xref: /linux/include/rdma/ib_verbs.h (revision 0bbb3b7496eabb6779962a998a8a91f4a8e589ff)
1 /*
2  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
3  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
4  * Copyright (c) 2004 Intel Corporation.  All rights reserved.
5  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
6  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
7  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
9  *
10  * This software is available to you under a choice of one of two
11  * licenses.  You may choose to be licensed under the terms of the GNU
12  * General Public License (GPL) Version 2, available from the file
13  * COPYING in the main directory of this source tree, or the
14  * OpenIB.org BSD license below:
15  *
16  *     Redistribution and use in source and binary forms, with or
17  *     without modification, are permitted provided that the following
18  *     conditions are met:
19  *
20  *      - Redistributions of source code must retain the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer.
23  *
24  *      - Redistributions in binary form must reproduce the above
25  *        copyright notice, this list of conditions and the following
26  *        disclaimer in the documentation and/or other materials
27  *        provided with the distribution.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36  * SOFTWARE.
37  */
38 
39 #if !defined(IB_VERBS_H)
40 #define IB_VERBS_H
41 
42 #include <linux/types.h>
43 #include <linux/device.h>
44 #include <linux/mm.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
54 #include <net/ipv6.h>
55 #include <net/ip.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 
59 #include <linux/if_link.h>
60 #include <linux/atomic.h>
61 #include <linux/mmu_notifier.h>
62 #include <linux/uaccess.h>
63 
64 extern struct workqueue_struct *ib_wq;
65 extern struct workqueue_struct *ib_comp_wq;
66 
67 union ib_gid {
68 	u8	raw[16];
69 	struct {
70 		__be64	subnet_prefix;
71 		__be64	interface_id;
72 	} global;
73 };
74 
75 extern union ib_gid zgid;
76 
77 enum ib_gid_type {
78 	/* If link layer is Ethernet, this is RoCE V1 */
79 	IB_GID_TYPE_IB        = 0,
80 	IB_GID_TYPE_ROCE      = 0,
81 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
82 	IB_GID_TYPE_SIZE
83 };
84 
85 #define ROCE_V2_UDP_DPORT      4791
86 struct ib_gid_attr {
87 	enum ib_gid_type	gid_type;
88 	struct net_device	*ndev;
89 };
90 
91 enum rdma_node_type {
92 	/* IB values map to NodeInfo:NodeType. */
93 	RDMA_NODE_IB_CA 	= 1,
94 	RDMA_NODE_IB_SWITCH,
95 	RDMA_NODE_IB_ROUTER,
96 	RDMA_NODE_RNIC,
97 	RDMA_NODE_USNIC,
98 	RDMA_NODE_USNIC_UDP,
99 };
100 
101 enum {
102 	/* set the local administered indication */
103 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
104 };
105 
106 enum rdma_transport_type {
107 	RDMA_TRANSPORT_IB,
108 	RDMA_TRANSPORT_IWARP,
109 	RDMA_TRANSPORT_USNIC,
110 	RDMA_TRANSPORT_USNIC_UDP
111 };
112 
113 enum rdma_protocol_type {
114 	RDMA_PROTOCOL_IB,
115 	RDMA_PROTOCOL_IBOE,
116 	RDMA_PROTOCOL_IWARP,
117 	RDMA_PROTOCOL_USNIC_UDP
118 };
119 
120 __attribute_const__ enum rdma_transport_type
121 rdma_node_get_transport(enum rdma_node_type node_type);
122 
123 enum rdma_network_type {
124 	RDMA_NETWORK_IB,
125 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
126 	RDMA_NETWORK_IPV4,
127 	RDMA_NETWORK_IPV6
128 };
129 
130 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
131 {
132 	if (network_type == RDMA_NETWORK_IPV4 ||
133 	    network_type == RDMA_NETWORK_IPV6)
134 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
135 
136 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
137 	return IB_GID_TYPE_IB;
138 }
139 
140 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
141 							    union ib_gid *gid)
142 {
143 	if (gid_type == IB_GID_TYPE_IB)
144 		return RDMA_NETWORK_IB;
145 
146 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
147 		return RDMA_NETWORK_IPV4;
148 	else
149 		return RDMA_NETWORK_IPV6;
150 }
151 
152 enum rdma_link_layer {
153 	IB_LINK_LAYER_UNSPECIFIED,
154 	IB_LINK_LAYER_INFINIBAND,
155 	IB_LINK_LAYER_ETHERNET,
156 };
157 
158 enum ib_device_cap_flags {
159 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
160 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
161 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
162 	IB_DEVICE_RAW_MULTI			= (1 << 3),
163 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
164 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
165 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
166 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
167 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
168 	IB_DEVICE_INIT_TYPE			= (1 << 9),
169 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
170 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
171 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
172 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
173 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
174 
175 	/*
176 	 * This device supports a per-device lkey or stag that can be
177 	 * used without performing a memory registration for the local
178 	 * memory.  Note that ULPs should never check this flag, but
179 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
180 	 * which will always contain a usable lkey.
181 	 */
182 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
183 	IB_DEVICE_RESERVED /* old SEND_W_INV */	= (1 << 16),
184 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
185 	/*
186 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
187 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
188 	 * messages and can verify the validity of checksum for
189 	 * incoming messages.  Setting this flag implies that the
190 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
191 	 */
192 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
193 	IB_DEVICE_UD_TSO			= (1 << 19),
194 	IB_DEVICE_XRC				= (1 << 20),
195 
196 	/*
197 	 * This device supports the IB "base memory management extension",
198 	 * which includes support for fast registrations (IB_WR_REG_MR,
199 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
200 	 * also be set by any iWarp device which must support FRs to comply
201 	 * to the iWarp verbs spec.  iWarp devices also support the
202 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
203 	 * stag.
204 	 */
205 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
206 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
207 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
208 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
209 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
210 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
211 	/*
212 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
213 	 * support execution of WQEs that involve synchronization
214 	 * of I/O operations with single completion queue managed
215 	 * by hardware.
216 	 */
217 	IB_DEVICE_CROSS_CHANNEL		= (1 << 27),
218 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
219 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
220 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
221 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
222 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
223 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
224 };
225 
226 enum ib_signature_prot_cap {
227 	IB_PROT_T10DIF_TYPE_1 = 1,
228 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
229 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
230 };
231 
232 enum ib_signature_guard_cap {
233 	IB_GUARD_T10DIF_CRC	= 1,
234 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
235 };
236 
237 enum ib_atomic_cap {
238 	IB_ATOMIC_NONE,
239 	IB_ATOMIC_HCA,
240 	IB_ATOMIC_GLOB
241 };
242 
243 enum ib_odp_general_cap_bits {
244 	IB_ODP_SUPPORT = 1 << 0,
245 };
246 
247 enum ib_odp_transport_cap_bits {
248 	IB_ODP_SUPPORT_SEND	= 1 << 0,
249 	IB_ODP_SUPPORT_RECV	= 1 << 1,
250 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
251 	IB_ODP_SUPPORT_READ	= 1 << 3,
252 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
253 };
254 
255 struct ib_odp_caps {
256 	uint64_t general_caps;
257 	struct {
258 		uint32_t  rc_odp_caps;
259 		uint32_t  uc_odp_caps;
260 		uint32_t  ud_odp_caps;
261 	} per_transport_caps;
262 };
263 
264 struct ib_rss_caps {
265 	/* Corresponding bit will be set if qp type from
266 	 * 'enum ib_qp_type' is supported, e.g.
267 	 * supported_qpts |= 1 << IB_QPT_UD
268 	 */
269 	u32 supported_qpts;
270 	u32 max_rwq_indirection_tables;
271 	u32 max_rwq_indirection_table_size;
272 };
273 
274 enum ib_cq_creation_flags {
275 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
276 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
277 };
278 
279 struct ib_cq_init_attr {
280 	unsigned int	cqe;
281 	int		comp_vector;
282 	u32		flags;
283 };
284 
285 struct ib_device_attr {
286 	u64			fw_ver;
287 	__be64			sys_image_guid;
288 	u64			max_mr_size;
289 	u64			page_size_cap;
290 	u32			vendor_id;
291 	u32			vendor_part_id;
292 	u32			hw_ver;
293 	int			max_qp;
294 	int			max_qp_wr;
295 	u64			device_cap_flags;
296 	int			max_sge;
297 	int			max_sge_rd;
298 	int			max_cq;
299 	int			max_cqe;
300 	int			max_mr;
301 	int			max_pd;
302 	int			max_qp_rd_atom;
303 	int			max_ee_rd_atom;
304 	int			max_res_rd_atom;
305 	int			max_qp_init_rd_atom;
306 	int			max_ee_init_rd_atom;
307 	enum ib_atomic_cap	atomic_cap;
308 	enum ib_atomic_cap	masked_atomic_cap;
309 	int			max_ee;
310 	int			max_rdd;
311 	int			max_mw;
312 	int			max_raw_ipv6_qp;
313 	int			max_raw_ethy_qp;
314 	int			max_mcast_grp;
315 	int			max_mcast_qp_attach;
316 	int			max_total_mcast_qp_attach;
317 	int			max_ah;
318 	int			max_fmr;
319 	int			max_map_per_fmr;
320 	int			max_srq;
321 	int			max_srq_wr;
322 	int			max_srq_sge;
323 	unsigned int		max_fast_reg_page_list_len;
324 	u16			max_pkeys;
325 	u8			local_ca_ack_delay;
326 	int			sig_prot_cap;
327 	int			sig_guard_cap;
328 	struct ib_odp_caps	odp_caps;
329 	uint64_t		timestamp_mask;
330 	uint64_t		hca_core_clock; /* in KHZ */
331 	struct ib_rss_caps	rss_caps;
332 	u32			max_wq_type_rq;
333 };
334 
335 enum ib_mtu {
336 	IB_MTU_256  = 1,
337 	IB_MTU_512  = 2,
338 	IB_MTU_1024 = 3,
339 	IB_MTU_2048 = 4,
340 	IB_MTU_4096 = 5
341 };
342 
343 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
344 {
345 	switch (mtu) {
346 	case IB_MTU_256:  return  256;
347 	case IB_MTU_512:  return  512;
348 	case IB_MTU_1024: return 1024;
349 	case IB_MTU_2048: return 2048;
350 	case IB_MTU_4096: return 4096;
351 	default: 	  return -1;
352 	}
353 }
354 
355 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 	IB_QP_RATE_LIMIT		= (1<<25),
1106 };
1107 
1108 enum ib_qp_state {
1109 	IB_QPS_RESET,
1110 	IB_QPS_INIT,
1111 	IB_QPS_RTR,
1112 	IB_QPS_RTS,
1113 	IB_QPS_SQD,
1114 	IB_QPS_SQE,
1115 	IB_QPS_ERR
1116 };
1117 
1118 enum ib_mig_state {
1119 	IB_MIG_MIGRATED,
1120 	IB_MIG_REARM,
1121 	IB_MIG_ARMED
1122 };
1123 
1124 enum ib_mw_type {
1125 	IB_MW_TYPE_1 = 1,
1126 	IB_MW_TYPE_2 = 2
1127 };
1128 
1129 struct ib_qp_attr {
1130 	enum ib_qp_state	qp_state;
1131 	enum ib_qp_state	cur_qp_state;
1132 	enum ib_mtu		path_mtu;
1133 	enum ib_mig_state	path_mig_state;
1134 	u32			qkey;
1135 	u32			rq_psn;
1136 	u32			sq_psn;
1137 	u32			dest_qp_num;
1138 	int			qp_access_flags;
1139 	struct ib_qp_cap	cap;
1140 	struct ib_ah_attr	ah_attr;
1141 	struct ib_ah_attr	alt_ah_attr;
1142 	u16			pkey_index;
1143 	u16			alt_pkey_index;
1144 	u8			en_sqd_async_notify;
1145 	u8			sq_draining;
1146 	u8			max_rd_atomic;
1147 	u8			max_dest_rd_atomic;
1148 	u8			min_rnr_timer;
1149 	u8			port_num;
1150 	u8			timeout;
1151 	u8			retry_cnt;
1152 	u8			rnr_retry;
1153 	u8			alt_port_num;
1154 	u8			alt_timeout;
1155 	u32			rate_limit;
1156 };
1157 
1158 enum ib_wr_opcode {
1159 	IB_WR_RDMA_WRITE,
1160 	IB_WR_RDMA_WRITE_WITH_IMM,
1161 	IB_WR_SEND,
1162 	IB_WR_SEND_WITH_IMM,
1163 	IB_WR_RDMA_READ,
1164 	IB_WR_ATOMIC_CMP_AND_SWP,
1165 	IB_WR_ATOMIC_FETCH_AND_ADD,
1166 	IB_WR_LSO,
1167 	IB_WR_SEND_WITH_INV,
1168 	IB_WR_RDMA_READ_WITH_INV,
1169 	IB_WR_LOCAL_INV,
1170 	IB_WR_REG_MR,
1171 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1172 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1173 	IB_WR_REG_SIG_MR,
1174 	/* reserve values for low level drivers' internal use.
