xref: /linux/include/rdma/ib_verbs.h (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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 
52 #include <linux/atomic.h>
53 #include <asm/uaccess.h>
54 
55 extern struct workqueue_struct *ib_wq;
56 
57 union ib_gid {
58 	u8	raw[16];
59 	struct {
60 		__be64	subnet_prefix;
61 		__be64	interface_id;
62 	} global;
63 };
64 
65 enum rdma_node_type {
66 	/* IB values map to NodeInfo:NodeType. */
67 	RDMA_NODE_IB_CA 	= 1,
68 	RDMA_NODE_IB_SWITCH,
69 	RDMA_NODE_IB_ROUTER,
70 	RDMA_NODE_RNIC
71 };
72 
73 enum rdma_transport_type {
74 	RDMA_TRANSPORT_IB,
75 	RDMA_TRANSPORT_IWARP
76 };
77 
78 enum rdma_transport_type
79 rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__;
80 
81 enum rdma_link_layer {
82 	IB_LINK_LAYER_UNSPECIFIED,
83 	IB_LINK_LAYER_INFINIBAND,
84 	IB_LINK_LAYER_ETHERNET,
85 };
86 
87 enum ib_device_cap_flags {
88 	IB_DEVICE_RESIZE_MAX_WR		= 1,
89 	IB_DEVICE_BAD_PKEY_CNTR		= (1<<1),
90 	IB_DEVICE_BAD_QKEY_CNTR		= (1<<2),
91 	IB_DEVICE_RAW_MULTI		= (1<<3),
92 	IB_DEVICE_AUTO_PATH_MIG		= (1<<4),
93 	IB_DEVICE_CHANGE_PHY_PORT	= (1<<5),
94 	IB_DEVICE_UD_AV_PORT_ENFORCE	= (1<<6),
95 	IB_DEVICE_CURR_QP_STATE_MOD	= (1<<7),
96 	IB_DEVICE_SHUTDOWN_PORT		= (1<<8),
97 	IB_DEVICE_INIT_TYPE		= (1<<9),
98 	IB_DEVICE_PORT_ACTIVE_EVENT	= (1<<10),
99 	IB_DEVICE_SYS_IMAGE_GUID	= (1<<11),
100 	IB_DEVICE_RC_RNR_NAK_GEN	= (1<<12),
101 	IB_DEVICE_SRQ_RESIZE		= (1<<13),
102 	IB_DEVICE_N_NOTIFY_CQ		= (1<<14),
103 	IB_DEVICE_LOCAL_DMA_LKEY	= (1<<15),
104 	IB_DEVICE_RESERVED		= (1<<16), /* old SEND_W_INV */
105 	IB_DEVICE_MEM_WINDOW		= (1<<17),
106 	/*
107 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
108 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
109 	 * messages and can verify the validity of checksum for
110 	 * incoming messages.  Setting this flag implies that the
111 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
112 	 */
113 	IB_DEVICE_UD_IP_CSUM		= (1<<18),
114 	IB_DEVICE_UD_TSO		= (1<<19),
115 	IB_DEVICE_XRC			= (1<<20),
116 	IB_DEVICE_MEM_MGT_EXTENSIONS	= (1<<21),
117 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
118 };
119 
120 enum ib_atomic_cap {
121 	IB_ATOMIC_NONE,
122 	IB_ATOMIC_HCA,
123 	IB_ATOMIC_GLOB
124 };
125 
126 struct ib_device_attr {
127 	u64			fw_ver;
128 	__be64			sys_image_guid;
129 	u64			max_mr_size;
130 	u64			page_size_cap;
131 	u32			vendor_id;
132 	u32			vendor_part_id;
133 	u32			hw_ver;
134 	int			max_qp;
135 	int			max_qp_wr;
136 	int			device_cap_flags;
137 	int			max_sge;
138 	int			max_sge_rd;
139 	int			max_cq;
140 	int			max_cqe;
141 	int			max_mr;
142 	int			max_pd;
143 	int			max_qp_rd_atom;
144 	int			max_ee_rd_atom;
145 	int			max_res_rd_atom;
146 	int			max_qp_init_rd_atom;
147 	int			max_ee_init_rd_atom;
148 	enum ib_atomic_cap	atomic_cap;
149 	enum ib_atomic_cap	masked_atomic_cap;
150 	int			max_ee;
151 	int			max_rdd;
152 	int			max_mw;
153 	int			max_raw_ipv6_qp;
154 	int			max_raw_ethy_qp;
155 	int			max_mcast_grp;
156 	int			max_mcast_qp_attach;
157 	int			max_total_mcast_qp_attach;
158 	int			max_ah;
159 	int			max_fmr;
160 	int			max_map_per_fmr;
161 	int			max_srq;
162 	int			max_srq_wr;
163 	int			max_srq_sge;
164 	unsigned int		max_fast_reg_page_list_len;
165 	u16			max_pkeys;
166 	u8			local_ca_ack_delay;
167 };
168 
169 enum ib_mtu {
170 	IB_MTU_256  = 1,
171 	IB_MTU_512  = 2,
172 	IB_MTU_1024 = 3,
173 	IB_MTU_2048 = 4,
174 	IB_MTU_4096 = 5
175 };
176 
177 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
178 {
179 	switch (mtu) {
180 	case IB_MTU_256:  return  256;
181 	case IB_MTU_512:  return  512;
182 	case IB_MTU_1024: return 1024;
183 	case IB_MTU_2048: return 2048;
184 	case IB_MTU_4096: return 4096;
185 	default: 	  return -1;
186 	}
187 }
188 
189 enum ib_port_state {
190 	IB_PORT_NOP		= 0,
191 	IB_PORT_DOWN		= 1,
192 	IB_PORT_INIT		= 2,
193 	IB_PORT_ARMED		= 3,
194 	IB_PORT_ACTIVE		= 4,
195 	IB_PORT_ACTIVE_DEFER	= 5
196 };
197 
198 enum ib_port_cap_flags {
199 	IB_PORT_SM				= 1 <<  1,
