xref: /openbmc/linux/include/rdma/ib_verbs.h (revision 56a0eccd)
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		= (1 << 31),
221 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
222 	IB_DEVICE_VIRTUAL_FUNCTION		= ((u64)1 << 33),
223 };
224 
225 enum ib_signature_prot_cap {
226 	IB_PROT_T10DIF_TYPE_1 = 1,
227 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
228 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
229 };
230 
231 enum ib_signature_guard_cap {
232 	IB_GUARD_T10DIF_CRC	= 1,
233 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
234 };
235 
236 enum ib_atomic_cap {
237 	IB_ATOMIC_NONE,
238 	IB_ATOMIC_HCA,
239 	IB_ATOMIC_GLOB
240 };
241 
242 enum ib_odp_general_cap_bits {
243 	IB_ODP_SUPPORT = 1 << 0,
244 };
245 
246 enum ib_odp_transport_cap_bits {
247 	IB_ODP_SUPPORT_SEND	= 1 << 0,
248 	IB_ODP_SUPPORT_RECV	= 1 << 1,
249 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
250 	IB_ODP_SUPPORT_READ	= 1 << 3,
251 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
252 };
253 
254 struct ib_odp_caps {
255 	uint64_t general_caps;
256 	struct {
257 		uint32_t  rc_odp_caps;
258 		uint32_t  uc_odp_caps;
259 		uint32_t  ud_odp_caps;
260 	} per_transport_caps;
261 };
262 
263 enum ib_cq_creation_flags {
264 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
265 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
266 };
267 
268 struct ib_cq_init_attr {
269 	unsigned int	cqe;
270 	int		comp_vector;
271 	u32		flags;
272 };
273 
274 struct ib_device_attr {
275 	u64			fw_ver;
276 	__be64			sys_image_guid;
277 	u64			max_mr_size;
278 	u64			page_size_cap;
279 	u32			vendor_id;
280 	u32			vendor_part_id;
281 	u32			hw_ver;
282 	int			max_qp;
283 	int			max_qp_wr;
284 	u64			device_cap_flags;
285 	int			max_sge;
286 	int			max_sge_rd;
287 	int			max_cq;
288 	int			max_cqe;
289 	int			max_mr;
290 	int			max_pd;
291 	int			max_qp_rd_atom;
292 	int			max_ee_rd_atom;
293 	int			max_res_rd_atom;
294 	int			max_qp_init_rd_atom;
295 	int			max_ee_init_rd_atom;
296 	enum ib_atomic_cap	atomic_cap;
297 	enum ib_atomic_cap	masked_atomic_cap;
298 	int			max_ee;
299 	int			max_rdd;
300 	int			max_mw;
301 	int			max_raw_ipv6_qp;
302 	int			max_raw_ethy_qp;
303 	int			max_mcast_grp;
304 	int			max_mcast_qp_attach;
305 	int			max_total_mcast_qp_attach;
306 	int			max_ah;
307 	int			max_fmr;
308 	int			max_map_per_fmr;
309 	int			max_srq;
310 	int			max_srq_wr;
311 	int			max_srq_sge;
312 	unsigned int		max_fast_reg_page_list_len;
313 	u16			max_pkeys;
314 	u8			local_ca_ack_delay;
315 	int			sig_prot_cap;
316 	int			sig_guard_cap;
317 	struct ib_odp_caps	odp_caps;
318 	uint64_t		timestamp_mask;
319 	uint64_t		hca_core_clock; /* in KHZ */
320 };
321 
322 enum ib_mtu {
323 	IB_MTU_256  = 1,
324 	IB_MTU_512  = 2,
325 	IB_MTU_1024 = 3,
326 	IB_MTU_2048 = 4,
327 	IB_MTU_4096 = 5
328 };
329 
330 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
331 {
332 	switch (mtu) {
333 	case IB_MTU_256:  return  256;
334 	case IB_MTU_512:  return  512;
335 	case IB_MTU_1024: return 1024;
336 	case IB_MTU_2048: return 2048;
337 	case IB_MTU_4096: return 4096;
338 	default: 	  return -1;
339 	}
340 }
341 
342 enum ib_port_state {
343 	IB_PORT_NOP		= 0,
344 	IB_PORT_DOWN		= 1,
345 	IB_PORT_INIT		= 2,
346 	IB_PORT_ARMED		= 3,
347 	IB_PORT_ACTIVE		= 4,
348 	IB_PORT_ACTIVE_DEFER	= 5
349 };
350 
351 enum ib_port_cap_flags {
352 	IB_PORT_SM				= 1 <<  1,
353 	IB_PORT_NOTICE_SUP			= 1 <<  2,
354 	IB_PORT_TRAP_SUP			= 1 <<  3,
355 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
356 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
357 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
358 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
359 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
360 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
361 	IB_PORT_SM_DISABLED			= 1 << 10,
362 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
363 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
364 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
365 	IB_PORT_CM_SUP				= 1 << 16,
366 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
367 	IB_PORT_REINIT_SUP			= 1 << 18,
368 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
369 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
370 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
371 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
372 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
373 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
374 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
375 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
376 };
377 
378 enum ib_port_width {
379 	IB_WIDTH_1X	= 1,
380 	IB_WIDTH_4X	= 2,
381 	IB_WIDTH_8X	= 4,
382 	IB_WIDTH_12X	= 8
383 };
384 
385 static inline int ib_width_enum_to_int(enum ib_port_width width)
386 {
387 	switch (width) {
388 	case IB_WIDTH_1X:  return  1;
389 	case IB_WIDTH_4X:  return  4;
390 	case IB_WIDTH_8X:  return  8;
391 	case IB_WIDTH_12X: return 12;
392 	default: 	  return -1;
393 	}
394 }
395 
396 enum ib_port_speed {
397 	IB_SPEED_SDR	= 1,
398 	IB_SPEED_DDR	= 2,
399 	IB_SPEED_QDR	= 4,
400 	IB_SPEED_FDR10	= 8,
401 	IB_SPEED_FDR	= 16,
402 	IB_SPEED_EDR	= 32
403 };
404 
405 struct ib_protocol_stats {
406 	/* TBD... */
407 };
408 
409 struct iw_protocol_stats {
410 	u64	ipInReceives;
411 	u64	ipInHdrErrors;
412 	u64	ipInTooBigErrors;
413 	u64	ipInNoRoutes;
414 	u64	ipInAddrErrors;
415 	u64	ipInUnknownProtos;
416 	u64	ipInTruncatedPkts;
417 	u64	ipInDiscards;
418 	u64	ipInDelivers;
419 	u64	ipOutForwDatagrams;
420 	u64	ipOutRequests;
421 	u64	ipOutDiscards;
422 	u64	ipOutNoRoutes;
423 	u64	ipReasmTimeout;
424 	u64	ipReasmReqds;
425 	u64	ipReasmOKs;
426 	u64	ipReasmFails;
427 	u64	ipFragOKs;
428 	u64	ipFragFails;
429 	u64	ipFragCreates;
430 	u64	ipInMcastPkts;
431 	u64	ipOutMcastPkts;
432 	u64	ipInBcastPkts;
433 	u64	ipOutBcastPkts;
434 
435 	u64	tcpRtoAlgorithm;
436 	u64	tcpRtoMin;
437 	u64	tcpRtoMax;
438 	u64	tcpMaxConn;
439 	u64	tcpActiveOpens;
440 	u64	tcpPassiveOpens;
441 	u64	tcpAttemptFails;
442 	u64	tcpEstabResets;
443 	u64	tcpCurrEstab;
444 	u64	tcpInSegs;
445 	u64	tcpOutSegs;
446 	u64	tcpRetransSegs;
447 	u64	tcpInErrs;
448 	u64	tcpOutRsts;
449 };
450 
451 union rdma_protocol_stats {
452 	struct ib_protocol_stats	ib;
453 	struct iw_protocol_stats	iw;
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 enum ib_sig_type {
937 	IB_SIGNAL_ALL_WR,
938 	IB_SIGNAL_REQ_WR
939 };
940 
941 enum ib_qp_type {
942 	/*
943 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
944 	 * here (and in that order) since the MAD layer uses them as
945 	 * indices into a 2-entry table.
946 	 */
947 	IB_QPT_SMI,
948 	IB_QPT_GSI,
949 
950 	IB_QPT_RC,
951 	IB_QPT_UC,
952 	IB_QPT_UD,
953 	IB_QPT_RAW_IPV6,
954 	IB_QPT_RAW_ETHERTYPE,
955 	IB_QPT_RAW_PACKET = 8,
956 	IB_QPT_XRC_INI = 9,
957 	IB_QPT_XRC_TGT,
958 	IB_QPT_MAX,
959 	/* Reserve a range for qp types internal to the low level driver.
960 	 * These qp types will not be visible at the IB core layer, so the
961 	 * IB_QPT_MAX usages should not be affected in the core layer
962 	 */
963 	IB_QPT_RESERVED1 = 0x1000,
964 	IB_QPT_RESERVED2,
965 	IB_QPT_RESERVED3,
966 	IB_QPT_RESERVED4,
967 	IB_QPT_RESERVED5,
968 	IB_QPT_RESERVED6,
969 	IB_QPT_RESERVED7,
970 	IB_QPT_RESERVED8,
971 	IB_QPT_RESERVED9,
972 	IB_QPT_RESERVED10,
973 };
974 
975 enum ib_qp_create_flags {
976 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
977 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
978 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
979 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
980 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
981 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
982 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
983 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
984 	/* reserve bits 26-31 for low level drivers' internal use */
985 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
986 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
987 };
988 
989 /*
990  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
991  * callback to destroy the passed in QP.