1175 	 * These values will not be used at all in the ib core layer.
1176 	 */
1177 	IB_WR_RESERVED1 = 0xf0,
1178 	IB_WR_RESERVED2,
1179 	IB_WR_RESERVED3,
1180 	IB_WR_RESERVED4,
1181 	IB_WR_RESERVED5,
1182 	IB_WR_RESERVED6,
1183 	IB_WR_RESERVED7,
1184 	IB_WR_RESERVED8,
1185 	IB_WR_RESERVED9,
1186 	IB_WR_RESERVED10,
1187 };
1188 
1189 enum ib_send_flags {
1190 	IB_SEND_FENCE		= 1,
1191 	IB_SEND_SIGNALED	= (1<<1),
1192 	IB_SEND_SOLICITED	= (1<<2),
1193 	IB_SEND_INLINE		= (1<<3),
1194 	IB_SEND_IP_CSUM		= (1<<4),
1195 
1196 	/* reserve bits 26-31 for low level drivers' internal use */
1197 	IB_SEND_RESERVED_START	= (1 << 26),
1198 	IB_SEND_RESERVED_END	= (1 << 31),
1199 };
1200 
1201 struct ib_sge {
1202 	u64	addr;
1203 	u32	length;
1204 	u32	lkey;
1205 };
1206 
1207 struct ib_cqe {
1208 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1209 };
1210 
1211 struct ib_send_wr {
1212 	struct ib_send_wr      *next;
1213 	union {
1214 		u64		wr_id;
1215 		struct ib_cqe	*wr_cqe;
1216 	};
1217 	struct ib_sge	       *sg_list;
1218 	int			num_sge;
1219 	enum ib_wr_opcode	opcode;
1220 	int			send_flags;
1221 	union {
1222 		__be32		imm_data;
1223 		u32		invalidate_rkey;
1224 	} ex;
1225 };
1226 
1227 struct ib_rdma_wr {
1228 	struct ib_send_wr	wr;
1229 	u64			remote_addr;
1230 	u32			rkey;
1231 };
1232 
1233 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1234 {
1235 	return container_of(wr, struct ib_rdma_wr, wr);
1236 }
1237 
1238 struct ib_atomic_wr {
1239 	struct ib_send_wr	wr;
1240 	u64			remote_addr;
1241 	u64			compare_add;
1242 	u64			swap;
1243 	u64			compare_add_mask;
1244 	u64			swap_mask;
1245 	u32			rkey;
1246 };
1247 
1248 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1249 {
1250 	return container_of(wr, struct ib_atomic_wr, wr);
1251 }
1252 
1253 struct ib_ud_wr {
1254 	struct ib_send_wr	wr;
1255 	struct ib_ah		*ah;
1256 	void			*header;
1257 	int			hlen;
1258 	int			mss;
1259 	u32			remote_qpn;
1260 	u32			remote_qkey;
1261 	u16			pkey_index; /* valid for GSI only */
1262 	u8			port_num;   /* valid for DR SMPs on switch only */
1263 };
1264 
1265 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1266 {
1267 	return container_of(wr, struct ib_ud_wr, wr);
1268 }
1269 
1270 struct ib_reg_wr {
1271 	struct ib_send_wr	wr;
1272 	struct ib_mr		*mr;
1273 	u32			key;
1274 	int			access;
1275 };
1276 
1277 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1278 {
1279 	return container_of(wr, struct ib_reg_wr, wr);
1280 }
1281 
1282 struct ib_sig_handover_wr {
1283 	struct ib_send_wr	wr;
1284 	struct ib_sig_attrs    *sig_attrs;
1285 	struct ib_mr	       *sig_mr;
1286 	int			access_flags;
1287 	struct ib_sge	       *prot;
1288 };
1289 
1290 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1291 {
1292 	return container_of(wr, struct ib_sig_handover_wr, wr);
1293 }
1294 
1295 struct ib_recv_wr {
1296 	struct ib_recv_wr      *next;
1297 	union {
1298 		u64		wr_id;
1299 		struct ib_cqe	*wr_cqe;
1300 	};
1301 	struct ib_sge	       *sg_list;
1302 	int			num_sge;
1303 };
1304 
1305 enum ib_access_flags {
1306 	IB_ACCESS_LOCAL_WRITE	= 1,
1307 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1308 	IB_ACCESS_REMOTE_READ	= (1<<2),
1309 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1310 	IB_ACCESS_MW_BIND	= (1<<4),
1311 	IB_ZERO_BASED		= (1<<5),
1312 	IB_ACCESS_ON_DEMAND     = (1<<6),
1313 };
1314 
1315 /*
1316  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1317  * are hidden here instead of a uapi header!
1318  */
1319 enum ib_mr_rereg_flags {
1320 	IB_MR_REREG_TRANS	= 1,
1321 	IB_MR_REREG_PD		= (1<<1),
1322 	IB_MR_REREG_ACCESS	= (1<<2),
1323 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1324 };
1325 
1326 struct ib_fmr_attr {
1327 	int	max_pages;
1328 	int	max_maps;
1329 	u8	page_shift;
1330 };
1331 
1332 struct ib_umem;
1333 
1334 struct ib_ucontext {
1335 	struct ib_device       *device;
1336 	struct list_head	pd_list;
1337 	struct list_head	mr_list;
1338 	struct list_head	mw_list;
1339 	struct list_head	cq_list;
1340 	struct list_head	qp_list;
1341 	struct list_head	srq_list;
1342 	struct list_head	ah_list;
1343 	struct list_head	xrcd_list;
1344 	struct list_head	rule_list;
1345 	struct list_head	wq_list;
1346 	struct list_head	rwq_ind_tbl_list;
1347 	int			closing;
1348 
1349 	struct pid             *tgid;
1350 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1351 	struct rb_root      umem_tree;
1352 	/*
1353 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1354 	 * mmu notifiers registration.