200 	IB_PORT_NOTICE_SUP			= 1 <<  2,
201 	IB_PORT_TRAP_SUP			= 1 <<  3,
202 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
203 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
204 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
205 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
206 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
207 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
208 	IB_PORT_SM_DISABLED			= 1 << 10,
209 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
210 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
211 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
212 	IB_PORT_CM_SUP				= 1 << 16,
213 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
214 	IB_PORT_REINIT_SUP			= 1 << 18,
215 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
216 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
217 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
218 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
219 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
220 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
221 	IB_PORT_CLIENT_REG_SUP			= 1 << 25
222 };
223 
224 enum ib_port_width {
225 	IB_WIDTH_1X	= 1,
226 	IB_WIDTH_4X	= 2,
227 	IB_WIDTH_8X	= 4,
228 	IB_WIDTH_12X	= 8
229 };
230 
231 static inline int ib_width_enum_to_int(enum ib_port_width width)
232 {
233 	switch (width) {
234 	case IB_WIDTH_1X:  return  1;
235 	case IB_WIDTH_4X:  return  4;
236 	case IB_WIDTH_8X:  return  8;
237 	case IB_WIDTH_12X: return 12;
238 	default: 	  return -1;
239 	}
240 }
241 
242 enum ib_port_speed {
243 	IB_SPEED_SDR	= 1,
244 	IB_SPEED_DDR	= 2,
245 	IB_SPEED_QDR	= 4,
246 	IB_SPEED_FDR10	= 8,
247 	IB_SPEED_FDR	= 16,
248 	IB_SPEED_EDR	= 32
249 };
250 
251 struct ib_protocol_stats {
252 	/* TBD... */
253 };
254 
255 struct iw_protocol_stats {
256 	u64	ipInReceives;
257 	u64	ipInHdrErrors;
258 	u64	ipInTooBigErrors;
259 	u64	ipInNoRoutes;
260 	u64	ipInAddrErrors;
261 	u64	ipInUnknownProtos;
262 	u64	ipInTruncatedPkts;
263 	u64	ipInDiscards;
264 	u64	ipInDelivers;
265 	u64	ipOutForwDatagrams;
266 	u64	ipOutRequests;
267 	u64	ipOutDiscards;
268 	u64	ipOutNoRoutes;
269 	u64	ipReasmTimeout;
270 	u64	ipReasmReqds;
271 	u64	ipReasmOKs;
272 	u64	ipReasmFails;
273 	u64	ipFragOKs;
274 	u64	ipFragFails;
275 	u64	ipFragCreates;
276 	u64	ipInMcastPkts;
277 	u64	ipOutMcastPkts;
278 	u64	ipInBcastPkts;
279 	u64	ipOutBcastPkts;
280 
281 	u64	tcpRtoAlgorithm;
282 	u64	tcpRtoMin;
283 	u64	tcpRtoMax;
284 	u64	tcpMaxConn;
285 	u64	tcpActiveOpens;
286 	u64	tcpPassiveOpens;
287 	u64	tcpAttemptFails;
288 	u64	tcpEstabResets;
289 	u64	tcpCurrEstab;
290 	u64	tcpInSegs;
291 	u64	tcpOutSegs;
292 	u64	tcpRetransSegs;
293 	u64	tcpInErrs;
294 	u64	tcpOutRsts;
295 };
296 
297 union rdma_protocol_stats {
298 	struct ib_protocol_stats	ib;
299 	struct iw_protocol_stats	iw;
300 };
301 
302 struct ib_port_attr {
303 	enum ib_port_state	state;
304 	enum ib_mtu		max_mtu;
305 	enum ib_mtu		active_mtu;
306 	int			gid_tbl_len;
307 	u32			port_cap_flags;
308 	u32			max_msg_sz;
309 	u32			bad_pkey_cntr;
310 	u32			qkey_viol_cntr;
311 	u16			pkey_tbl_len;
312 	u16			lid;
313 	u16			sm_lid;
314 	u8			lmc;
315 	u8			max_vl_num;
316 	u8			sm_sl;
317 	u8			subnet_timeout;
318 	u8			init_type_reply;
319 	u8			active_width;
320 	u8			active_speed;
321 	u8                      phys_state;
322 };
323 
324 enum ib_device_modify_flags {
325 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
326 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
327 };
328 
329 struct ib_device_modify {
330 	u64	sys_image_guid;
331 	char	node_desc[64];
332 };
333 
334 enum ib_port_modify_flags {
335 	IB_PORT_SHUTDOWN		= 1,
336 	IB_PORT_INIT_TYPE		= (1<<2),
337 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
338 };
339 
340 struct ib_port_modify {
341 	u32	set_port_cap_mask;
342 	u32	clr_port_cap_mask;
343 	u8	init_type;
344 };
345 
346 enum ib_event_type {
347 	IB_EVENT_CQ_ERR,
348 	IB_EVENT_QP_FATAL,
