xref: /linux/include/rdma/ib_verbs.h (revision 0883c2c06fb5bcf5b9e008270827e63c09a88c1e)
1 /*
2  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
3  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
4  * Copyright (c) 2004 Intel Corporation.  All rights reserved.
5  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
6  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
7  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
9  *
10  * This software is available to you under a choice of one of two
11  * licenses.  You may choose to be licensed under the terms of the GNU
12  * General Public License (GPL) Version 2, available from the file
13  * COPYING in the main directory of this source tree, or the
14  * OpenIB.org BSD license below:
15  *
16  *     Redistribution and use in source and binary forms, with or
17  *     without modification, are permitted provided that the following
18  *     conditions are met:
19  *
20  *      - Redistributions of source code must retain the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer.
23  *
24  *      - Redistributions in binary form must reproduce the above
25  *        copyright notice, this list of conditions and the following
26  *        disclaimer in the documentation and/or other materials
27  *        provided with the distribution.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36  * SOFTWARE.
37  */
38 
39 #if !defined(IB_VERBS_H)
40 #define IB_VERBS_H
41 
42 #include <linux/types.h>
43 #include <linux/device.h>
44 #include <linux/mm.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
54 #include <net/ipv6.h>
55 #include <net/ip.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 
59 #include <linux/if_link.h>
60 #include <linux/atomic.h>
61 #include <linux/mmu_notifier.h>
62 #include <asm/uaccess.h>
63 
64 extern struct workqueue_struct *ib_wq;
65 extern struct workqueue_struct *ib_comp_wq;
66 
67 union ib_gid {
68 	u8	raw[16];
69 	struct {
70 		__be64	subnet_prefix;
71 		__be64	interface_id;
72 	} global;
73 };
74 
75 extern union ib_gid zgid;
76 
77 enum ib_gid_type {
78 	/* If link layer is Ethernet, this is RoCE V1 */
79 	IB_GID_TYPE_IB        = 0,
80 	IB_GID_TYPE_ROCE      = 0,
81 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
82 	IB_GID_TYPE_SIZE
83 };
84 
85 #define ROCE_V2_UDP_DPORT      4791
86 struct ib_gid_attr {
87 	enum ib_gid_type	gid_type;
88 	struct net_device	*ndev;
89 };
90 
91 enum rdma_node_type {
92 	/* IB values map to NodeInfo:NodeType. */
93 	RDMA_NODE_IB_CA 	= 1,
94 	RDMA_NODE_IB_SWITCH,
95 	RDMA_NODE_IB_ROUTER,
96 	RDMA_NODE_RNIC,
97 	RDMA_NODE_USNIC,
98 	RDMA_NODE_USNIC_UDP,
99 };
100 
101 enum {
102 	/* set the local administered indication */
103 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
104 };
105 
106 enum rdma_transport_type {
107 	RDMA_TRANSPORT_IB,
108 	RDMA_TRANSPORT_IWARP,
109 	RDMA_TRANSPORT_USNIC,
110 	RDMA_TRANSPORT_USNIC_UDP
111 };
112 
113 enum rdma_protocol_type {
114 	RDMA_PROTOCOL_IB,
115 	RDMA_PROTOCOL_IBOE,
116 	RDMA_PROTOCOL_IWARP,
117 	RDMA_PROTOCOL_USNIC_UDP
118 };
119 
120 __attribute_const__ enum rdma_transport_type
121 rdma_node_get_transport(enum rdma_node_type node_type);
122 
123 enum rdma_network_type {
124 	RDMA_NETWORK_IB,
125 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
126 	RDMA_NETWORK_IPV4,
127 	RDMA_NETWORK_IPV6
128 };
129 
130 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
131 {
132 	if (network_type == RDMA_NETWORK_IPV4 ||
133 	    network_type == RDMA_NETWORK_IPV6)
134 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
135 
136 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
137 	return IB_GID_TYPE_IB;
138 }
139 
140 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
141 							    union ib_gid *gid)
142 {
143 	if (gid_type == IB_GID_TYPE_IB)
144 		return RDMA_NETWORK_IB;
145 
146 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
147 		return RDMA_NETWORK_IPV4;
148 	else
149 		return RDMA_NETWORK_IPV6;
150 }
151 
152 enum rdma_link_layer {
153 	IB_LINK_LAYER_UNSPECIFIED,
154 	IB_LINK_LAYER_INFINIBAND,
155 	IB_LINK_LAYER_ETHERNET,
156 };
157 
158 enum ib_device_cap_flags {
159 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
160 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
161 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
162 	IB_DEVICE_RAW_MULTI			= (1 << 3),
163 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
164 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
165 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
166 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
167 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
168 	IB_DEVICE_INIT_TYPE			= (1 << 9),
169 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
170 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
171 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
172 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
173 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
174 
175 	/*
176 	 * This device supports a per-device lkey or stag that can be
177 	 * used without performing a memory registration for the local
178 	 * memory.  Note that ULPs should never check this flag, but
179 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
180 	 * which will always contain a usable lkey.
181 	 */
182 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
183 	IB_DEVICE_RESERVED /* old SEND_W_INV */	= (1 << 16),
184 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
185 	/*
186 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
187 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
188 	 * messages and can verify the validity of checksum for
189 	 * incoming messages.  Setting this flag implies that the
190 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
191 	 */
192 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
193 	IB_DEVICE_UD_TSO			= (1 << 19),
194 	IB_DEVICE_XRC				= (1 << 20),
195 
196 	/*
197 	 * This device supports the IB "base memory management extension",
198 	 * which includes support for fast registrations (IB_WR_REG_MR,
199 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
200 	 * also be set by any iWarp device which must support FRs to comply
201 	 * to the iWarp verbs spec.  iWarp devices also support the
202 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
203 	 * stag.
204 	 */
205 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
206 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
207 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
208 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
209 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
210 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
211 	/*
212 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
213 	 * support execution of WQEs that involve synchronization
214 	 * of I/O operations with single completion queue managed
215 	 * by hardware.
216 	 */
217 	IB_DEVICE_CROSS_CHANNEL		= (1 << 27),
218 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
219 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
220 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
221 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
222 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
223 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
224 };
225 
226 enum ib_signature_prot_cap {
227 	IB_PROT_T10DIF_TYPE_1 = 1,
228 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
229 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
230 };
231 
232 enum ib_signature_guard_cap {
233 	IB_GUARD_T10DIF_CRC	= 1,
234 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
235 };
236 
237 enum ib_atomic_cap {
238 	IB_ATOMIC_NONE,
239 	IB_ATOMIC_HCA,
240 	IB_ATOMIC_GLOB
241 };
242 
243 enum ib_odp_general_cap_bits {
244 	IB_ODP_SUPPORT = 1 << 0,
245 };
246 
247 enum ib_odp_transport_cap_bits {
248 	IB_ODP_SUPPORT_SEND	= 1 << 0,
249 	IB_ODP_SUPPORT_RECV	= 1 << 1,
250 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
251 	IB_ODP_SUPPORT_READ	= 1 << 3,
252 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
253 };
254 
255 struct ib_odp_caps {
256 	uint64_t general_caps;
257 	struct {
258 		uint32_t  rc_odp_caps;
259 		uint32_t  uc_odp_caps;
260 		uint32_t  ud_odp_caps;
261 	} per_transport_caps;
262 };
263 
264 enum ib_cq_creation_flags {
265 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
266 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
267 };
268 
269 struct ib_cq_init_attr {
270 	unsigned int	cqe;
271 	int		comp_vector;
272 	u32		flags;
273 };
274 
275 struct ib_device_attr {
276 	u64			fw_ver;
277 	__be64			sys_image_guid;
278 	u64			max_mr_size;
279 	u64			page_size_cap;
280 	u32			vendor_id;
281 	u32			vendor_part_id;
282 	u32			hw_ver;
283 	int			max_qp;
284 	int			max_qp_wr;
285 	u64			device_cap_flags;
286 	int			max_sge;
287 	int			max_sge_rd;
288 	int			max_cq;
289 	int			max_cqe;
290 	int			max_mr;
291 	int			max_pd;
292 	int			max_qp_rd_atom;
293 	int			max_ee_rd_atom;
294 	int			max_res_rd_atom;
295 	int			max_qp_init_rd_atom;
296 	int			max_ee_init_rd_atom;
297 	enum ib_atomic_cap	atomic_cap;
298 	enum ib_atomic_cap	masked_atomic_cap;
299 	int			max_ee;
300 	int			max_rdd;
301 	int			max_mw;
302 	int			max_raw_ipv6_qp;
303 	int			max_raw_ethy_qp;
304 	int			max_mcast_grp;
305 	int			max_mcast_qp_attach;
306 	int			max_total_mcast_qp_attach;
307 	int			max_ah;
308 	int			max_fmr;
309 	int			max_map_per_fmr;
310 	int			max_srq;
311 	int			max_srq_wr;
312 	int			max_srq_sge;
313 	unsigned int		max_fast_reg_page_list_len;
314 	u16			max_pkeys;
315 	u8			local_ca_ack_delay;
316 	int			sig_prot_cap;
317 	int			sig_guard_cap;
318 	struct ib_odp_caps	odp_caps;
319 	uint64_t		timestamp_mask;
320 	uint64_t		hca_core_clock; /* in KHZ */
321 };
322 
323 enum ib_mtu {
324 	IB_MTU_256  = 1,
325 	IB_MTU_512  = 2,
326 	IB_MTU_1024 = 3,
327 	IB_MTU_2048 = 4,
328 	IB_MTU_4096 = 5
329 };
330 
331 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
332 {
333 	switch (mtu) {
334 	case IB_MTU_256:  return  256;
335 	case IB_MTU_512:  return  512;
336 	case IB_MTU_1024: return 1024;
337 	case IB_MTU_2048: return 2048;
338 	case IB_MTU_4096: return 4096;
339 	default: 	  return -1;
340 	}
341 }
342 
343 enum ib_port_state {
344 	IB_PORT_NOP		= 0,
345 	IB_PORT_DOWN		= 1,
346 	IB_PORT_INIT		= 2,
347 	IB_PORT_ARMED		= 3,
348 	IB_PORT_ACTIVE		= 4,
349 	IB_PORT_ACTIVE_DEFER	= 5
350 };
351 
352 enum ib_port_cap_flags {
353 	IB_PORT_SM				= 1 <<  1,
354 	IB_PORT_NOTICE_SUP			= 1 <<  2,
355 	IB_PORT_TRAP_SUP			= 1 <<  3,
356 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
357 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
358 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
359 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
360 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
361 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
362 	IB_PORT_SM_DISABLED			= 1 << 10,
363 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
364 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
365 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
366 	IB_PORT_CM_SUP				= 1 << 16,
367 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
368 	IB_PORT_REINIT_SUP			= 1 << 18,
369 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
370 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
371 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
372 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
373 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
374 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
375 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
376 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
377 };
378 
379 enum ib_port_width {
380 	IB_WIDTH_1X	= 1,
381 	IB_WIDTH_4X	= 2,
382 	IB_WIDTH_8X	= 4,
383 	IB_WIDTH_12X	= 8
384 };
385 
386 static inline int ib_width_enum_to_int(enum ib_port_width width)
387 {
388 	switch (width) {
389 	case IB_WIDTH_1X:  return  1;
390 	case IB_WIDTH_4X:  return  4;
391 	case IB_WIDTH_8X:  return  8;
392 	case IB_WIDTH_12X: return 12;
393 	default: 	  return -1;
394 	}
395 }
396 
397 enum ib_port_speed {
398 	IB_SPEED_SDR	= 1,
399 	IB_SPEED_DDR	= 2,
400 	IB_SPEED_QDR	= 4,
401 	IB_SPEED_FDR10	= 8,
402 	IB_SPEED_FDR	= 16,
403 	IB_SPEED_EDR	= 32
404 };
405 
406 /**
407  * struct rdma_hw_stats
408  * @timestamp - Used by the core code to track when the last update was
409  * @lifespan - Used by the core code to determine how old the counters
410  *   should be before being updated again.  Stored in jiffies, defaults
411  *   to 10 milliseconds, drivers can override the default be specifying
412  *   their own value during their allocation routine.
413  * @name - Array of pointers to static names used for the counters in
414  *   directory.
415  * @num_counters - How many hardware counters there are.  If name is
416  *   shorter than this number, a kernel oops will result.  Driver authors
417  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
418  *   in their code to prevent this.
419  * @value - Array of u64 counters that are accessed by the sysfs code and
420  *   filled in by the drivers get_stats routine
421  */
422 struct rdma_hw_stats {
423 	unsigned long	timestamp;
424 	unsigned long	lifespan;
425 	const char * const *names;
426 	int		num_counters;
427 	u64		value[];
428 };
429 
430 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
431 /**
432  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
433  *   for drivers.
434  * @names - Array of static const char *
435  * @num_counters - How many elements in array
436  * @lifespan - How many milliseconds between updates
437  */
438 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
439 		const char * const *names, int num_counters,
440 		unsigned long lifespan)
441 {
442 	struct rdma_hw_stats *stats;
443 
444 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
445 			GFP_KERNEL);
446 	if (!stats)
447 		return NULL;
448 	stats->names = names;
449 	stats->num_counters = num_counters;
450 	stats->lifespan = msecs_to_jiffies(lifespan);
451 
452 	return stats;
453 }
454 
455 
456 /* Define bits for the various functionality this port needs to be supported by
457  * the core.