1355 	 */
1356 	struct rw_semaphore	umem_rwsem;
1357 	void (*invalidate_range)(struct ib_umem *umem,
1358 				 unsigned long start, unsigned long end);
1359 
1360 	struct mmu_notifier	mn;
1361 	atomic_t		notifier_count;
1362 	/* A list of umems that don't have private mmu notifier counters yet. */
1363 	struct list_head	no_private_counters;
1364 	int                     odp_mrs_count;
1365 #endif
1366 };
1367 
1368 struct ib_uobject {
1369 	u64			user_handle;	/* handle given to us by userspace */
1370 	struct ib_ucontext     *context;	/* associated user context */
1371 	void		       *object;		/* containing object */
1372 	struct list_head	list;		/* link to context's list */
1373 	int			id;		/* index into kernel idr */
1374 	struct kref		ref;
1375 	struct rw_semaphore	mutex;		/* protects .live */
1376 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1377 	int			live;
1378 };
1379 
1380 struct ib_udata {
1381 	const void __user *inbuf;
1382 	void __user *outbuf;
1383 	size_t       inlen;
1384 	size_t       outlen;
1385 };
1386 
1387 struct ib_pd {
1388 	u32			local_dma_lkey;
1389 	u32			flags;
1390 	struct ib_device       *device;
1391 	struct ib_uobject      *uobject;
1392 	atomic_t          	usecnt; /* count all resources */
1393 
1394 	u32			unsafe_global_rkey;
1395 
1396 	/*
1397 	 * Implementation details of the RDMA core, don't use in drivers:
1398 	 */
1399 	struct ib_mr	       *__internal_mr;
1400 };
1401 
1402 struct ib_xrcd {
1403 	struct ib_device       *device;
1404 	atomic_t		usecnt; /* count all exposed resources */
1405 	struct inode	       *inode;
1406 
1407 	struct mutex		tgt_qp_mutex;
1408 	struct list_head	tgt_qp_list;
1409 };
1410 
1411 struct ib_ah {
1412 	struct ib_device	*device;
1413 	struct ib_pd		*pd;
1414 	struct ib_uobject	*uobject;
1415 };
1416 
1417 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1418 
1419 enum ib_poll_context {
1420 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1421 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1422 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1423 };
1424 
1425 struct ib_cq {
1426 	struct ib_device       *device;
1427 	struct ib_uobject      *uobject;
1428 	ib_comp_handler   	comp_handler;
1429 	void                  (*event_handler)(struct ib_event *, void *);
1430 	void                   *cq_context;
1431 	int               	cqe;
1432 	atomic_t          	usecnt; /* count number of work queues */
1433 	enum ib_poll_context	poll_ctx;
1434 	struct ib_wc		*wc;
1435 	union {
1436 		struct irq_poll		iop;
1437 		struct work_struct	work;
1438 	};
1439 };
1440 
1441 struct ib_srq {
1442 	struct ib_device       *device;
1443 	struct ib_pd	       *pd;
1444 	struct ib_uobject      *uobject;
1445 	void		      (*event_handler)(struct ib_event *, void *);
1446 	void		       *srq_context;
1447 	enum ib_srq_type	srq_type;
1448 	atomic_t		usecnt;
1449 
1450 	union {
1451 		struct {
1452 			struct ib_xrcd *xrcd;
1453 			struct ib_cq   *cq;
1454 			u32		srq_num;
1455 		} xrc;
1456 	} ext;
1457 };
1458 
1459 enum ib_wq_type {
1460 	IB_WQT_RQ
1461 };
1462 
1463 enum ib_wq_state {
1464 	IB_WQS_RESET,
1465 	IB_WQS_RDY,
1466 	IB_WQS_ERR
1467 };
1468 
1469 struct ib_wq {
1470 	struct ib_device       *device;
1471 	struct ib_uobject      *uobject;
1472 	void		    *wq_context;
1473 	void		    (*event_handler)(struct ib_event *, void *);
1474 	struct ib_pd	       *pd;
1475 	struct ib_cq	       *cq;
1476 	u32		wq_num;
1477 	enum ib_wq_state       state;
1478 	enum ib_wq_type	wq_type;
1479 	atomic_t		usecnt;
1480 };
1481 
1482 struct ib_wq_init_attr {
1483 	void		       *wq_context;
1484 	enum ib_wq_type	wq_type;
1485 	u32		max_wr;
1486 	u32		max_sge;
1487 	struct	ib_cq	       *cq;
1488 	void		    (*event_handler)(struct ib_event *, void *);
1489 };
1490 
1491 enum ib_wq_attr_mask {
1492 	IB_WQ_STATE	= 1 << 0,
1493 	IB_WQ_CUR_STATE	= 1 << 1,
1494 };
1495 
1496 struct ib_wq_attr {
1497 	enum	ib_wq_state	wq_state;
1498 	enum	ib_wq_state	curr_wq_state;
1499 };
1500 
1501 struct ib_rwq_ind_table {
1502 	struct ib_device	*device;
1503 	struct ib_uobject      *uobject;
1504 	atomic_t		usecnt;
1505 	u32		ind_tbl_num;
1506 	u32		log_ind_tbl_size;
1507 	struct ib_wq	**ind_tbl;
1508 };
1509 
1510 struct ib_rwq_ind_table_init_attr {
1511 	u32		log_ind_tbl_size;
1512 	/* Each entry is a pointer to Receive Work Queue */
1513 	struct ib_wq	**ind_tbl;
1514 };
1515 
1516 /*
1517  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1518  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1519  */
1520 struct ib_qp {
1521 	struct ib_device       *device;
1522 	struct ib_pd	       *pd;
1523 	struct ib_cq	       *send_cq;
1524 	struct ib_cq	       *recv_cq;
1525 	spinlock_t		mr_lock;
1526 	int			mrs_used;
1527 	struct list_head	rdma_mrs;
1528 	struct list_head	sig_mrs;
1529 	struct ib_srq	       *srq;
1530 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1531 	struct list_head	xrcd_list;
1532 
1533 	/* count times opened, mcast attaches, flow attaches */
1534 	atomic_t		usecnt;
1535 	struct list_head	open_list;
1536 	struct ib_qp           *real_qp;
1537 	struct ib_uobject      *uobject;
1538 	void                  (*event_handler)(struct ib_event *, void *);
1539 	void		       *qp_context;
1540 	u32			qp_num;
1541 	u32			max_write_sge;
1542 	u32			max_read_sge;
1543 	enum ib_qp_type		qp_type;
1544 	struct ib_rwq_ind_table *rwq_ind_tbl;
1545 };
1546 
1547 struct ib_mr {
1548 	struct ib_device  *device;
1549 	struct ib_pd	  *pd;
1550 	u32		   lkey;
1551 	u32		   rkey;
1552 	u64		   iova;
1553 	u32		   length;
1554 	unsigned int	   page_size;
1555 	bool		   need_inval;
1556 	union {
1557 		struct ib_uobject	*uobject;	/* user */
1558 		struct list_head	qp_entry;	/* FR */
1559 	};
1560 };
1561 
1562 struct ib_mw {
1563 	struct ib_device	*device;
1564 	struct ib_pd		*pd;
1565 	struct ib_uobject	*uobject;
1566 	u32			rkey;
1567 	enum ib_mw_type         type;
1568 };
1569 
1570 struct ib_fmr {
1571 	struct ib_device	*device;
1572 	struct ib_pd		*pd;
1573 	struct list_head	list;
1574 	u32			lkey;
1575 	u32			rkey;
1576 };
1577 
1578 /* Supported steering options */
1579 enum ib_flow_attr_type {
1580 	/* steering according to rule specifications */
1581 	IB_FLOW_ATTR_NORMAL		= 0x0,
1582 	/* default unicast and multicast rule -
1583 	 * receive all Eth traffic which isn't steered to any QP
1584 	 */
1585 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1586 	/* default multicast rule -
1587 	 * receive all Eth multicast traffic which isn't steered to any QP
1588 	 */
1589 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1590 	/* sniffer rule - receive all port traffic */
1591 	IB_FLOW_ATTR_SNIFFER		= 0x3
1592 };
1593 
1594 /* Supported steering header types */
1595 enum ib_flow_spec_type {
1596 	/* L2 headers*/
1597 	IB_FLOW_SPEC_ETH		= 0x20,
1598 	IB_FLOW_SPEC_IB			= 0x22,
1599 	/* L3 header*/
1600 	IB_FLOW_SPEC_IPV4		= 0x30,
1601 	IB_FLOW_SPEC_IPV6		= 0x31,
1602 	/* L4 headers*/
1603 	IB_FLOW_SPEC_TCP		= 0x40,
1604 	IB_FLOW_SPEC_UDP		= 0x41,
1605 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1606 	IB_FLOW_SPEC_INNER		= 0x100,
1607 };
1608 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1609 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1610 
1611 /* Flow steering rule priority is set according to it's domain.
1612  * Lower domain value means higher priority.
1613  */
1614 enum ib_flow_domain {
1615 	IB_FLOW_DOMAIN_USER,
1616 	IB_FLOW_DOMAIN_ETHTOOL,
1617 	IB_FLOW_DOMAIN_RFS,
1618 	IB_FLOW_DOMAIN_NIC,
1619 	IB_FLOW_DOMAIN_NUM /* Must be last */
1620 };
1621 
1622 enum ib_flow_flags {
1623 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1624 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1625 };
1626 
1627 struct ib_flow_eth_filter {
1628 	u8	dst_mac[6];
1629 	u8	src_mac[6];
1630 	__be16	ether_type;
1631 	__be16	vlan_tag;
1632 	/* Must be last */
1633 	u8	real_sz[0];
1634 };
1635 
1636 struct ib_flow_spec_eth {
1637 	u32			  type;
1638 	u16			  size;
1639 	struct ib_flow_eth_filter val;
1640 	struct ib_flow_eth_filter mask;
1641 };
1642 
1643 struct ib_flow_ib_filter {
1644 	__be16 dlid;
1645 	__u8   sl;
1646 	/* Must be last */
1647 	u8	real_sz[0];
1648 };
1649 
1650 struct ib_flow_spec_ib {
1651 	u32			 type;
1652 	u16			 size;
1653 	struct ib_flow_ib_filter val;
1654 	struct ib_flow_ib_filter mask;
1655 };
1656 
1657 /* IPv4 header flags */
1658 enum ib_ipv4_flags {
1659 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1660 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1661 				    last have this flag set */
1662 };
1663 
1664 struct ib_flow_ipv4_filter {
1665 	__be32	src_ip;
1666 	__be32	dst_ip;
1667 	u8	proto;
1668 	u8	tos;
1669 	u8	ttl;
1670 	u8	flags;
1671 	/* Must be last */
1672 	u8	real_sz[0];
1673 };
1674 
1675 struct ib_flow_spec_ipv4 {
1676 	u32			   type;
1677 	u16			   size;
1678 	struct ib_flow_ipv4_filter val;
1679 	struct ib_flow_ipv4_filter mask;
1680 };
1681 
1682 struct ib_flow_ipv6_filter {
1683 	u8	src_ip[16];
1684 	u8	dst_ip[16];
1685 	__be32	flow_label;
1686 	u8	next_hdr;
1687 	u8	traffic_class;
1688 	u8	hop_limit;
1689 	/* Must be last */
1690 	u8	real_sz[0];
1691 };
1692 
1693 struct ib_flow_spec_ipv6 {
1694 	u32			   type;
1695 	u16			   size;
1696 	struct ib_flow_ipv6_filter val;
1697 	struct ib_flow_ipv6_filter mask;
1698 };
1699 
1700 struct ib_flow_tcp_udp_filter {
1701 	__be16	dst_port;
1702 	__be16	src_port;
1703 	/* Must be last */
1704 	u8	real_sz[0];
1705 };
1706 
1707 struct ib_flow_spec_tcp_udp {
1708 	u32			      type;
1709 	u16			      size;
1710 	struct ib_flow_tcp_udp_filter val;
1711 	struct ib_flow_tcp_udp_filter mask;
1712 };
1713 
1714 struct ib_flow_tunnel_filter {
1715 	__be32	tunnel_id;
1716 	u8	real_sz[0];
1717 };
1718 
1719 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1720  * the tunnel_id from val has the vni value
1721  */
1722 struct ib_flow_spec_tunnel {
1723 	u32			      type;
1724 	u16			      size;
1725 	struct ib_flow_tunnel_filter  val;
1726 	struct ib_flow_tunnel_filter  mask;
1727 };
1728 
1729 union ib_flow_spec {
1730 	struct {
1731 		u32			type;
1732 		u16			size;
1733 	};
1734 	struct ib_flow_spec_eth		eth;
1735 	struct ib_flow_spec_ib		ib;
1736 	struct ib_flow_spec_ipv4        ipv4;
1737 	struct ib_flow_spec_tcp_udp	tcp_udp;
1738 	struct ib_flow_spec_ipv6        ipv6;
1739 	struct ib_flow_spec_tunnel      tunnel;
1740 };
1741 
1742 struct ib_flow_attr {
1743 	enum ib_flow_attr_type type;
1744 	u16	     size;
1745 	u16	     priority;
1746 	u32	     flags;
1747 	u8	     num_of_specs;
1748 	u8	     port;
1749 	/* Following are the optional layers according to user request
1750 	 * struct ib_flow_spec_xxx
1751 	 * struct ib_flow_spec_yyy
1752 	 */
1753 };
1754 
1755 struct ib_flow {
1756 	struct ib_qp		*qp;
1757 	struct ib_uobject	*uobject;
1758 };
1759 
1760 struct ib_mad_hdr;
1761 struct ib_grh;
1762 
1763 enum ib_process_mad_flags {
1764 	IB_MAD_IGNORE_MKEY	= 1,
1765 	IB_MAD_IGNORE_BKEY	= 2,
1766 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1767 };
1768 
1769 enum ib_mad_result {
1770 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1771 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1772 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1773 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1774 };
1775 
1776 #define IB_DEVICE_NAME_MAX 64
1777 
1778 struct ib_cache {
1779 	rwlock_t                lock;
1780 	struct ib_event_handler event_handler;
1781 	struct ib_pkey_cache  **pkey_cache;
1782 	struct ib_gid_table   **gid_cache;
1783 	u8                     *lmc_cache;
1784 	enum ib_port_state     *port_state_cache;
1785 };
1786 
1787 struct iw_cm_verbs;
1788 
1789 struct ib_port_immutable {
1790 	int                           pkey_tbl_len;
1791 	int                           gid_tbl_len;
1792 	u32                           core_cap_flags;
1793 	u32                           max_mad_size;
1794 };
1795 
1796 struct ib_device {
1797 	char                          name[IB_DEVICE_NAME_MAX];
1798 
1799 	struct list_head              event_handler_list;
1800 	spinlock_t                    event_handler_lock;
1801 
1802 	spinlock_t                    client_data_lock;
1803 	struct list_head              core_list;
1804 	/* Access to the client_data_list is protected by the client_data_lock
1805 	 * spinlock and the lists_rwsem read-write semaphore */
1806 	struct list_head              client_data_list;
1807 
1808 	struct ib_cache               cache;
1809 	/**
1810 	 * port_immutable is indexed by port number
1811 	 */
1812 	struct ib_port_immutable     *port_immutable;
1813 
1814 	int			      num_comp_vectors;
1815 
1816 	struct iw_cm_verbs	     *iwcm;
1817 
1818 	/**
1819 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1820 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
1821 	 *   core when the device is removed.  A lifespan of -1 in the return
1822 	 *   struct tells the core to set a default lifespan.