349 	IB_EVENT_QP_REQ_ERR,
350 	IB_EVENT_QP_ACCESS_ERR,
351 	IB_EVENT_COMM_EST,
352 	IB_EVENT_SQ_DRAINED,
353 	IB_EVENT_PATH_MIG,
354 	IB_EVENT_PATH_MIG_ERR,
355 	IB_EVENT_DEVICE_FATAL,
356 	IB_EVENT_PORT_ACTIVE,
357 	IB_EVENT_PORT_ERR,
358 	IB_EVENT_LID_CHANGE,
359 	IB_EVENT_PKEY_CHANGE,
360 	IB_EVENT_SM_CHANGE,
361 	IB_EVENT_SRQ_ERR,
362 	IB_EVENT_SRQ_LIMIT_REACHED,
363 	IB_EVENT_QP_LAST_WQE_REACHED,
364 	IB_EVENT_CLIENT_REREGISTER,
365 	IB_EVENT_GID_CHANGE,
366 };
367 
368 struct ib_event {
369 	struct ib_device	*device;
370 	union {
371 		struct ib_cq	*cq;
372 		struct ib_qp	*qp;
373 		struct ib_srq	*srq;
374 		u8		port_num;
375 	} element;
376 	enum ib_event_type	event;
377 };
378 
379 struct ib_event_handler {
380 	struct ib_device *device;
381 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
382 	struct list_head  list;
383 };
384 
385 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
386 	do {							\
387 		(_ptr)->device  = _device;			\
388 		(_ptr)->handler = _handler;			\
389 		INIT_LIST_HEAD(&(_ptr)->list);			\
390 	} while (0)
391 
392 struct ib_global_route {
393 	union ib_gid	dgid;
394 	u32		flow_label;
395 	u8		sgid_index;
396 	u8		hop_limit;
397 	u8		traffic_class;
398 };
399 
400 struct ib_grh {
401 	__be32		version_tclass_flow;
402 	__be16		paylen;
403 	u8		next_hdr;
404 	u8		hop_limit;
405 	union ib_gid	sgid;
406 	union ib_gid	dgid;
407 };
408 
409 enum {
410 	IB_MULTICAST_QPN = 0xffffff
411 };
412 
413 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
414 
415 enum ib_ah_flags {
416 	IB_AH_GRH	= 1
417 };
418 
419 enum ib_rate {
420 	IB_RATE_PORT_CURRENT = 0,
421 	IB_RATE_2_5_GBPS = 2,
422 	IB_RATE_5_GBPS   = 5,
423 	IB_RATE_10_GBPS  = 3,
424 	IB_RATE_20_GBPS  = 6,
425 	IB_RATE_30_GBPS  = 4,
426 	IB_RATE_40_GBPS  = 7,
427 	IB_RATE_60_GBPS  = 8,
428 	IB_RATE_80_GBPS  = 9,
429 	IB_RATE_120_GBPS = 10,
430 	IB_RATE_14_GBPS  = 11,
431 	IB_RATE_56_GBPS  = 12,
432 	IB_RATE_112_GBPS = 13,
433 	IB_RATE_168_GBPS = 14,
434 	IB_RATE_25_GBPS  = 15,
435 	IB_RATE_100_GBPS = 16,
436 	IB_RATE_200_GBPS = 17,
437 	IB_RATE_300_GBPS = 18
438 };
439 
440 /**
441  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
442  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
443  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
444  * @rate: rate to convert.
445  */
446 int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
447 
448 /**
449  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
450  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
451  * @rate: rate to convert.
452  */
453 int ib_rate_to_mbps(enum ib_rate rate) __attribute_const__;
454 
455 /**
456  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
457  * enum.
458  * @mult: multiple to convert.
459  */
460 enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
461 
462 struct ib_ah_attr {
463 	struct ib_global_route	grh;
464 	u16			dlid;
465 	u8			sl;
466 	u8			src_path_bits;
467 	u8			static_rate;
468 	u8			ah_flags;
469 	u8			port_num;
470 };
471 
472 enum ib_wc_status {
473 	IB_WC_SUCCESS,
474 	IB_WC_LOC_LEN_ERR,
475 	IB_WC_LOC_QP_OP_ERR,
476 	IB_WC_LOC_EEC_OP_ERR,
477 	IB_WC_LOC_PROT_ERR,
478 	IB_WC_WR_FLUSH_ERR,
479 	IB_WC_MW_BIND_ERR,
480 	IB_WC_BAD_RESP_ERR,
481 	IB_WC_LOC_ACCESS_ERR,
482 	IB_WC_REM_INV_REQ_ERR,
483 	IB_WC_REM_ACCESS_ERR,
484 	IB_WC_REM_OP_ERR,
485 	IB_WC_RETRY_EXC_ERR,
486 	IB_WC_RNR_RETRY_EXC_ERR,
487 	IB_WC_LOC_RDD_VIOL_ERR,
488 	IB_WC_REM_INV_RD_REQ_ERR,
489 	IB_WC_REM_ABORT_ERR,
490 	IB_WC_INV_EECN_ERR,
491 	IB_WC_INV_EEC_STATE_ERR,
492 	IB_WC_FATAL_ERR,
493 	IB_WC_RESP_TIMEOUT_ERR,
494 	IB_WC_GENERAL_ERR
495 };
496 
497 enum ib_wc_opcode {
498 	IB_WC_SEND,
499 	IB_WC_RDMA_WRITE,
500 	IB_WC_RDMA_READ,
501 	IB_WC_COMP_SWAP,
502 	IB_WC_FETCH_ADD,
503 	IB_WC_BIND_MW,
504 	IB_WC_LSO,
505 	IB_WC_LOCAL_INV,
506 	IB_WC_FAST_REG_MR,
507 	IB_WC_MASKED_COMP_SWAP,
508 	IB_WC_MASKED_FETCH_ADD,
509 /*
510  * Set value of IB_WC_RECV so consumers can test if a completion is a
511  * receive by testing (opcode & IB_WC_RECV).