458  */
459 /* Management                           0x00000FFF */
460 #define RDMA_CORE_CAP_IB_MAD            0x00000001
461 #define RDMA_CORE_CAP_IB_SMI            0x00000002
462 #define RDMA_CORE_CAP_IB_CM             0x00000004
463 #define RDMA_CORE_CAP_IW_CM             0x00000008
464 #define RDMA_CORE_CAP_IB_SA             0x00000010
465 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
466 
467 /* Address format                       0x000FF000 */
468 #define RDMA_CORE_CAP_AF_IB             0x00001000
469 #define RDMA_CORE_CAP_ETH_AH            0x00002000
470 
471 /* Protocol                             0xFFF00000 */
472 #define RDMA_CORE_CAP_PROT_IB           0x00100000
473 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
474 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
475 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
476 
477 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
478 					| RDMA_CORE_CAP_IB_MAD \
479 					| RDMA_CORE_CAP_IB_SMI \
480 					| RDMA_CORE_CAP_IB_CM  \
481 					| RDMA_CORE_CAP_IB_SA  \
482 					| RDMA_CORE_CAP_AF_IB)
483 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
484 					| RDMA_CORE_CAP_IB_MAD  \
485 					| RDMA_CORE_CAP_IB_CM   \
486 					| RDMA_CORE_CAP_AF_IB   \
487 					| RDMA_CORE_CAP_ETH_AH)
488 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
489 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
490 					| RDMA_CORE_CAP_IB_MAD  \
491 					| RDMA_CORE_CAP_IB_CM   \
492 					| RDMA_CORE_CAP_AF_IB   \
493 					| RDMA_CORE_CAP_ETH_AH)
494 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
495 					| RDMA_CORE_CAP_IW_CM)
496 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
497 					| RDMA_CORE_CAP_OPA_MAD)
498 
499 struct ib_port_attr {
500 	u64			subnet_prefix;
501 	enum ib_port_state	state;
502 	enum ib_mtu		max_mtu;
503 	enum ib_mtu		active_mtu;
504 	int			gid_tbl_len;
505 	u32			port_cap_flags;
506 	u32			max_msg_sz;
507 	u32			bad_pkey_cntr;
508 	u32			qkey_viol_cntr;
509 	u16			pkey_tbl_len;
510 	u16			lid;
511 	u16			sm_lid;
512 	u8			lmc;
513 	u8			max_vl_num;
514 	u8			sm_sl;
515 	u8			subnet_timeout;
516 	u8			init_type_reply;
517 	u8			active_width;
518 	u8			active_speed;
519 	u8                      phys_state;
520 	bool			grh_required;
521 };
522 
523 enum ib_device_modify_flags {
524 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
525 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
526 };
527 
528 struct ib_device_modify {
529 	u64	sys_image_guid;
530 	char	node_desc[64];
531 };
532 
533 enum ib_port_modify_flags {
534 	IB_PORT_SHUTDOWN		= 1,
535 	IB_PORT_INIT_TYPE		= (1<<2),
536 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
537 };
538 
539 struct ib_port_modify {
540 	u32	set_port_cap_mask;
541 	u32	clr_port_cap_mask;
542 	u8	init_type;
543 };
544 
545 enum ib_event_type {
546 	IB_EVENT_CQ_ERR,
547 	IB_EVENT_QP_FATAL,
548 	IB_EVENT_QP_REQ_ERR,
549 	IB_EVENT_QP_ACCESS_ERR,
550 	IB_EVENT_COMM_EST,
551 	IB_EVENT_SQ_DRAINED,
552 	IB_EVENT_PATH_MIG,
553 	IB_EVENT_PATH_MIG_ERR,
554 	IB_EVENT_DEVICE_FATAL,
555 	IB_EVENT_PORT_ACTIVE,
556 	IB_EVENT_PORT_ERR,
557 	IB_EVENT_LID_CHANGE,
558 	IB_EVENT_PKEY_CHANGE,
559 	IB_EVENT_SM_CHANGE,
560 	IB_EVENT_SRQ_ERR,
561 	IB_EVENT_SRQ_LIMIT_REACHED,
562 	IB_EVENT_QP_LAST_WQE_REACHED,
563 	IB_EVENT_CLIENT_REREGISTER,
564 	IB_EVENT_GID_CHANGE,
565 };
566 
567 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
568 
569 struct ib_event {
570 	struct ib_device	*device;
571 	union {
572 		struct ib_cq	*cq;
573 		struct ib_qp	*qp;
574 		struct ib_srq	*srq;
575 		u8		port_num;
576 	} element;
577 	enum ib_event_type	event;
578 };
579 
580 struct ib_event_handler {
581 	struct ib_device *device;
582 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
583 	struct list_head  list;
584 };
585 
586 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
587 	do {							\
588 		(_ptr)->device  = _device;			\
589 		(_ptr)->handler = _handler;			\
590 		INIT_LIST_HEAD(&(_ptr)->list);			\
591 	} while (0)
592 
593 struct ib_global_route {
594 	union ib_gid	dgid;
595 	u32		flow_label;
596 	u8		sgid_index;
597 	u8		hop_limit;
598 	u8		traffic_class;
599 };
600 
601 struct ib_grh {
602 	__be32		version_tclass_flow;
603 	__be16		paylen;
604 	u8		next_hdr;
605 	u8		hop_limit;
606 	union ib_gid	sgid;
607 	union ib_gid	dgid;
608 };
609 
610 union rdma_network_hdr {
611 	struct ib_grh ibgrh;
612 	struct {
613 		/* The IB spec states that if it's IPv4, the header
614 		 * is located in the last 20 bytes of the header.
615 		 */
616 		u8		reserved[20];
617 		struct iphdr	roce4grh;
618 	};
619 };
620 
621 enum {
622 	IB_MULTICAST_QPN = 0xffffff
623 };
624 
625 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
626 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
627 
628 enum ib_ah_flags {
629 	IB_AH_GRH	= 1
630 };
631 
632 enum ib_rate {
633 	IB_RATE_PORT_CURRENT = 0,
634 	IB_RATE_2_5_GBPS = 2,
635 	IB_RATE_5_GBPS   = 5,
636 	IB_RATE_10_GBPS  = 3,
637 	IB_RATE_20_GBPS  = 6,
638 	IB_RATE_30_GBPS  = 4,
639 	IB_RATE_40_GBPS  = 7,
640 	IB_RATE_60_GBPS  = 8,
641 	IB_RATE_80_GBPS  = 9,
642 	IB_RATE_120_GBPS = 10,
643 	IB_RATE_14_GBPS  = 11,
644 	IB_RATE_56_GBPS  = 12,
645 	IB_RATE_112_GBPS = 13,
646 	IB_RATE_168_GBPS = 14,
647 	IB_RATE_25_GBPS  = 15,
648 	IB_RATE_100_GBPS = 16,
649 	IB_RATE_200_GBPS = 17,
650 	IB_RATE_300_GBPS = 18
651 };
652 
653 /**
654  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
655  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
656  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
657  * @rate: rate to convert.
658  */
659 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
660 
661 /**
662  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
663  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
664  * @rate: rate to convert.
665  */
666 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
667 
668 
669 /**
670  * enum ib_mr_type - memory region type
671  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
672  *                            normal registration
673  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
674  *                            signature operations (data-integrity
675  *                            capable regions)
676  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
677  *                            register any arbitrary sg lists (without
678  *                            the normal mr constraints - see
679  *                            ib_map_mr_sg)
680  */
681 enum ib_mr_type {
682 	IB_MR_TYPE_MEM_REG,
683 	IB_MR_TYPE_SIGNATURE,
684 	IB_MR_TYPE_SG_GAPS,
685 };
686 
687 /**
688  * Signature types
689  * IB_SIG_TYPE_NONE: Unprotected.
690  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
691  */
692 enum ib_signature_type {
693 	IB_SIG_TYPE_NONE,
694 	IB_SIG_TYPE_T10_DIF,
695 };
696 
697 /**
698  * Signature T10-DIF block-guard types
699  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
700  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
701  */
702 enum ib_t10_dif_bg_type {
703 	IB_T10DIF_CRC,
704 	IB_T10DIF_CSUM
705 };
706 
707 /**
708  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
709  *     domain.
710  * @bg_type: T10-DIF block guard type (CRC|CSUM)
711  * @pi_interval: protection information interval.
712  * @bg: seed of guard computation.
713  * @app_tag: application tag of guard block
714  * @ref_tag: initial guard block reference tag.
715  * @ref_remap: Indicate wethear the reftag increments each block
716  * @app_escape: Indicate to skip block check if apptag=0xffff
717  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
718  * @apptag_check_mask: check bitmask of application tag.
719  */
720 struct ib_t10_dif_domain {
721 	enum ib_t10_dif_bg_type bg_type;
722 	u16			pi_interval;
723 	u16			bg;
724 	u16			app_tag;
725 	u32			ref_tag;
726 	bool			ref_remap;
727 	bool			app_escape;
728 	bool			ref_escape;
729 	u16			apptag_check_mask;
730 };
731 
732 /**
733  * struct ib_sig_domain - Parameters for signature domain
734  * @sig_type: specific signauture type
735  * @sig: union of all signature domain attributes that may
736  *     be used to set domain layout.
737  */
738 struct ib_sig_domain {
739 	enum ib_signature_type sig_type;
740 	union {
741 		struct ib_t10_dif_domain dif;
742 	} sig;
743 };
744 
745 /**
746  * struct ib_sig_attrs - Parameters for signature handover operation
747  * @check_mask: bitmask for signature byte check (8 bytes)
748  * @mem: memory domain layout desciptor.
749  * @wire: wire domain layout desciptor.
750  */
751 struct ib_sig_attrs {
752 	u8			check_mask;
753 	struct ib_sig_domain	mem;
754 	struct ib_sig_domain	wire;
755 };
756 
757 enum ib_sig_err_type {
758 	IB_SIG_BAD_GUARD,
759 	IB_SIG_BAD_REFTAG,
760 	IB_SIG_BAD_APPTAG,
761 };
762 
763 /**
764  * struct ib_sig_err - signature error descriptor
765  */
766 struct ib_sig_err {
767 	enum ib_sig_err_type	err_type;
768 	u32			expected;
769 	u32			actual;
770 	u64			sig_err_offset;
771 	u32			key;
772 };
773 
774 enum ib_mr_status_check {
775 	IB_MR_CHECK_SIG_STATUS = 1,
776 };
777 
778 /**
779  * struct ib_mr_status - Memory region status container
780  *
781  * @fail_status: Bitmask of MR checks status. For each
782  *     failed check a corresponding status bit is set.
783  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
784  *     failure.
785  */
786 struct ib_mr_status {
787 	u32		    fail_status;
788 	struct ib_sig_err   sig_err;
789 };
790 
791 /**
792  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
793  * enum.
794  * @mult: multiple to convert.
795  */
796 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
797 
798 struct ib_ah_attr {
799 	struct ib_global_route	grh;
800 	u16			dlid;
801 	u8			sl;
802 	u8			src_path_bits;
803 	u8			static_rate;
804 	u8			ah_flags;
805 	u8			port_num;
806 	u8			dmac[ETH_ALEN];
807 };
808 
809 enum ib_wc_status {
810 	IB_WC_SUCCESS,
811 	IB_WC_LOC_LEN_ERR,
812 	IB_WC_LOC_QP_OP_ERR,
813 	IB_WC_LOC_EEC_OP_ERR,
814 	IB_WC_LOC_PROT_ERR,
815 	IB_WC_WR_FLUSH_ERR,
816 	IB_WC_MW_BIND_ERR,
817 	IB_WC_BAD_RESP_ERR,
818 	IB_WC_LOC_ACCESS_ERR,
819 	IB_WC_REM_INV_REQ_ERR,
820 	IB_WC_REM_ACCESS_ERR,
821 	IB_WC_REM_OP_ERR,
822 	IB_WC_RETRY_EXC_ERR,
823 	IB_WC_RNR_RETRY_EXC_ERR,
824 	IB_WC_LOC_RDD_VIOL_ERR,
825 	IB_WC_REM_INV_RD_REQ_ERR,
826 	IB_WC_REM_ABORT_ERR,
827 	IB_WC_INV_EECN_ERR,
828 	IB_WC_INV_EEC_STATE_ERR,
829 	IB_WC_FATAL_ERR,
830 	IB_WC_RESP_TIMEOUT_ERR,
831 	IB_WC_GENERAL_ERR
832 };
833 
834 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
835 
836 enum ib_wc_opcode {
837 	IB_WC_SEND,
838 	IB_WC_RDMA_WRITE,
839 	IB_WC_RDMA_READ,
840 	IB_WC_COMP_SWAP,
841 	IB_WC_FETCH_ADD,
842 	IB_WC_LSO,
843 	IB_WC_LOCAL_INV,
844 	IB_WC_REG_MR,
845 	IB_WC_MASKED_COMP_SWAP,
846 	IB_WC_MASKED_FETCH_ADD,
847 /*
848  * Set value of IB_WC_RECV so consumers can test if a completion is a
849  * receive by testing (opcode & IB_WC_RECV).