992  */
993 
994 struct ib_qp_init_attr {
995 	void                  (*event_handler)(struct ib_event *, void *);
996 	void		       *qp_context;
997 	struct ib_cq	       *send_cq;
998 	struct ib_cq	       *recv_cq;
999 	struct ib_srq	       *srq;
1000 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1001 	struct ib_qp_cap	cap;
1002 	enum ib_sig_type	sq_sig_type;
1003 	enum ib_qp_type		qp_type;
1004 	enum ib_qp_create_flags	create_flags;
1005 	u8			port_num; /* special QP types only */
1006 };
1007 
1008 struct ib_qp_open_attr {
1009 	void                  (*event_handler)(struct ib_event *, void *);
1010 	void		       *qp_context;
1011 	u32			qp_num;
1012 	enum ib_qp_type		qp_type;
1013 };
1014 
1015 enum ib_rnr_timeout {
1016 	IB_RNR_TIMER_655_36 =  0,
1017 	IB_RNR_TIMER_000_01 =  1,
1018 	IB_RNR_TIMER_000_02 =  2,
1019 	IB_RNR_TIMER_000_03 =  3,
1020 	IB_RNR_TIMER_000_04 =  4,
1021 	IB_RNR_TIMER_000_06 =  5,
1022 	IB_RNR_TIMER_000_08 =  6,
1023 	IB_RNR_TIMER_000_12 =  7,
1024 	IB_RNR_TIMER_000_16 =  8,
1025 	IB_RNR_TIMER_000_24 =  9,
1026 	IB_RNR_TIMER_000_32 = 10,
1027 	IB_RNR_TIMER_000_48 = 11,
1028 	IB_RNR_TIMER_000_64 = 12,
1029 	IB_RNR_TIMER_000_96 = 13,
1030 	IB_RNR_TIMER_001_28 = 14,
1031 	IB_RNR_TIMER_001_92 = 15,
1032 	IB_RNR_TIMER_002_56 = 16,
1033 	IB_RNR_TIMER_003_84 = 17,
1034 	IB_RNR_TIMER_005_12 = 18,
1035 	IB_RNR_TIMER_007_68 = 19,
1036 	IB_RNR_TIMER_010_24 = 20,
1037 	IB_RNR_TIMER_015_36 = 21,
1038 	IB_RNR_TIMER_020_48 = 22,
1039 	IB_RNR_TIMER_030_72 = 23,
1040 	IB_RNR_TIMER_040_96 = 24,
1041 	IB_RNR_TIMER_061_44 = 25,
1042 	IB_RNR_TIMER_081_92 = 26,
1043 	IB_RNR_TIMER_122_88 = 27,
1044 	IB_RNR_TIMER_163_84 = 28,
1045 	IB_RNR_TIMER_245_76 = 29,
1046 	IB_RNR_TIMER_327_68 = 30,
1047 	IB_RNR_TIMER_491_52 = 31
1048 };
1049 
1050 enum ib_qp_attr_mask {
1051 	IB_QP_STATE			= 1,
1052 	IB_QP_CUR_STATE			= (1<<1),
1053 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1054 	IB_QP_ACCESS_FLAGS		= (1<<3),
1055 	IB_QP_PKEY_INDEX		= (1<<4),
1056 	IB_QP_PORT			= (1<<5),
1057 	IB_QP_QKEY			= (1<<6),
1058 	IB_QP_AV			= (1<<7),
1059 	IB_QP_PATH_MTU			= (1<<8),
1060 	IB_QP_TIMEOUT			= (1<<9),
1061 	IB_QP_RETRY_CNT			= (1<<10),
1062 	IB_QP_RNR_RETRY			= (1<<11),
1063 	IB_QP_RQ_PSN			= (1<<12),
1064 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1065 	IB_QP_ALT_PATH			= (1<<14),
1066 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1067 	IB_QP_SQ_PSN			= (1<<16),
1068 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1069 	IB_QP_PATH_MIG_STATE		= (1<<18),
1070 	IB_QP_CAP			= (1<<19),
1071 	IB_QP_DEST_QPN			= (1<<20),
1072 	IB_QP_RESERVED1			= (1<<21),
1073 	IB_QP_RESERVED2			= (1<<22),
1074 	IB_QP_RESERVED3			= (1<<23),
1075 	IB_QP_RESERVED4			= (1<<24),
1076 };
1077 
1078 enum ib_qp_state {
1079 	IB_QPS_RESET,
1080 	IB_QPS_INIT,
1081 	IB_QPS_RTR,
1082 	IB_QPS_RTS,
1083 	IB_QPS_SQD,
1084 	IB_QPS_SQE,
1085 	IB_QPS_ERR
1086 };
1087 
1088 enum ib_mig_state {
1089 	IB_MIG_MIGRATED,
1090 	IB_MIG_REARM,
1091 	IB_MIG_ARMED
1092 };
1093 
1094 enum ib_mw_type {
1095 	IB_MW_TYPE_1 = 1,
1096 	IB_MW_TYPE_2 = 2
1097 };
1098 
1099 struct ib_qp_attr {
1100 	enum ib_qp_state	qp_state;
1101 	enum ib_qp_state	cur_qp_state;
1102 	enum ib_mtu		path_mtu;
1103 	enum ib_mig_state	path_mig_state;
1104 	u32			qkey;
1105 	u32			rq_psn;
1106 	u32			sq_psn;
1107 	u32			dest_qp_num;
1108 	int			qp_access_flags;
1109 	struct ib_qp_cap	cap;
1110 	struct ib_ah_attr	ah_attr;
1111 	struct ib_ah_attr	alt_ah_attr;
1112 	u16			pkey_index;
1113 	u16			alt_pkey_index;
1114 	u8			en_sqd_async_notify;
1115 	u8			sq_draining;
1116 	u8			max_rd_atomic;
1117 	u8			max_dest_rd_atomic;
1118 	u8			min_rnr_timer;
1119 	u8			port_num;
1120 	u8			timeout;
1121 	u8			retry_cnt;
1122 	u8			rnr_retry;
1123 	u8			alt_port_num;
1124 	u8			alt_timeout;
1125 };
1126 
1127 enum ib_wr_opcode {
1128 	IB_WR_RDMA_WRITE,
1129 	IB_WR_RDMA_WRITE_WITH_IMM,
1130 	IB_WR_SEND,
1131 	IB_WR_SEND_WITH_IMM,
1132 	IB_WR_RDMA_READ,
1133 	IB_WR_ATOMIC_CMP_AND_SWP,
1134 	IB_WR_ATOMIC_FETCH_AND_ADD,
1135 	IB_WR_LSO,
1136 	IB_WR_SEND_WITH_INV,
1137 	IB_WR_RDMA_READ_WITH_INV,
1138 	IB_WR_LOCAL_INV,
1139 	IB_WR_REG_MR,
1140 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1141 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1142 	IB_WR_REG_SIG_MR,
1143 	/* reserve values for low level drivers' internal use.
1144 	 * These values will not be used at all in the ib core layer.
1145 	 */
1146 	IB_WR_RESERVED1 = 0xf0,
1147 	IB_WR_RESERVED2,
1148 	IB_WR_RESERVED3,
1149 	IB_WR_RESERVED4,
1150 	IB_WR_RESERVED5,
1151 	IB_WR_RESERVED6,
1152 	IB_WR_RESERVED7,
1153 	IB_WR_RESERVED8,
1154 	IB_WR_RESERVED9,
1155 	IB_WR_RESERVED10,
1156 };
1157 
1158 enum ib_send_flags {
1159 	IB_SEND_FENCE		= 1,
1160 	IB_SEND_SIGNALED	= (1<<1),
1161 	IB_SEND_SOLICITED	= (1<<2),
1162 	IB_SEND_INLINE		= (1<<3),
1163 	IB_SEND_IP_CSUM		= (1<<4),
1164 
1165 	/* reserve bits 26-31 for low level drivers' internal use */
1166 	IB_SEND_RESERVED_START	= (1 << 26),
1167 	IB_SEND_RESERVED_END	= (1 << 31),
1168 };
1169 
1170 struct ib_sge {
1171 	u64	addr;
1172 	u32	length;
1173 	u32	lkey;
1174 };
1175 
1176 struct ib_cqe {
1177 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1178 };
1179 
1180 struct ib_send_wr {
1181 	struct ib_send_wr      *next;
1182 	union {
1183 		u64		wr_id;
1184 		struct ib_cqe	*wr_cqe;
1185 	};
1186 	struct ib_sge	       *sg_list;
1187 	int			num_sge;
1188 	enum ib_wr_opcode	opcode;
1189 	int			send_flags;
1190 	union {
1191 		__be32		imm_data;
1192 		u32		invalidate_rkey;
1193 	} ex;
1194 };
1195 
1196 struct ib_rdma_wr {
1197 	struct ib_send_wr	wr;
1198 	u64			remote_addr;
1199 	u32			rkey;
1200 };
1201 
1202 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1203 {
1204 	return container_of(wr, struct ib_rdma_wr, wr);
1205 }
1206 
1207 struct ib_atomic_wr {
1208 	struct ib_send_wr	wr;
1209 	u64			remote_addr;
1210 	u64			compare_add;
1211 	u64			swap;
1212 	u64			compare_add_mask;
1213 	u64			swap_mask;
1214 	u32			rkey;
1215 };
1216 
1217 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1218 {
1219 	return container_of(wr, struct ib_atomic_wr, wr);
1220 }
1221 
1222 struct ib_ud_wr {
1223 	struct ib_send_wr	wr;
1224 	struct ib_ah		*ah;
1225 	void			*header;
1226 	int			hlen;
1227 	int			mss;
1228 	u32			remote_qpn;
1229 	u32			remote_qkey;
1230 	u16			pkey_index; /* valid for GSI only */
1231 	u8			port_num;   /* valid for DR SMPs on switch only */
1232 };
1233 
1234 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1235 {
1236 	return container_of(wr, struct ib_ud_wr, wr);
1237 }
1238 
1239 struct ib_reg_wr {
1240 	struct ib_send_wr	wr;
1241 	struct ib_mr		*mr;
1242 	u32			key;
1243 	int			access;
1244 };
1245 
1246 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1247 {
1248 	return container_of(wr, struct ib_reg_wr, wr);
1249 }
1250 
1251 struct ib_sig_handover_wr {
1252 	struct ib_send_wr	wr;
1253 	struct ib_sig_attrs    *sig_attrs;
1254 	struct ib_mr	       *sig_mr;
1255 	int			access_flags;
1256 	struct ib_sge	       *prot;
1257 };
1258 
1259 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1260 {
1261 	return container_of(wr, struct ib_sig_handover_wr, wr);
1262 }
1263 
1264 struct ib_recv_wr {
1265 	struct ib_recv_wr      *next;
1266 	union {
1267 		u64		wr_id;
1268 		struct ib_cqe	*wr_cqe;
1269 	};
1270 	struct ib_sge	       *sg_list;
1271 	int			num_sge;
1272 };
1273 
1274 enum ib_access_flags {
1275 	IB_ACCESS_LOCAL_WRITE	= 1,
1276 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1277 	IB_ACCESS_REMOTE_READ	= (1<<2),
1278 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1279 	IB_ACCESS_MW_BIND	= (1<<4),
1280 	IB_ZERO_BASED		= (1<<5),
1281 	IB_ACCESS_ON_DEMAND     = (1<<6),
1282 };
1283 
1284 /*
1285  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1286  * are hidden here instead of a uapi header!