512  */
513 	IB_WC_RECV			= 1 << 7,
514 	IB_WC_RECV_RDMA_WITH_IMM
515 };
516 
517 enum ib_wc_flags {
518 	IB_WC_GRH		= 1,
519 	IB_WC_WITH_IMM		= (1<<1),
520 	IB_WC_WITH_INVALIDATE	= (1<<2),
521 	IB_WC_IP_CSUM_OK	= (1<<3),
522 };
523 
524 struct ib_wc {
525 	u64			wr_id;
526 	enum ib_wc_status	status;
527 	enum ib_wc_opcode	opcode;
528 	u32			vendor_err;
529 	u32			byte_len;
530 	struct ib_qp	       *qp;
531 	union {
532 		__be32		imm_data;
533 		u32		invalidate_rkey;
534 	} ex;
535 	u32			src_qp;
536 	int			wc_flags;
537 	u16			pkey_index;
538 	u16			slid;
539 	u8			sl;
540 	u8			dlid_path_bits;
541 	u8			port_num;	/* valid only for DR SMPs on switches */
542 };
543 
544 enum ib_cq_notify_flags {
545 	IB_CQ_SOLICITED			= 1 << 0,
546 	IB_CQ_NEXT_COMP			= 1 << 1,
547 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
548 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
549 };
550 
551 enum ib_srq_type {
552 	IB_SRQT_BASIC,
553 	IB_SRQT_XRC
554 };
555 
556 enum ib_srq_attr_mask {
557 	IB_SRQ_MAX_WR	= 1 << 0,
558 	IB_SRQ_LIMIT	= 1 << 1,
559 };
560 
561 struct ib_srq_attr {
562 	u32	max_wr;
563 	u32	max_sge;
564 	u32	srq_limit;
565 };
566 
567 struct ib_srq_init_attr {
568 	void		      (*event_handler)(struct ib_event *, void *);
569 	void		       *srq_context;
570 	struct ib_srq_attr	attr;
571 	enum ib_srq_type	srq_type;
572 
573 	union {
574 		struct {
575 			struct ib_xrcd *xrcd;
576 			struct ib_cq   *cq;
577 		} xrc;
578 	} ext;
579 };
580 
581 struct ib_qp_cap {
582 	u32	max_send_wr;
583 	u32	max_recv_wr;
584 	u32	max_send_sge;
585 	u32	max_recv_sge;
586 	u32	max_inline_data;
587 };
588 
589 enum ib_sig_type {
590 	IB_SIGNAL_ALL_WR,
591 	IB_SIGNAL_REQ_WR
592 };
593 
594 enum ib_qp_type {
595 	/*
596 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
597 	 * here (and in that order) since the MAD layer uses them as
598 	 * indices into a 2-entry table.
599 	 */
600 	IB_QPT_SMI,
601 	IB_QPT_GSI,
602 
603 	IB_QPT_RC,
604 	IB_QPT_UC,
605 	IB_QPT_UD,
606 	IB_QPT_RAW_IPV6,
607 	IB_QPT_RAW_ETHERTYPE,
608 	IB_QPT_RAW_PACKET = 8,
609 	IB_QPT_XRC_INI = 9,
610 	IB_QPT_XRC_TGT,
611 	IB_QPT_MAX
612 };
613 
614 enum ib_qp_create_flags {
615 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
616 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
617 	/* reserve bits 26-31 for low level drivers' internal use */
618 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
619 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
620 };
621 
622 struct ib_qp_init_attr {
623 	void                  (*event_handler)(struct ib_event *, void *);
624 	void		       *qp_context;
625 	struct ib_cq	       *send_cq;
626 	struct ib_cq	       *recv_cq;
627 	struct ib_srq	       *srq;
628 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
629 	struct ib_qp_cap	cap;
630 	enum ib_sig_type	sq_sig_type;
631 	enum ib_qp_type		qp_type;
632 	enum ib_qp_create_flags	create_flags;
633 	u8			port_num; /* special QP types only */
634 };
635 
636 struct ib_qp_open_attr {
637 	void                  (*event_handler)(struct ib_event *, void *);
638 	void		       *qp_context;
639 	u32			qp_num;
640 	enum ib_qp_type		qp_type;
641 };
642 
643 enum ib_rnr_timeout {
644 	IB_RNR_TIMER_655_36 =  0,
645 	IB_RNR_TIMER_000_01 =  1,
646 	IB_RNR_TIMER_000_02 =  2,
647 	IB_RNR_TIMER_000_03 =  3,
648 	IB_RNR_TIMER_000_04 =  4,
649 	IB_RNR_TIMER_000_06 =  5,
650 	IB_RNR_TIMER_000_08 =  6,
651 	IB_RNR_TIMER_000_12 =  7,
652 	IB_RNR_TIMER_000_16 =  8,
653 	IB_RNR_TIMER_000_24 =  9,
654 	IB_RNR_TIMER_000_32 = 10,
655 	IB_RNR_TIMER_000_48 = 11,
656 	IB_RNR_TIMER_000_64 = 12,
657 	IB_RNR_TIMER_000_96 = 13,
658 	IB_RNR_TIMER_001_28 = 14,
659 	IB_RNR_TIMER_001_92 = 15,
660 	IB_RNR_TIMER_002_56 = 16,
661 	IB_RNR_TIMER_003_84 = 17,
662 	IB_RNR_TIMER_005_12 = 18,
663 	IB_RNR_TIMER_007_68 = 19,
664 	IB_RNR_TIMER_010_24 = 20,
665 	IB_RNR_TIMER_015_36 = 21,
666 	IB_RNR_TIMER_020_48 = 22,
667 	IB_RNR_TIMER_030_72 = 23,
668 	IB_RNR_TIMER_040_96 = 24,
669 	IB_RNR_TIMER_061_44 = 25,
670 	IB_RNR_TIMER_081_92 = 26,
671 	IB_RNR_TIMER_122_88 = 27,
672 	IB_RNR_TIMER_163_84 = 28,
673 	IB_RNR_TIMER_245_76 = 29,
674 	IB_RNR_TIMER_327_68 = 30,
675 	IB_RNR_TIMER_491_52 = 31
676 };
677 
678 enum ib_qp_attr_mask {