1823 	 */
1824 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
1825 						     u8 port_num);
1826 	/**
1827 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1828 	 * @index - The index in the value array we wish to have updated, or
1829 	 *   num_counters if we want all stats updated
1830 	 * Return codes -
1831 	 *   < 0 - Error, no counters updated
1832 	 *   index - Updated the single counter pointed to by index
1833 	 *   num_counters - Updated all counters (will reset the timestamp
1834 	 *     and prevent further calls for lifespan milliseconds)
1835 	 * Drivers are allowed to update all counters in leiu of just the
1836 	 *   one given in index at their option
1837 	 */
1838 	int		           (*get_hw_stats)(struct ib_device *device,
1839 						   struct rdma_hw_stats *stats,
1840 						   u8 port, int index);
1841 	int		           (*query_device)(struct ib_device *device,
1842 						   struct ib_device_attr *device_attr,
1843 						   struct ib_udata *udata);
1844 	int		           (*query_port)(struct ib_device *device,
1845 						 u8 port_num,
1846 						 struct ib_port_attr *port_attr);
1847 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1848 						     u8 port_num);
1849 	/* When calling get_netdev, the HW vendor's driver should return the
1850 	 * net device of device @device at port @port_num or NULL if such
1851 	 * a net device doesn't exist. The vendor driver should call dev_hold
1852 	 * on this net device. The HW vendor's device driver must guarantee
1853 	 * that this function returns NULL before the net device reaches
1854 	 * NETDEV_UNREGISTER_FINAL state.
1855 	 */
1856 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1857 						 u8 port_num);
1858 	int		           (*query_gid)(struct ib_device *device,
1859 						u8 port_num, int index,
1860 						union ib_gid *gid);
1861 	/* When calling add_gid, the HW vendor's driver should
1862 	 * add the gid of device @device at gid index @index of
1863 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1864 	 * the network device related to this gid is available
1865 	 * at @attr. @context allows the HW vendor driver to store extra
1866 	 * information together with a GID entry. The HW vendor may allocate
1867 	 * memory to contain this information and store it in @context when a
1868 	 * new GID entry is written to. Params are consistent until the next
1869 	 * call of add_gid or delete_gid. The function should return 0 on
1870 	 * success or error otherwise. The function could be called
1871 	 * concurrently for different ports. This function is only called
1872 	 * when roce_gid_table is used.
1873 	 */
1874 	int		           (*add_gid)(struct ib_device *device,
1875 					      u8 port_num,
1876 					      unsigned int index,
1877 					      const union ib_gid *gid,
1878 					      const struct ib_gid_attr *attr,
1879 					      void **context);
1880 	/* When calling del_gid, the HW vendor's driver should delete the
1881 	 * gid of device @device at gid index @index of port @port_num.
1882 	 * Upon the deletion of a GID entry, the HW vendor must free any
1883 	 * allocated memory. The caller will clear @context afterwards.
1884 	 * This function is only called when roce_gid_table is used.
1885 	 */
1886 	int		           (*del_gid)(struct ib_device *device,
1887 					      u8 port_num,
1888 					      unsigned int index,
1889 					      void **context);
1890 	int		           (*query_pkey)(struct ib_device *device,
1891 						 u8 port_num, u16 index, u16 *pkey);
1892 	int		           (*modify_device)(struct ib_device *device,
1893 						    int device_modify_mask,
1894 						    struct ib_device_modify *device_modify);
1895 	int		           (*modify_port)(struct ib_device *device,
1896 						  u8 port_num, int port_modify_mask,
1897 						  struct ib_port_modify *port_modify);
1898 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1899 						     struct ib_udata *udata);
1900 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1901 	int                        (*mmap)(struct ib_ucontext *context,
1902 					   struct vm_area_struct *vma);
1903 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1904 					       struct ib_ucontext *context,
1905 					       struct ib_udata *udata);
1906 	int                        (*dealloc_pd)(struct ib_pd *pd);
1907 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1908 						struct ib_ah_attr *ah_attr,
1909 						struct ib_udata *udata);
1910 	int                        (*modify_ah)(struct ib_ah *ah,
1911 						struct ib_ah_attr *ah_attr);
1912 	int                        (*query_ah)(struct ib_ah *ah,
1913 					       struct ib_ah_attr *ah_attr);
1914 	int                        (*destroy_ah)(struct ib_ah *ah);
1915 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1916 						 struct ib_srq_init_attr *srq_init_attr,
1917 						 struct ib_udata *udata);
1918 	int                        (*modify_srq)(struct ib_srq *srq,
1919 						 struct ib_srq_attr *srq_attr,
1920 						 enum ib_srq_attr_mask srq_attr_mask,
1921 						 struct ib_udata *udata);
1922 	int                        (*query_srq)(struct ib_srq *srq,
1923 						struct ib_srq_attr *srq_attr);
1924 	int                        (*destroy_srq)(struct ib_srq *srq);
1925 	int                        (*post_srq_recv)(struct ib_srq *srq,
1926 						    struct ib_recv_wr *recv_wr,
1927 						    struct ib_recv_wr **bad_recv_wr);
1928 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1929 						struct ib_qp_init_attr *qp_init_attr,
1930 						struct ib_udata *udata);
1931 	int                        (*modify_qp)(struct ib_qp *qp,
1932 						struct ib_qp_attr *qp_attr,
1933 						int qp_attr_mask,
1934 						struct ib_udata *udata);
1935 	int                        (*query_qp)(struct ib_qp *qp,
1936 					       struct ib_qp_attr *qp_attr,
1937 					       int qp_attr_mask,
1938 					       struct ib_qp_init_attr *qp_init_attr);
1939 	int                        (*destroy_qp)(struct ib_qp *qp);
1940 	int                        (*post_send)(struct ib_qp *qp,
1941 						struct ib_send_wr *send_wr,
1942 						struct ib_send_wr **bad_send_wr);
1943 	int                        (*post_recv)(struct ib_qp *qp,
1944 						struct ib_recv_wr *recv_wr,
1945 						struct ib_recv_wr **bad_recv_wr);
1946 	struct ib_cq *             (*create_cq)(struct ib_device *device,
1947 						const struct ib_cq_init_attr *attr,
1948 						struct ib_ucontext *context,
1949 						struct ib_udata *udata);
1950 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1951 						u16 cq_period);
1952 	int                        (*destroy_cq)(struct ib_cq *cq);
1953 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1954 						struct ib_udata *udata);
1955 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1956 					      struct ib_wc *wc);
1957 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1958 	int                        (*req_notify_cq)(struct ib_cq *cq,
1959 						    enum ib_cq_notify_flags flags);
1960 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1961 						      int wc_cnt);
1962 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1963 						 int mr_access_flags);
1964 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1965 						  u64 start, u64 length,
1966 						  u64 virt_addr,
1967 						  int mr_access_flags,
1968 						  struct ib_udata *udata);
1969 	int			   (*rereg_user_mr)(struct ib_mr *mr,
1970 						    int flags,
1971 						    u64 start, u64 length,
1972 						    u64 virt_addr,
1973 						    int mr_access_flags,
1974 						    struct ib_pd *pd,
1975 						    struct ib_udata *udata);
1976 	int                        (*dereg_mr)(struct ib_mr *mr);
1977 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
1978 					       enum ib_mr_type mr_type,
1979 					       u32 max_num_sg);
1980 	int                        (*map_mr_sg)(struct ib_mr *mr,
1981 						struct scatterlist *sg,
1982 						int sg_nents,
1983 						unsigned int *sg_offset);
1984 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
1985 					       enum ib_mw_type type,
1986 					       struct ib_udata *udata);
1987 	int                        (*dealloc_mw)(struct ib_mw *mw);
1988 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1989 						int mr_access_flags,
1990 						struct ib_fmr_attr *fmr_attr);
1991 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1992 						   u64 *page_list, int list_len,
1993 						   u64 iova);
1994 	int		           (*unmap_fmr)(struct list_head *fmr_list);
1995 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1996 	int                        (*attach_mcast)(struct ib_qp *qp,
1997 						   union ib_gid *gid,
1998 						   u16 lid);
1999 	int                        (*detach_mcast)(struct ib_qp *qp,
2000 						   union ib_gid *gid,
2001 						   u16 lid);
2002 	int                        (*process_mad)(struct ib_device *device,
2003 						  int process_mad_flags,
2004 						  u8 port_num,
2005 						  const struct ib_wc *in_wc,
2006 						  const struct ib_grh *in_grh,
2007 						  const struct ib_mad_hdr *in_mad,
2008 						  size_t in_mad_size,
2009 						  struct ib_mad_hdr *out_mad,
2010 						  size_t *out_mad_size,
2011 						  u16 *out_mad_pkey_index);
2012 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2013 						 struct ib_ucontext *ucontext,
2014 						 struct ib_udata *udata);
2015 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2016 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2017 						  struct ib_flow_attr
2018 						  *flow_attr,
2019 						  int domain);
2020 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2021 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2022 						      struct ib_mr_status *mr_status);
2023 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2024 	void			   (*drain_rq)(struct ib_qp *qp);
2025 	void			   (*drain_sq)(struct ib_qp *qp);
2026 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2027 							int state);
2028 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2029 						   struct ifla_vf_info *ivf);
2030 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2031 						   struct ifla_vf_stats *stats);
2032 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2033 						  int type);
2034 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2035 						struct ib_wq_init_attr *init_attr,
2036 						struct ib_udata *udata);
2037 	int			   (*destroy_wq)(struct ib_wq *wq);
2038 	int			   (*modify_wq)(struct ib_wq *wq,
2039 						struct ib_wq_attr *attr,
2040 						u32 wq_attr_mask,
2041 						struct ib_udata *udata);
2042 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2043 							   struct ib_rwq_ind_table_init_attr *init_attr,
2044 							   struct ib_udata *udata);
2045 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2046 
2047 	struct module               *owner;
2048 	struct device                dev;
2049 	struct kobject               *ports_parent;
2050 	struct list_head             port_list;
2051 
2052 	enum {
2053 		IB_DEV_UNINITIALIZED,
2054 		IB_DEV_REGISTERED,
2055 		IB_DEV_UNREGISTERED
2056 	}                            reg_state;
2057 
2058 	int			     uverbs_abi_ver;
2059 	u64			     uverbs_cmd_mask;
2060 	u64			     uverbs_ex_cmd_mask;
2061 
2062 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2063 	__be64			     node_guid;
2064 	u32			     local_dma_lkey;
2065 	u16                          is_switch:1;
2066 	u8                           node_type;
2067 	u8                           phys_port_cnt;
2068 	struct ib_device_attr        attrs;
2069 	struct attribute_group	     *hw_stats_ag;
2070 	struct rdma_hw_stats         *hw_stats;
2071 
2072 	/**
2073 	 * The following mandatory functions are used only at device
2074 	 * registration.  Keep functions such as these at the end of this
2075 	 * structure to avoid cache line misses when accessing struct ib_device
2076 	 * in fast paths.