850  */
851 	IB_WC_RECV			= 1 << 7,
852 	IB_WC_RECV_RDMA_WITH_IMM
853 };
854 
855 enum ib_wc_flags {
856 	IB_WC_GRH		= 1,
857 	IB_WC_WITH_IMM		= (1<<1),
858 	IB_WC_WITH_INVALIDATE	= (1<<2),
859 	IB_WC_IP_CSUM_OK	= (1<<3),
860 	IB_WC_WITH_SMAC		= (1<<4),
861 	IB_WC_WITH_VLAN		= (1<<5),
862 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
863 };
864 
865 struct ib_wc {
866 	union {
867 		u64		wr_id;
868 		struct ib_cqe	*wr_cqe;
869 	};
870 	enum ib_wc_status	status;
871 	enum ib_wc_opcode	opcode;
872 	u32			vendor_err;
873 	u32			byte_len;
874 	struct ib_qp	       *qp;
875 	union {
876 		__be32		imm_data;
877 		u32		invalidate_rkey;
878 	} ex;
879 	u32			src_qp;
880 	int			wc_flags;
881 	u16			pkey_index;
882 	u16			slid;
883 	u8			sl;
884 	u8			dlid_path_bits;
885 	u8			port_num;	/* valid only for DR SMPs on switches */
886 	u8			smac[ETH_ALEN];
887 	u16			vlan_id;
888 	u8			network_hdr_type;
889 };
890 
891 enum ib_cq_notify_flags {
892 	IB_CQ_SOLICITED			= 1 << 0,
893 	IB_CQ_NEXT_COMP			= 1 << 1,
894 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
895 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
896 };
897 
898 enum ib_srq_type {
899 	IB_SRQT_BASIC,
900 	IB_SRQT_XRC
901 };
902 
903 enum ib_srq_attr_mask {
904 	IB_SRQ_MAX_WR	= 1 << 0,
905 	IB_SRQ_LIMIT	= 1 << 1,
906 };
907 
908 struct ib_srq_attr {
909 	u32	max_wr;
910 	u32	max_sge;
911 	u32	srq_limit;
912 };
913 
914 struct ib_srq_init_attr {
915 	void		      (*event_handler)(struct ib_event *, void *);
916 	void		       *srq_context;
917 	struct ib_srq_attr	attr;
918 	enum ib_srq_type	srq_type;
919 
920 	union {
921 		struct {
922 			struct ib_xrcd *xrcd;
923 			struct ib_cq   *cq;
924 		} xrc;
925 	} ext;
926 };
927 
928 struct ib_qp_cap {
929 	u32	max_send_wr;
930 	u32	max_recv_wr;
931 	u32	max_send_sge;
932 	u32	max_recv_sge;
933 	u32	max_inline_data;
934 
935 	/*
936 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
937 	 * ib_create_qp() will calculate the right amount of neededed WRs
938 	 * and MRs based on this.
939 	 */
940 	u32	max_rdma_ctxs;
941 };
942 
943 enum ib_sig_type {
944 	IB_SIGNAL_ALL_WR,
945 	IB_SIGNAL_REQ_WR
946 };
947 
948 enum ib_qp_type {
949 	/*
950 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
951 	 * here (and in that order) since the MAD layer uses them as
952 	 * indices into a 2-entry table.
953 	 */
954 	IB_QPT_SMI,
955 	IB_QPT_GSI,
956 
957 	IB_QPT_RC,
958 	IB_QPT_UC,
959 	IB_QPT_UD,
960 	IB_QPT_RAW_IPV6,
961 	IB_QPT_RAW_ETHERTYPE,
962 	IB_QPT_RAW_PACKET = 8,
963 	IB_QPT_XRC_INI = 9,
964 	IB_QPT_XRC_TGT,
965 	IB_QPT_MAX,
966 	/* Reserve a range for qp types internal to the low level driver.
967 	 * These qp types will not be visible at the IB core layer, so the
968 	 * IB_QPT_MAX usages should not be affected in the core layer
969 	 */
970 	IB_QPT_RESERVED1 = 0x1000,
971 	IB_QPT_RESERVED2,
972 	IB_QPT_RESERVED3,
973 	IB_QPT_RESERVED4,
974 	IB_QPT_RESERVED5,
975 	IB_QPT_RESERVED6,
976 	IB_QPT_RESERVED7,
977 	IB_QPT_RESERVED8,
978 	IB_QPT_RESERVED9,
979 	IB_QPT_RESERVED10,
980 };
981 
982 enum ib_qp_create_flags {
983 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
984 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
985 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
986 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
987 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
988 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
989 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
990 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
991 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
992 	/* reserve bits 26-31 for low level drivers' internal use */
993 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
994 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
995 };
996 
997 /*
998  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
999  * callback to destroy the passed in QP.
1000  */
1001 
1002 struct ib_qp_init_attr {
1003 	void                  (*event_handler)(struct ib_event *, void *);
1004 	void		       *qp_context;
1005 	struct ib_cq	       *send_cq;
1006 	struct ib_cq	       *recv_cq;
1007 	struct ib_srq	       *srq;
1008 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1009 	struct ib_qp_cap	cap;
1010 	enum ib_sig_type	sq_sig_type;
1011 	enum ib_qp_type		qp_type;
1012 	enum ib_qp_create_flags	create_flags;
1013 
1014 	/*
1015 	 * Only needed for special QP types, or when using the RW API.
1016 	 */
1017 	u8			port_num;
1018 };
1019 
1020 struct ib_qp_open_attr {
1021 	void                  (*event_handler)(struct ib_event *, void *);
1022 	void		       *qp_context;
1023 	u32			qp_num;
1024 	enum ib_qp_type		qp_type;
1025 };
1026 
1027 enum ib_rnr_timeout {
1028 	IB_RNR_TIMER_655_36 =  0,
1029 	IB_RNR_TIMER_000_01 =  1,
1030 	IB_RNR_TIMER_000_02 =  2,
1031 	IB_RNR_TIMER_000_03 =  3,
1032 	IB_RNR_TIMER_000_04 =  4,
1033 	IB_RNR_TIMER_000_06 =  5,
1034 	IB_RNR_TIMER_000_08 =  6,
1035 	IB_RNR_TIMER_000_12 =  7,
1036 	IB_RNR_TIMER_000_16 =  8,
1037 	IB_RNR_TIMER_000_24 =  9,
1038 	IB_RNR_TIMER_000_32 = 10,
1039 	IB_RNR_TIMER_000_48 = 11,
1040 	IB_RNR_TIMER_000_64 = 12,
1041 	IB_RNR_TIMER_000_96 = 13,
1042 	IB_RNR_TIMER_001_28 = 14,
1043 	IB_RNR_TIMER_001_92 = 15,
1044 	IB_RNR_TIMER_002_56 = 16,
1045 	IB_RNR_TIMER_003_84 = 17,
1046 	IB_RNR_TIMER_005_12 = 18,
1047 	IB_RNR_TIMER_007_68 = 19,
1048 	IB_RNR_TIMER_010_24 = 20,
1049 	IB_RNR_TIMER_015_36 = 21,
1050 	IB_RNR_TIMER_020_48 = 22,
1051 	IB_RNR_TIMER_030_72 = 23,
1052 	IB_RNR_TIMER_040_96 = 24,
1053 	IB_RNR_TIMER_061_44 = 25,
1054 	IB_RNR_TIMER_081_92 = 26,
1055 	IB_RNR_TIMER_122_88 = 27,
1056 	IB_RNR_TIMER_163_84 = 28,
1057 	IB_RNR_TIMER_245_76 = 29,
1058 	IB_RNR_TIMER_327_68 = 30,
1059 	IB_RNR_TIMER_491_52 = 31
1060 };
1061 
1062 enum ib_qp_attr_mask {
1063 	IB_QP_STATE			= 1,
1064 	IB_QP_CUR_STATE			= (1<<1),
1065 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1066 	IB_QP_ACCESS_FLAGS		= (1<<3),
1067 	IB_QP_PKEY_INDEX		= (1<<4),
1068 	IB_QP_PORT			= (1<<5),
1069 	IB_QP_QKEY			= (1<<6),
1070 	IB_QP_AV			= (1<<7),
1071 	IB_QP_PATH_MTU			= (1<<8),
1072 	IB_QP_TIMEOUT			= (1<<9),
1073 	IB_QP_RETRY_CNT			= (1<<10),
1074 	IB_QP_RNR_RETRY			= (1<<11),
1075 	IB_QP_RQ_PSN			= (1<<12),
1076 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1077 	IB_QP_ALT_PATH			= (1<<14),
1078 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1079 	IB_QP_SQ_PSN			= (1<<16),
1080 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1081 	IB_QP_PATH_MIG_STATE		= (1<<18),
1082 	IB_QP_CAP			= (1<<19),
1083 	IB_QP_DEST_QPN			= (1<<20),
1084 	IB_QP_RESERVED1			= (1<<21),
1085 	IB_QP_RESERVED2			= (1<<22),
1086 	IB_QP_RESERVED3			= (1<<23),
1087 	IB_QP_RESERVED4			= (1<<24),
1088 };
1089 
1090 enum ib_qp_state {
1091 	IB_QPS_RESET,
1092 	IB_QPS_INIT,
1093 	IB_QPS_RTR,
1094 	IB_QPS_RTS,
1095 	IB_QPS_SQD,
1096 	IB_QPS_SQE,
1097 	IB_QPS_ERR
1098 };
1099 
1100 enum ib_mig_state {
1101 	IB_MIG_MIGRATED,
1102 	IB_MIG_REARM,
1103 	IB_MIG_ARMED
1104 };
1105 
1106 enum ib_mw_type {
1107 	IB_MW_TYPE_1 = 1,
1108 	IB_MW_TYPE_2 = 2
1109 };
1110 
1111 struct ib_qp_attr {
1112 	enum ib_qp_state	qp_state;
1113 	enum ib_qp_state	cur_qp_state;
1114 	enum ib_mtu		path_mtu;
1115 	enum ib_mig_state	path_mig_state;
1116 	u32			qkey;
1117 	u32			rq_psn;
1118 	u32			sq_psn;
1119 	u32			dest_qp_num;
1120 	int			qp_access_flags;
1121 	struct ib_qp_cap	cap;
1122 	struct ib_ah_attr	ah_attr;
1123 	struct ib_ah_attr	alt_ah_attr;
1124 	u16			pkey_index;
1125 	u16			alt_pkey_index;
1126 	u8			en_sqd_async_notify;
1127 	u8			sq_draining;
1128 	u8			max_rd_atomic;
1129 	u8			max_dest_rd_atomic;
1130 	u8			min_rnr_timer;
1131 	u8			port_num;
1132 	u8			timeout;
1133 	u8			retry_cnt;
1134 	u8			rnr_retry;
1135 	u8			alt_port_num;
1136 	u8			alt_timeout;
1137 };
1138 
1139 enum ib_wr_opcode {
1140 	IB_WR_RDMA_WRITE,
1141 	IB_WR_RDMA_WRITE_WITH_IMM,
1142 	IB_WR_SEND,
1143 	IB_WR_SEND_WITH_IMM,
1144 	IB_WR_RDMA_READ,
1145 	IB_WR_ATOMIC_CMP_AND_SWP,
1146 	IB_WR_ATOMIC_FETCH_AND_ADD,
1147 	IB_WR_LSO,
1148 	IB_WR_SEND_WITH_INV,
1149 	IB_WR_RDMA_READ_WITH_INV,
1150 	IB_WR_LOCAL_INV,
1151 	IB_WR_REG_MR,
1152 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1153 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1154 	IB_WR_REG_SIG_MR,
1155 	/* reserve values for low level drivers' internal use.
1156 	 * These values will not be used at all in the ib core layer.
