  * ib_modify_ah - Modifies the address vector associated with an address
2615  *   handle.
2616  * @ah: The address handle to modify.
2617  * @ah_attr: The new address vector attributes to associate with the
2618  *   address handle.
2619  */
2620 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2621 
2622 /**
2623  * ib_query_ah - Queries the address vector associated with an address
2624  *   handle.
2625  * @ah: The address handle to query.
2626  * @ah_attr: The address vector attributes associated with the address
2627  *   handle.
2628  */
2629 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2630 
2631 /**
2632  * ib_destroy_ah - Destroys an address handle.
2633  * @ah: The address handle to destroy.
2634  */
2635 int ib_destroy_ah(struct ib_ah *ah);
2636 
2637 /**
2638  * ib_create_srq - Creates a SRQ associated with the specified protection
2639  *   domain.
2640  * @pd: The protection domain associated with the SRQ.
2641  * @srq_init_attr: A list of initial attributes required to create the
2642  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2643  *   the actual capabilities of the created SRQ.
2644  *
2645  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2646  * requested size of the SRQ, and set to the actual values allocated
2647  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2648  * will always be at least as large as the requested values.
2649  */
2650 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2651 			     struct ib_srq_init_attr *srq_init_attr);
2652 
2653 /**
2654  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2655  * @srq: The SRQ to modify.
2656  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2657  *   the current values of selected SRQ attributes are returned.
2658  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2659  *   are being modified.
2660  *
2661  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2662  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2663  * the number of receives queued drops below the limit.
2664  */
2665 int ib_modify_srq(struct ib_srq *srq,
2666 		  struct ib_srq_attr *srq_attr,
2667 		  enum ib_srq_attr_mask srq_attr_mask);
2668 
2669 /**
2670  * ib_query_srq - Returns the attribute list and current values for the
2671  *   specified SRQ.
2672  * @srq: The SRQ to query.
2673  * @srq_attr: The attributes of the specified SRQ.
2674  */
2675 int ib_query_srq(struct ib_srq *srq,
2676 		 struct ib_srq_attr *srq_attr);
2677 
2678 /**
2679  * ib_destroy_srq - Destroys the specified SRQ.
2680  * @srq: The SRQ to destroy.
2681  */
2682 int ib_destroy_srq(struct ib_srq *srq);
2683 
2684 /**
2685  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2686  * @srq: The SRQ to post the work request on.
2687  * @recv_wr: A list of work requests to post on the receive queue.
2688  * @bad_recv_wr: On an immediate failure, this parameter will reference
2689  *   the work request that failed to be posted on the QP.
2690  */
2691 static inline int ib_post_srq_recv(struct ib_srq *srq,
2692 				   struct ib_recv_wr *recv_wr,
2693 				   struct ib_recv_wr **bad_recv_wr)
2694 {
2695 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2696 }
2697 
2698 /**
2699  * ib_create_qp - Creates a QP associated with the specified protection
2700  *   domain.
2701  * @pd: The protection domain associated with the QP.
2702  * @qp_init_attr: A list of initial attributes required to create the
2703  *   QP.  If QP creation succeeds, then the attributes are updated to
2704  *   the actual capabilities of the created QP.
2705  */
2706 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2707 			   struct ib_qp_init_attr *qp_init_attr);
2708 
2709 /**
2710  * ib_modify_qp - Modifies the attributes for the specified QP and then
2711  *   transitions the QP to the given state.
2712  * @qp: The QP to modify.
2713  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2714  *   the current values of selected QP attributes are returned.
2715  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2716  *   are being modified.
2717  */
2718 int ib_modify_qp(struct ib_qp *qp,
2719 		 struct ib_qp_attr *qp_attr,
2720 		 int qp_attr_mask);
2721 
2722 /**
2723  * ib_query_qp - Returns the attribute list and current values for the
2724  *   specified QP.
2725  * @qp: The QP to query.
2726  * @qp_attr: The attributes of the specified QP.
2727  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2728  * @qp_init_attr: Additional attributes of the selected QP.
2729  *
2730  * The qp_attr_mask may be used to limit the query to gathering only the
2731  * selected attributes.