1287  */
1288 enum ib_mr_rereg_flags {
1289 	IB_MR_REREG_TRANS	= 1,
1290 	IB_MR_REREG_PD		= (1<<1),
1291 	IB_MR_REREG_ACCESS	= (1<<2),
1292 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1293 };
1294 
1295 struct ib_fmr_attr {
1296 	int	max_pages;
1297 	int	max_maps;
1298 	u8	page_shift;
1299 };
1300 
1301 struct ib_umem;
1302 
1303 struct ib_ucontext {
1304 	struct ib_device       *device;
1305 	struct list_head	pd_list;
1306 	struct list_head	mr_list;
1307 	struct list_head	mw_list;
1308 	struct list_head	cq_list;
1309 	struct list_head	qp_list;
1310 	struct list_head	srq_list;
1311 	struct list_head	ah_list;
1312 	struct list_head	xrcd_list;
1313 	struct list_head	rule_list;
1314 	int			closing;
1315 
1316 	struct pid             *tgid;
1317 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1318 	struct rb_root      umem_tree;
1319 	/*
1320 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1321 	 * mmu notifiers registration.
1322 	 */
1323 	struct rw_semaphore	umem_rwsem;
1324 	void (*invalidate_range)(struct ib_umem *umem,
1325 				 unsigned long start, unsigned long end);
1326 
1327 	struct mmu_notifier	mn;
1328 	atomic_t		notifier_count;
1329 	/* A list of umems that don't have private mmu notifier counters yet. */
1330 	struct list_head	no_private_counters;
1331 	int                     odp_mrs_count;
1332 #endif
1333 };
1334 
1335 struct ib_uobject {
1336 	u64			user_handle;	/* handle given to us by userspace */
1337 	struct ib_ucontext     *context;	/* associated user context */
1338 	void		       *object;		/* containing object */
1339 	struct list_head	list;		/* link to context's list */
1340 	int			id;		/* index into kernel idr */
1341 	struct kref		ref;
1342 	struct rw_semaphore	mutex;		/* protects .live */
1343 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1344 	int			live;
1345 };
1346 
1347 struct ib_udata {
1348 	const void __user *inbuf;
1349 	void __user *outbuf;
1350 	size_t       inlen;
1351 	size_t       outlen;
1352 };
1353 
1354 struct ib_pd {
1355 	u32			local_dma_lkey;
1356 	struct ib_device       *device;
1357 	struct ib_uobject      *uobject;
1358 	atomic_t          	usecnt; /* count all resources */
1359 	struct ib_mr	       *local_mr;
1360 };
1361 
1362 struct ib_xrcd {
1363 	struct ib_device       *device;
1364 	atomic_t		usecnt; /* count all exposed resources */
1365 	struct inode	       *inode;
1366 
1367 	struct mutex		tgt_qp_mutex;
1368 	struct list_head	tgt_qp_list;
1369 };
1370 
1371 struct ib_ah {
1372 	struct ib_device	*device;
1373 	struct ib_pd		*pd;
1374 	struct ib_uobject	*uobject;
1375 };
1376 
1377 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1378 
1379 enum ib_poll_context {
1380 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1381 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1382 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1383 };
1384 
1385 struct ib_cq {
1386 	struct ib_device       *device;
1387 	struct ib_uobject      *uobject;
1388 	ib_comp_handler   	comp_handler;
1389 	void                  (*event_handler)(struct ib_event *, void *);
1390 	void                   *cq_context;
1391 	int               	cqe;
1392 	atomic_t          	usecnt; /* count number of work queues */
1393 	enum ib_poll_context	poll_ctx;
1394 	struct ib_wc		*wc;
1395 	union {
1396 		struct irq_poll		iop;
1397 		struct work_struct	work;
1398 	};
1399 };
1400 
1401 struct ib_srq {
1402 	struct ib_device       *device;
1403 	struct ib_pd	       *pd;
1404 	struct ib_uobject      *uobject;
1405 	void		      (*event_handler)(struct ib_event *, void *);
1406 	void		       *srq_context;
1407 	enum ib_srq_type	srq_type;
1408 	atomic_t		usecnt;
1409 
1410 	union {
1411 		struct {
1412 			struct ib_xrcd *xrcd;
1413 			struct ib_cq   *cq;
1414 			u32		srq_num;
1415 		} xrc;
1416 	} ext;
1417 };
1418 
1419 struct ib_qp {
1420 	struct ib_device       *device;
1421 	struct ib_pd	       *pd;
1422 	struct ib_cq	       *send_cq;
1423 	struct ib_cq	       *recv_cq;
1424 	struct ib_srq	       *srq;
1425 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1426 	struct list_head	xrcd_list;
1427 	/* count times opened, mcast attaches, flow attaches */
1428 	atomic_t		usecnt;
1429 	struct list_head	open_list;
1430 	struct ib_qp           *real_qp;
1431 	struct ib_uobject      *uobject;
1432 	void                  (*event_handler)(struct ib_event *, void *);
1433 	void		       *qp_context;
1434 	u32			qp_num;
1435 	enum ib_qp_type		qp_type;
1436 };
1437 
1438 struct ib_mr {
1439 	struct ib_device  *device;
1440 	struct ib_pd	  *pd;
1441 	struct ib_uobject *uobject;
1442 	u32		   lkey;
1443 	u32		   rkey;
1444 	u64		   iova;
1445 	u32		   length;
1446 	unsigned int	   page_size;
1447 };
1448 
1449 struct ib_mw {
1450 	struct ib_device	*device;
1451 	struct ib_pd		*pd;
1452 	struct ib_uobject	*uobject;
1453 	u32			rkey;
1454 	enum ib_mw_type         type;
1455 };
1456 
1457 struct ib_fmr {
1458 	struct ib_device	*device;
1459 	struct ib_pd		*pd;
1460 	struct list_head	list;
1461 	u32			lkey;
1462 	u32			rkey;
1463 };
1464 
1465 /* Supported steering options */
1466 enum ib_flow_attr_type {
1467 	/* steering according to rule specifications */
1468 	IB_FLOW_ATTR_NORMAL		= 0x0,
1469 	/* default unicast and multicast rule -
1470 	 * receive all Eth traffic which isn't steered to any QP
1471 	 */
1472 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1473 	/* default multicast rule -
1474 	 * receive all Eth multicast traffic which isn't steered to any QP
1475 	 */
1476 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1477 	/* sniffer rule - receive all port traffic */
1478 	IB_FLOW_ATTR_SNIFFER		= 0x3
1479 };
1480 
1481 /* Supported steering header types */
1482 enum ib_flow_spec_type {
1483 	/* L2 headers*/
1484 	IB_FLOW_SPEC_ETH	= 0x20,
1485 	IB_FLOW_SPEC_IB		= 0x22,
1486 	/* L3 header*/
1487 	IB_FLOW_SPEC_IPV4	= 0x30,
1488 	/* L4 headers*/
1489 	IB_FLOW_SPEC_TCP	= 0x40,
1490 	IB_FLOW_SPEC_UDP	= 0x41
1491 };
1492 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1493 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1494 
1495 /* Flow steering rule priority is set according to it's domain.
1496  * Lower domain value means higher priority.
1497  */
1498 enum ib_flow_domain {
1499 	IB_FLOW_DOMAIN_USER,
1500 	IB_FLOW_DOMAIN_ETHTOOL,
1501 	IB_FLOW_DOMAIN_RFS,
1502 	IB_FLOW_DOMAIN_NIC,
1503 	IB_FLOW_DOMAIN_NUM /* Must be last */
1504 };
1505 
1506 enum ib_flow_flags {
1507 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1508 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1509 };
1510 
1511 struct ib_flow_eth_filter {
1512 	u8	dst_mac[6];
1513 	u8	src_mac[6];
1514 	__be16	ether_type;
1515 	__be16	vlan_tag;
1516 };
1517 
1518 struct ib_flow_spec_eth {
1519 	enum ib_flow_spec_type	  type;
1520 	u16			  size;
1521 	struct ib_flow_eth_filter val;
1522 	struct ib_flow_eth_filter mask;
1523 };
1524 
1525 struct ib_flow_ib_filter {
1526 	__be16 dlid;
1527 	__u8   sl;
1528 };
1529 
1530 struct ib_flow_spec_ib {
1531 	enum ib_flow_spec_type	 type;
1532 	u16			 size;
1533 	struct ib_flow_ib_filter val;
1534 	struct ib_flow_ib_filter mask;
1535 };
1536 
1537 struct ib_flow_ipv4_filter {
1538 	__be32	src_ip;
1539 	__be32	dst_ip;
1540 };
1541 
1542 struct ib_flow_spec_ipv4 {
1543 	enum ib_flow_spec_type	   type;
1544 	u16			   size;
1545 	struct ib_flow_ipv4_filter val;
1546 	struct ib_flow_ipv4_filter mask;
1547 };
1548 
1549 struct ib_flow_tcp_udp_filter {
1550 	__be16	dst_port;
1551 	__be16	src_port;
1552 };
1553 
1554 struct ib_flow_spec_tcp_udp {
1555 	enum ib_flow_spec_type	      type;
1556 	u16			      size;
1557 	struct ib_flow_tcp_udp_filter val;
1558 	struct ib_flow_tcp_udp_filter mask;
1559 };
1560 
1561 union ib_flow_spec {
1562 	struct {
1563 		enum ib_flow_spec_type	type;
1564 		u16			size;
1565 	};
1566 	struct ib_flow_spec_eth		eth;
1567 	struct ib_flow_spec_ib		ib;
1568 	struct ib_flow_spec_ipv4        ipv4;
1569 	struct ib_flow_spec_tcp_udp	tcp_udp;
1570 };
1571 
1572 struct ib_flow_attr {
1573 	enum ib_flow_attr_type type;
1574 	u16	     size;
1575 	u16	     priority;
1576 	u32	     flags;
1577 	u8	     num_of_specs;
1578 	u8	     port;
1579 	/* Following are the optional layers according to user request
1580 	 * struct ib_flow_spec_xxx
1581 	 * struct ib_flow_spec_yyy
1582 	 */
1583 };
1584 
1585 struct ib_flow {
1586 	struct ib_qp		*qp;
1587 	struct ib_uobject	*uobject;
1588 };
1589 
1590 struct ib_mad_hdr;
1591 struct ib_grh;
1592 
1593 enum ib_process_mad_flags {
1594 	IB_MAD_IGNORE_MKEY	= 1,
1595 	IB_MAD_IGNORE_BKEY	= 2,
1596 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1597 };
1598 
1599 enum ib_mad_result {
1600 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1601 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1602 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1603 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1604 };
1605 
1606 #define IB_DEVICE_NAME_MAX 64
1607 
1608 struct ib_cache {
1609 	rwlock_t                lock;
1610 	struct ib_event_handler event_handler;
1611 	struct ib_pkey_cache  **pkey_cache;
1612 	struct ib_gid_table   **gid_cache;
1613 	u8                     *lmc_cache;
1614 };
1615 
1616 struct ib_dma_mapping_ops {
1617 	int		(*mapping_error)(struct ib_device *dev,
1618 					 u64 dma_addr);
1619 	u64		(*map_single)(struct ib_device *dev,
1620 				      void *ptr, size_t size,
1621 				      enum dma_data_direction direction);
1622 	void		(*unmap_single)(struct ib_device *dev,
1623 					u64 addr, size_t size,
1624 					enum dma_data_direction direction);
1625 	u64		(*map_page)(struct ib_device *dev,
1626 				    struct page *page, unsigned long offset,
1627 				    size_t size,
1628 				    enum dma_data_direction direction);
1629 	void		(*unmap_page)(struct ib_device *dev,
1630 				      u64 addr, size_t size,
1631 				      enum dma_data_direction direction);
1632 	int		(*map_sg)(struct ib_device *dev,
1633 				  struct scatterlist *sg, int nents,
1634 				  enum dma_data_direction direction);
1635 	void		(*unmap_sg)(struct ib_device *dev,
1636 				    struct scatterlist *sg, int nents,
1637 				    enum dma_data_direction direction);
1638 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1639 					       u64 dma_handle,
1640 					       size_t size,
1641 					       enum dma_data_direction dir);
1642 	void		(*sync_single_for_device)(struct ib_device *dev,
1643 						  u64 dma_handle,
1644 						  size_t size,
1645 						  enum dma_data_direction dir);
1646 	void		*(*alloc_coherent)(struct ib_device *dev,
1647 					   size_t size,
1648 					   u64 *dma_handle,
1649 					   gfp_t flag);
1650 	void		(*free_coherent)(struct ib_device *dev,
1651 					 size_t size, void *cpu_addr,
1652 					 u64 dma_handle);
1653 };
1654 
1655 struct iw_cm_verbs;
1656 
1657 struct ib_port_immutable {
1658 	int                           pkey_tbl_len;
1659 	int                           gid_tbl_len;
1660 	u32                           core_cap_flags;
1661 	u32                           max_mad_size;
1662 };
1663 
1664 struct ib_device {
1665 	struct device                *dma_device;
1666 
1667 	char                          name[IB_DEVICE_NAME_MAX];