1157 	 */
1158 	IB_WR_RESERVED1 = 0xf0,
1159 	IB_WR_RESERVED2,
1160 	IB_WR_RESERVED3,
1161 	IB_WR_RESERVED4,
1162 	IB_WR_RESERVED5,
1163 	IB_WR_RESERVED6,
1164 	IB_WR_RESERVED7,
1165 	IB_WR_RESERVED8,
1166 	IB_WR_RESERVED9,
1167 	IB_WR_RESERVED10,
1168 };
1169 
1170 enum ib_send_flags {
1171 	IB_SEND_FENCE		= 1,
1172 	IB_SEND_SIGNALED	= (1<<1),
1173 	IB_SEND_SOLICITED	= (1<<2),
1174 	IB_SEND_INLINE		= (1<<3),
1175 	IB_SEND_IP_CSUM		= (1<<4),
1176 
1177 	/* reserve bits 26-31 for low level drivers' internal use */
1178 	IB_SEND_RESERVED_START	= (1 << 26),
1179 	IB_SEND_RESERVED_END	= (1 << 31),
1180 };
1181 
1182 struct ib_sge {
1183 	u64	addr;
1184 	u32	length;
1185 	u32	lkey;
1186 };
1187 
1188 struct ib_cqe {
1189 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1190 };
1191 
1192 struct ib_send_wr {
1193 	struct ib_send_wr      *next;
1194 	union {
1195 		u64		wr_id;
1196 		struct ib_cqe	*wr_cqe;
1197 	};
1198 	struct ib_sge	       *sg_list;
1199 	int			num_sge;
1200 	enum ib_wr_opcode	opcode;
1201 	int			send_flags;
1202 	union {
1203 		__be32		imm_data;
1204 		u32		invalidate_rkey;
1205 	} ex;
1206 };
1207 
1208 struct ib_rdma_wr {
1209 	struct ib_send_wr	wr;
1210 	u64			remote_addr;
1211 	u32			rkey;
1212 };
1213 
1214 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1215 {
1216 	return container_of(wr, struct ib_rdma_wr, wr);
1217 }
1218 
1219 struct ib_atomic_wr {
1220 	struct ib_send_wr	wr;
1221 	u64			remote_addr;
1222 	u64			compare_add;
1223 	u64			swap;
1224 	u64			compare_add_mask;
1225 	u64			swap_mask;
1226 	u32			rkey;
1227 };
1228 
1229 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1230 {
1231 	return container_of(wr, struct ib_atomic_wr, wr);
1232 }
1233 
1234 struct ib_ud_wr {
1235 	struct ib_send_wr	wr;
1236 	struct ib_ah		*ah;
1237 	void			*header;
1238 	int			hlen;
1239 	int			mss;
1240 	u32			remote_qpn;
1241 	u32			remote_qkey;
1242 	u16			pkey_index; /* valid for GSI only */
1243 	u8			port_num;   /* valid for DR SMPs on switch only */
1244 };
1245 
1246 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1247 {
1248 	return container_of(wr, struct ib_ud_wr, wr);
1249 }
1250 
1251 struct ib_reg_wr {
1252 	struct ib_send_wr	wr;
1253 	struct ib_mr		*mr;
1254 	u32			key;
1255 	int			access;
1256 };
1257 
1258 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1259 {
1260 	return container_of(wr, struct ib_reg_wr, wr);
1261 }
1262 
1263 struct ib_sig_handover_wr {
1264 	struct ib_send_wr	wr;
1265 	struct ib_sig_attrs    *sig_attrs;
1266 	struct ib_mr	       *sig_mr;
1267 	int			access_flags;
1268 	struct ib_sge	       *prot;
1269 };
1270 
1271 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1272 {
1273 	return container_of(wr, struct ib_sig_handover_wr, wr);
1274 }
1275 
1276 struct ib_recv_wr {
1277 	struct ib_recv_wr      *next;
1278 	union {
1279 		u64		wr_id;
1280 		struct ib_cqe	*wr_cqe;
1281 	};
1282 	struct ib_sge	       *sg_list;
1283 	int			num_sge;
1284 };
1285 
1286 enum ib_access_flags {
1287 	IB_ACCESS_LOCAL_WRITE	= 1,
1288 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1289 	IB_ACCESS_REMOTE_READ	= (1<<2),
1290 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1291 	IB_ACCESS_MW_BIND	= (1<<4),
1292 	IB_ZERO_BASED		= (1<<5),
1293 	IB_ACCESS_ON_DEMAND     = (1<<6),
1294 };
1295 
1296 /*
1297  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1298  * are hidden here instead of a uapi header!
1299  */
1300 enum ib_mr_rereg_flags {
1301 	IB_MR_REREG_TRANS	= 1,
1302 	IB_MR_REREG_PD		= (1<<1),
1303 	IB_MR_REREG_ACCESS	= (1<<2),
1304 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1305 };
1306 
1307 struct ib_fmr_attr {
1308 	int	max_pages;
1309 	int	max_maps;
1310 	u8	page_shift;
1311 };
1312 
1313 struct ib_umem;
1314 
1315 struct ib_ucontext {
1316 	struct ib_device       *device;
1317 	struct list_head	pd_list;
1318 	struct list_head	mr_list;
1319 	struct list_head	mw_list;
1320 	struct list_head	cq_list;
1321 	struct list_head	qp_list;
1322 	struct list_head	srq_list;
1323 	struct list_head	ah_list;
1324 	struct list_head	xrcd_list;
1325 	struct list_head	rule_list;
1326 	int			closing;
1327 
1328 	struct pid             *tgid;
1329 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1330 	struct rb_root      umem_tree;
1331 	/*
1332 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1333 	 * mmu notifiers registration.
1334 	 */
1335 	struct rw_semaphore	umem_rwsem;
1336 	void (*invalidate_range)(struct ib_umem *umem,
1337 				 unsigned long start, unsigned long end);
1338 
1339 	struct mmu_notifier	mn;
1340 	atomic_t		notifier_count;
1341 	/* A list of umems that don't have private mmu notifier counters yet. */
1342 	struct list_head	no_private_counters;
1343 	int                     odp_mrs_count;
1344 #endif
1345 };
1346 
1347 struct ib_uobject {
1348 	u64			user_handle;	/* handle given to us by userspace */
1349 	struct ib_ucontext     *context;	/* associated user context */
1350 	void		       *object;		/* containing object */
1351 	struct list_head	list;		/* link to context's list */
1352 	int			id;		/* index into kernel idr */
1353 	struct kref		ref;
1354 	struct rw_semaphore	mutex;		/* protects .live */
1355 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1356 	int			live;
1357 };
1358 
1359 struct ib_udata {
1360 	const void __user *inbuf;
1361 	void __user *outbuf;
1362 	size_t       inlen;
1363 	size_t       outlen;
1364 };
1365 
1366 struct ib_pd {
1367 	u32			local_dma_lkey;
1368 	struct ib_device       *device;
1369 	struct ib_uobject      *uobject;
1370 	atomic_t          	usecnt; /* count all resources */
1371 	struct ib_mr	       *local_mr;
1372 };
1373 
1374 struct ib_xrcd {
1375 	struct ib_device       *device;
1376 	atomic_t		usecnt; /* count all exposed resources */
1377 	struct inode	       *inode;
1378 
1379 	struct mutex		tgt_qp_mutex;
1380 	struct list_head	tgt_qp_list;
1381 };
1382 
1383 struct ib_ah {
1384 	struct ib_device	*device;
1385 	struct ib_pd		*pd;
1386 	struct ib_uobject	*uobject;
1387 };
1388 
1389 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1390 
1391 enum ib_poll_context {
1392 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1393 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1394 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1395 };
1396 
1397 struct ib_cq {
1398 	struct ib_device       *device;
1399 	struct ib_uobject      *uobject;
1400 	ib_comp_handler   	comp_handler;
1401 	void                  (*event_handler)(struct ib_event *, void *);
1402 	void                   *cq_context;
1403 	int               	cqe;
1404 	atomic_t          	usecnt; /* count number of work queues */
1405 	enum ib_poll_context	poll_ctx;
1406 	struct ib_wc		*wc;
1407 	union {
1408 		struct irq_poll		iop;
1409 		struct work_struct	work;
1410 	};
1411 };
1412 
1413 struct ib_srq {
1414 	struct ib_device       *device;
1415 	struct ib_pd	       *pd;
1416 	struct ib_uobject      *uobject;
1417 	void		      (*event_handler)(struct ib_event *, void *);
1418 	void		       *srq_context;
1419 	enum ib_srq_type	srq_type;
1420 	atomic_t		usecnt;
1421 
1422 	union {
1423 		struct {
1424 			struct ib_xrcd *xrcd;
1425 			struct ib_cq   *cq;
1426 			u32		srq_num;
1427 		} xrc;
1428 	} ext;
1429 };
1430 
1431 struct ib_qp {
1432 	struct ib_device       *device;
1433 	struct ib_pd	       *pd;
1434 	struct ib_cq	       *send_cq;
1435 	struct ib_cq	       *recv_cq;
1436 	spinlock_t		mr_lock;
1437 	int			mrs_used;
1438 	struct list_head	rdma_mrs;
1439 	struct list_head	sig_mrs;
1440 	struct ib_srq	       *srq;
1441 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1442 	struct list_head	xrcd_list;
1443 
1444 	/* count times opened, mcast attaches, flow attaches */
1445 	atomic_t		usecnt;
1446 	struct list_head	open_list;
1447 	struct ib_qp           *real_qp;
1448 	struct ib_uobject      *uobject;
1449 	void                  (*event_handler)(struct ib_event *, void *);
1450 	void		       *qp_context;
1451 	u32			qp_num;
1452 	enum ib_qp_type		qp_type;
1453 };
1454 
1455 struct ib_mr {
1456 	struct ib_device  *device;
1457 	struct ib_pd	  *pd;
1458 	u32		   lkey;
1459 	u32		   rkey;
1460 	u64		   iova;
1461 	u32		   length;
1462 	unsigned int	   page_size;
1463 	bool		   need_inval;
1464 	union {
1465 		struct ib_uobject	*uobject;	/* user */
1466 		struct list_head	qp_entry;	/* FR */
1467 	};
1468 };
1469 
1470 struct ib_mw {
1471 	struct ib_device	*device;
1472 	struct ib_pd		*pd;
1473 	struct ib_uobject	*uobject;
1474 	u32			rkey;
1475 	enum ib_mw_type         type;
1476 };
1477 
1478 struct ib_fmr {
1479 	struct ib_device	*device;
1480 	struct ib_pd		*pd;
1481 	struct list_head	list;
1482 	u32			lkey;
1483 	u32			rkey;
1484 };
1485 
1486 /* Supported steering options */
1487 enum ib_flow_attr_type {
1488 	/* steering according to rule specifications */
1489 	IB_FLOW_ATTR_NORMAL		= 0x0,
1490 	/* default unicast and multicast rule -
1491 	 * receive all Eth traffic which isn't steered to any QP
1492 	 */
1493 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1494 	/* default multicast rule -
1495 	 * receive all Eth multicast traffic which isn't steered to any QP
1496 	 */
1497 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1498 	/* sniffer rule - receive all port traffic */
1499 	IB_FLOW_ATTR_SNIFFER		= 0x3
1500 };
1501 
1502 /* Supported steering header types */
1503 enum ib_flow_spec_type {
1504 	/* L2 headers*/
1505 	IB_FLOW_SPEC_ETH	= 0x20,
1506 	IB_FLOW_SPEC_IB		= 0x22,
1507 	/* L3 header*/
1508 	IB_FLOW_SPEC_IPV4	= 0x30,
1509 	/* L4 headers*/
1510 	IB_FLOW_SPEC_TCP	= 0x40,
1511 	IB_FLOW_SPEC_UDP	= 0x41
1512 };
1513 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1514 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1515 
1516 /* Flow steering rule priority is set according to it's domain.
1517  * Lower domain value means higher priority.