2732  */
2733 int ib_query_qp(struct ib_qp *qp,
2734 		struct ib_qp_attr *qp_attr,
2735 		int qp_attr_mask,
2736 		struct ib_qp_init_attr *qp_init_attr);
2737 
2738 /**
2739  * ib_destroy_qp - Destroys the specified QP.
2740  * @qp: The QP to destroy.
2741  */
2742 int ib_destroy_qp(struct ib_qp *qp);
2743 
2744 /**
2745  * ib_open_qp - Obtain a reference to an existing sharable QP.
2746  * @xrcd - XRC domain
2747  * @qp_open_attr: Attributes identifying the QP to open.
2748  *
2749  * Returns a reference to a sharable QP.
2750  */
2751 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2752 			 struct ib_qp_open_attr *qp_open_attr);
2753 
2754 /**
2755  * ib_close_qp - Release an external reference to a QP.
2756  * @qp: The QP handle to release
2757  *
2758  * The opened QP handle is released by the caller.  The underlying
2759  * shared QP is not destroyed until all internal references are released.
2760  */
2761 int ib_close_qp(struct ib_qp *qp);
2762 
2763 /**
2764  * ib_post_send - Posts a list of work requests to the send queue of
2765  *   the specified QP.
2766  * @qp: The QP to post the work request on.
2767  * @send_wr: A list of work requests to post on the send queue.
2768  * @bad_send_wr: On an immediate failure, this parameter will reference
2769  *   the work request that failed to be posted on the QP.
2770  *
2771  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2772  * error is returned, the QP state shall not be affected,
2773  * ib_post_send() will return an immediate error after queueing any
2774  * earlier work requests in the list.
2775  */
2776 static inline int ib_post_send(struct ib_qp *qp,
2777 			       struct ib_send_wr *send_wr,
2778 			       struct ib_send_wr **bad_send_wr)
2779 {
2780 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2781 }
2782 
2783 /**
2784  * ib_post_recv - Posts a list of work requests to the receive queue of
2785  *   the specified QP.
2786  * @qp: The QP to post the work request on.
2787  * @recv_wr: A list of work requests to post on the receive queue.
2788  * @bad_recv_wr: On an immediate failure, this parameter will reference
2789  *   the work request that failed to be posted on the QP.
2790  */
2791 static inline int ib_post_recv(struct ib_qp *qp,
2792 			       struct ib_recv_wr *recv_wr,
2793 			       struct ib_recv_wr **bad_recv_wr)
2794 {
2795 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2796 }
2797 
2798 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2799 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2800 void ib_free_cq(struct ib_cq *cq);
2801 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2802 
2803 /**
2804  * ib_create_cq - Creates a CQ on the specified device.
2805  * @device: The device on which to create the CQ.
2806  * @comp_handler: A user-specified callback that is invoked when a
2807  *   completion event occurs on the CQ.
2808  * @event_handler: A user-specified callback that is invoked when an
2809  *   asynchronous event not associated with a completion occurs on the CQ.
2810  * @cq_context: Context associated with the CQ returned to the user via
2811  *   the associated completion and event handlers.
2812  * @cq_attr: The attributes the CQ should be created upon.
2813  *
2814  * Users can examine the cq structure to determine the actual CQ size.
2815  */
2816 struct ib_cq *ib_create_cq(struct ib_device *device,
2817 			   ib_comp_handler comp_handler,
2818 			   void (*event_handler)(struct ib_event *, void *),
2819 			   void *cq_context,
2820 			   const struct ib_cq_init_attr *cq_attr);
2821 
2822 /**
2823  * ib_resize_cq - Modifies the capacity of the CQ.
2824  * @cq: The CQ to resize.
2825  * @cqe: The minimum size of the CQ.
2826  *
2827  * Users can examine the cq structure to determine the actual CQ size.
2828  */
2829 int ib_resize_cq(struct ib_cq *cq, int cqe);
2830 
2831 /**
2832  * ib_modify_cq - Modifies moderation params of the CQ
2833  * @cq: The CQ to modify.
2834  * @cq_count: number of CQEs that will trigger an event
2835  * @cq_period: max period of time in usec before triggering an event
2836  *
2837  */
2838 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2839 
2840 /**
2841  * ib_destroy_cq - Destroys the specified CQ.