679 	IB_QP_STATE			= 1,
680 	IB_QP_CUR_STATE			= (1<<1),
681 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
682 	IB_QP_ACCESS_FLAGS		= (1<<3),
683 	IB_QP_PKEY_INDEX		= (1<<4),
684 	IB_QP_PORT			= (1<<5),
685 	IB_QP_QKEY			= (1<<6),
686 	IB_QP_AV			= (1<<7),
687 	IB_QP_PATH_MTU			= (1<<8),
688 	IB_QP_TIMEOUT			= (1<<9),
689 	IB_QP_RETRY_CNT			= (1<<10),
690 	IB_QP_RNR_RETRY			= (1<<11),
691 	IB_QP_RQ_PSN			= (1<<12),
692 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
693 	IB_QP_ALT_PATH			= (1<<14),
694 	IB_QP_MIN_RNR_TIMER		= (1<<15),
695 	IB_QP_SQ_PSN			= (1<<16),
696 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
697 	IB_QP_PATH_MIG_STATE		= (1<<18),
698 	IB_QP_CAP			= (1<<19),
699 	IB_QP_DEST_QPN			= (1<<20)
700 };
701 
702 enum ib_qp_state {
703 	IB_QPS_RESET,
704 	IB_QPS_INIT,
705 	IB_QPS_RTR,
706 	IB_QPS_RTS,
707 	IB_QPS_SQD,
708 	IB_QPS_SQE,
709 	IB_QPS_ERR
710 };
711 
712 enum ib_mig_state {
713 	IB_MIG_MIGRATED,
714 	IB_MIG_REARM,
715 	IB_MIG_ARMED
716 };
717 
718 struct ib_qp_attr {
719 	enum ib_qp_state	qp_state;
720 	enum ib_qp_state	cur_qp_state;
721 	enum ib_mtu		path_mtu;
722 	enum ib_mig_state	path_mig_state;
723 	u32			qkey;
724 	u32			rq_psn;
725 	u32			sq_psn;
726 	u32			dest_qp_num;
727 	int			qp_access_flags;
728 	struct ib_qp_cap	cap;
729 	struct ib_ah_attr	ah_attr;
730 	struct ib_ah_attr	alt_ah_attr;
731 	u16			pkey_index;
732 	u16			alt_pkey_index;
733 	u8			en_sqd_async_notify;
734 	u8			sq_draining;
735 	u8			max_rd_atomic;
736 	u8			max_dest_rd_atomic;
737 	u8			min_rnr_timer;
738 	u8			port_num;
739 	u8			timeout;
740 	u8			retry_cnt;
741 	u8			rnr_retry;
742 	u8			alt_port_num;
743 	u8			alt_timeout;
744 };
745 
746 enum ib_wr_opcode {
747 	IB_WR_RDMA_WRITE,
748 	IB_WR_RDMA_WRITE_WITH_IMM,
749 	IB_WR_SEND,
750 	IB_WR_SEND_WITH_IMM,
751 	IB_WR_RDMA_READ,
752 	IB_WR_ATOMIC_CMP_AND_SWP,
753 	IB_WR_ATOMIC_FETCH_AND_ADD,
754 	IB_WR_LSO,
755 	IB_WR_SEND_WITH_INV,
756 	IB_WR_RDMA_READ_WITH_INV,
757 	IB_WR_LOCAL_INV,
758 	IB_WR_FAST_REG_MR,
759 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
760 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
761 };
762 
763 enum ib_send_flags {
764 	IB_SEND_FENCE		= 1,
765 	IB_SEND_SIGNALED	= (1<<1),
766 	IB_SEND_SOLICITED	= (1<<2),
767 	IB_SEND_INLINE		= (1<<3),
768 	IB_SEND_IP_CSUM		= (1<<4)
769 };
770 
771 struct ib_sge {
772 	u64	addr;
773 	u32	length;
774 	u32	lkey;
775 };
776 
777 struct ib_fast_reg_page_list {
778 	struct ib_device       *device;
779 	u64		       *page_list;
780 	unsigned int		max_page_list_len;
781 };
782 
783 struct ib_send_wr {
784 	struct ib_send_wr      *next;
785 	u64			wr_id;
786 	struct ib_sge	       *sg_list;
787 	int			num_sge;
788 	enum ib_wr_opcode	opcode;
789 	int			send_flags;
790 	union {
791 		__be32		imm_data;
792 		u32		invalidate_rkey;
793 	} ex;
794 	union {
795 		struct {
796 			u64	remote_addr;
797 			u32	rkey;
798 		} rdma;
799 		struct {
800 			u64	remote_addr;
801 			u64	compare_add;
802 			u64	swap;
803 			u64	compare_add_mask;
804 			u64	swap_mask;
805 			u32	rkey;
806 		} atomic;
807 		struct {
808 			struct ib_ah *ah;
809 			void   *header;
810 			int     hlen;
811 			int     mss;
812 			u32	remote_qpn;
813 			u32	remote_qkey;
814 			u16	pkey_index; /* valid for GSI only */
815 			u8	port_num;   /* valid for DR SMPs on switch only */
816 		} ud;
817 		struct {
818 			u64				iova_start;
819 			struct ib_fast_reg_page_list   *page_list;
820 			unsigned int			page_shift;
821 			unsigned int			page_list_len;
822 			u32				length;
823 			int				access_flags;
824 			u32				rkey;
825 		} fast_reg;
826 	} wr;
827 	u32			xrc_remote_srq_num;	/* XRC TGT QPs only */
828 };
829 
830 struct ib_recv_wr {
831 	struct ib_recv_wr      *next;
832 	u64			wr_id;
833 	struct ib_sge	       *sg_list;
834 	int			num_sge;
835 };
836 
837 enum ib_access_flags {
838 	IB_ACCESS_LOCAL_WRITE	= 1,
839 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
840 	IB_ACCESS_REMOTE_READ	= (1<<2),
841 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
842 	IB_ACCESS_MW_BIND	= (1<<4)
843 };
844 
845 struct ib_phys_buf {
846 	u64      addr;
847 	u64      size;
848 };
849 
850 struct ib_mr_attr {
851 	struct ib_pd	*pd;
852 	u64		device_virt_addr;