1518  */
1519 enum ib_flow_domain {
1520 	IB_FLOW_DOMAIN_USER,
1521 	IB_FLOW_DOMAIN_ETHTOOL,
1522 	IB_FLOW_DOMAIN_RFS,
1523 	IB_FLOW_DOMAIN_NIC,
1524 	IB_FLOW_DOMAIN_NUM /* Must be last */
1525 };
1526 
1527 enum ib_flow_flags {
1528 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1529 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1530 };
1531 
1532 struct ib_flow_eth_filter {
1533 	u8	dst_mac[6];
1534 	u8	src_mac[6];
1535 	__be16	ether_type;
1536 	__be16	vlan_tag;
1537 };
1538 
1539 struct ib_flow_spec_eth {
1540 	enum ib_flow_spec_type	  type;
1541 	u16			  size;
1542 	struct ib_flow_eth_filter val;
1543 	struct ib_flow_eth_filter mask;
1544 };
1545 
1546 struct ib_flow_ib_filter {
1547 	__be16 dlid;
1548 	__u8   sl;
1549 };
1550 
1551 struct ib_flow_spec_ib {
1552 	enum ib_flow_spec_type	 type;
1553 	u16			 size;
1554 	struct ib_flow_ib_filter val;
1555 	struct ib_flow_ib_filter mask;
1556 };
1557 
1558 struct ib_flow_ipv4_filter {
1559 	__be32	src_ip;
1560 	__be32	dst_ip;
1561 };
1562 
1563 struct ib_flow_spec_ipv4 {
1564 	enum ib_flow_spec_type	   type;
1565 	u16			   size;
1566 	struct ib_flow_ipv4_filter val;
1567 	struct ib_flow_ipv4_filter mask;
1568 };
1569 
1570 struct ib_flow_tcp_udp_filter {
1571 	__be16	dst_port;
1572 	__be16	src_port;
1573 };
1574 
1575 struct ib_flow_spec_tcp_udp {
1576 	enum ib_flow_spec_type	      type;
1577 	u16			      size;
1578 	struct ib_flow_tcp_udp_filter val;
1579 	struct ib_flow_tcp_udp_filter mask;
1580 };
1581 
1582 union ib_flow_spec {
1583 	struct {
1584 		enum ib_flow_spec_type	type;
1585 		u16			size;
1586 	};
1587 	struct ib_flow_spec_eth		eth;
1588 	struct ib_flow_spec_ib		ib;
1589 	struct ib_flow_spec_ipv4        ipv4;
1590 	struct ib_flow_spec_tcp_udp	tcp_udp;
1591 };
1592 
1593 struct ib_flow_attr {
1594 	enum ib_flow_attr_type type;
1595 	u16	     size;
1596 	u16	     priority;
1597 	u32	     flags;
1598 	u8	     num_of_specs;
1599 	u8	     port;
1600 	/* Following are the optional layers according to user request
1601 	 * struct ib_flow_spec_xxx
1602 	 * struct ib_flow_spec_yyy
1603 	 */
1604 };
1605 
1606 struct ib_flow {
1607 	struct ib_qp		*qp;
1608 	struct ib_uobject	*uobject;
1609 };
1610 
1611 struct ib_mad_hdr;
1612 struct ib_grh;
1613 
1614 enum ib_process_mad_flags {
1615 	IB_MAD_IGNORE_MKEY	= 1,
1616 	IB_MAD_IGNORE_BKEY	= 2,
1617 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1618 };
1619 
1620 enum ib_mad_result {
1621 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1622 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1623 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1624 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1625 };
1626 
1627 #define IB_DEVICE_NAME_MAX 64
1628 
1629 struct ib_cache {
1630 	rwlock_t                lock;
1631 	struct ib_event_handler event_handler;
1632 	struct ib_pkey_cache  **pkey_cache;
1633 	struct ib_gid_table   **gid_cache;
1634 	u8                     *lmc_cache;
1635 };
1636 
1637 struct ib_dma_mapping_ops {
1638 	int		(*mapping_error)(struct ib_device *dev,
1639 					 u64 dma_addr);
1640 	u64		(*map_single)(struct ib_device *dev,
1641 				      void *ptr, size_t size,
1642 				      enum dma_data_direction direction);
1643 	void		(*unmap_single)(struct ib_device *dev,
1644 					u64 addr, size_t size,
1645 					enum dma_data_direction direction);
1646 	u64		(*map_page)(struct ib_device *dev,
1647 				    struct page *page, unsigned long offset,
1648 				    size_t size,
1649 				    enum dma_data_direction direction);
1650 	void		(*unmap_page)(struct ib_device *dev,
1651 				      u64 addr, size_t size,
1652 				      enum dma_data_direction direction);
1653 	int		(*map_sg)(struct ib_device *dev,
1654 				  struct scatterlist *sg, int nents,
1655 				  enum dma_data_direction direction);
1656 	void		(*unmap_sg)(struct ib_device *dev,
1657 				    struct scatterlist *sg, int nents,
1658 				    enum dma_data_direction direction);
1659 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1660 					       u64 dma_handle,
1661 					       size_t size,
1662 					       enum dma_data_direction dir);
1663 	void		(*sync_single_for_device)(struct ib_device *dev,
1664 						  u64 dma_handle,
1665 						  size_t size,
1666 						  enum dma_data_direction dir);
1667 	void		*(*alloc_coherent)(struct ib_device *dev,
1668 					   size_t size,
1669 					   u64 *dma_handle,
1670 					   gfp_t flag);
1671 	void		(*free_coherent)(struct ib_device *dev,
1672 					 size_t size, void *cpu_addr,
1673 					 u64 dma_handle);
1674 };
1675 
1676 struct iw_cm_verbs;
1677 
1678 struct ib_port_immutable {
1679 	int                           pkey_tbl_len;
1680 	int                           gid_tbl_len;
1681 	u32                           core_cap_flags;
1682 	u32                           max_mad_size;
1683 };
1684 
1685 struct ib_device {
1686 	struct device                *dma_device;
1687 
1688 	char                          name[IB_DEVICE_NAME_MAX];
1689 
1690 	struct list_head              event_handler_list;
1691 	spinlock_t                    event_handler_lock;
1692 
1693 	spinlock_t                    client_data_lock;
1694 	struct list_head              core_list;
1695 	/* Access to the client_data_list is protected by the client_data_lock
1696 	 * spinlock and the lists_rwsem read-write semaphore */
1697 	struct list_head              client_data_list;
1698 
1699 	struct ib_cache               cache;
1700 	/**
1701 	 * port_immutable is indexed by port number
1702 	 */
1703 	struct ib_port_immutable     *port_immutable;
1704 
1705 	int			      num_comp_vectors;
1706 
1707 	struct iw_cm_verbs	     *iwcm;
1708 
1709 	/**
1710 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1711 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
1712 	 *   core when the device is removed.  A lifespan of -1 in the return
1713 	 *   struct tells the core to set a default lifespan.
1714 	 */
1715 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
1716 						     u8 port_num);
1717 	/**
1718 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1719 	 * @index - The index in the value array we wish to have updated, or
1720 	 *   num_counters if we want all stats updated
1721 	 * Return codes -
1722 	 *   < 0 - Error, no counters updated
1723 	 *   index - Updated the single counter pointed to by index
1724 	 *   num_counters - Updated all counters (will reset the timestamp
1725 	 *     and prevent further calls for lifespan milliseconds)
1726 	 * Drivers are allowed to update all counters in leiu of just the
1727 	 *   one given in index at their option
1728 	 */
1729 	int		           (*get_hw_stats)(struct ib_device *device,
1730 						   struct rdma_hw_stats *stats,
1731 						   u8 port, int index);
1732 	int		           (*query_device)(struct ib_device *device,
1733 						   struct ib_device_attr *device_attr,
1734 						   struct ib_udata *udata);
1735 	int		           (*query_port)(struct ib_device *device,
1736 						 u8 port_num,
1737 						 struct ib_port_attr *port_attr);
1738 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1739 						     u8 port_num);
1740 	/* When calling get_netdev, the HW vendor's driver should return the
1741 	 * net device of device @device at port @port_num or NULL if such
1742 	 * a net device doesn't exist. The vendor driver should call dev_hold
1743 	 * on this net device. The HW vendor's device driver must guarantee
1744 	 * that this function returns NULL before the net device reaches
1745 	 * NETDEV_UNREGISTER_FINAL state.
1746 	 */
1747 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1748 						 u8 port_num);
1749 	int		           (*query_gid)(struct ib_device *device,
1750 						u8 port_num, int index,
1751 						union ib_gid *gid);
1752 	/* When calling add_gid, the HW vendor's driver should
1753 	 * add the gid of device @device at gid index @index of
1754 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1755 	 * the network device related to this gid is available
1756 	 * at @attr. @context allows the HW vendor driver to store extra
1757 	 * information together with a GID entry. The HW vendor may allocate
1758 	 * memory to contain this information and store it in @context when a
1759 	 * new GID entry is written to. Params are consistent until the next
1760 	 * call of add_gid or delete_gid. The function should return 0 on
1761 	 * success or error otherwise. The function could be called
1762 	 * concurrently for different ports. This function is only called
1763 	 * when roce_gid_table is used.
1764 	 */
1765 	int		           (*add_gid)(struct ib_device *device,
1766 					      u8 port_num,
1767 					      unsigned int index,
1768 					      const union ib_gid *gid,
1769 					      const struct ib_gid_attr *attr,
1770 					      void **context);
1771 	/* When calling del_gid, the HW vendor's driver should delete the
1772 	 * gid of device @device at gid index @index of port @port_num.
1773 	 * Upon the deletion of a GID entry, the HW vendor must free any
1774 	 * allocated memory. The caller will clear @context afterwards.
1775 	 * This function is only called when roce_gid_table is used.
1776 	 */
1777 	int		           (*del_gid)(struct ib_device *device,
1778 					      u8 port_num,
1779 					      unsigned int index,
1780 					      void **context);
1781 	int		           (*query_pkey)(struct ib_device *device,
1782 						 u8 port_num, u16 index, u16 *pkey);
1783 	int		           (*modify_device)(struct ib_device *device,
1784 						    int device_modify_mask,
1785 						    struct ib_device_modify *device_modify);
1786 	int		           (*modify_port)(struct ib_device *device,
1787 						  u8 port_num, int port_modify_mask,
1788 						  struct ib_port_modify *port_modify);
1789 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1790 						     struct ib_udata *udata);
1791 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1792 	int                        (*mmap)(struct ib_ucontext *context,
1793 					   struct vm_area_struct *vma);
1794 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1795 					       struct ib_ucontext *context,
1796 					       struct ib_udata *udata);
1797 	int                        (*dealloc_pd)(struct ib_pd *pd);
1798 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1799 						struct ib_ah_attr *ah_attr);
1800 	int                        (*modify_ah)(struct ib_ah *ah,
1801 						struct ib_ah_attr *ah_attr);
1802 	int                        (*query_ah)(struct ib_ah *ah,
1803 					       struct ib_ah_attr *ah_attr);
1804 	int                        (*destroy_ah)(struct ib_ah *ah);
1805 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1806 						 struct ib_srq_init_attr *srq_init_attr,
1807 						 struct ib_udata *udata);
1808 	int                        (*modify_srq)(struct ib_srq *srq,
1809 						 struct ib_srq_attr *srq_attr,
1810 						 enum ib_srq_attr_mask srq_attr_mask,
1811 						 struct ib_udata *udata);
1812 	int                        (*query_srq)(struct ib_srq *srq,
1813 						struct ib_srq_attr *srq_attr);
1814 	int                        (*destroy_srq)(struct ib_srq *srq);
1815 	int                        (*post_srq_recv)(struct ib_srq *srq,
1816 						    struct ib_recv_wr *recv_wr,
1817 						    struct ib_recv_wr **bad_recv_wr);
1818 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1819 						struct ib_qp_init_attr *qp_init_attr,
1820 						struct ib_udata *udata);
1821 	int                        (*modify_qp)(struct ib_qp *qp,
1822 						struct ib_qp_attr *qp_attr,
1823 						int qp_attr_mask,
1824 						struct ib_udata *udata);
1825 	int                        (*query_qp)(struct ib_qp *qp,
1826 					       struct ib_qp_attr *qp_attr,
1827 					       int qp_attr_mask,
1828 					       struct ib_qp_init_attr *qp_init_attr);
1829 	int                        (*destroy_qp)(struct ib_qp *qp);
1830 	int                        (*post_send)(struct ib_qp *qp,
1831 						struct ib_send_wr *send_wr,
1832 						struct ib_send_wr **bad_send_wr);
1833 	int                        (*post_recv)(struct ib_qp *qp,
1834 						struct ib_recv_wr *recv_wr,
1835 						struct ib_recv_wr **bad_recv_wr);
1836 	struct ib_cq *             (*create_cq)(struct ib_device *device,
1837 						const struct ib_cq_init_attr *attr,
1838 						struct ib_ucontext *context,
1839 						struct ib_udata *udata);
1840 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1841 						u16 cq_period);
1842 	int                        (*destroy_cq)(struct ib_cq *cq);
1843 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1844 						struct ib_udata *udata);
1845 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1846 					      struct ib_wc *wc);
1847 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1848 	int                        (*req_notify_cq)(struct ib_cq *cq,
1849 						    enum ib_cq_notify_flags flags);
1850 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1851 						      int wc_cnt);
1852 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1853 						 int mr_access_flags);
1854 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1855 						  u64 start, u64 length,
1856 						  u64 virt_addr,
1857 						  int mr_access_flags,
1858 						  struct ib_udata *udata);
1859 	int			   (*rereg_user_mr)(struct ib_mr *mr,
1860 						    int flags,
1861 						    u64 start, u64 length,
1862 						    u64 virt_addr,
1863 						    int mr_access_flags,
1864 						    struct ib_pd *pd,
1865 						    struct ib_udata *udata);
1866 	int                        (*dereg_mr)(struct ib_mr *mr);
1867 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
1868 					       enum ib_mr_type mr_type,
1869 					       u32 max_num_sg);
1870 	int                        (*map_mr_sg)(struct ib_mr *mr,
1871 						struct scatterlist *sg,
1872 						int sg_nents,
1873 						unsigned int *sg_offset);
1874 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
1875 					       enum ib_mw_type type,
1876 					       struct ib_udata *udata);
1877 	int                        (*dealloc_mw)(struct ib_mw *mw);
1878 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1879 						int mr_access_flags,
1880 						struct ib_fmr_attr *fmr_attr);
1881 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1882 						   u64 *page_list, int list_len,
1883 						   u64 iova);
1884 	int		           (*unmap_fmr)(struct list_head *fmr_list);
1885 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1886 	int                        (*attach_mcast)(struct ib_qp *qp,
1887 						   union ib_gid *gid,
1888 						   u16 lid);
1889 	int                        (*detach_mcast)(struct ib_qp *qp,
1890 						   union ib_gid *gid,
1891 						   u16 lid);
1892 	int                        (*process_mad)(struct ib_device *device,
1893 						  int process_mad_flags,
1894 						  u8 port_num,
1895 						  const struct ib_wc *in_wc,
1896 						  const struct ib_grh *in_grh,
1897 						  const struct ib_mad_hdr *in_mad,
1898 						  size_t in_mad_size,
1899 						  struct ib_mad_hdr *out_mad,
1900 						  size_t *out_mad_size,
1901 						  u16 *out_mad_pkey_index);
1902 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
1903 						 struct ib_ucontext *ucontext,
1904 						 struct ib_udata *udata);
1905 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1906 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
1907 						  struct ib_flow_attr
1908 						  *flow_attr,
1909 						  int domain);
1910 	int			   (*destroy_flow)(struct ib_flow *flow_id);
1911 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1912 						      struct ib_mr_status *mr_status);
1913 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
1914 	void			   (*drain_rq)(struct ib_qp *qp);
1915 	void			   (*drain_sq)(struct ib_qp *qp);
1916 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
1917 							int state);
1918 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
1919 						   struct ifla_vf_info *ivf);
1920 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
1921 						   struct ifla_vf_stats *stats);
1922 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
1923 						  int type);
1924 
1925 	struct ib_dma_mapping_ops   *dma_ops;
1926 
1927 	struct module               *owner;
1928 	struct device                dev;
1929 	struct kobject               *ports_parent;
1930 	struct list_head             port_list;
1931 
1932 	enum {
1933 		IB_DEV_UNINITIALIZED,
1934 		IB_DEV_REGISTERED,
1935 		IB_DEV_UNREGISTERED
1936 	}                            reg_state;
1937 
1938 	int			     uverbs_abi_ver;
1939 	u64			     uverbs_cmd_mask;
1940 	u64			     uverbs_ex_cmd_mask;
1941 
1942 	char			     node_desc[64];
1943 	__be64			     node_guid;
1944 	u32			     local_dma_lkey;
1945 	u16                          is_switch:1;
1946 	u8                           node_type;
1947 	u8                           phys_port_cnt;
1948 	struct ib_device_attr        attrs;
1949 	struct attribute_group	     *hw_stats_ag;
1950 	struct rdma_hw_stats         *hw_stats;
1951 
1952 	/**
1953 	 * The following mandatory functions are used only at device
1954 	 * registration.  Keep functions such as these at the end of this
1955 	 * structure to avoid cache line misses when accessing struct ib_device
1956 	 * in fast paths.