2842  * @cq: The CQ to destroy.
2843  */
2844 int ib_destroy_cq(struct ib_cq *cq);
2845 
2846 /**
2847  * ib_poll_cq - poll a CQ for completion(s)
2848  * @cq:the CQ being polled
2849  * @num_entries:maximum number of completions to return
2850  * @wc:array of at least @num_entries &struct ib_wc where completions
2851  *   will be returned
2852  *
2853  * Poll a CQ for (possibly multiple) completions.  If the return value
2854  * is < 0, an error occurred.  If the return value is >= 0, it is the
2855  * number of completions returned.  If the return value is
2856  * non-negative and < num_entries, then the CQ was emptied.
2857  */
2858 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2859 			     struct ib_wc *wc)
2860 {
2861 	return cq->device->poll_cq(cq, num_entries, wc);
2862 }
2863 
2864 /**
2865  * ib_peek_cq - Returns the number of unreaped completions currently
2866  *   on the specified CQ.
2867  * @cq: The CQ to peek.
2868  * @wc_cnt: A minimum number of unreaped completions to check for.
2869  *
2870  * If the number of unreaped completions is greater than or equal to wc_cnt,
2871  * this function returns wc_cnt, otherwise, it returns the actual number of
2872  * unreaped completions.
2873  */
2874 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2875 
2876 /**
2877  * ib_req_notify_cq - Request completion notification on a CQ.
2878  * @cq: The CQ to generate an event for.
2879  * @flags:
2880  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2881  *   to request an event on the next solicited event or next work
2882  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2883  *   may also be |ed in to request a hint about missed events, as
2884  *   described below.
2885  *
2886  * Return Value:
2887  *    < 0 means an error occurred while requesting notification
2888  *   == 0 means notification was requested successfully, and if
2889  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2890  *        were missed and it is safe to wait for another event.  In
2891  *        this case is it guaranteed that any work completions added
2892  *        to the CQ since the last CQ poll will trigger a completion
2893  *        notification event.
2894  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2895  *        in.  It means that the consumer must poll the CQ again to
2896  *        make sure it is empty to avoid missing an event because of a
2897  *        race between requesting notification and an entry being
2898  *        added to the CQ.  This return value means it is possible
2899  *        (but not guaranteed) that a work completion has been added
2900  *        to the CQ since the last poll without triggering a
2901  *        completion notification event.
2902  */
2903 static inline int ib_req_notify_cq(struct ib_cq *cq,
2904 				   enum ib_cq_notify_flags flags)
2905 {
2906 	return cq->device->req_notify_cq(cq, flags);
2907 }
2908 
2909 /**
2910  * ib_req_ncomp_notif - Request completion notification when there are
2911  *   at least the specified number of unreaped completions on the CQ.
2912  * @cq: The CQ to generate an event for.
2913  * @wc_cnt: The number of unreaped completions that should be on the
2914  *   CQ before an event is generated.
2915  */
2916 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2917 {
2918 	return cq->device->req_ncomp_notif ?