2077 	 */
2078 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2079 	void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2080 };
2081 
2082 struct ib_client {
2083 	char  *name;
2084 	void (*add)   (struct ib_device *);
2085 	void (*remove)(struct ib_device *, void *client_data);
2086 
2087 	/* Returns the net_dev belonging to this ib_client and matching the
2088 	 * given parameters.
2089 	 * @dev:	 An RDMA device that the net_dev use for communication.
2090 	 * @port:	 A physical port number on the RDMA device.
2091 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2092 	 * @gid:	 A GID that the net_dev uses to communicate.
2093 	 * @addr:	 An IP address the net_dev is configured with.
2094 	 * @client_data: The device's client data set by ib_set_client_data().
2095 	 *
2096 	 * An ib_client that implements a net_dev on top of RDMA devices
2097 	 * (such as IP over IB) should implement this callback, allowing the
2098 	 * rdma_cm module to find the right net_dev for a given request.
2099 	 *
2100 	 * The caller is responsible for calling dev_put on the returned
2101 	 * netdev. */
2102 	struct net_device *(*get_net_dev_by_params)(
2103 			struct ib_device *dev,
2104 			u8 port,
2105 			u16 pkey,
2106 			const union ib_gid *gid,
2107 			const struct sockaddr *addr,
2108 			void *client_data);
2109 	struct list_head list;
2110 };
2111 
2112 struct ib_device *ib_alloc_device(size_t size);
2113 void ib_dealloc_device(struct ib_device *device);
2114 
2115 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2116 
2117 int ib_register_device(struct ib_device *device,
2118 		       int (*port_callback)(struct ib_device *,
2119 					    u8, struct kobject *));
2120 void ib_unregister_device(struct ib_device *device);
2121 
2122 int ib_register_client   (struct ib_client *client);
2123 void ib_unregister_client(struct ib_client *client);
2124 
2125 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2126 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2127 			 void *data);
2128 
2129 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2130 {
2131 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2132 }
2133 
2134 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2135 {
2136 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2137 }
2138 
2139 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2140 				       size_t offset,
2141 				       size_t len)
2142 {
2143 	const void __user *p = udata->inbuf + offset;
2144 	bool ret;
2145 	u8 *buf;
2146 
2147 	if (len > USHRT_MAX)
2148 		return false;
2149 
2150 	buf = memdup_user(p, len);
2151 	if (IS_ERR(buf))
2152 		return false;
2153 
2154 	ret = !memchr_inv(buf, 0, len);
2155 	kfree(buf);
2156 	return ret;
2157 }
2158 
2159 /**
2160  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2161  * contains all required attributes and no attributes not allowed for
2162  * the given QP state transition.
2163  * @cur_state: Current QP state
2164  * @next_state: Next QP state
2165  * @type: QP type
2166  * @mask: Mask of supplied QP attributes
2167  * @ll : link layer of port
2168  *
2169  * This function is a helper function that a low-level driver's
2170  * modify_qp method can use to validate the consumer's input.  It
2171  * checks that cur_state and next_state are valid QP states, that a
2172  * transition from cur_state to next_state is allowed by the IB spec,
2173  * and that the attribute mask supplied is allowed for the transition.
2174  */
2175 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2176 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2177 		       enum rdma_link_layer ll);
2178 
2179 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2180 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2181 void ib_dispatch_event(struct ib_event *event);
2182 
2183 int ib_query_port(struct ib_device *device,
2184 		  u8 port_num, struct ib_port_attr *port_attr);
2185 
2186 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2187 					       u8 port_num);
2188 
2189 /**
2190  * rdma_cap_ib_switch - Check if the device is IB switch
2191  * @device: Device to check
2192  *
2193  * Device driver is responsible for setting is_switch bit on
2194  * in ib_device structure at init time.
2195  *
2196  * Return: true if the device is IB switch.
2197  */
2198 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2199 {
2200 	return device->is_switch;
2201 }
2202 
2203 /**
2204  * rdma_start_port - Return the first valid port number for the device
2205  * specified
2206  *
2207  * @device: Device to be checked
2208  *
2209  * Return start port number
2210  */
2211 static inline u8 rdma_start_port(const struct ib_device *device)
2212 {
2213 	return rdma_cap_ib_switch(device) ? 0 : 1;
2214 }
2215 
2216 /**
2217  * rdma_end_port - Return the last valid port number for the device
2218  * specified
2219  *
2220  * @device: Device to be checked
2221  *
2222  * Return last port number
2223  */
2224 static inline u8 rdma_end_port(const struct ib_device *device)
2225 {
2226 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2227 }
2228 
2229 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2230 {
2231 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2232 }
2233 
2234 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2235 {
2236 	return device->port_immutable[port_num].core_cap_flags &
2237 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2238 }
2239 
2240 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2241 {
2242 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2243 }
2244 
2245 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2246 {
2247 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2248 }
2249 
2250 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2251 {
2252 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2253 }
2254 
2255 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2256 {
2257 	return rdma_protocol_ib(device, port_num) ||
2258 		rdma_protocol_roce(device, port_num);
2259 }
2260 
2261 /**
2262  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2263  * Management Datagrams.
2264  * @device: Device to check
2265  * @port_num: Port number to check
2266  *
2267  * Management Datagrams (MAD) are a required part of the InfiniBand
2268  * specification and are supported on all InfiniBand devices.  A slightly
2269  * extended version are also supported on OPA interfaces.
2270  *
2271  * Return: true if the port supports sending/receiving of MAD packets.
2272  */
2273 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2274 {
2275 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2276 }
2277 
2278 /**
2279  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2280  * Management Datagrams.
2281  * @device: Device to check
2282  * @port_num: Port number to check
2283  *
2284  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2285  * datagrams with their own versions.  These OPA MADs share many but not all of
2286  * the characteristics of InfiniBand MADs.
2287  *
2288  * OPA MADs differ in the following ways:
2289  *
2290  *    1) MADs are variable size up to 2K
2291  *       IBTA defined MADs remain fixed at 256 bytes
2292  *    2) OPA SMPs must carry valid PKeys
2293  *    3) OPA SMP packets are a different format
2294  *
2295  * Return: true if the port supports OPA MAD packet formats.
2296  */
2297 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2298 {
2299 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2300 		== RDMA_CORE_CAP_OPA_MAD;
2301 }
2302 
2303 /**
2304  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2305  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2306  * @device: Device to check
2307  * @port_num: Port number to check
2308  *
2309  * Each InfiniBand node is required to provide a Subnet Management Agent
2310  * that the subnet manager can access.  Prior to the fabric being fully
2311  * configured by the subnet manager, the SMA is accessed via a well known
2312  * interface called the Subnet Management Interface (SMI).  This interface
2313  * uses directed route packets to communicate with the SM to get around the
2314  * chicken and egg problem of the SM needing to know what's on the fabric
2315  * in order to configure the fabric, and needing to configure the fabric in
2316  * order to send packets to the devices on the fabric.  These directed
2317  * route packets do not need the fabric fully configured in order to reach
2318  * their destination.  The SMI is the only method allowed to send
2319  * directed route packets on an InfiniBand fabric.
2320  *
2321  * Return: true if the port provides an SMI.
2322  */
2323 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2324 {
2325 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2326 }
2327 
2328 /**
2329  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2330  * Communication Manager.
2331  * @device: Device to check
2332  * @port_num: Port number to check
2333  *
2334  * The InfiniBand Communication Manager is one of many pre-defined General
2335  * Service Agents (GSA) that are accessed via the General Service
2336  * Interface (GSI).  It's role is to facilitate establishment of connections
2337  * between nodes as well as other management related tasks for established
2338  * connections.