853 	u64		size;
854 	int		mr_access_flags;
855 	u32		lkey;
856 	u32		rkey;
857 };
858 
859 enum ib_mr_rereg_flags {
860 	IB_MR_REREG_TRANS	= 1,
861 	IB_MR_REREG_PD		= (1<<1),
862 	IB_MR_REREG_ACCESS	= (1<<2)
863 };
864 
865 struct ib_mw_bind {
866 	struct ib_mr   *mr;
867 	u64		wr_id;
868 	u64		addr;
869 	u32		length;
870 	int		send_flags;
871 	int		mw_access_flags;
872 };
873 
874 struct ib_fmr_attr {
875 	int	max_pages;
876 	int	max_maps;
877 	u8	page_shift;
878 };
879 
880 struct ib_ucontext {
881 	struct ib_device       *device;
882 	struct list_head	pd_list;
883 	struct list_head	mr_list;
884 	struct list_head	mw_list;
885 	struct list_head	cq_list;
886 	struct list_head	qp_list;
887 	struct list_head	srq_list;
888 	struct list_head	ah_list;
889 	struct list_head	xrcd_list;
890 	int			closing;
891 };
892 
893 struct ib_uobject {
894 	u64			user_handle;	/* handle given to us by userspace */
895 	struct ib_ucontext     *context;	/* associated user context */
896 	void		       *object;		/* containing object */
897 	struct list_head	list;		/* link to context's list */
898 	int			id;		/* index into kernel idr */
899 	struct kref		ref;
900 	struct rw_semaphore	mutex;		/* protects .live */
901 	int			live;
902 };
903 
904 struct ib_udata {
905 	void __user *inbuf;
906 	void __user *outbuf;
907 	size_t       inlen;
908 	size_t       outlen;
909 };
910 
911 struct ib_pd {
912 	struct ib_device       *device;
913 	struct ib_uobject      *uobject;
914 	atomic_t          	usecnt; /* count all resources */
915 };
916 
917 struct ib_xrcd {
918 	struct ib_device       *device;
919 	atomic_t		usecnt; /* count all exposed resources */
920 	struct inode	       *inode;
921 
922 	struct mutex		tgt_qp_mutex;
923 	struct list_head	tgt_qp_list;
924 };
925 
926 struct ib_ah {
927 	struct ib_device	*device;
928 	struct ib_pd		*pd;
929 	struct ib_uobject	*uobject;
930 };
931 
932 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
933 
934 struct ib_cq {
935 	struct ib_device       *device;
936 	struct ib_uobject      *uobject;
937 	ib_comp_handler   	comp_handler;
938 	void                  (*event_handler)(struct ib_event *, void *);
939 	void                   *cq_context;
940 	int               	cqe;
941 	atomic_t          	usecnt; /* count number of work queues */
942 };
943 
944 struct ib_srq {
945 	struct ib_device       *device;
946 	struct ib_pd	       *pd;
947 	struct ib_uobject      *uobject;
948 	void		      (*event_handler)(struct ib_event *, void *);
949 	void		       *srq_context;
950 	enum ib_srq_type	srq_type;
951 	atomic_t		usecnt;
952 
953 	union {
954 		struct {
955 			struct ib_xrcd *xrcd;
956 			struct ib_cq   *cq;
957 			u32		srq_num;
958 		} xrc;
959 	} ext;
960 };
961 
962 struct ib_qp {
963 	struct ib_device       *device;
964 	struct ib_pd	       *pd;
965 	struct ib_cq	       *send_cq;
966 	struct ib_cq	       *recv_cq;
967 	struct ib_srq	       *srq;
968 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
969 	struct list_head	xrcd_list;
970 	atomic_t		usecnt; /* count times opened, mcast attaches */
971 	struct list_head	open_list;
972 	struct ib_qp           *real_qp;
973 	struct ib_uobject      *uobject;
974 	void                  (*event_handler)(struct ib_event *, void *);
975 	void		       *qp_context;
976 	u32			qp_num;
977 	enum ib_qp_type		qp_type;
978 };
979 
980 struct ib_mr {
981 	struct ib_device  *device;
982 	struct ib_pd	  *pd;
983 	struct ib_uobject *uobject;
984 	u32		   lkey;
985 	u32		   rkey;
986 	atomic_t	   usecnt; /* count number of MWs */
987 };
988 
989 struct ib_mw {
990 	struct ib_device	*device;
991 	struct ib_pd		*pd;
992 	struct ib_uobject	*uobject;
993 	u32			rkey;
994 };
995 
996 struct ib_fmr {
997 	struct ib_device	*device;
998 	struct ib_pd		*pd;
999 	struct list_head	list;
1000 	u32			lkey;
1001 	u32			rkey;
1002 };
1003 
1004 struct ib_mad;
1005 struct ib_grh;
1006 
1007 enum ib_process_mad_flags {
1008 	IB_MAD_IGNORE_MKEY	= 1,
1009 	IB_MAD_IGNORE_BKEY	= 2,
1010 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1011 };
1012 
1013 enum ib_mad_result {
1014 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1015 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1016 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1017 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1018 };