2919 		cq->device->req_ncomp_notif(cq, wc_cnt) :
2920 		-ENOSYS;
2921 }
2922 
2923 /**
2924  * ib_dma_mapping_error - check a DMA addr for error
2925  * @dev: The device for which the dma_addr was created
2926  * @dma_addr: The DMA address to check
2927  */
2928 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2929 {
2930 	if (dev->dma_ops)
2931 		return dev->dma_ops->mapping_error(dev, dma_addr);
2932 	return dma_mapping_error(dev->dma_device, dma_addr);
2933 }
2934 
2935 /**
2936  * ib_dma_map_single - Map a kernel virtual address to DMA address
2937  * @dev: The device for which the dma_addr is to be created
2938  * @cpu_addr: The kernel virtual address
2939  * @size: The size of the region in bytes
2940  * @direction: The direction of the DMA
2941  */
2942 static inline u64 ib_dma_map_single(struct ib_device *dev,
2943 				    void *cpu_addr, size_t size,
2944 				    enum dma_data_direction direction)
2945 {
2946 	if (dev->dma_ops)
2947 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2948 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2949 }
2950 
2951 /**
2952  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2953  * @dev: The device for which the DMA address was created
2954  * @addr: The DMA address
2955  * @size: The size of the region in bytes
2956  * @direction: The direction of the DMA
2957  */
2958 static inline void ib_dma_unmap_single(struct ib_device *dev,
2959 				       u64 addr, size_t size,
2960 				       enum dma_data_direction direction)
2961 {
2962 	if (dev->dma_ops)
2963 		dev->dma_ops->unmap_single(dev, addr, size, direction);
2964 	else
2965 		dma_unmap_single(dev->dma_device, addr, size, direction);
2966 }
2967 
2968 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2969 					  void *cpu_addr, size_t size,
2970 					  enum dma_data_direction direction,
2971 					  unsigned long dma_attrs)
2972 {
2973 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2974 				    direction, dma_attrs);
2975 }
2976 
2977 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2978 					     u64 addr, size_t size,
2979 					     enum dma_data_direction direction,
2980 					     unsigned long dma_attrs)
2981 {
2982 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
2983 				      direction, dma_attrs);
2984 }
2985 
2986 /**
2987  * ib_dma_map_page - Map a physical page to DMA address
2988  * @dev: The device for which the dma_addr is to be created
2989  * @page: The page to be mapped
2990  * @offset: The offset within the page
2991  * @size: The size of the region in bytes
2992  * @direction: The direction of the DMA
2993  */
2994 static inline u64 ib_dma_map_page(struct ib_device *dev,
2995 				  struct page *page,
2996 				  unsigned long offset,
2997 				  size_t size,
2998 					 enum dma_data_direction direction)
2999 {
3000 	if (dev->dma_ops)
3001 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
3002 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3003 }
3004 
3005 /**
3006  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3007  * @dev: The device for which the DMA address was created
3008  * @addr: The DMA address
3009  * @size: The size of the region in bytes
3010  * @direction: The direction of the DMA
3011  */
3012 static inline void ib_dma_unmap_page(struct ib_device *dev,
3013 				     u64 addr, size_t size,
3014 				     enum dma_data_direction direction)
3015 {
3016 	if (dev->dma_ops)
3017 		dev->dma_ops->unmap_page(dev, addr, size, direction);
3018 	else
3019 		dma_unmap_page(dev->dma_device, addr, size, direction);
3020 }
3021 
3022 /**
3023  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3024  * @dev: The device for which the DMA addresses are to be created
3025  * @sg: The array of scatter/gather entries
3026  * @nents: The number of scatter/gather entries
3027  * @direction: The direction of the DMA
3028  */
3029 static inline int ib_dma_map_sg(struct ib_device *dev,
3030 				struct scatterlist *sg, int nents,
3031 				enum dma_data_direction direction)
3032 {
3033 	if (dev->dma_ops)
3034 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
3035 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3036 }
3037 
3038 /**
3039  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3040  * @dev: The device for which the DMA addresses were created
3041  * @sg: The array of scatter/gather entries
3042  * @nents: The number of scatter/gather entries
3043  * @direction: The direction of the DMA
3044  */
3045 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3046 				   struct scatterlist *sg, int nents,
3047 				   enum dma_data_direction direction)
3048 {
3049 	if (dev->dma_ops)
3050 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
3051 	else
3052 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
3053 }
3054 
3055 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3056 				      struct scatterlist *sg, int nents,
3057 				      enum dma_data_direction direction,
3058 				      unsigned long dma_attrs)
3059 {
3060 	if (dev->dma_ops)
3061 		return dev->dma_ops->map_sg_attrs(dev, sg, nents, direction,
3062 						  dma_attrs);
3063 	else
3064 		return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3065 					dma_attrs);
3066 }
3067 
3068 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3069 					 struct scatterlist *sg, int nents,
3070 					 enum dma_data_direction direction,
3071 					 unsigned long dma_attrs)
3072 {
3073 	if (dev->dma_ops)
3074 		return dev->dma_ops->unmap_sg_attrs(dev, sg, nents, direction,
3075 						  dma_attrs);
3076 	else
3077 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
3078 				   dma_attrs);
3079 }
3080 /**
3081  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3082  * @dev: The device for which the DMA addresses were created
3083  * @sg: The scatter/gather entry
3084  *
3085  * Note: this function is obsolete. To do: change all occurrences of
3086  * ib_sg_dma_address() into sg_dma_address().