1957 	 */
1958 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
1959 };
1960 
1961 struct ib_client {
1962 	char  *name;
1963 	void (*add)   (struct ib_device *);
1964 	void (*remove)(struct ib_device *, void *client_data);
1965 
1966 	/* Returns the net_dev belonging to this ib_client and matching the
1967 	 * given parameters.
1968 	 * @dev:	 An RDMA device that the net_dev use for communication.
1969 	 * @port:	 A physical port number on the RDMA device.
1970 	 * @pkey:	 P_Key that the net_dev uses if applicable.
1971 	 * @gid:	 A GID that the net_dev uses to communicate.
1972 	 * @addr:	 An IP address the net_dev is configured with.
1973 	 * @client_data: The device's client data set by ib_set_client_data().
1974 	 *
1975 	 * An ib_client that implements a net_dev on top of RDMA devices
1976 	 * (such as IP over IB) should implement this callback, allowing the
1977 	 * rdma_cm module to find the right net_dev for a given request.
1978 	 *
1979 	 * The caller is responsible for calling dev_put on the returned
1980 	 * netdev. */
1981 	struct net_device *(*get_net_dev_by_params)(
1982 			struct ib_device *dev,
1983 			u8 port,
1984 			u16 pkey,
1985 			const union ib_gid *gid,
1986 			const struct sockaddr *addr,
1987 			void *client_data);
1988 	struct list_head list;
1989 };
1990 
1991 struct ib_device *ib_alloc_device(size_t size);
1992 void ib_dealloc_device(struct ib_device *device);
1993 
1994 int ib_register_device(struct ib_device *device,
1995 		       int (*port_callback)(struct ib_device *,
1996 					    u8, struct kobject *));
1997 void ib_unregister_device(struct ib_device *device);
1998 
1999 int ib_register_client   (struct ib_client *client);
2000 void ib_unregister_client(struct ib_client *client);
2001 
2002 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2003 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2004 			 void *data);
2005 
2006 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2007 {
2008 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2009 }
2010 
2011 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2012 {
2013 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2014 }
2015 
2016 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2017 				       size_t offset,
2018 				       size_t len)
2019 {
2020 	const void __user *p = udata->inbuf + offset;
2021 	bool ret = false;
2022 	u8 *buf;
2023 
2024 	if (len > USHRT_MAX)
2025 		return false;
2026 
2027 	buf = kmalloc(len, GFP_KERNEL);
2028 	if (!buf)
2029 		return false;
2030 
2031 	if (copy_from_user(buf, p, len))
2032 		goto free;
2033 
2034 	ret = !memchr_inv(buf, 0, len);
2035 
2036 free:
2037 	kfree(buf);
2038 	return ret;
2039 }
2040 
2041 /**
2042  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2043  * contains all required attributes and no attributes not allowed for
2044  * the given QP state transition.
2045  * @cur_state: Current QP state
2046  * @next_state: Next QP state
2047  * @type: QP type
2048  * @mask: Mask of supplied QP attributes
2049  * @ll : link layer of port
2050  *
2051  * This function is a helper function that a low-level driver's
2052  * modify_qp method can use to validate the consumer's input.  It
2053  * checks that cur_state and next_state are valid QP states, that a
2054  * transition from cur_state to next_state is allowed by the IB spec,
2055  * and that the attribute mask supplied is allowed for the transition.
2056  */
2057 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2058 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2059 		       enum rdma_link_layer ll);
2060 
2061 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2062 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2063 void ib_dispatch_event(struct ib_event *event);
2064 
2065 int ib_query_port(struct ib_device *device,
2066 		  u8 port_num, struct ib_port_attr *port_attr);
2067 
2068 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2069 					       u8 port_num);
2070 
2071 /**
2072  * rdma_cap_ib_switch - Check if the device is IB switch
2073  * @device: Device to check
2074  *
2075  * Device driver is responsible for setting is_switch bit on
2076  * in ib_device structure at init time.
2077  *
2078  * Return: true if the device is IB switch.
2079  */
2080 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2081 {
2082 	return device->is_switch;
2083 }
2084 
2085 /**
2086  * rdma_start_port - Return the first valid port number for the device
2087  * specified
2088  *
2089  * @device: Device to be checked
2090  *
2091  * Return start port number
2092  */
2093 static inline u8 rdma_start_port(const struct ib_device *device)
2094 {
2095 	return rdma_cap_ib_switch(device) ? 0 : 1;
2096 }
2097 
2098 /**
2099  * rdma_end_port - Return the last valid port number for the device
2100  * specified
2101  *
2102  * @device: Device to be checked
2103  *
2104  * Return last port number
2105  */
2106 static inline u8 rdma_end_port(const struct ib_device *device)
2107 {
2108 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2109 }
2110 
2111 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2112 {
2113 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2114 }
2115 
2116 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2117 {
2118 	return device->port_immutable[port_num].core_cap_flags &
2119 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2120 }
2121 
2122 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2123 {
2124 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2125 }
2126 
2127 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2128 {
2129 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2130 }
2131 
2132 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2133 {
2134 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2135 }
2136 
2137 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2138 {
2139 	return rdma_protocol_ib(device, port_num) ||
2140 		rdma_protocol_roce(device, port_num);
2141 }
2142 
2143 /**
2144  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2145  * Management Datagrams.
2146  * @device: Device to check
2147  * @port_num: Port number to check
2148  *
2149  * Management Datagrams (MAD) are a required part of the InfiniBand
2150  * specification and are supported on all InfiniBand devices.  A slightly
2151  * extended version are also supported on OPA interfaces.
2152  *
2153  * Return: true if the port supports sending/receiving of MAD packets.
2154  */
2155 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2156 {
2157 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2158 }
2159 
2160 /**
2161  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2162  * Management Datagrams.
2163  * @device: Device to check
2164  * @port_num: Port number to check
2165  *
2166  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2167  * datagrams with their own versions.  These OPA MADs share many but not all of
2168  * the characteristics of InfiniBand MADs.
2169  *
2170  * OPA MADs differ in the following ways:
2171  *
2172  *    1) MADs are variable size up to 2K
2173  *       IBTA defined MADs remain fixed at 256 bytes
2174  *    2) OPA SMPs must carry valid PKeys
2175  *    3) OPA SMP packets are a different format
2176  *
2177  * Return: true if the port supports OPA MAD packet formats.
2178  */
2179 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2180 {
2181 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2182 		== RDMA_CORE_CAP_OPA_MAD;
2183 }
2184 
2185 /**
2186  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2187  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2188  * @device: Device to check
2189  * @port_num: Port number to check
2190  *
2191  * Each InfiniBand node is required to provide a Subnet Management Agent
2192  * that the subnet manager can access.  Prior to the fabric being fully
2193  * configured by the subnet manager, the SMA is accessed via a well known
2194  * interface called the Subnet Management Interface (SMI).  This interface
2195  * uses directed route packets to communicate with the SM to get around the
2196  * chicken and egg problem of the SM needing to know what's on the fabric
2197  * in order to configure the fabric, and needing to configure the fabric in
2198  * order to send packets to the devices on the fabric.  These directed
2199  * route packets do not need the fabric fully configured in order to reach
2200  * their destination.  The SMI is the only method allowed to send
2201  * directed route packets on an InfiniBand fabric.
2202  *
2203  * Return: true if the port provides an SMI.
2204  */
2205 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2206 {
2207 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2208 }
2209 
2210 /**
2211  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2212  * Communication Manager.
2213  * @device: Device to check
2214  * @port_num: Port number to check
2215  *
2216  * The InfiniBand Communication Manager is one of many pre-defined General
2217  * Service Agents (GSA) that are accessed via the General Service
2218  * Interface (GSI).  It's role is to facilitate establishment of connections
2219  * between nodes as well as other management related tasks for established
2220  * connections.
2221  *
2222  * Return: true if the port supports an IB CM (this does not guarantee that
2223  * a CM is actually running however).
2224  */
2225 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2226 {
2227 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2228 }
2229 
2230 /**
2231  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2232  * Communication Manager.
2233  * @device: Device to check
2234  * @port_num: Port number to check
2235  *
2236  * Similar to above, but specific to iWARP connections which have a different
2237  * managment protocol than InfiniBand.
2238  *
2239  * Return: true if the port supports an iWARP CM (this does not guarantee that
2240  * a CM is actually running however).
2241  */
2242 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2243 {
2244 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2245 }
2246 
2247 /**
2248  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2249  * Subnet Administration.
2250  * @device: Device to check
2251  * @port_num: Port number to check
2252  *
2253  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2254  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2255  * fabrics, devices should resolve routes to other hosts by contacting the
2256  * SA to query the proper route.
2257  *
2258  * Return: true if the port should act as a client to the fabric Subnet
2259  * Administration interface.  This does not imply that the SA service is
2260  * running locally.
2261  */
2262 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2263 {
2264 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2265 }
2266 
2267 /**
2268  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2269  * Multicast.
2270  * @device: Device to check
2271  * @port_num: Port number to check
2272  *
2273  * InfiniBand multicast registration is more complex than normal IPv4 or
2274  * IPv6 multicast registration.  Each Host Channel Adapter must register
2275  * with the Subnet Manager when it wishes to join a multicast group.  It
2276  * should do so only once regardless of how many queue pairs it subscribes
2277  * to this group.  And it should leave the group only after all queue pairs
2278  * attached to the group have been detached.
2279  *
2280  * Return: true if the port must undertake the additional adminstrative
2281  * overhead of registering/unregistering with the SM and tracking of the
2282  * total number of queue pairs attached to the multicast group.
2283  */
2284 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2285 {
2286 	return rdma_cap_ib_sa(device, port_num);
2287 }
2288 
2289 /**
2290  * rdma_cap_af_ib - Check if the port of device has the capability
2291  * Native Infiniband Address.
2292  * @device: Device to check
2293  * @port_num: Port number to check
2294  *
2295  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2296  * GID.  RoCE uses a different mechanism, but still generates a GID via
2297  * a prescribed mechanism and port specific data.
2298  *
2299  * Return: true if the port uses a GID address to identify devices on the
2300  * network.
2301  */
2302 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2303 {
2304 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2305 }
2306 
2307 /**
2308  * rdma_cap_eth_ah - Check if the port of device has the capability
2309  * Ethernet Address Handle.
2310  * @device: Device to check
2311  * @port_num: Port number to check
2312  *
2313  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2314  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2315  * port.  Normally, packet headers are generated by the sending host
2316  * adapter, but when sending connectionless datagrams, we must manually
2317  * inject the proper headers for the fabric we are communicating over.
2318  *
2319  * Return: true if we are running as a RoCE port and must force the
2320  * addition of a Global Route Header built from our Ethernet Address
2321  * Handle into our header list for connectionless packets.
2322  */
2323 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2324 {
2325 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2326 }
2327 
2328 /**
2329  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2330  *
2331  * @device: Device
2332  * @port_num: Port number
2333  *
2334  * This MAD size includes the MAD headers and MAD payload.  No other headers
2335  * are included.
2336  *
2337  * Return the max MAD size required by the Port.  Will return 0 if the port
2338  * does not support MADs
2339  */
2340 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2341 {
2342 	return device->port_immutable[port_num].max_mad_size;
2343 }
2344 
2345 /**
2346  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2347  * @device: Device to check
2348  * @port_num: Port number to check
2349  *
2350  * RoCE GID table mechanism manages the various GIDs for a device.
2351  *
2352  * NOTE: if allocating the port's GID table has failed, this call will still
2353  * return true, but any RoCE GID table API will fail.
2354  *
2355  * Return: true if the port uses RoCE GID table mechanism in order to manage
2356  * its GIDs.
2357  */
2358 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2359 					   u8 port_num)
2360 {
2361 	return rdma_protocol_roce(device, port_num) &&
2362 		device->add_gid && device->del_gid;
2363 }
2364 
2365 /*
2366  * Check if the device supports READ W/ INVALIDATE.
2367  */
2368 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2369 {
2370 	/*
2371 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2372 	 * has support for it yet.