1668 
1669 	struct list_head              event_handler_list;
1670 	spinlock_t                    event_handler_lock;
1671 
1672 	spinlock_t                    client_data_lock;
1673 	struct list_head              core_list;
1674 	/* Access to the client_data_list is protected by the client_data_lock
1675 	 * spinlock and the lists_rwsem read-write semaphore */
1676 	struct list_head              client_data_list;
1677 
1678 	struct ib_cache               cache;
1679 	/**
1680 	 * port_immutable is indexed by port number
1681 	 */
1682 	struct ib_port_immutable     *port_immutable;
1683 
1684 	int			      num_comp_vectors;
1685 
1686 	struct iw_cm_verbs	     *iwcm;
1687 
1688 	int		           (*get_protocol_stats)(struct ib_device *device,
1689 							 union rdma_protocol_stats *stats);
1690 	int		           (*query_device)(struct ib_device *device,
1691 						   struct ib_device_attr *device_attr,
1692 						   struct ib_udata *udata);
1693 	int		           (*query_port)(struct ib_device *device,
1694 						 u8 port_num,
1695 						 struct ib_port_attr *port_attr);
1696 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1697 						     u8 port_num);
1698 	/* When calling get_netdev, the HW vendor's driver should return the
1699 	 * net device of device @device at port @port_num or NULL if such
1700 	 * a net device doesn't exist. The vendor driver should call dev_hold
1701 	 * on this net device. The HW vendor's device driver must guarantee
1702 	 * that this function returns NULL before the net device reaches
1703 	 * NETDEV_UNREGISTER_FINAL state.
1704 	 */
1705 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1706 						 u8 port_num);
1707 	int		           (*query_gid)(struct ib_device *device,
1708 						u8 port_num, int index,
1709 						union ib_gid *gid);
1710 	/* When calling add_gid, the HW vendor's driver should
1711 	 * add the gid of device @device at gid index @index of
1712 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1713 	 * the network device related to this gid is available
1714 	 * at @attr. @context allows the HW vendor driver to store extra
1715 	 * information together with a GID entry. The HW vendor may allocate
1716 	 * memory to contain this information and store it in @context when a
1717 	 * new GID entry is written to. Params are consistent until the next
1718 	 * call of add_gid or delete_gid. The function should return 0 on
1719 	 * success or error otherwise. The function could be called
1720 	 * concurrently for different ports. This function is only called
1721 	 * when roce_gid_table is used.
1722 	 */
1723 	int		           (*add_gid)(struct ib_device *device,
1724 					      u8 port_num,
1725 					      unsigned int index,
1726 					      const union ib_gid *gid,
1727 					      const struct ib_gid_attr *attr,
1728 					      void **context);
1729 	/* When calling del_gid, the HW vendor's driver should delete the
1730 	 * gid of device @device at gid index @index of port @port_num.
1731 	 * Upon the deletion of a GID entry, the HW vendor must free any
1732 	 * allocated memory. The caller will clear @context afterwards.
1733 	 * This function is only called when roce_gid_table is used.
1734 	 */
1735 	int		           (*del_gid)(struct ib_device *device,
1736 					      u8 port_num,
1737 					      unsigned int index,
1738 					      void **context);
1739 	int		           (*query_pkey)(struct ib_device *device,
1740 						 u8 port_num, u16 index, u16 *pkey);
1741 	int		           (*modify_device)(struct ib_device *device,
1742 						    int device_modify_mask,
1743 						    struct ib_device_modify *device_modify);
1744 	int		           (*modify_port)(struct ib_device *device,
1745 						  u8 port_num, int port_modify_mask,
1746 						  struct ib_port_modify *port_modify);
1747 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1748 						     struct ib_udata *udata);
1749 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1750 	int                        (*mmap)(struct ib_ucontext *context,
1751 					   struct vm_area_struct *vma);
1752 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1753 					       struct ib_ucontext *context,
1754 					       struct ib_udata *udata);
1755 	int                        (*dealloc_pd)(struct ib_pd *pd);
1756 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1757 						struct ib_ah_attr *ah_attr);
1758 	int                        (*modify_ah)(struct ib_ah *ah,
1759 						struct ib_ah_attr *ah_attr);
1760 	int                        (*query_ah)(struct ib_ah *ah,
1761 					       struct ib_ah_attr *ah_attr);
1762 	int                        (*destroy_ah)(struct ib_ah *ah);
1763 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1764 						 struct ib_srq_init_attr *srq_init_attr,
1765 						 struct ib_udata *udata);
1766 	int                        (*modify_srq)(struct ib_srq *srq,
1767 						 struct ib_srq_attr *srq_attr,
1768 						 enum ib_srq_attr_mask srq_attr_mask,
1769 						 struct ib_udata *udata);
1770 	int                        (*query_srq)(struct ib_srq *srq,
1771 						struct ib_srq_attr *srq_attr);
1772 	int                        (*destroy_srq)(struct ib_srq *srq);
1773 	int                        (*post_srq_recv)(struct ib_srq *srq,
1774 						    struct ib_recv_wr *recv_wr,
1775 						    struct ib_recv_wr **bad_recv_wr);
1776 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1777 						struct ib_qp_init_attr *qp_init_attr,
1778 						struct ib_udata *udata);
1779 	int                        (*modify_qp)(struct ib_qp *qp,
1780 						struct ib_qp_attr *qp_attr,
1781 						int qp_attr_mask,
1782 						struct ib_udata *udata);
1783 	int                        (*query_qp)(struct ib_qp *qp,
1784 					       struct ib_qp_attr *qp_attr,
1785 					       int qp_attr_mask,
1786 					       struct ib_qp_init_attr *qp_init_attr);
1787 	int                        (*destroy_qp)(struct ib_qp *qp);
1788 	int                        (*post_send)(struct ib_qp *qp,
1789 						struct ib_send_wr *send_wr,
1790 						struct ib_send_wr **bad_send_wr);
1791 	int                        (*post_recv)(struct ib_qp *qp,
1792 						struct ib_recv_wr *recv_wr,
1793 						struct ib_recv_wr **bad_recv_wr);
1794 	struct ib_cq *             (*create_cq)(struct ib_device *device,
1795 						const struct ib_cq_init_attr *attr,
1796 						struct ib_ucontext *context,
1797 						struct ib_udata *udata);
1798 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1799 						u16 cq_period);
1800 	int                        (*destroy_cq)(struct ib_cq *cq);
1801 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1802 						struct ib_udata *udata);
1803 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1804 					      struct ib_wc *wc);
1805 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1806 	int                        (*req_notify_cq)(struct ib_cq *cq,
1807 						    enum ib_cq_notify_flags flags);
1808 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1809 						      int wc_cnt);
1810 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1811 						 int mr_access_flags);
1812 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1813 						  u64 start, u64 length,
1814 						  u64 virt_addr,
1815 						  int mr_access_flags,
1816 						  struct ib_udata *udata);
1817 	int			   (*rereg_user_mr)(struct ib_mr *mr,
1818 						    int flags,
1819 						    u64 start, u64 length,
1820 						    u64 virt_addr,
1821 						    int mr_access_flags,
1822 						    struct ib_pd *pd,
1823 						    struct ib_udata *udata);
1824 	int                        (*dereg_mr)(struct ib_mr *mr);
1825 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
1826 					       enum ib_mr_type mr_type,
1827 					       u32 max_num_sg);
1828 	int                        (*map_mr_sg)(struct ib_mr *mr,
1829 						struct scatterlist *sg,
1830 						int sg_nents);
1831 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
1832 					       enum ib_mw_type type,
1833 					       struct ib_udata *udata);
1834 	int                        (*dealloc_mw)(struct ib_mw *mw);
1835 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1836 						int mr_access_flags,
1837 						struct ib_fmr_attr *fmr_attr);
1838 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1839 						   u64 *page_list, int list_len,
1840 						   u64 iova);
1841 	int		           (*unmap_fmr)(struct list_head *fmr_list);
1842 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1843 	int                        (*attach_mcast)(struct ib_qp *qp,
1844 						   union ib_gid *gid,
1845 						   u16 lid);
1846 	int                        (*detach_mcast)(struct ib_qp *qp,
1847 						   union ib_gid *gid,
1848 						   u16 lid);
1849 	int                        (*process_mad)(struct ib_device *device,
1850 						  int process_mad_flags,
1851 						  u8 port_num,
1852 						  const struct ib_wc *in_wc,
1853 						  const struct ib_grh *in_grh,
1854 						  const struct ib_mad_hdr *in_mad,
1855 						  size_t in_mad_size,
1856 						  struct ib_mad_hdr *out_mad,
1857 						  size_t *out_mad_size,
1858 						  u16 *out_mad_pkey_index);
1859 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
1860 						 struct ib_ucontext *ucontext,
1861 						 struct ib_udata *udata);
1862 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1863 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
1864 						  struct ib_flow_attr
1865 						  *flow_attr,
1866 						  int domain);
1867 	int			   (*destroy_flow)(struct ib_flow *flow_id);
1868 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1869 						      struct ib_mr_status *mr_status);
1870 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
1871 	void			   (*drain_rq)(struct ib_qp *qp);
1872 	void			   (*drain_sq)(struct ib_qp *qp);
1873 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
1874 							int state);
1875 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
1876 						   struct ifla_vf_info *ivf);
1877 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
1878 						   struct ifla_vf_stats *stats);
1879 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
1880 						  int type);
1881 
1882 	struct ib_dma_mapping_ops   *dma_ops;
1883 
1884 	struct module               *owner;
1885 	struct device                dev;
1886 	struct kobject               *ports_parent;
1887 	struct list_head             port_list;
1888 
1889 	enum {
1890 		IB_DEV_UNINITIALIZED,
1891 		IB_DEV_REGISTERED,
1892 		IB_DEV_UNREGISTERED
1893 	}                            reg_state;
1894 
1895 	int			     uverbs_abi_ver;
1896 	u64			     uverbs_cmd_mask;
1897 	u64			     uverbs_ex_cmd_mask;
1898 
1899 	char			     node_desc[64];
1900 	__be64			     node_guid;
1901 	u32			     local_dma_lkey;
1902 	u16                          is_switch:1;
1903 	u8                           node_type;
1904 	u8                           phys_port_cnt;
1905 	struct ib_device_attr        attrs;
1906 
1907 	/**
1908 	 * The following mandatory functions are used only at device
1909 	 * registration.  Keep functions such as these at the end of this
1910 	 * structure to avoid cache line misses when accessing struct ib_device
1911 	 * in fast paths.