3087  */
3088 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3089 				    struct scatterlist *sg)
3090 {
3091 	return sg_dma_address(sg);
3092 }
3093 
3094 /**
3095  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3096  * @dev: The device for which the DMA addresses were created
3097  * @sg: The scatter/gather entry
3098  *
3099  * Note: this function is obsolete. To do: change all occurrences of
3100  * ib_sg_dma_len() into sg_dma_len().
3101  */
3102 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3103 					 struct scatterlist *sg)
3104 {
3105 	return sg_dma_len(sg);
3106 }
3107 
3108 /**
3109  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3110  * @dev: The device for which the DMA address was created
3111  * @addr: The DMA address
3112  * @size: The size of the region in bytes
3113  * @dir: The direction of the DMA
3114  */
3115 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3116 					      u64 addr,
3117 					      size_t size,
3118 					      enum dma_data_direction dir)
3119 {
3120 	if (dev->dma_ops)
3121 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
3122 	else
3123 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3124 }
3125 
3126 /**
3127  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3128  * @dev: The device for which the DMA address was created
3129  * @addr: The DMA address
3130  * @size: The size of the region in bytes
3131  * @dir: The direction of the DMA
3132  */
3133 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3134 						 u64 addr,
3135 						 size_t size,
3136 						 enum dma_data_direction dir)
3137 {
3138 	if (dev->dma_ops)
3139 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
3140 	else
3141 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3142 }
3143 
3144 /**
3145  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3146  * @dev: The device for which the DMA address is requested
3147  * @size: The size of the region to allocate in bytes
3148  * @dma_handle: A pointer for returning the DMA address of the region
3149  * @flag: memory allocator flags
3150  */
3151 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3152 					   size_t size,
3153 					   u64 *dma_handle,
3154 					   gfp_t flag)
3155 {
3156 	if (dev->dma_ops)
3157 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
3158 	else {
3159 		dma_addr_t handle;
3160 		void *ret;
3161 
3162 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3163 		*dma_handle = handle;
3164 		return ret;
3165 	}
3166 }
3167 
3168 /**
3169  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3170  * @dev: The device for which the DMA addresses were allocated
3171  * @size: The size of the region
3172  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3173  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3174  */
3175 static inline void ib_dma_free_coherent(struct ib_device *dev,
3176 					size_t size, void *cpu_addr,
3177 					u64 dma_handle)
3178 {
3179 	if (dev->dma_ops)
3180 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3181 	else
3182 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3183 }
3184 
3185 /**
3186  * ib_dereg_mr - Deregisters a memory region and removes it from the
3187  *   HCA translation table.
3188  * @mr: The memory region to deregister.
3189  *
3190  * This function can fail, if the memory region has memory windows bound to it.
3191  */
3192 int ib_dereg_mr(struct ib_mr *mr);
3193 
3194 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3195 			  enum ib_mr_type mr_type,
3196 			  u32 max_num_sg);
3197 
3198 /**
3199  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3200  *   R_Key and L_Key.
3201  * @mr - struct ib_mr pointer to be updated.
3202  * @newkey - new key to be used.
3203  */
3204 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3205 {
3206 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3207 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3208 }
3209 
3210 /**
3211  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3212  * for calculating a new rkey for type 2 memory windows.
3213  * @rkey - the rkey to increment.
3214  */
3215 static inline u32 ib_inc_rkey(u32 rkey)
3216 {
3217 	const u32 mask = 0x000000ff;
3218 	return ((rkey + 1) & mask) | (rkey & ~mask);
3219 }
3220 
3221 /**
3222  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3223  * @pd: The protection domain associated with the unmapped region.
3224  * @mr_access_flags: Specifies the memory access rights.
3225  * @fmr_attr: Attributes of the unmapped region.
3226  *
3227  * A fast memory region must be mapped before it can be used as part of
3228  * a work request.