1019 
1020 #define IB_DEVICE_NAME_MAX 64
1021 
1022 struct ib_cache {
1023 	rwlock_t                lock;
1024 	struct ib_event_handler event_handler;
1025 	struct ib_pkey_cache  **pkey_cache;
1026 	struct ib_gid_cache   **gid_cache;
1027 	u8                     *lmc_cache;
1028 };
1029 
1030 struct ib_dma_mapping_ops {
1031 	int		(*mapping_error)(struct ib_device *dev,
1032 					 u64 dma_addr);
1033 	u64		(*map_single)(struct ib_device *dev,
1034 				      void *ptr, size_t size,
1035 				      enum dma_data_direction direction);
1036 	void		(*unmap_single)(struct ib_device *dev,
1037 					u64 addr, size_t size,
1038 					enum dma_data_direction direction);
1039 	u64		(*map_page)(struct ib_device *dev,
1040 				    struct page *page, unsigned long offset,
1041 				    size_t size,
1042 				    enum dma_data_direction direction);
1043 	void		(*unmap_page)(struct ib_device *dev,
1044 				      u64 addr, size_t size,
1045 				      enum dma_data_direction direction);
1046 	int		(*map_sg)(struct ib_device *dev,
1047 				  struct scatterlist *sg, int nents,
1048 				  enum dma_data_direction direction);
1049 	void		(*unmap_sg)(struct ib_device *dev,
1050 				    struct scatterlist *sg, int nents,
1051 				    enum dma_data_direction direction);
1052 	u64		(*dma_address)(struct ib_device *dev,
1053 				       struct scatterlist *sg);
1054 	unsigned int	(*dma_len)(struct ib_device *dev,
1055 				   struct scatterlist *sg);
1056 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1057 					       u64 dma_handle,
1058 					       size_t size,
1059 					       enum dma_data_direction dir);
1060 	void		(*sync_single_for_device)(struct ib_device *dev,
1061 						  u64 dma_handle,
1062 						  size_t size,
1063 						  enum dma_data_direction dir);
1064 	void		*(*alloc_coherent)(struct ib_device *dev,
1065 					   size_t size,
1066 					   u64 *dma_handle,
1067 					   gfp_t flag);
1068 	void		(*free_coherent)(struct ib_device *dev,
1069 					 size_t size, void *cpu_addr,
1070 					 u64 dma_handle);
1071 };
1072 
1073 struct iw_cm_verbs;
1074 
1075 struct ib_device {
1076 	struct device                *dma_device;
1077 
1078 	char                          name[IB_DEVICE_NAME_MAX];
1079 
1080 	struct list_head              event_handler_list;
1081 	spinlock_t                    event_handler_lock;
1082 
1083 	spinlock_t                    client_data_lock;
1084 	struct list_head              core_list;
1085 	struct list_head              client_data_list;
1086 
1087 	struct ib_cache               cache;
1088 	int                          *pkey_tbl_len;
1089 	int                          *gid_tbl_len;
1090 
1091 	int			      num_comp_vectors;
1092 
1093 	struct iw_cm_verbs	     *iwcm;
1094 
1095 	int		           (*get_protocol_stats)(struct ib_device *device,
1096 							 union rdma_protocol_stats *stats);
1097 	int		           (*query_device)(struct ib_device *device,
1098 						   struct ib_device_attr *device_attr);
1099 	int		           (*query_port)(struct ib_device *device,
1100 						 u8 port_num,
1101 						 struct ib_port_attr *port_attr);
1102 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1103 						     u8 port_num);
1104 	int		           (*query_gid)(struct ib_device *device,
1105 						u8 port_num, int index,
1106 						union ib_gid *gid);
1107 	int		           (*query_pkey)(struct ib_device *device,
1108 						 u8 port_num, u16 index, u16 *pkey);
1109 	int		           (*modify_device)(struct ib_device *device,
1110 						    int device_modify_mask,
1111 						    struct ib_device_modify *device_modify);
1112 	int		           (*modify_port)(struct ib_device *device,
1113 						  u8 port_num, int port_modify_mask,
1114 						  struct ib_port_modify *port_modify);
1115 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1116 						     struct ib_udata *udata);
1117 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1118 	int                        (*mmap)(struct ib_ucontext *context,
1119 					   struct vm_area_struct *vma);
1120 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1121 					       struct ib_ucontext *context,
1122 					       struct ib_udata *udata);