1912 	 */
1913 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
1914 };
1915 
1916 struct ib_client {
1917 	char  *name;
1918 	void (*add)   (struct ib_device *);
1919 	void (*remove)(struct ib_device *, void *client_data);
1920 
1921 	/* Returns the net_dev belonging to this ib_client and matching the
1922 	 * given parameters.
1923 	 * @dev:	 An RDMA device that the net_dev use for communication.
1924 	 * @port:	 A physical port number on the RDMA device.
1925 	 * @pkey:	 P_Key that the net_dev uses if applicable.
1926 	 * @gid:	 A GID that the net_dev uses to communicate.
1927 	 * @addr:	 An IP address the net_dev is configured with.
1928 	 * @client_data: The device's client data set by ib_set_client_data().
1929 	 *
1930 	 * An ib_client that implements a net_dev on top of RDMA devices
1931 	 * (such as IP over IB) should implement this callback, allowing the
1932 	 * rdma_cm module to find the right net_dev for a given request.
1933 	 *
1934 	 * The caller is responsible for calling dev_put on the returned
1935 	 * netdev. */
1936 	struct net_device *(*get_net_dev_by_params)(
1937 			struct ib_device *dev,
1938 			u8 port,
1939 			u16 pkey,
1940 			const union ib_gid *gid,
1941 			const struct sockaddr *addr,
1942 			void *client_data);
1943 	struct list_head list;
1944 };
1945 
1946 struct ib_device *ib_alloc_device(size_t size);
1947 void ib_dealloc_device(struct ib_device *device);
1948 
1949 int ib_register_device(struct ib_device *device,
1950 		       int (*port_callback)(struct ib_device *,
1951 					    u8, struct kobject *));
1952 void ib_unregister_device(struct ib_device *device);
1953 
1954 int ib_register_client   (struct ib_client *client);
1955 void ib_unregister_client(struct ib_client *client);
1956 
1957 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1958 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
1959 			 void *data);
1960 
1961 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1962 {
1963 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1964 }
1965 
1966 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1967 {
1968 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1969 }
1970 
1971 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
1972 				       size_t offset,
1973 				       size_t len)
1974 {
1975 	const void __user *p = udata->inbuf + offset;
1976 	bool ret = false;
1977 	u8 *buf;
1978 
1979 	if (len > USHRT_MAX)
1980 		return false;
1981 
1982 	buf = kmalloc(len, GFP_KERNEL);
1983 	if (!buf)
1984 		return false;
1985 
1986 	if (copy_from_user(buf, p, len))
1987 		goto free;
1988 
1989 	ret = !memchr_inv(buf, 0, len);
1990 
1991 free:
1992 	kfree(buf);
1993 	return ret;
1994 }
1995 
1996 /**
1997  * ib_modify_qp_is_ok - Check that the supplied attribute mask
1998  * contains all required attributes and no attributes not allowed for
1999  * the given QP state transition.
2000  * @cur_state: Current QP state
2001  * @next_state: Next QP state
2002  * @type: QP type
2003  * @mask: Mask of supplied QP attributes
2004  * @ll : link layer of port
2005  *
2006  * This function is a helper function that a low-level driver's
2007  * modify_qp method can use to validate the consumer's input.  It
2008  * checks that cur_state and next_state are valid QP states, that a
2009  * transition from cur_state to next_state is allowed by the IB spec,
2010  * and that the attribute mask supplied is allowed for the transition.
2011  */
2012 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2013 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2014 		       enum rdma_link_layer ll);
2015 
2016 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2017 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2018 void ib_dispatch_event(struct ib_event *event);
2019 
2020 int ib_query_port(struct ib_device *device,
2021 		  u8 port_num, struct ib_port_attr *port_attr);
2022 
2023 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2024 					       u8 port_num);
2025 
2026 /**
2027  * rdma_cap_ib_switch - Check if the device is IB switch
2028  * @device: Device to check
2029  *
2030  * Device driver is responsible for setting is_switch bit on
2031  * in ib_device structure at init time.
2032  *
2033  * Return: true if the device is IB switch.
2034  */
2035 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2036 {
2037 	return device->is_switch;
2038 }
2039 
2040 /**
2041  * rdma_start_port - Return the first valid port number for the device
2042  * specified
2043  *
2044  * @device: Device to be checked
2045  *
2046  * Return start port number
2047  */
2048 static inline u8 rdma_start_port(const struct ib_device *device)
2049 {
2050 	return rdma_cap_ib_switch(device) ? 0 : 1;
2051 }
2052 
2053 /**
2054  * rdma_end_port - Return the last valid port number for the device
2055  * specified
2056  *
2057  * @device: Device to be checked
2058  *
2059  * Return last port number
2060  */
2061 static inline u8 rdma_end_port(const struct ib_device *device)
2062 {
2063 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2064 }
2065 
2066 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2067 {
2068 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2069 }
2070 
2071 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2072 {
2073 	return device->port_immutable[port_num].core_cap_flags &
2074 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2075 }
2076 
2077 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2078 {
2079 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2080 }
2081 
2082 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2083 {
2084 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2085 }
2086 
2087 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2088 {
2089 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2090 }
2091 
2092 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2093 {
2094 	return rdma_protocol_ib(device, port_num) ||
2095 		rdma_protocol_roce(device, port_num);
2096 }
2097 
2098 /**
2099  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2100  * Management Datagrams.
2101  * @device: Device to check
2102  * @port_num: Port number to check
2103  *
2104  * Management Datagrams (MAD) are a required part of the InfiniBand
2105  * specification and are supported on all InfiniBand devices.  A slightly
2106  * extended version are also supported on OPA interfaces.
2107  *
2108  * Return: true if the port supports sending/receiving of MAD packets.
2109  */
2110 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2111 {
2112 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2113 }
2114 
2115 /**
2116  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2117  * Management Datagrams.
2118  * @device: Device to check
2119  * @port_num: Port number to check
2120  *
2121  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2122  * datagrams with their own versions.  These OPA MADs share many but not all of
2123  * the characteristics of InfiniBand MADs.
2124  *
2125  * OPA MADs differ in the following ways:
2126  *
2127  *    1) MADs are variable size up to 2K
2128  *       IBTA defined MADs remain fixed at 256 bytes
2129  *    2) OPA SMPs must carry valid PKeys
2130  *    3) OPA SMP packets are a different format
2131  *
2132  * Return: true if the port supports OPA MAD packet formats.
2133  */
2134 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2135 {
2136 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2137 		== RDMA_CORE_CAP_OPA_MAD;
2138 }
2139 
2140 /**
2141  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2142  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2143  * @device: Device to check
2144  * @port_num: Port number to check
2145  *
2146  * Each InfiniBand node is required to provide a Subnet Management Agent
2147  * that the subnet manager can access.  Prior to the fabric being fully
2148  * configured by the subnet manager, the SMA is accessed via a well known
2149  * interface called the Subnet Management Interface (SMI).  This interface
2150  * uses directed route packets to communicate with the SM to get around the
2151  * chicken and egg problem of the SM needing to know what's on the fabric
2152  * in order to configure the fabric, and needing to configure the fabric in
2153  * order to send packets to the devices on the fabric.  These directed
2154  * route packets do not need the fabric fully configured in order to reach
2155  * their destination.  The SMI is the only method allowed to send
2156  * directed route packets on an InfiniBand fabric.
2157  *
2158  * Return: true if the port provides an SMI.
2159  */
2160 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2161 {
2162 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2163 }
2164 
2165 /**
2166  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2167  * Communication Manager.
2168  * @device: Device to check
2169  * @port_num: Port number to check
2170  *
2171  * The InfiniBand Communication Manager is one of many pre-defined General
2172  * Service Agents (GSA) that are accessed via the General Service
2173  * Interface (GSI).  It's role is to facilitate establishment of connections
2174  * between nodes as well as other management related tasks for established
2175  * connections.