2339  *
2340  * Return: true if the port supports an IB CM (this does not guarantee that
2341  * a CM is actually running however).
2342  */
2343 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2344 {
2345 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2346 }
2347 
2348 /**
2349  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2350  * Communication Manager.
2351  * @device: Device to check
2352  * @port_num: Port number to check
2353  *
2354  * Similar to above, but specific to iWARP connections which have a different
2355  * managment protocol than InfiniBand.
2356  *
2357  * Return: true if the port supports an iWARP CM (this does not guarantee that
2358  * a CM is actually running however).
2359  */
2360 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2361 {
2362 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2363 }
2364 
2365 /**
2366  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2367  * Subnet Administration.
2368  * @device: Device to check
2369  * @port_num: Port number to check
2370  *
2371  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2372  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2373  * fabrics, devices should resolve routes to other hosts by contacting the
2374  * SA to query the proper route.
2375  *
2376  * Return: true if the port should act as a client to the fabric Subnet
2377  * Administration interface.  This does not imply that the SA service is
2378  * running locally.
2379  */
2380 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2381 {
2382 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2383 }
2384 
2385 /**
2386  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2387  * Multicast.
2388  * @device: Device to check
2389  * @port_num: Port number to check
2390  *
2391  * InfiniBand multicast registration is more complex than normal IPv4 or
2392  * IPv6 multicast registration.  Each Host Channel Adapter must register
2393  * with the Subnet Manager when it wishes to join a multicast group.  It
2394  * should do so only once regardless of how many queue pairs it subscribes
2395  * to this group.  And it should leave the group only after all queue pairs
2396  * attached to the group have been detached.
2397  *
2398  * Return: true if the port must undertake the additional adminstrative
2399  * overhead of registering/unregistering with the SM and tracking of the
2400  * total number of queue pairs attached to the multicast group.
2401  */
2402 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2403 {
2404 	return rdma_cap_ib_sa(device, port_num);
2405 }
2406 
2407 /**
2408  * rdma_cap_af_ib - Check if the port of device has the capability
2409  * Native Infiniband Address.
2410  * @device: Device to check
2411  * @port_num: Port number to check
2412  *
2413  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2414  * GID.  RoCE uses a different mechanism, but still generates a GID via
2415  * a prescribed mechanism and port specific data.
2416  *
2417  * Return: true if the port uses a GID address to identify devices on the
2418  * network.
2419  */
2420 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2421 {
2422 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2423 }
2424 
2425 /**
2426  * rdma_cap_eth_ah - Check if the port of device has the capability
2427  * Ethernet Address Handle.
2428  * @device: Device to check
2429  * @port_num: Port number to check
2430  *
2431  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2432  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2433  * port.  Normally, packet headers are generated by the sending host
2434  * adapter, but when sending connectionless datagrams, we must manually
2435  * inject the proper headers for the fabric we are communicating over.
2436  *
2437  * Return: true if we are running as a RoCE port and must force the
2438  * addition of a Global Route Header built from our Ethernet Address
2439  * Handle into our header list for connectionless packets.
2440  */
2441 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2442 {
2443 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2444 }
2445 
2446 /**
2447  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2448  *
2449  * @device: Device
2450  * @port_num: Port number
2451  *
2452  * This MAD size includes the MAD headers and MAD payload.  No other headers
2453  * are included.
2454  *
2455  * Return the max MAD size required by the Port.  Will return 0 if the port
2456  * does not support MADs
2457  */
2458 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2459 {
2460 	return device->port_immutable[port_num].max_mad_size;
2461 }
2462 
2463 /**
2464  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2465  * @device: Device to check
2466  * @port_num: Port number to check
2467  *
2468  * RoCE GID table mechanism manages the various GIDs for a device.
2469  *
2470  * NOTE: if allocating the port's GID table has failed, this call will still
2471  * return true, but any RoCE GID table API will fail.
2472  *
2473  * Return: true if the port uses RoCE GID table mechanism in order to manage
2474  * its GIDs.
2475  */
2476 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2477 					   u8 port_num)
2478 {
2479 	return rdma_protocol_roce(device, port_num) &&
2480 		device->add_gid && device->del_gid;
2481 }
2482 
2483 /*
2484  * Check if the device supports READ W/ INVALIDATE.
2485  */
2486 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2487 {
2488 	/*
2489 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2490 	 * has support for it yet.
2491 	 */
2492 	return rdma_protocol_iwarp(dev, port_num);
2493 }
2494 
2495 int ib_query_gid(struct ib_device *device,
2496 		 u8 port_num, int index, union ib_gid *gid,
2497 		 struct ib_gid_attr *attr);
2498 
2499 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2500 			 int state);
2501 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2502 		     struct ifla_vf_info *info);
2503 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2504 		    struct ifla_vf_stats *stats);
2505 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2506 		   int type);
2507 
2508 int ib_query_pkey(struct ib_device *device,
2509 		  u8 port_num, u16 index, u16 *pkey);
2510 
2511 int ib_modify_device(struct ib_device *device,
2512 		     int device_modify_mask,
2513 		     struct ib_device_modify *device_modify);
2514 
2515 int ib_modify_port(struct ib_device *device,
2516 		   u8 port_num, int port_modify_mask,
2517 		   struct ib_port_modify *port_modify);
2518 
2519 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2520 		enum ib_gid_type gid_type, struct net_device *ndev,
2521 		u8 *port_num, u16 *index);
2522 
2523 int ib_find_pkey(struct ib_device *device,
2524 		 u8 port_num, u16 pkey, u16 *index);
2525 
2526 enum ib_pd_flags {
2527 	/*
2528 	 * Create a memory registration for all memory in the system and place
2529 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2530 	 * ULPs to avoid the overhead of dynamic MRs.
2531 	 *
2532 	 * This flag is generally considered unsafe and must only be used in
2533 	 * extremly trusted environments.  Every use of it will log a warning
2534 	 * in the kernel log.
2535 	 */
2536 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2537 };
2538 
2539 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2540 		const char *caller);
2541 #define ib_alloc_pd(device, flags) \
2542 	__ib_alloc_pd((device), (flags), __func__)
2543 void ib_dealloc_pd(struct ib_pd *pd);
2544 
2545 /**
2546  * ib_create_ah - Creates an address handle for the given address vector.
2547  * @pd: The protection domain associated with the address handle.
2548  * @ah_attr: The attributes of the address vector.
2549  *
2550  * The address handle is used to reference a local or global destination
2551  * in all UD QP post sends.
2552  */
2553 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2554 
2555 /**
2556  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2557  *   work completion.
2558  * @hdr: the L3 header to parse
2559  * @net_type: type of header to parse
2560  * @sgid: place to store source gid
2561  * @dgid: place to store destination gid
2562  */
2563 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2564 			      enum rdma_network_type net_type,
2565 			      union ib_gid *sgid, union ib_gid *dgid);
2566 
2567 /**
2568  * ib_get_rdma_header_version - Get the header version
2569  * @hdr: the L3 header to parse
2570  */
2571 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2572 
2573 /**
2574  * ib_init_ah_from_wc - Initializes address handle attributes from a
2575  *   work completion.
2576  * @device: Device on which the received message arrived.
2577  * @port_num: Port on which the received message arrived.
2578  * @wc: Work completion associated with the received message.
2579  * @grh: References the received global route header.  This parameter is
2580  *   ignored unless the work completion indicates that the GRH is valid.
2581  * @ah_attr: Returned attributes that can be used when creating an address
2582  *   handle for replying to the message.
2583  */
2584 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2585 		       const struct ib_wc *wc, const struct ib_grh *grh,
2586 		       struct ib_ah_attr *ah_attr);
2587 
2588 /**
2589  * ib_create_ah_from_wc - Creates an address handle associated with the
2590  *   sender of the specified work completion.
2591  * @pd: The protection domain associated with the address handle.
2592  * @wc: Work completion information associated with a received message.
2593  * @grh: References the received global route header.  This parameter is
2594  *   ignored unless the work completion indicates that the GRH is valid.
2595  * @port_num: The outbound port number to associate with the address.
2596  *
2597  * The address handle is used to reference a local or global destination
2598  * in all UD QP post sends.
2599  */
2600 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2601 				   const struct ib_grh *grh, u8 port_num);
2602 
2603 /**
2604  * ib_modify_ah - Modifies the address vector associated with an address
2605  *   handle.
2606  * @ah: The address handle to modify.
2607  * @ah_attr: The new address vector attributes to associate with the
2608  *   address handle.
2609  */
2610 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2611 
2612 /**
2613  * ib_query_ah - Queries the address vector associated with an address
2614  *   handle.
2615  * @ah: The address handle to query.
2616  * @ah_attr: The address vector attributes associated with the address
2617  *   handle.
2618  */
2619 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2620 
2621 /**
2622  * ib_destroy_ah - Destroys an address handle.
2623  * @ah: The address handle to destroy.
2624  */
2625 int ib_destroy_ah(struct ib_ah *ah);
2626 
2627 /**
2628  * ib_create_srq - Creates a SRQ associated with the specified protection
2629  *   domain.
2630  * @pd: The protection domain associated with the SRQ.
2631  * @srq_init_attr: A list of initial attributes required to create the
2632  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2633  *   the actual capabilities of the created SRQ.
2634  *
2635  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2636  * requested size of the SRQ, and set to the actual values allocated
2637  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2638  * will always be at least as large as the requested values.
2639  */
2640 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2641 			     struct ib_srq_init_attr *srq_init_attr);
2642 
2643 /**
2644  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2645  * @srq: The SRQ to modify.
2646  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2647  *   the current values of selected SRQ attributes are returned.
2648  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2649  *   are being modified.
2650  *
2651  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2652  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2653  * the number of receives queued drops below the limit.
2654  */
2655 int ib_modify_srq(struct ib_srq *srq,
2656 		  struct ib_srq_attr *srq_attr,
2657 		  enum ib_srq_attr_mask srq_attr_mask);
2658 
2659 /**
2660  * ib_query_srq - Returns the attribute list and current values for the
2661  *   specified SRQ.
2662  * @srq: The SRQ to query.
2663  * @srq_attr: The attributes of the specified SRQ.
2664  */
2665 int ib_query_srq(struct ib_srq *srq,
2666 		 struct ib_srq_attr *srq_attr);
2667 
2668 /**
2669  * ib_destroy_srq - Destroys the specified SRQ.
2670  * @srq: The SRQ to destroy.