3229  */
3230 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3231 			    int mr_access_flags,
3232 			    struct ib_fmr_attr *fmr_attr);
3233 
3234 /**
3235  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3236  * @fmr: The fast memory region to associate with the pages.
3237  * @page_list: An array of physical pages to map to the fast memory region.
3238  * @list_len: The number of pages in page_list.
3239  * @iova: The I/O virtual address to use with the mapped region.
3240  */
3241 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3242 				  u64 *page_list, int list_len,
3243 				  u64 iova)
3244 {
3245 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3246 }
3247 
3248 /**
3249  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3250  * @fmr_list: A linked list of fast memory regions to unmap.
3251  */
3252 int ib_unmap_fmr(struct list_head *fmr_list);
3253 
3254 /**
3255  * ib_dealloc_fmr - Deallocates a fast memory region.
3256  * @fmr: The fast memory region to deallocate.
3257  */
3258 int ib_dealloc_fmr(struct ib_fmr *fmr);
3259 
3260 /**
3261  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3262  * @qp: QP to attach to the multicast group.  The QP must be type
3263  *   IB_QPT_UD.
3264  * @gid: Multicast group GID.
3265  * @lid: Multicast group LID in host byte order.
3266  *
3267  * In order to send and receive multicast packets, subnet
3268  * administration must have created the multicast group and configured
3269  * the fabric appropriately.  The port associated with the specified
3270  * QP must also be a member of the multicast group.
3271  */
3272 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3273 
3274 /**
3275  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3276  * @qp: QP to detach from the multicast group.
3277  * @gid: Multicast group GID.
3278  * @lid: Multicast group LID in host byte order.
3279  */
3280 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3281 
3282 /**
3283  * ib_alloc_xrcd - Allocates an XRC domain.
3284  * @device: The device on which to allocate the XRC domain.
3285  */
3286 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3287 
3288 /**
3289  * ib_dealloc_xrcd - Deallocates an XRC domain.
3290  * @xrcd: The XRC domain to deallocate.
3291  */
3292 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3293 
3294 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3295 			       struct ib_flow_attr *flow_attr, int domain);
3296 int ib_destroy_flow(struct ib_flow *flow_id);
3297 
3298 static inline int ib_check_mr_access(int flags)
3299 {
3300 	/*
3301 	 * Local write permission is required if remote write or
3302 	 * remote atomic permission is also requested.
3303 	 */
3304 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3305 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3306 		return -EINVAL;
3307 
3308 	return 0;
3309 }
3310 
3311 /**
3312  * ib_check_mr_status: lightweight check of MR status.
3313  *     This routine may provide status checks on a selected
3314  *     ib_mr. first use is for signature status check.
3315  *
3316  * @mr: A memory region.
3317  * @check_mask: Bitmask of which checks to perform from
3318  *     ib_mr_status_check enumeration.
3319  * @mr_status: The container of relevant status checks.
3320  *     failed checks will be indicated in the status bitmask
3321  *     and the relevant info shall be in the error item.
3322  */
3323 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3324 		       struct ib_mr_status *mr_status);
3325 
3326 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3327 					    u16 pkey, const union ib_gid *gid,
3328 					    const struct sockaddr *addr);
3329 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3330 			   struct ib_wq_init_attr *init_attr);
3331 int ib_destroy_wq(struct ib_wq *wq);
3332 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3333 		 u32 wq_attr_mask);
3334 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3335 						 struct ib_rwq_ind_table_init_attr*
3336 						 wq_ind_table_init_attr);
3337 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3338 
3339 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3340 		 unsigned int *sg_offset, unsigned int page_size);
3341 
3342 static inline int
3343 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3344 		  unsigned int *sg_offset, unsigned int page_size)
3345 {
3346 	int n;
3347 
3348 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3349 	mr->iova = 0;
3350 
3351 	return n;
3352 }
3353 
3354 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3355 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3356 
3357 void ib_drain_rq(struct ib_qp *qp);
3358 void ib_drain_sq(struct ib_qp *qp);
3359 void ib_drain_qp(struct ib_qp *qp);
3360 #endif /* IB_VERBS_H */
3361