2176  *
2177  * Return: true if the port supports an IB CM (this does not guarantee that
2178  * a CM is actually running however).
2179  */
2180 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2181 {
2182 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2183 }
2184 
2185 /**
2186  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2187  * Communication Manager.
2188  * @device: Device to check
2189  * @port_num: Port number to check
2190  *
2191  * Similar to above, but specific to iWARP connections which have a different
2192  * managment protocol than InfiniBand.
2193  *
2194  * Return: true if the port supports an iWARP CM (this does not guarantee that
2195  * a CM is actually running however).
2196  */
2197 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2198 {
2199 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2200 }
2201 
2202 /**
2203  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2204  * Subnet Administration.
2205  * @device: Device to check
2206  * @port_num: Port number to check
2207  *
2208  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2209  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2210  * fabrics, devices should resolve routes to other hosts by contacting the
2211  * SA to query the proper route.
2212  *
2213  * Return: true if the port should act as a client to the fabric Subnet
2214  * Administration interface.  This does not imply that the SA service is
2215  * running locally.
2216  */
2217 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2218 {
2219 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2220 }
2221 
2222 /**
2223  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2224  * Multicast.
2225  * @device: Device to check
2226  * @port_num: Port number to check
2227  *
2228  * InfiniBand multicast registration is more complex than normal IPv4 or
2229  * IPv6 multicast registration.  Each Host Channel Adapter must register
2230  * with the Subnet Manager when it wishes to join a multicast group.  It
2231  * should do so only once regardless of how many queue pairs it subscribes
2232  * to this group.  And it should leave the group only after all queue pairs
2233  * attached to the group have been detached.
2234  *
2235  * Return: true if the port must undertake the additional adminstrative
2236  * overhead of registering/unregistering with the SM and tracking of the
2237  * total number of queue pairs attached to the multicast group.
2238  */
2239 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2240 {
2241 	return rdma_cap_ib_sa(device, port_num);
2242 }
2243 
2244 /**
2245  * rdma_cap_af_ib - Check if the port of device has the capability
2246  * Native Infiniband Address.
2247  * @device: Device to check
2248  * @port_num: Port number to check
2249  *
2250  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2251  * GID.  RoCE uses a different mechanism, but still generates a GID via
2252  * a prescribed mechanism and port specific data.
2253  *
2254  * Return: true if the port uses a GID address to identify devices on the
2255  * network.
2256  */
2257 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2258 {
2259 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2260 }
2261 
2262 /**
2263  * rdma_cap_eth_ah - Check if the port of device has the capability
2264  * Ethernet Address Handle.
2265  * @device: Device to check
2266  * @port_num: Port number to check
2267  *
2268  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2269  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2270  * port.  Normally, packet headers are generated by the sending host
2271  * adapter, but when sending connectionless datagrams, we must manually
2272  * inject the proper headers for the fabric we are communicating over.
2273  *
2274  * Return: true if we are running as a RoCE port and must force the
2275  * addition of a Global Route Header built from our Ethernet Address
2276  * Handle into our header list for connectionless packets.
2277  */
2278 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2279 {
2280 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2281 }
2282 
2283 /**
2284  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2285  *
2286  * @device: Device
2287  * @port_num: Port number
2288  *
2289  * This MAD size includes the MAD headers and MAD payload.  No other headers
2290  * are included.
2291  *
2292  * Return the max MAD size required by the Port.  Will return 0 if the port
2293  * does not support MADs
2294  */
2295 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2296 {
2297 	return device->port_immutable[port_num].max_mad_size;
2298 }
2299 
2300 /**
2301  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2302  * @device: Device to check
2303  * @port_num: Port number to check
2304  *
2305  * RoCE GID table mechanism manages the various GIDs for a device.
2306  *
2307  * NOTE: if allocating the port's GID table has failed, this call will still
2308  * return true, but any RoCE GID table API will fail.
2309  *
2310  * Return: true if the port uses RoCE GID table mechanism in order to manage
2311  * its GIDs.
2312  */
2313 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2314 					   u8 port_num)
2315 {
2316 	return rdma_protocol_roce(device, port_num) &&
2317 		device->add_gid && device->del_gid;
2318 }
2319 
2320 int ib_query_gid(struct ib_device *device,
2321 		 u8 port_num, int index, union ib_gid *gid,
2322 		 struct ib_gid_attr *attr);
2323 
2324 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2325 			 int state);
2326 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2327 		     struct ifla_vf_info *info);
2328 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2329 		    struct ifla_vf_stats *stats);
2330 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2331 		   int type);
2332 
2333 int ib_query_pkey(struct ib_device *device,
2334 		  u8 port_num, u16 index, u16 *pkey);
2335 
2336 int ib_modify_device(struct ib_device *device,
2337 		     int device_modify_mask,
2338 		     struct ib_device_modify *device_modify);
2339 
2340 int ib_modify_port(struct ib_device *device,
2341 		   u8 port_num, int port_modify_mask,
2342 		   struct ib_port_modify *port_modify);
2343 
2344 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2345 		enum ib_gid_type gid_type, struct net_device *ndev,
2346 		u8 *port_num, u16 *index);
2347 
2348 int ib_find_pkey(struct ib_device *device,
2349 		 u8 port_num, u16 pkey, u16 *index);
2350 
2351 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2352 
2353 void ib_dealloc_pd(struct ib_pd *pd);
2354 
2355 /**
2356  * ib_create_ah - Creates an address handle for the given address vector.
2357  * @pd: The protection domain associated with the address handle.
2358  * @ah_attr: The attributes of the address vector.
2359  *
2360  * The address handle is used to reference a local or global destination
2361  * in all UD QP post sends.
2362  */
2363 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2364 
2365 /**
2366  * ib_init_ah_from_wc - Initializes address handle attributes from a
2367  *   work completion.
2368  * @device: Device on which the received message arrived.
2369  * @port_num: Port on which the received message arrived.
2370  * @wc: Work completion associated with the received message.
2371  * @grh: References the received global route header.  This parameter is
2372  *   ignored unless the work completion indicates that the GRH is valid.
2373  * @ah_attr: Returned attributes that can be used when creating an address
2374  *   handle for replying to the message.
2375  */
2376 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2377 		       const struct ib_wc *wc, const struct ib_grh *grh,
2378 		       struct ib_ah_attr *ah_attr);
2379 
2380 /**
2381  * ib_create_ah_from_wc - Creates an address handle associated with the
2382  *   sender of the specified work completion.
2383  * @pd: The protection domain associated with the address handle.
2384  * @wc: Work completion information associated with a received message.
2385  * @grh: References the received global route header.  This parameter is
2386  *   ignored unless the work completion indicates that the GRH is valid.
2387  * @port_num: The outbound port number to associate with the address.
2388  *
2389  * The address handle is used to reference a local or global destination
2390  * in all UD QP post sends.
2391  */
2392 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2393 				   const struct ib_grh *grh, u8 port_num);
2394 
2395 /**
2396  * ib_modify_ah - Modifies the address vector associated with an address
2397  *   handle.
2398  * @ah: The address handle to modify.
2399  * @ah_attr: The new address vector attributes to associate with the
2400  *   address handle.
2401  */
2402 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2403 
2404 /**
2405  * ib_query_ah - Queries the address vector associated with an address
2406  *   handle.
2407  * @ah: The address handle to query.
2408  * @ah_attr: The address vector attributes associated with the address
2409  *   handle.
2410  */
2411 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2412 
2413 /**
2414  * ib_destroy_ah - Destroys an address handle.
2415  * @ah: The address handle to destroy.
2416  */
2417 int ib_destroy_ah(struct ib_ah *ah);
2418 
2419 /**
2420  * ib_create_srq - Creates a SRQ associated with the specified protection
2421  *   domain.
2422  * @pd: The protection domain associated with the SRQ.
2423  * @srq_init_attr: A list of initial attributes required to create the
2424  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2425  *   the actual capabilities of the created SRQ.
2426  *
2427  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2428  * requested size of the SRQ, and set to the actual values allocated
2429  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2430  * will always be at least as large as the requested values.
2431  */
2432 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2433 			     struct ib_srq_init_attr *srq_init_attr);
2434 
2435 /**
2436  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2437  * @srq: The SRQ to modify.
2438  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2439  *   the current values of selected SRQ attributes are returned.
2440  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2441  *   are being modified.
2442  *
2443  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2444  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2445  * the number of receives queued drops below the limit.
2446  */
2447 int ib_modify_srq(struct ib_srq *srq,
2448 		  struct ib_srq_attr *srq_attr,
2449 		  enum ib_srq_attr_mask srq_attr_mask);
2450 
2451 /**
2452  * ib_query_srq - Returns the attribute list and current values for the
2453  *   specified SRQ.
2454  * @srq: The SRQ to query.
2455  * @srq_attr: The attributes of the specified SRQ.
2456  */
2457 int ib_query_srq(struct ib_srq *srq,
2458 		 struct ib_srq_attr *srq_attr);
2459 
2460 /**
2461  * ib_destroy_srq - Destroys the specified SRQ.
2462  * @srq: The SRQ to destroy.
2463  */
2464 int ib_destroy_srq(struct ib_srq *srq);
2465 
2466 /**
2467  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2468  * @srq: The SRQ to post the work request on.
2469  * @recv_wr: A list of work requests to post on the receive queue.
2470  * @bad_recv_wr: On an immediate failure, this parameter will reference
2471  *   the work request that failed to be posted on the QP.
2472  */
2473 static inline int ib_post_srq_recv(struct ib_srq *srq,
2474 				   struct ib_recv_wr *recv_wr,
2475 				   struct ib_recv_wr **bad_recv_wr)
2476 {
2477 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2478 }
2479 
2480 /**
2481  * ib_create_qp - Creates a QP associated with the specified protection
2482  *   domain.
2483  * @pd: The protection domain associated with the QP.
2484  * @qp_init_attr: A list of initial attributes required to create the
2485  *   QP.  If QP creation succeeds, then the attributes are updated to
2486  *   the actual capabilities of the created QP.
2487  */
2488 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2489 			   struct ib_qp_init_attr *qp_init_attr);
2490 
2491 /**
2492  * ib_modify_qp - Modifies the attributes for the specified QP and then
2493  *   transitions the QP to the given state.
2494  * @qp: The QP to modify.
2495  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2496  *   the current values of selected QP attributes are returned.
2497  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2498  *   are being modified.
2499  */
2500 int ib_modify_qp(struct ib_qp *qp,
2501 		 struct ib_qp_attr *qp_attr,
2502 		 int qp_attr_mask);
2503 
2504 /**
2505  * ib_query_qp - Returns the attribute list and current values for the
2506  *   specified QP.