2671  */
2672 int ib_destroy_srq(struct ib_srq *srq);
2673 
2674 /**
2675  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2676  * @srq: The SRQ to post the work request on.
2677  * @recv_wr: A list of work requests to post on the receive queue.
2678  * @bad_recv_wr: On an immediate failure, this parameter will reference
2679  *   the work request that failed to be posted on the QP.
2680  */
2681 static inline int ib_post_srq_recv(struct ib_srq *srq,
2682 				   struct ib_recv_wr *recv_wr,
2683 				   struct ib_recv_wr **bad_recv_wr)
2684 {
2685 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2686 }
2687 
2688 /**
2689  * ib_create_qp - Creates a QP associated with the specified protection
2690  *   domain.
2691  * @pd: The protection domain associated with the QP.
2692  * @qp_init_attr: A list of initial attributes required to create the
2693  *   QP.  If QP creation succeeds, then the attributes are updated to
2694  *   the actual capabilities of the created QP.
2695  */
2696 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2697 			   struct ib_qp_init_attr *qp_init_attr);
2698 
2699 /**
2700  * ib_modify_qp - Modifies the attributes for the specified QP and then
2701  *   transitions the QP to the given state.
2702  * @qp: The QP to modify.
2703  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2704  *   the current values of selected QP attributes are returned.
2705  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2706  *   are being modified.
2707  */
2708 int ib_modify_qp(struct ib_qp *qp,
2709 		 struct ib_qp_attr *qp_attr,
2710 		 int qp_attr_mask);
2711 
2712 /**
2713  * ib_query_qp - Returns the attribute list and current values for the
2714  *   specified QP.
2715  * @qp: The QP to query.
2716  * @qp_attr: The attributes of the specified QP.
2717  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2718  * @qp_init_attr: Additional attributes of the selected QP.
2719  *
2720  * The qp_attr_mask may be used to limit the query to gathering only the
2721  * selected attributes.
2722  */
2723 int ib_query_qp(struct ib_qp *qp,
2724 		struct ib_qp_attr *qp_attr,
2725 		int qp_attr_mask,
2726 		struct ib_qp_init_attr *qp_init_attr);
2727 
2728 /**
2729  * ib_destroy_qp - Destroys the specified QP.
2730  * @qp: The QP to destroy.
2731  */
2732 int ib_destroy_qp(struct ib_qp *qp);
2733 
2734 /**
2735  * ib_open_qp - Obtain a reference to an existing sharable QP.
2736  * @xrcd - XRC domain
2737  * @qp_open_attr: Attributes identifying the QP to open.
2738  *
2739  * Returns a reference to a sharable QP.
2740  */
2741 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2742 			 struct ib_qp_open_attr *qp_open_attr);
2743 
2744 /**
2745  * ib_close_qp - Release an external reference to a QP.
2746  * @qp: The QP handle to release
2747  *
2748  * The opened QP handle is released by the caller.  The underlying
2749  * shared QP is not destroyed until all internal references are released.
2750  */
2751 int ib_close_qp(struct ib_qp *qp);
2752 
2753 /**
2754  * ib_post_send - Posts a list of work requests to the send queue of
2755  *   the specified QP.
2756  * @qp: The QP to post the work request on.
2757  * @send_wr: A list of work requests to post on the send queue.
2758  * @bad_send_wr: On an immediate failure, this parameter will reference
2759  *   the work request that failed to be posted on the QP.
2760  *
2761  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2762  * error is returned, the QP state shall not be affected,
2763  * ib_post_send() will return an immediate error after queueing any
2764  * earlier work requests in the list.
2765  */
2766 static inline int ib_post_send(struct ib_qp *qp,
2767 			       struct ib_send_wr *send_wr,
2768 			       struct ib_send_wr **bad_send_wr)
2769 {
2770 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2771 }
2772 
2773 /**
2774  * ib_post_recv - Posts a list of work requests to the receive queue of
2775  *   the specified QP.
2776  * @qp: The QP to post the work request on.
2777  * @recv_wr: A list of work requests to post on the receive queue.
2778  * @bad_recv_wr: On an immediate failure, this parameter will reference
2779  *   the work request that failed to be posted on the QP.
2780  */
2781 static inline int ib_post_recv(struct ib_qp *qp,
2782 			       struct ib_recv_wr *recv_wr,
2783 			       struct ib_recv_wr **bad_recv_wr)
2784 {
2785 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2786 }
2787 
2788 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2789 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2790 void ib_free_cq(struct ib_cq *cq);
2791 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2792 
2793 /**
2794  * ib_create_cq - Creates a CQ on the specified device.
2795  * @device: The device on which to create the CQ.
2796  * @comp_handler: A user-specified callback that is invoked when a
2797  *   completion event occurs on the CQ.
2798  * @event_handler: A user-specified callback that is invoked when an
2799  *   asynchronous event not associated with a completion occurs on the CQ.
2800  * @cq_context: Context associated with the CQ returned to the user via
2801  *   the associated completion and event handlers.
2802  * @cq_attr: The attributes the CQ should be created upon.
2803  *
2804  * Users can examine the cq structure to determine the actual CQ size.
2805  */
2806 struct ib_cq *ib_create_cq(struct ib_device *device,
2807 			   ib_comp_handler comp_handler,
2808 			   void (*event_handler)(struct ib_event *, void *),
2809 			   void *cq_context,
2810 			   const struct ib_cq_init_attr *cq_attr);
2811 
2812 /**
2813  * ib_resize_cq - Modifies the capacity of the CQ.
2814  * @cq: The CQ to resize.
2815  * @cqe: The minimum size of the CQ.
2816  *
2817  * Users can examine the cq structure to determine the actual CQ size.
2818  */
2819 int ib_resize_cq(struct ib_cq *cq, int cqe);
2820 
2821 /**
2822  * ib_modify_cq - Modifies moderation params of the CQ
2823  * @cq: The CQ to modify.
2824  * @cq_count: number of CQEs that will trigger an event
2825  * @cq_period: max period of time in usec before triggering an event
2826  *
2827  */
2828 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2829 
2830 /**
2831  * ib_destroy_cq - Destroys the specified CQ.
2832  * @cq: The CQ to destroy.
2833  */
2834 int ib_destroy_cq(struct ib_cq *cq);
2835 
2836 /**
2837  * ib_poll_cq - poll a CQ for completion(s)
2838  * @cq:the CQ being polled
2839  * @num_entries:maximum number of completions to return
2840  * @wc:array of at least @num_entries &struct ib_wc where completions
2841  *   will be returned
2842  *
2843  * Poll a CQ for (possibly multiple) completions.  If the return value
2844  * is < 0, an error occurred.  If the return value is >= 0, it is the
2845  * number of completions returned.  If the return value is
2846  * non-negative and < num_entries, then the CQ was emptied.
2847  */
2848 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2849 			     struct ib_wc *wc)
2850 {
2851 	return cq->device->poll_cq(cq, num_entries, wc);
2852 }
2853 
2854 /**
2855  * ib_peek_cq - Returns the number of unreaped completions currently
2856  *   on the specified CQ.
2857  * @cq: The CQ to peek.
2858  * @wc_cnt: A minimum number of unreaped completions to check for.
2859  *
2860  * If the number of unreaped completions is greater than or equal to wc_cnt,
2861  * this function returns wc_cnt, otherwise, it returns the actual number of
2862  * unreaped completions.
2863  */
2864 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2865 
2866 /**
2867  * ib_req_notify_cq - Request completion notification on a CQ.
2868  * @cq: The CQ to generate an event for.
2869  * @flags:
2870  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2871  *   to request an event on the next solicited event or next work
2872  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2873  *   may also be |ed in to request a hint about missed events, as
2874  *   described below.
2875  *
2876  * Return Value:
2877  *    < 0 means an error occurred while requesting notification
2878  *   == 0 means notification was requested successfully, and if
2879  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2880  *        were missed and it is safe to wait for another event.  In
2881  *        this case is it guaranteed that any work completions added
2882  *        to the CQ since the last CQ poll will trigger a completion
2883  *        notification event.
2884  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2885  *        in.  It means that the consumer must poll the CQ again to
2886  *        make sure it is empty to avoid missing an event because of a
2887  *        race between requesting notification and an entry being
2888  *        added to the CQ.  This return value means it is possible
2889  *        (but not guaranteed) that a work completion has been added
2890  *        to the CQ since the last poll without triggering a
2891  *        completion notification event.
2892  */
2893 static inline int ib_req_notify_cq(struct ib_cq *cq,
2894 				   enum ib_cq_notify_flags flags)
2895 {
2896 	return cq->device->req_notify_cq(cq, flags);
2897 }
2898 
2899 /**
2900  * ib_req_ncomp_notif - Request completion notification when there are
2901  *   at least the specified number of unreaped completions on the CQ.
2902  * @cq: The CQ to generate an event for.
2903  * @wc_cnt: The number of unreaped completions that should be on the
2904  *   CQ before an event is generated.
2905  */
2906 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2907 {
2908 	return cq->device->req_ncomp_notif ?