2373 	 */
2374 	return rdma_protocol_iwarp(dev, port_num);
2375 }
2376 
2377 int ib_query_gid(struct ib_device *device,
2378 		 u8 port_num, int index, union ib_gid *gid,
2379 		 struct ib_gid_attr *attr);
2380 
2381 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2382 			 int state);
2383 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2384 		     struct ifla_vf_info *info);
2385 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2386 		    struct ifla_vf_stats *stats);
2387 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2388 		   int type);
2389 
2390 int ib_query_pkey(struct ib_device *device,
2391 		  u8 port_num, u16 index, u16 *pkey);
2392 
2393 int ib_modify_device(struct ib_device *device,
2394 		     int device_modify_mask,
2395 		     struct ib_device_modify *device_modify);
2396 
2397 int ib_modify_port(struct ib_device *device,
2398 		   u8 port_num, int port_modify_mask,
2399 		   struct ib_port_modify *port_modify);
2400 
2401 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2402 		enum ib_gid_type gid_type, struct net_device *ndev,
2403 		u8 *port_num, u16 *index);
2404 
2405 int ib_find_pkey(struct ib_device *device,
2406 		 u8 port_num, u16 pkey, u16 *index);
2407 
2408 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2409 
2410 void ib_dealloc_pd(struct ib_pd *pd);
2411 
2412 /**
2413  * ib_create_ah - Creates an address handle for the given address vector.
2414  * @pd: The protection domain associated with the address handle.
2415  * @ah_attr: The attributes of the address vector.
2416  *
2417  * The address handle is used to reference a local or global destination
2418  * in all UD QP post sends.
2419  */
2420 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2421 
2422 /**
2423  * ib_init_ah_from_wc - Initializes address handle attributes from a
2424  *   work completion.
2425  * @device: Device on which the received message arrived.
2426  * @port_num: Port on which the received message arrived.
2427  * @wc: Work completion associated with the received message.
2428  * @grh: References the received global route header.  This parameter is
2429  *   ignored unless the work completion indicates that the GRH is valid.
2430  * @ah_attr: Returned attributes that can be used when creating an address
2431  *   handle for replying to the message.
2432  */
2433 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2434 		       const struct ib_wc *wc, const struct ib_grh *grh,
2435 		       struct ib_ah_attr *ah_attr);
2436 
2437 /**
2438  * ib_create_ah_from_wc - Creates an address handle associated with the
2439  *   sender of the specified work completion.
2440  * @pd: The protection domain associated with the address handle.
2441  * @wc: Work completion information associated with a received message.
2442  * @grh: References the received global route header.  This parameter is
2443  *   ignored unless the work completion indicates that the GRH is valid.
2444  * @port_num: The outbound port number to associate with the address.
2445  *
2446  * The address handle is used to reference a local or global destination
2447  * in all UD QP post sends.
2448  */
2449 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2450 				   const struct ib_grh *grh, u8 port_num);
2451 
2452 /**
2453  * ib_modify_ah - Modifies the address vector associated with an address
2454  *   handle.
2455  * @ah: The address handle to modify.
2456  * @ah_attr: The new address vector attributes to associate with the
2457  *   address handle.
2458  */
2459 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2460 
2461 /**
2462  * ib_query_ah - Queries the address vector associated with an address
2463  *   handle.
2464  * @ah: The address handle to query.
2465  * @ah_attr: The address vector attributes associated with the address
2466  *   handle.
2467  */
2468 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2469 
2470 /**
2471  * ib_destroy_ah - Destroys an address handle.
2472  * @ah: The address handle to destroy.
2473  */
2474 int ib_destroy_ah(struct ib_ah *ah);
2475 
2476 /**
2477  * ib_create_srq - Creates a SRQ associated with the specified protection
2478  *   domain.
2479  * @pd: The protection domain associated with the SRQ.
2480  * @srq_init_attr: A list of initial attributes required to create the
2481  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2482  *   the actual capabilities of the created SRQ.
2483  *
2484  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2485  * requested size of the SRQ, and set to the actual values allocated
2486  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2487  * will always be at least as large as the requested values.
2488  */
2489 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2490 			     struct ib_srq_init_attr *srq_init_attr);
2491 
2492 /**
2493  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2494  * @srq: The SRQ to modify.
2495  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2496  *   the current values of selected SRQ attributes are returned.
2497  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2498  *   are being modified.
2499  *
2500  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2501  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2502  * the number of receives queued drops below the limit.
2503  */
2504 int ib_modify_srq(struct ib_srq *srq,
2505 		  struct ib_srq_attr *srq_attr,
2506 		  enum ib_srq_attr_mask srq_attr_mask);
2507 
2508 /**
2509  * ib_query_srq - Returns the attribute list and current values for the
2510  *   specified SRQ.
2511  * @srq: The SRQ to query.
2512  * @srq_attr: The attributes of the specified SRQ.
2513  */
2514 int ib_query_srq(struct ib_srq *srq,
2515 		 struct ib_srq_attr *srq_attr);
2516 
2517 /**
2518  * ib_destroy_srq - Destroys the specified SRQ.
2519  * @srq: The SRQ to destroy.
2520  */
2521 int ib_destroy_srq(struct ib_srq *srq);
2522 
2523 /**
2524  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2525  * @srq: The SRQ to post the work request on.
2526  * @recv_wr: A list of work requests to post on the receive queue.
2527  * @bad_recv_wr: On an immediate failure, this parameter will reference
2528  *   the work request that failed to be posted on the QP.
2529  */
2530 static inline int ib_post_srq_recv(struct ib_srq *srq,
2531 				   struct ib_recv_wr *recv_wr,
2532 				   struct ib_recv_wr **bad_recv_wr)
2533 {
2534 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2535 }
2536 
2537 /**
2538  * ib_create_qp - Creates a QP associated with the specified protection
2539  *   domain.
2540  * @pd: The protection domain associated with the QP.
2541  * @qp_init_attr: A list of initial attributes required to create the
2542  *   QP.  If QP creation succeeds, then the attributes are updated to
2543  *   the actual capabilities of the created QP.
2544  */
2545 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2546 			   struct ib_qp_init_attr *qp_init_attr);
2547 
2548 /**
2549  * ib_modify_qp - Modifies the attributes for the specified QP and then
2550  *   transitions the QP to the given state.
2551  * @qp: The QP to modify.
2552  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2553  *   the current values of selected QP attributes are returned.
2554  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2555  *   are being modified.
2556  */
2557 int ib_modify_qp(struct ib_qp *qp,
2558 		 struct ib_qp_attr *qp_attr,
2559 		 int qp_attr_mask);
2560 
2561 /**
2562  * ib_query_qp - Returns the attribute list and current values for the
2563  *   specified QP.
2564  * @qp: The QP to query.
2565  * @qp_attr: The attributes of the specified QP.
2566  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2567  * @qp_init_attr: Additional attributes of the selected QP.
2568  *
2569  * The qp_attr_mask may be used to limit the query to gathering only the
2570  * selected attributes.
2571  */
2572 int ib_query_qp(struct ib_qp *qp,
2573 		struct ib_qp_attr *qp_attr,
2574 		int qp_attr_mask,
2575 		struct ib_qp_init_attr *qp_init_attr);
2576 
2577 /**
2578  * ib_destroy_qp - Destroys the specified QP.
2579  * @qp: The QP to destroy.
2580  */
2581 int ib_destroy_qp(struct ib_qp *qp);
2582 
2583 /**
2584  * ib_open_qp - Obtain a reference to an existing sharable QP.
2585  * @xrcd - XRC domain
2586  * @qp_open_attr: Attributes identifying the QP to open.
2587  *
2588  * Returns a reference to a sharable QP.
2589  */
2590 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2591 			 struct ib_qp_open_attr *qp_open_attr);
2592 
2593 /**
2594  * ib_close_qp - Release an external reference to a QP.
2595  * @qp: The QP handle to release
2596  *
2597  * The opened QP handle is released by the caller.  The underlying
2598  * shared QP is not destroyed until all internal references are released.
2599  */
2600 int ib_close_qp(struct ib_qp *qp);
2601 
2602 /**
2603  * ib_post_send - Posts a list of work requests to the send queue of
2604  *   the specified QP.
2605  * @qp: The QP to post the work request on.
2606  * @send_wr: A list of work requests to post on the send queue.
2607  * @bad_send_wr: On an immediate failure, this parameter will reference
2608  *   the work request that failed to be posted on the QP.
2609  *
2610  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2611  * error is returned, the QP state shall not be affected,
2612  * ib_post_send() will return an immediate error after queueing any
2613  * earlier work requests in the list.
2614  */
2615 static inline int ib_post_send(struct ib_qp *qp,
2616 			       struct ib_send_wr *send_wr,
2617 			       struct ib_send_wr **bad_send_wr)
2618 {
2619 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2620 }
2621 
2622 /**
2623  * ib_post_recv - Posts a list of work requests to the receive queue of
2624  *   the specified QP.
2625  * @qp: The QP to post the work request on.
2626  * @recv_wr: A list of work requests to post on the receive queue.
2627  * @bad_recv_wr: On an immediate failure, this parameter will reference
2628  *   the work request that failed to be posted on the QP.
2629  */
2630 static inline int ib_post_recv(struct ib_qp *qp,
2631 			       struct ib_recv_wr *recv_wr,
2632 			       struct ib_recv_wr **bad_recv_wr)
2633 {
2634 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2635 }
2636 
2637 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2638 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2639 void ib_free_cq(struct ib_cq *cq);
2640 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2641 
2642 /**
2643  * ib_create_cq - Creates a CQ on the specified device.
2644  * @device: The device on which to create the CQ.
2645  * @comp_handler: A user-specified callback that is invoked when a
2646  *   completion event occurs on the CQ.
2647  * @event_handler: A user-specified callback that is invoked when an
2648  *   asynchronous event not associated with a completion occurs on the CQ.
2649  * @cq_context: Context associated with the CQ returned to the user via
2650  *   the associated completion and event handlers.
2651  * @cq_attr: The attributes the CQ should be created upon.
2652  *
2653  * Users can examine the cq structure to determine the actual CQ size.
2654  */
2655 struct ib_cq *ib_create_cq(struct ib_device *device,
2656 			   ib_comp_handler comp_handler,
2657 			   void (*event_handler)(struct ib_event *, void *),
2658 			   void *cq_context,
2659 			   const struct ib_cq_init_attr *cq_attr);
2660 
2661 /**
2662  * ib_resize_cq - Modifies the capacity of the CQ.
2663  * @cq: The CQ to resize.
2664  * @cqe: The minimum size of the CQ.
2665  *
2666  * Users can examine the cq structure to determine the actual CQ size.
2667  */
2668 int ib_resize_cq(struct ib_cq *cq, int cqe);
2669 
2670 /**
2671  * ib_modify_cq - Modifies moderation params of the CQ
2672  * @cq: The CQ to modify.
2673  * @cq_count: number of CQEs that will trigger an event
2674  * @cq_period: max period of time in usec before triggering an event
2675  *
2676  */
2677 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2678 
2679 /**
2680  * ib_destroy_cq - Destroys the specified CQ.
2681  * @cq: The CQ to destroy.
2682  */
2683 int ib_destroy_cq(struct ib_cq *cq);
2684 
2685 /**
2686  * ib_poll_cq - poll a CQ for completion(s)
2687  * @cq:the CQ being polled
2688  * @num_entries:maximum number of completions to return
2689  * @wc:array of at least @num_entries &struct ib_wc where completions
2690  *   will be returned
2691  *
2692  * Poll a CQ for (possibly multiple) completions.  If the return value
2693  * is < 0, an error occurred.  If the return value is >= 0, it is the
2694  * number of completions returned.  If the return value is
2695  * non-negative and < num_entries, then the CQ was emptied.
2696  */
2697 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2698 			     struct ib_wc *wc)
2699 {
2700 	return cq->device->poll_cq(cq, num_entries, wc);
2701 }
2702 
2703 /**
2704  * ib_peek_cq - Returns the number of unreaped completions currently
2705  *   on the specified CQ.
2706  * @cq: The CQ to peek.
2707  * @wc_cnt: A minimum number of unreaped completions to check for.
2708  *
2709  * If the number of unreaped completions is greater than or equal to wc_cnt,
2710  * this function returns wc_cnt, otherwise, it returns the actual number of
2711  * unreaped completions.
2712  */
2713 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2714 
2715 /**
2716  * ib_req_notify_cq - Request completion notification on a CQ.
2717  * @cq: The CQ to generate an event for.
2718  * @flags:
2719  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2720  *   to request an event on the next solicited event or next work
2721  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2722  *   may also be |ed in to request a hint about missed events, as
2723  *   described below.
2724  *
2725  * Return Value:
2726  *    < 0 means an error occurred while requesting notification
2727  *   == 0 means notification was requested successfully, and if
2728  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2729  *        were missed and it is safe to wait for another event.  In
2730  *        this case is it guaranteed that any work completions added
2731  *        to the CQ since the last CQ poll will trigger a completion
2732  *        notification event.
2733  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2734  *        in.  It means that the consumer must poll the CQ again to
2735  *        make sure it is empty to avoid missing an event because of a
2736  *        race between requesting notification and an entry being
2737  *        added to the CQ.  This return value means it is possible
2738  *        (but not guaranteed) that a work completion has been added
2739  *        to the CQ since the last poll without triggering a
2740  *        completion notification event.
2741  */
2742 static inline int ib_req_notify_cq(struct ib_cq *cq,
2743 				   enum ib_cq_notify_flags flags)
2744 {
2745 	return cq->device->req_notify_cq(cq, flags);
2746 }
2747 
2748 /**
2749  * ib_req_ncomp_notif - Request completion notification when there are
2750  *   at least the specified number of unreaped completions on the CQ.