2507  * @qp: The QP to query.
2508  * @qp_attr: The attributes of the specified QP.
2509  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2510  * @qp_init_attr: Additional attributes of the selected QP.
2511  *
2512  * The qp_attr_mask may be used to limit the query to gathering only the
2513  * selected attributes.
2514  */
2515 int ib_query_qp(struct ib_qp *qp,
2516 		struct ib_qp_attr *qp_attr,
2517 		int qp_attr_mask,
2518 		struct ib_qp_init_attr *qp_init_attr);
2519 
2520 /**
2521  * ib_destroy_qp - Destroys the specified QP.
2522  * @qp: The QP to destroy.
2523  */
2524 int ib_destroy_qp(struct ib_qp *qp);
2525 
2526 /**
2527  * ib_open_qp - Obtain a reference to an existing sharable QP.
2528  * @xrcd - XRC domain
2529  * @qp_open_attr: Attributes identifying the QP to open.
2530  *
2531  * Returns a reference to a sharable QP.
2532  */
2533 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2534 			 struct ib_qp_open_attr *qp_open_attr);
2535 
2536 /**
2537  * ib_close_qp - Release an external reference to a QP.
2538  * @qp: The QP handle to release
2539  *
2540  * The opened QP handle is released by the caller.  The underlying
2541  * shared QP is not destroyed until all internal references are released.
2542  */
2543 int ib_close_qp(struct ib_qp *qp);
2544 
2545 /**
2546  * ib_post_send - Posts a list of work requests to the send queue of
2547  *   the specified QP.
2548  * @qp: The QP to post the work request on.
2549  * @send_wr: A list of work requests to post on the send queue.
2550  * @bad_send_wr: On an immediate failure, this parameter will reference
2551  *   the work request that failed to be posted on the QP.
2552  *
2553  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2554  * error is returned, the QP state shall not be affected,
2555  * ib_post_send() will return an immediate error after queueing any
2556  * earlier work requests in the list.
2557  */
2558 static inline int ib_post_send(struct ib_qp *qp,
2559 			       struct ib_send_wr *send_wr,
2560 			       struct ib_send_wr **bad_send_wr)
2561 {
2562 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2563 }
2564 
2565 /**
2566  * ib_post_recv - Posts a list of work requests to the receive queue of
2567  *   the specified QP.
2568  * @qp: The QP to post the work request on.
2569  * @recv_wr: A list of work requests to post on the receive queue.
2570  * @bad_recv_wr: On an immediate failure, this parameter will reference
2571  *   the work request that failed to be posted on the QP.
2572  */
2573 static inline int ib_post_recv(struct ib_qp *qp,
2574 			       struct ib_recv_wr *recv_wr,
2575 			       struct ib_recv_wr **bad_recv_wr)
2576 {
2577 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2578 }
2579 
2580 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2581 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2582 void ib_free_cq(struct ib_cq *cq);
2583 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2584 
2585 /**
2586  * ib_create_cq - Creates a CQ on the specified device.
2587  * @device: The device on which to create the CQ.
2588  * @comp_handler: A user-specified callback that is invoked when a
2589  *   completion event occurs on the CQ.
2590  * @event_handler: A user-specified callback that is invoked when an
2591  *   asynchronous event not associated with a completion occurs on the CQ.
2592  * @cq_context: Context associated with the CQ returned to the user via
2593  *   the associated completion and event handlers.
2594  * @cq_attr: The attributes the CQ should be created upon.
2595  *
2596  * Users can examine the cq structure to determine the actual CQ size.
2597  */
2598 struct ib_cq *ib_create_cq(struct ib_device *device,
2599 			   ib_comp_handler comp_handler,
2600 			   void (*event_handler)(struct ib_event *, void *),
2601 			   void *cq_context,
2602 			   const struct ib_cq_init_attr *cq_attr);
2603 
2604 /**
2605  * ib_resize_cq - Modifies the capacity of the CQ.
2606  * @cq: The CQ to resize.
2607  * @cqe: The minimum size of the CQ.
2608  *
2609  * Users can examine the cq structure to determine the actual CQ size.
2610  */
2611 int ib_resize_cq(struct ib_cq *cq, int cqe);
2612 
2613 /**
2614  * ib_modify_cq - Modifies moderation params of the CQ
2615  * @cq: The CQ to modify.
2616  * @cq_count: number of CQEs that will trigger an event
2617  * @cq_period: max period of time in usec before triggering an event
2618  *
2619  */
2620 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2621 
2622 /**
2623  * ib_destroy_cq - Destroys the specified CQ.
2624  * @cq: The CQ to destroy.
2625  */
2626 int ib_destroy_cq(struct ib_cq *cq);
2627 
2628 /**
2629  * ib_poll_cq - poll a CQ for completion(s)
2630  * @cq:the CQ being polled
2631  * @num_entries:maximum number of completions to return
2632  * @wc:array of at least @num_entries &struct ib_wc where completions
2633  *   will be returned
2634  *
2635  * Poll a CQ for (possibly multiple) completions.  If the return value
2636  * is < 0, an error occurred.  If the return value is >= 0, it is the
2637  * number of completions returned.  If the return value is
2638  * non-negative and < num_entries, then the CQ was emptied.
2639  */
2640 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2641 			     struct ib_wc *wc)
2642 {
2643 	return cq->device->poll_cq(cq, num_entries, wc);
2644 }
2645 
2646 /**
2647  * ib_peek_cq - Returns the number of unreaped completions currently
2648  *   on the specified CQ.
2649  * @cq: The CQ to peek.
2650  * @wc_cnt: A minimum number of unreaped completions to check for.
2651  *
2652  * If the number of unreaped completions is greater than or equal to wc_cnt,
2653  * this function returns wc_cnt, otherwise, it returns the actual number of
2654  * unreaped completions.
2655  */
2656 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2657 
2658 /**
2659  * ib_req_notify_cq - Request completion notification on a CQ.
2660  * @cq: The CQ to generate an event for.
2661  * @flags:
2662  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2663  *   to request an event on the next solicited event or next work
2664  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2665  *   may also be |ed in to request a hint about missed events, as
2666  *   described below.
2667  *
2668  * Return Value:
2669  *    < 0 means an error occurred while requesting notification
2670  *   == 0 means notification was requested successfully, and if
2671  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2672  *        were missed and it is safe to wait for another event.  In
2673  *        this case is it guaranteed that any work completions added
2674  *        to the CQ since the last CQ poll will trigger a completion
2675  *        notification event.
2676  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2677  *        in.  It means that the consumer must poll the CQ again to
2678  *        make sure it is empty to avoid missing an event because of a
2679  *        race between requesting notification and an entry being
2680  *        added to the CQ.  This return value means it is possible
2681  *        (but not guaranteed) that a work completion has been added
2682  *        to the CQ since the last poll without triggering a
2683  *        completion notification event.
2684  */
2685 static inline int ib_req_notify_cq(struct ib_cq *cq,
2686 				   enum ib_cq_notify_flags flags)
2687 {
2688 	return cq->device->req_notify_cq(cq, flags);
2689 }
2690 
2691 /**
2692  * ib_req_ncomp_notif - Request completion notification when there are
2693  *   at least the specified number of unreaped completions on the CQ.
2694  * @cq: The CQ to generate an event for.
2695  * @wc_cnt: The number of unreaped completions that should be on the
2696  *   CQ before an event is generated.
2697  */
2698 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2699 {
2700 	return cq->device->req_ncomp_notif ?
2701 		cq->device->req_ncomp_notif(cq, wc_cnt) :
2702 		-ENOSYS;
2703 }
2704 
2705 /**
2706  * ib_get_dma_mr - Returns a memory region for system memory that is
2707  *   usable for DMA.
2708  * @pd: The protection domain associated with the memory region.
2709  * @mr_access_flags: Specifies the memory access rights.
2710  *
2711  * Note that the ib_dma_*() functions defined below must be used
2712  * to create/destroy addresses used with the Lkey or Rkey returned
2713  * by ib_get_dma_mr().