2909 		cq->device->req_ncomp_notif(cq, wc_cnt) :
2910 		-ENOSYS;
2911 }
2912 
2913 /**
2914  * ib_dma_mapping_error - check a DMA addr for error
2915  * @dev: The device for which the dma_addr was created
2916  * @dma_addr: The DMA address to check
2917  */
2918 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2919 {
2920 	return dma_mapping_error(&dev->dev, dma_addr);
2921 }
2922 
2923 /**
2924  * ib_dma_map_single - Map a kernel virtual address to DMA address
2925  * @dev: The device for which the dma_addr is to be created
2926  * @cpu_addr: The kernel virtual address
2927  * @size: The size of the region in bytes
2928  * @direction: The direction of the DMA
2929  */
2930 static inline u64 ib_dma_map_single(struct ib_device *dev,
2931 				    void *cpu_addr, size_t size,
2932 				    enum dma_data_direction direction)
2933 {
2934 	return dma_map_single(&dev->dev, cpu_addr, size, direction);
2935 }
2936 
2937 /**
2938  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2939  * @dev: The device for which the DMA address was created
2940  * @addr: The DMA address
2941  * @size: The size of the region in bytes
2942  * @direction: The direction of the DMA
2943  */
2944 static inline void ib_dma_unmap_single(struct ib_device *dev,
2945 				       u64 addr, size_t size,
2946 				       enum dma_data_direction direction)
2947 {
2948 	dma_unmap_single(&dev->dev, addr, size, direction);
2949 }
2950 
2951 /**
2952  * ib_dma_map_page - Map a physical page to DMA address
2953  * @dev: The device for which the dma_addr is to be created
2954  * @page: The page to be mapped
2955  * @offset: The offset within the page
2956  * @size: The size of the region in bytes
2957  * @direction: The direction of the DMA
2958  */
2959 static inline u64 ib_dma_map_page(struct ib_device *dev,
2960 				  struct page *page,
2961 				  unsigned long offset,
2962 				  size_t size,
2963 					 enum dma_data_direction direction)
2964 {
2965 	return dma_map_page(&dev->dev, page, offset, size, direction);
2966 }
2967 
2968 /**
2969  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2970  * @dev: The device for which the DMA address was created
2971  * @addr: The DMA address
2972  * @size: The size of the region in bytes
2973  * @direction: The direction of the DMA
2974  */
2975 static inline void ib_dma_unmap_page(struct ib_device *dev,
2976 				     u64 addr, size_t size,
2977 				     enum dma_data_direction direction)
2978 {
2979 	dma_unmap_page(&dev->dev, addr, size, direction);
2980 }
2981 
2982 /**
2983  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2984  * @dev: The device for which the DMA addresses are to be created
2985  * @sg: The array of scatter/gather entries
2986  * @nents: The number of scatter/gather entries
2987  * @direction: The direction of the DMA
2988  */
2989 static inline int ib_dma_map_sg(struct ib_device *dev,
2990 				struct scatterlist *sg, int nents,
2991 				enum dma_data_direction direction)
2992 {
2993 	return dma_map_sg(&dev->dev, sg, nents, direction);
2994 }
2995 
2996 /**
2997  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2998  * @dev: The device for which the DMA addresses were created
2999  * @sg: The array of scatter/gather entries
3000  * @nents: The number of scatter/gather entries
3001  * @direction: The direction of the DMA
3002  */
3003 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3004 				   struct scatterlist *sg, int nents,
3005 				   enum dma_data_direction direction)
3006 {
3007 	dma_unmap_sg(&dev->dev, sg, nents, direction);
3008 }
3009 
3010 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3011 				      struct scatterlist *sg, int nents,
3012 				      enum dma_data_direction direction,
3013 				      unsigned long dma_attrs)
3014 {
3015 	return dma_map_sg_attrs(&dev->dev, sg, nents, direction, dma_attrs);
3016 }
3017 
3018 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3019 					 struct scatterlist *sg, int nents,
3020 					 enum dma_data_direction direction,
3021 					 unsigned long dma_attrs)
3022 {
3023 	dma_unmap_sg_attrs(&dev->dev, sg, nents, direction, dma_attrs);
3024 }
3025 /**
3026  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3027  * @dev: The device for which the DMA addresses were created
3028  * @sg: The scatter/gather entry
3029  *
3030  * Note: this function is obsolete. To do: change all occurrences of
3031  * ib_sg_dma_address() into sg_dma_address().
3032  */
3033 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3034 				    struct scatterlist *sg)
3035 {
3036 	return sg_dma_address(sg);
3037 }
3038 
3039 /**
3040  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3041  * @dev: The device for which the DMA addresses were created
3042  * @sg: The scatter/gather entry
3043  *
3044  * Note: this function is obsolete. To do: change all occurrences of
3045  * ib_sg_dma_len() into sg_dma_len().
3046  */
3047 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3048 					 struct scatterlist *sg)
3049 {
3050 	return sg_dma_len(sg);
3051 }
3052 
3053 /**
3054  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3055  * @dev: The device for which the DMA address was created
3056  * @addr: The DMA address
3057  * @size: The size of the region in bytes
3058  * @dir: The direction of the DMA
3059  */
3060 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3061 					      u64 addr,
3062 					      size_t size,
3063 					      enum dma_data_direction dir)
3064 {
3065 	dma_sync_single_for_cpu(&dev->dev, addr, size, dir);
3066 }
3067 
3068 /**
3069  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3070  * @dev: The device for which the DMA address was created
3071  * @addr: The DMA address
3072  * @size: The size of the region in bytes
3073  * @dir: The direction of the DMA
3074  */
3075 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3076 						 u64 addr,
3077 						 size_t size,
3078 						 enum dma_data_direction dir)
3079 {
3080 	dma_sync_single_for_device(&dev->dev, addr, size, dir);
3081 }
3082 
3083 /**
3084  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3085  * @dev: The device for which the DMA address is requested
3086  * @size: The size of the region to allocate in bytes
3087  * @dma_handle: A pointer for returning the DMA address of the region
3088  * @flag: memory allocator flags
3089  */
3090 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3091 					   size_t size,
3092 					   dma_addr_t *dma_handle,
3093 					   gfp_t flag)
3094 {
3095 	return dma_alloc_coherent(&dev->dev, size, dma_handle, flag);
3096 }
3097 
3098 /**
3099  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3100  * @dev: The device for which the DMA addresses were allocated
3101  * @size: The size of the region
3102  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3103  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3104  */
3105 static inline void ib_dma_free_coherent(struct ib_device *dev,
3106 					size_t size, void *cpu_addr,
3107 					dma_addr_t dma_handle)
3108 {
3109 	dma_free_coherent(&dev->dev, size, cpu_addr, dma_handle);
3110 }
3111 
3112 /**
3113  * ib_dereg_mr - Deregisters a memory region and removes it from the
3114  *   HCA translation table.
3115  * @mr: The memory region to deregister.
3116  *
3117  * This function can fail, if the memory region has memory windows bound to it.
3118  */
3119 int ib_dereg_mr(struct ib_mr *mr);
3120 
3121 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3122 			  enum ib_mr_type mr_type,
3123 			  u32 max_num_sg);
3124 
3125 /**
3126  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3127  *   R_Key and L_Key.
3128  * @mr - struct ib_mr pointer to be updated.
3129  * @newkey - new key to be used.
3130  */
3131 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3132 {
3133 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3134 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3135 }
3136 
3137 /**
3138  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3139  * for calculating a new rkey for type 2 memory windows.
3140  * @rkey - the rkey to increment.
3141  */
3142 static inline u32 ib_inc_rkey(u32 rkey)
3143 {
3144 	const u32 mask = 0x000000ff;
3145 	return ((rkey + 1) & mask) | (rkey & ~mask);
3146 }
3147 
3148 /**
3149  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3150  * @pd: The protection domain associated with the unmapped region.
3151  * @mr_access_flags: Specifies the memory access rights.
3152  * @fmr_attr: Attributes of the unmapped region.
3153  *
3154  * A fast memory region must be mapped before it can be used as part of
3155  * a work request.
3156  */
3157 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3158 			    int mr_access_flags,
3159 			    struct ib_fmr_attr *fmr_attr);
3160 
3161 /**
3162  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3163  * @fmr: The fast memory region to associate with the pages.
3164  * @page_list: An array of physical pages to map to the fast memory region.
3165  * @list_len: The number of pages in page_list.
3166  * @iova: The I/O virtual address to use with the mapped region.
3167  */
3168 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3169 				  u64 *page_list, int list_len,
3170 				  u64 iova)
3171 {
3172 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3173 }
3174 
3175 /**
3176  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3177  * @fmr_list: A linked list of fast memory regions to unmap.
3178  */
3179 int ib_unmap_fmr(struct list_head *fmr_list);
3180 
3181 /**
3182  * ib_dealloc_fmr - Deallocates a fast memory region.
3183  * @fmr: The fast memory region to deallocate.
3184  */
3185 int ib_dealloc_fmr(struct ib_fmr *fmr);
3186 
3187 /**
3188  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3189  * @qp: QP to attach to the multicast group.  The QP must be type
3190  *   IB_QPT_UD.
3191  * @gid: Multicast group GID.
3192  * @lid: Multicast group LID in host byte order.
3193  *
3194  * In order to send and receive multicast packets, subnet
3195  * administration must have created the multicast group and configured
3196  * the fabric appropriately.  The port associated with the specified
3197  * QP must also be a member of the multicast group.
3198  */
3199 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3200 
3201 /**
3202  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3203  * @qp: QP to detach from the multicast group.
3204  * @gid: Multicast group GID.
3205  * @lid: Multicast group LID in host byte order.
3206  */
3207 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3208 
3209 /**
3210  * ib_alloc_xrcd - Allocates an XRC domain.
3211  * @device: The device on which to allocate the XRC domain.
3212  */
3213 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3214 
3215 /**
3216  * ib_dealloc_xrcd - Deallocates an XRC domain.
3217  * @xrcd: The XRC domain to deallocate.
3218  */
3219 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3220 
3221 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3222 			       struct ib_flow_attr *flow_attr, int domain);
3223 int ib_destroy_flow(struct ib_flow *flow_id);
3224 
3225 static inline int ib_check_mr_access(int flags)
3226 {
3227 	/*
3228 	 * Local write permission is required if remote write or
3229 	 * remote atomic permission is also requested.
3230 	 */
3231 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3232 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3233 		return -EINVAL;
3234 
3235 	return 0;
3236 }
3237 
3238 /**
3239  * ib_check_mr_status: lightweight check of MR status.
3240  *     This routine may provide status checks on a selected
3241  *     ib_mr. first use is for signature status check.
3242  *
3243  * @mr: A memory region.
3244  * @check_mask: Bitmask of which checks to perform from
3245  *     ib_mr_status_check enumeration.
3246  * @mr_status: The container of relevant status checks.
3247  *     failed checks will be indicated in the status bitmask
3248  *     and the relevant info shall be in the error item.
3249  */
3250 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3251 		       struct ib_mr_status *mr_status);
3252 
3253 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3254 					    u16 pkey, const union ib_gid *gid,
3255 					    const struct sockaddr *addr);
3256 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3257 			   struct ib_wq_init_attr *init_attr);
3258 int ib_destroy_wq(struct ib_wq *wq);
3259 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3260 		 u32 wq_attr_mask);
3261 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3262 						 struct ib_rwq_ind_table_init_attr*
3263 						 wq_ind_table_init_attr);
3264 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3265 
3266 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3267 		 unsigned int *sg_offset, unsigned int page_size);
3268 
3269 static inline int
3270 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3271 		  unsigned int *sg_offset, unsigned int page_size)
3272 {
3273 	int n;
3274 
3275 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3276 	mr->iova = 0;
3277 
3278 	return n;
3279 }
3280 
3281 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3282 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3283 
3284 void ib_drain_rq(struct ib_qp *qp);
3285 void ib_drain_sq(struct ib_qp *qp);
3286 void ib_drain_qp(struct ib_qp *qp);
3287 
3288 int ib_resolve_eth_dmac(struct ib_device *device,
3289 			struct ib_ah_attr *ah_attr);
3290 #endif /* IB_VERBS_H */
3291