2751  * @cq: The CQ to generate an event for.
2752  * @wc_cnt: The number of unreaped completions that should be on the
2753  *   CQ before an event is generated.
2754  */
2755 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2756 {
2757 	return cq->device->req_ncomp_notif ?
2758 		cq->device->req_ncomp_notif(cq, wc_cnt) :
2759 		-ENOSYS;
2760 }
2761 
2762 /**
2763  * ib_get_dma_mr - Returns a memory region for system memory that is
2764  *   usable for DMA.
2765  * @pd: The protection domain associated with the memory region.
2766  * @mr_access_flags: Specifies the memory access rights.
2767  *
2768  * Note that the ib_dma_*() functions defined below must be used
2769  * to create/destroy addresses used with the Lkey or Rkey returned
2770  * by ib_get_dma_mr().
2771  */
2772 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2773 
2774 /**
2775  * ib_dma_mapping_error - check a DMA addr for error
2776  * @dev: The device for which the dma_addr was created
2777  * @dma_addr: The DMA address to check
2778  */
2779 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2780 {
2781 	if (dev->dma_ops)
2782 		return dev->dma_ops->mapping_error(dev, dma_addr);
2783 	return dma_mapping_error(dev->dma_device, dma_addr);
2784 }
2785 
2786 /**
2787  * ib_dma_map_single - Map a kernel virtual address to DMA address
2788  * @dev: The device for which the dma_addr is to be created
2789  * @cpu_addr: The kernel virtual address
2790  * @size: The size of the region in bytes
2791  * @direction: The direction of the DMA
2792  */
2793 static inline u64 ib_dma_map_single(struct ib_device *dev,
2794 				    void *cpu_addr, size_t size,
2795 				    enum dma_data_direction direction)
2796 {
2797 	if (dev->dma_ops)
2798 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2799 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2800 }
2801 
2802 /**
2803  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2804  * @dev: The device for which the DMA address was created
2805  * @addr: The DMA address
2806  * @size: The size of the region in bytes
2807  * @direction: The direction of the DMA
2808  */
2809 static inline void ib_dma_unmap_single(struct ib_device *dev,
2810 				       u64 addr, size_t size,
2811 				       enum dma_data_direction direction)
2812 {
2813 	if (dev->dma_ops)
2814 		dev->dma_ops->unmap_single(dev, addr, size, direction);
2815 	else
2816 		dma_unmap_single(dev->dma_device, addr, size, direction);
2817 }
2818 
2819 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2820 					  void *cpu_addr, size_t size,
2821 					  enum dma_data_direction direction,
2822 					  struct dma_attrs *attrs)
2823 {
2824 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2825 				    direction, attrs);
2826 }
2827 
2828 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2829 					     u64 addr, size_t size,
2830 					     enum dma_data_direction direction,
2831 					     struct dma_attrs *attrs)
2832 {
2833 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
2834 				      direction, attrs);
2835 }
2836 
2837 /**
2838  * ib_dma_map_page - Map a physical page to DMA address
2839  * @dev: The device for which the dma_addr is to be created
2840  * @page: The page to be mapped
2841  * @offset: The offset within the page
2842  * @size: The size of the region in bytes
2843  * @direction: The direction of the DMA
2844  */
2845 static inline u64 ib_dma_map_page(struct ib_device *dev,
2846 				  struct page *page,
2847 				  unsigned long offset,
2848 				  size_t size,
2849 					 enum dma_data_direction direction)
2850 {
2851 	if (dev->dma_ops)
2852 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
2853 	return dma_map_page(dev->dma_device, page, offset, size, direction);
2854 }
2855 
2856 /**
2857  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2858  * @dev: The device for which the DMA address was created
2859  * @addr: The DMA address
2860  * @size: The size of the region in bytes
2861  * @direction: The direction of the DMA
2862  */
2863 static inline void ib_dma_unmap_page(struct ib_device *dev,
2864 				     u64 addr, size_t size,
2865 				     enum dma_data_direction direction)
2866 {
2867 	if (dev->dma_ops)
2868 		dev->dma_ops->unmap_page(dev, addr, size, direction);
2869 	else
2870 		dma_unmap_page(dev->dma_device, addr, size, direction);
2871 }
2872 
2873 /**
2874  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2875  * @dev: The device for which the DMA addresses are to be created
2876  * @sg: The array of scatter/gather entries
2877  * @nents: The number of scatter/gather entries
2878  * @direction: The direction of the DMA
2879  */
2880 static inline int ib_dma_map_sg(struct ib_device *dev,
2881 				struct scatterlist *sg, int nents,
2882 				enum dma_data_direction direction)
2883 {
2884 	if (dev->dma_ops)
2885 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
2886 	return dma_map_sg(dev->dma_device, sg, nents, direction);
2887 }
2888 
2889 /**
2890  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2891  * @dev: The device for which the DMA addresses were created
2892  * @sg: The array of scatter/gather entries
2893  * @nents: The number of scatter/gather entries
2894  * @direction: The direction of the DMA
2895  */
2896 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2897 				   struct scatterlist *sg, int nents,
2898 				   enum dma_data_direction direction)
2899 {
2900 	if (dev->dma_ops)
2901 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2902 	else
2903 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
2904 }
2905 
2906 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2907 				      struct scatterlist *sg, int nents,
2908 				      enum dma_data_direction direction,
2909 				      struct dma_attrs *attrs)
2910 {
2911 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2912 }
2913 
2914 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2915 					 struct scatterlist *sg, int nents,
2916 					 enum dma_data_direction direction,
2917 					 struct dma_attrs *attrs)
2918 {
2919 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2920 }
2921 /**
2922  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2923  * @dev: The device for which the DMA addresses were created
2924  * @sg: The scatter/gather entry
2925  *
2926  * Note: this function is obsolete. To do: change all occurrences of
2927  * ib_sg_dma_address() into sg_dma_address().
2928  */
2929 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2930 				    struct scatterlist *sg)
2931 {
2932 	return sg_dma_address(sg);
2933 }
2934 
2935 /**
2936  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2937  * @dev: The device for which the DMA addresses were created
2938  * @sg: The scatter/gather entry
2939  *
2940  * Note: this function is obsolete. To do: change all occurrences of
2941  * ib_sg_dma_len() into sg_dma_len().
2942  */
2943 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2944 					 struct scatterlist *sg)
2945 {
2946 	return sg_dma_len(sg);
2947 }
2948 
2949 /**
2950  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2951  * @dev: The device for which the DMA address was created
2952  * @addr: The DMA address
2953  * @size: The size of the region in bytes
2954  * @dir: The direction of the DMA
2955  */
2956 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2957 					      u64 addr,
2958 					      size_t size,
2959 					      enum dma_data_direction dir)
2960 {
2961 	if (dev->dma_ops)
2962 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2963 	else
2964 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2965 }
2966 
2967 /**
2968  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2969  * @dev: The device for which the DMA address was created
2970  * @addr: The DMA address
2971  * @size: The size of the region in bytes
2972  * @dir: The direction of the DMA
2973  */
2974 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2975 						 u64 addr,
2976 						 size_t size,
2977 						 enum dma_data_direction dir)
2978 {
2979 	if (dev->dma_ops)
2980 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2981 	else
2982 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2983 }
2984 
2985 /**
2986  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2987  * @dev: The device for which the DMA address is requested
2988  * @size: The size of the region to allocate in bytes
2989  * @dma_handle: A pointer for returning the DMA address of the region
2990  * @flag: memory allocator flags
2991  */
2992 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2993 					   size_t size,
2994 					   u64 *dma_handle,
2995 					   gfp_t flag)
2996 {
2997 	if (dev->dma_ops)
2998 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2999 	else {
3000 		dma_addr_t handle;
3001 		void *ret;
3002 
3003 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3004 		*dma_handle = handle;
3005 		return ret;
3006 	}
3007 }
3008 
3009 /**
3010  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3011  * @dev: The device for which the DMA addresses were allocated
3012  * @size: The size of the region
3013  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3014  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3015  */
3016 static inline void ib_dma_free_coherent(struct ib_device *dev,
3017 					size_t size, void *cpu_addr,
3018 					u64 dma_handle)
3019 {
3020 	if (dev->dma_ops)
3021 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3022 	else
3023 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3024 }
3025 
3026 /**
3027  * ib_dereg_mr - Deregisters a memory region and removes it from the
3028  *   HCA translation table.
3029  * @mr: The memory region to deregister.
3030  *
3031  * This function can fail, if the memory region has memory windows bound to it.
3032  */
3033 int ib_dereg_mr(struct ib_mr *mr);
3034 
3035 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3036 			  enum ib_mr_type mr_type,
3037 			  u32 max_num_sg);
3038 
3039 /**
3040  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3041  *   R_Key and L_Key.
3042  * @mr - struct ib_mr pointer to be updated.
3043  * @newkey - new key to be used.
3044  */
3045 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3046 {
3047 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3048 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3049 }
3050 
3051 /**
3052  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3053  * for calculating a new rkey for type 2 memory windows.
3054  * @rkey - the rkey to increment.
3055  */
3056 static inline u32 ib_inc_rkey(u32 rkey)
3057 {
3058 	const u32 mask = 0x000000ff;
3059 	return ((rkey + 1) & mask) | (rkey & ~mask);
3060 }
3061 
3062 /**
3063  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3064  * @pd: The protection domain associated with the unmapped region.
3065  * @mr_access_flags: Specifies the memory access rights.
3066  * @fmr_attr: Attributes of the unmapped region.
3067  *
3068  * A fast memory region must be mapped before it can be used as part of
3069  * a work request.
3070  */
3071 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3072 			    int mr_access_flags,
3073 			    struct ib_fmr_attr *fmr_attr);
3074 
3075 /**
3076  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3077  * @fmr: The fast memory region to associate with the pages.
3078  * @page_list: An array of physical pages to map to the fast memory region.
3079  * @list_len: The number of pages in page_list.
3080  * @iova: The I/O virtual address to use with the mapped region.
3081  */
3082 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3083 				  u64 *page_list, int list_len,
3084 				  u64 iova)
3085 {
3086 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3087 }
3088 
3089 /**
3090  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3091  * @fmr_list: A linked list of fast memory regions to unmap.
3092  */
3093 int ib_unmap_fmr(struct list_head *fmr_list);
3094 
3095 /**
3096  * ib_dealloc_fmr - Deallocates a fast memory region.
3097  * @fmr: The fast memory region to deallocate.
3098  */
3099 int ib_dealloc_fmr(struct ib_fmr *fmr);
3100 
3101 /**
3102  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3103  * @qp: QP to attach to the multicast group.  The QP must be type
3104  *   IB_QPT_UD.
3105  * @gid: Multicast group GID.
3106  * @lid: Multicast group LID in host byte order.
3107  *
3108  * In order to send and receive multicast packets, subnet
3109  * administration must have created the multicast group and configured
3110  * the fabric appropriately.  The port associated with the specified
3111  * QP must also be a member of the multicast group.
3112  */
3113 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3114 
3115 /**
3116  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3117  * @qp: QP to detach from the multicast group.
3118  * @gid: Multicast group GID.
3119  * @lid: Multicast group LID in host byte order.
3120  */
3121 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3122 
3123 /**
3124  * ib_alloc_xrcd - Allocates an XRC domain.
3125  * @device: The device on which to allocate the XRC domain.
3126  */
3127 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3128 
3129 /**
3130  * ib_dealloc_xrcd - Deallocates an XRC domain.
3131  * @xrcd: The XRC domain to deallocate.
3132  */
3133 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3134 
3135 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3136 			       struct ib_flow_attr *flow_attr, int domain);
3137 int ib_destroy_flow(struct ib_flow *flow_id);
3138 
3139 static inline int ib_check_mr_access(int flags)
3140 {
3141 	/*
3142 	 * Local write permission is required if remote write or
3143 	 * remote atomic permission is also requested.
3144 	 */
3145 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3146 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3147 		return -EINVAL;
3148 
3149 	return 0;
3150 }
3151 
3152 /**
3153  * ib_check_mr_status: lightweight check of MR status.
3154  *     This routine may provide status checks on a selected
3155  *     ib_mr. first use is for signature status check.
3156  *
3157  * @mr: A memory region.
3158  * @check_mask: Bitmask of which checks to perform from
3159  *     ib_mr_status_check enumeration.
3160  * @mr_status: The container of relevant status checks.
3161  *     failed checks will be indicated in the status bitmask
3162  *     and the relevant info shall be in the error item.
3163  */
3164 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3165 		       struct ib_mr_status *mr_status);
3166 
3167 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3168 					    u16 pkey, const union ib_gid *gid,
3169 					    const struct sockaddr *addr);
3170 
3171 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3172 		 unsigned int *sg_offset, unsigned int page_size);
3173 
3174 static inline int
3175 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3176 		  unsigned int *sg_offset, unsigned int page_size)
3177 {
3178 	int n;
3179 
3180 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3181 	mr->iova = 0;
3182 
3183 	return n;
3184 }
3185 
3186 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3187 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3188 
3189 void ib_drain_rq(struct ib_qp *qp);
3190 void ib_drain_sq(struct ib_qp *qp);
3191 void ib_drain_qp(struct ib_qp *qp);
3192 #endif /* IB_VERBS_H */
3193