2714  */
2715 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2716 
2717 /**
2718  * ib_dma_mapping_error - check a DMA addr for error
2719  * @dev: The device for which the dma_addr was created
2720  * @dma_addr: The DMA address to check
2721  */
2722 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2723 {
2724 	if (dev->dma_ops)
2725 		return dev->dma_ops->mapping_error(dev, dma_addr);
2726 	return dma_mapping_error(dev->dma_device, dma_addr);
2727 }
2728 
2729 /**
2730  * ib_dma_map_single - Map a kernel virtual address to DMA address
2731  * @dev: The device for which the dma_addr is to be created
2732  * @cpu_addr: The kernel virtual address
2733  * @size: The size of the region in bytes
2734  * @direction: The direction of the DMA
2735  */
2736 static inline u64 ib_dma_map_single(struct ib_device *dev,
2737 				    void *cpu_addr, size_t size,
2738 				    enum dma_data_direction direction)
2739 {
2740 	if (dev->dma_ops)
2741 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2742 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2743 }
2744 
2745 /**
2746  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2747  * @dev: The device for which the DMA address was created
2748  * @addr: The DMA address
2749  * @size: The size of the region in bytes
2750  * @direction: The direction of the DMA
2751  */
2752 static inline void ib_dma_unmap_single(struct ib_device *dev,
2753 				       u64 addr, size_t size,
2754 				       enum dma_data_direction direction)
2755 {
2756 	if (dev->dma_ops)
2757 		dev->dma_ops->unmap_single(dev, addr, size, direction);
2758 	else
2759 		dma_unmap_single(dev->dma_device, addr, size, direction);
2760 }
2761 
2762 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2763 					  void *cpu_addr, size_t size,
2764 					  enum dma_data_direction direction,
2765 					  struct dma_attrs *attrs)
2766 {
2767 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2768 				    direction, attrs);
2769 }
2770 
2771 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2772 					     u64 addr, size_t size,
2773 					     enum dma_data_direction direction,
2774 					     struct dma_attrs *attrs)
2775 {
2776 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
2777 				      direction, attrs);
2778 }
2779 
2780 /**
2781  * ib_dma_map_page - Map a physical page to DMA address
2782  * @dev: The device for which the dma_addr is to be created
2783  * @page: The page to be mapped
2784  * @offset: The offset within the page
2785  * @size: The size of the region in bytes
2786  * @direction: The direction of the DMA
2787  */
2788 static inline u64 ib_dma_map_page(struct ib_device *dev,
2789 				  struct page *page,
2790 				  unsigned long offset,
2791 				  size_t size,
2792 					 enum dma_data_direction direction)
2793 {
2794 	if (dev->dma_ops)
2795 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
2796 	return dma_map_page(dev->dma_device, page, offset, size, direction);
2797 }
2798 
2799 /**
2800  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2801  * @dev: The device for which the DMA address was created
2802  * @addr: The DMA address
2803  * @size: The size of the region in bytes
2804  * @direction: The direction of the DMA
2805  */
2806 static inline void ib_dma_unmap_page(struct ib_device *dev,
2807 				     u64 addr, size_t size,
2808 				     enum dma_data_direction direction)
2809 {
2810 	if (dev->dma_ops)
2811 		dev->dma_ops->unmap_page(dev, addr, size, direction);
2812 	else
2813 		dma_unmap_page(dev->dma_device, addr, size, direction);
2814 }
2815 
2816 /**
2817  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2818  * @dev: The device for which the DMA addresses are to be created
2819  * @sg: The array of scatter/gather entries
2820  * @nents: The number of scatter/gather entries
2821  * @direction: The direction of the DMA
2822  */
2823 static inline int ib_dma_map_sg(struct ib_device *dev,
2824 				struct scatterlist *sg, int nents,
2825 				enum dma_data_direction direction)
2826 {
2827 	if (dev->dma_ops)
2828 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
2829 	return dma_map_sg(dev->dma_device, sg, nents, direction);
2830 }
2831 
2832 /**
2833  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2834  * @dev: The device for which the DMA addresses were created
2835  * @sg: The array of scatter/gather entries
2836  * @nents: The number of scatter/gather entries
2837  * @direction: The direction of the DMA
2838  */
2839 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2840 				   struct scatterlist *sg, int nents,
2841 				   enum dma_data_direction direction)
2842 {
2843 	if (dev->dma_ops)
2844 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2845 	else
2846 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
2847 }
2848 
2849 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2850 				      struct scatterlist *sg, int nents,
2851 				      enum dma_data_direction direction,
2852 				      struct dma_attrs *attrs)
2853 {
2854 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2855 }
2856 
2857 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2858 					 struct scatterlist *sg, int nents,
2859 					 enum dma_data_direction direction,
2860 					 struct dma_attrs *attrs)
2861 {
2862 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2863 }
2864 /**
2865  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2866  * @dev: The device for which the DMA addresses were created
2867  * @sg: The scatter/gather entry
2868  *
2869  * Note: this function is obsolete. To do: change all occurrences of
2870  * ib_sg_dma_address() into sg_dma_address().
2871  */
2872 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2873 				    struct scatterlist *sg)
2874 {
2875 	return sg_dma_address(sg);
2876 }
2877 
2878 /**
2879  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2880  * @dev: The device for which the DMA addresses were created
2881  * @sg: The scatter/gather entry
2882  *
2883  * Note: this function is obsolete. To do: change all occurrences of
2884  * ib_sg_dma_len() into sg_dma_len().
2885  */
2886 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2887 					 struct scatterlist *sg)
2888 {
2889 	return sg_dma_len(sg);
2890 }
2891 
2892 /**
2893  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2894  * @dev: The device for which the DMA address was created
2895  * @addr: The DMA address
2896  * @size: The size of the region in bytes
2897  * @dir: The direction of the DMA
2898  */
2899 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2900 					      u64 addr,
2901 					      size_t size,
2902 					      enum dma_data_direction dir)
2903 {
2904 	if (dev->dma_ops)
2905 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2906 	else
2907 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2908 }
2909 
2910 /**
2911  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2912  * @dev: The device for which the DMA address was created
2913  * @addr: The DMA address
2914  * @size: The size of the region in bytes
2915  * @dir: The direction of the DMA
2916  */
2917 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2918 						 u64 addr,
2919 						 size_t size,
2920 						 enum dma_data_direction dir)
2921 {
2922 	if (dev->dma_ops)
2923 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2924 	else
2925 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2926 }
2927 
2928 /**
2929  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2930  * @dev: The device for which the DMA address is requested
2931  * @size: The size of the region to allocate in bytes
2932  * @dma_handle: A pointer for returning the DMA address of the region
2933  * @flag: memory allocator flags
2934  */
2935 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2936 					   size_t size,
2937 					   u64 *dma_handle,
2938 					   gfp_t flag)
2939 {
2940 	if (dev->dma_ops)
2941 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2942 	else {
2943 		dma_addr_t handle;
2944 		void *ret;
2945 
2946 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2947 		*dma_handle = handle;
2948 		return ret;
2949 	}
2950 }
2951 
2952 /**
2953  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2954  * @dev: The device for which the DMA addresses were allocated
2955  * @size: The size of the region
2956  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2957  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2958  */
2959 static inline void ib_dma_free_coherent(struct ib_device *dev,
2960 					size_t size, void *cpu_addr,
2961 					u64 dma_handle)
2962 {
2963 	if (dev->dma_ops)
2964 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2965 	else
2966 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2967 }
2968 
2969 /**
2970  * ib_dereg_mr - Deregisters a memory region and removes it from the
2971  *   HCA translation table.
2972  * @mr: The memory region to deregister.
2973  *
2974  * This function can fail, if the memory region has memory windows bound to it.
2975  */
2976 int ib_dereg_mr(struct ib_mr *mr);
2977 
2978 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
2979 			  enum ib_mr_type mr_type,
2980 			  u32 max_num_sg);
2981 
2982 /**
2983  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2984  *   R_Key and L_Key.
2985  * @mr - struct ib_mr pointer to be updated.
2986  * @newkey - new key to be used.
2987  */
2988 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2989 {
2990 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2991 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2992 }
2993 
2994 /**
2995  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2996  * for calculating a new rkey for type 2 memory windows.
2997  * @rkey - the rkey to increment.
2998  */
2999 static inline u32 ib_inc_rkey(u32 rkey)
3000 {
3001 	const u32 mask = 0x000000ff;
3002 	return ((rkey + 1) & mask) | (rkey & ~mask);
3003 }
3004 
3005 /**
3006  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3007  * @pd: The protection domain associated with the unmapped region.
3008  * @mr_access_flags: Specifies the memory access rights.
3009  * @fmr_attr: Attributes of the unmapped region.
3010  *
3011  * A fast memory region must be mapped before it can be used as part of
3012  * a work request.
3013  */
3014 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3015 			    int mr_access_flags,
3016 			    struct ib_fmr_attr *fmr_attr);
3017 
3018 /**
3019  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3020  * @fmr: The fast memory region to associate with the pages.
3021  * @page_list: An array of physical pages to map to the fast memory region.
3022  * @list_len: The number of pages in page_list.
3023  * @iova: The I/O virtual address to use with the mapped region.
3024  */
3025 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3026 				  u64 *page_list, int list_len,
3027 				  u64 iova)
3028 {
3029 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3030 }
3031 
3032 /**
3033  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3034  * @fmr_list: A linked list of fast memory regions to unmap.
3035  */
3036 int ib_unmap_fmr(struct list_head *fmr_list);
3037 
3038 /**
3039  * ib_dealloc_fmr - Deallocates a fast memory region.
3040  * @fmr: The fast memory region to deallocate.
3041  */
3042 int ib_dealloc_fmr(struct ib_fmr *fmr);
3043 
3044 /**
3045  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3046  * @qp: QP to attach to the multicast group.  The QP must be type
3047  *   IB_QPT_UD.
3048  * @gid: Multicast group GID.
3049  * @lid: Multicast group LID in host byte order.
3050  *
3051  * In order to send and receive multicast packets, subnet
3052  * administration must have created the multicast group and configured
3053  * the fabric appropriately.  The port associated with the specified
3054  * QP must also be a member of the multicast group.
3055  */
3056 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3057 
3058 /**
3059  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3060  * @qp: QP to detach from the multicast group.
3061  * @gid: Multicast group GID.
3062  * @lid: Multicast group LID in host byte order.
3063  */
3064 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3065 
3066 /**
3067  * ib_alloc_xrcd - Allocates an XRC domain.
3068  * @device: The device on which to allocate the XRC domain.
3069  */
3070 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3071 
3072 /**
3073  * ib_dealloc_xrcd - Deallocates an XRC domain.
3074  * @xrcd: The XRC domain to deallocate.
3075  */
3076 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3077 
3078 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3079 			       struct ib_flow_attr *flow_attr, int domain);
3080 int ib_destroy_flow(struct ib_flow *flow_id);
3081 
3082 static inline int ib_check_mr_access(int flags)
3083 {
3084 	/*
3085 	 * Local write permission is required if remote write or
3086 	 * remote atomic permission is also requested.
3087 	 */
3088 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3089 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3090 		return -EINVAL;
3091 
3092 	return 0;
3093 }
3094 
3095 /**
3096  * ib_check_mr_status: lightweight check of MR status.
3097  *     This routine may provide status checks on a selected
3098  *     ib_mr. first use is for signature status check.
3099  *
3100  * @mr: A memory region.
3101  * @check_mask: Bitmask of which checks to perform from
3102  *     ib_mr_status_check enumeration.
3103  * @mr_status: The container of relevant status checks.
3104  *     failed checks will be indicated in the status bitmask
3105  *     and the relevant info shall be in the error item.
3106  */
3107 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3108 		       struct ib_mr_status *mr_status);
3109 
3110 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3111 					    u16 pkey, const union ib_gid *gid,
3112 					    const struct sockaddr *addr);
3113 
3114 int ib_map_mr_sg(struct ib_mr *mr,
3115 		 struct scatterlist *sg,
3116 		 int sg_nents,
3117 		 unsigned int page_size);
3118 
3119 static inline int
3120 ib_map_mr_sg_zbva(struct ib_mr *mr,
3121 		  struct scatterlist *sg,
3122 		  int sg_nents,
3123 		  unsigned int page_size)
3124 {
3125 	int n;
3126 
3127 	n = ib_map_mr_sg(mr, sg, sg_nents, page_size);
3128 	mr->iova = 0;
3129 
3130 	return n;
3131 }
3132 
3133 int ib_sg_to_pages(struct ib_mr *mr,
3134 		   struct scatterlist *sgl,
3135 		   int sg_nents,
3136 		   int (*set_page)(struct ib_mr *, u64));
3137 
3138 void ib_drain_rq(struct ib_qp *qp);
3139 void ib_drain_sq(struct ib_qp *qp);
3140 void ib_drain_qp(struct ib_qp *qp);
3141 #endif /* IB_VERBS_H */
3142