1123 	int                        (*dealloc_pd)(struct ib_pd *pd);
1124 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1125 						struct ib_ah_attr *ah_attr);
1126 	int                        (*modify_ah)(struct ib_ah *ah,
1127 						struct ib_ah_attr *ah_attr);
1128 	int                        (*query_ah)(struct ib_ah *ah,
1129 					       struct ib_ah_attr *ah_attr);
1130 	int                        (*destroy_ah)(struct ib_ah *ah);
1131 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1132 						 struct ib_srq_init_attr *srq_init_attr,
1133 						 struct ib_udata *udata);
1134 	int                        (*modify_srq)(struct ib_srq *srq,
1135 						 struct ib_srq_attr *srq_attr,
1136 						 enum ib_srq_attr_mask srq_attr_mask,
1137 						 struct ib_udata *udata);
1138 	int                        (*query_srq)(struct ib_srq *srq,
1139 						struct ib_srq_attr *srq_attr);
1140 	int                        (*destroy_srq)(struct ib_srq *srq);
1141 	int                        (*post_srq_recv)(struct ib_srq *srq,
1142 						    struct ib_recv_wr *recv_wr,
1143 						    struct ib_recv_wr **bad_recv_wr);
1144 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1145 						struct ib_qp_init_attr *qp_init_attr,
1146 						struct ib_udata *udata);
1147 	int                        (*modify_qp)(struct ib_qp *qp,
1148 						struct ib_qp_attr *qp_attr,
1149 						int qp_attr_mask,
1150 						struct ib_udata *udata);
1151 	int                        (*query_qp)(struct ib_qp *qp,
1152 					       struct ib_qp_attr *qp_attr,
1153 					       int qp_attr_mask,
1154 					       struct ib_qp_init_attr *qp_init_attr);
1155 	int                        (*destroy_qp)(struct ib_qp *qp);
1156 	int                        (*post_send)(struct ib_qp *qp,
1157 						struct ib_send_wr *send_wr,
1158 						struct ib_send_wr **bad_send_wr);
1159 	int                        (*post_recv)(struct ib_qp *qp,
1160 						struct ib_recv_wr *recv_wr,
1161 						struct ib_recv_wr **bad_recv_wr);
1162 	struct ib_cq *             (*create_cq)(struct ib_device *device, int cqe,
1163 						int comp_vector,
1164 						struct ib_ucontext *context,
1165 						struct ib_udata *udata);
1166 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1167 						u16 cq_period);
1168 	int                        (*destroy_cq)(struct ib_cq *cq);
1169 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1170 						struct ib_udata *udata);
1171 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1172 					      struct ib_wc *wc);
1173 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1174 	int                        (*req_notify_cq)(struct ib_cq *cq,
1175 						    enum ib_cq_notify_flags flags);
1176 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1177 						      int wc_cnt);
1178 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1179 						 int mr_access_flags);
1180 	struct ib_mr *             (*reg_phys_mr)(struct ib_pd *pd,
1181 						  struct ib_phys_buf *phys_buf_array,
1182 						  int num_phys_buf,
1183 						  int mr_access_flags,
1184 						  u64 *iova_start);
1185 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1186 						  u64 start, u64 length,
1187 						  u64 virt_addr,
1188 						  int mr_access_flags,
1189 						  struct ib_udata *udata);
1190 	int                        (*query_mr)(struct ib_mr *mr,
1191 					       struct ib_mr_attr *mr_attr);
1192 	int                        (*dereg_mr)(struct ib_mr *mr);
1193 	struct ib_mr *		   (*alloc_fast_reg_mr)(struct ib_pd *pd,
1194 					       int max_page_list_len);
1195 	struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1196 								   int page_list_len);
1197 	void			   (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1198 	int                        (*rereg_phys_mr)(struct ib_mr *mr,
1199 						    int mr_rereg_mask,
1200 						    struct ib_pd *pd,
1201 						    struct ib_phys_buf *phys_buf_array,
1202 						    int num_phys_buf,
1203 						    int mr_access_flags,
1204 						    u64 *iova_start);
1205 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd);
1206 	int                        (*bind_mw)(struct ib_qp *qp,
1207 					      struct ib_mw *mw,
1208 					      struct ib_mw_bind *mw_bind);
1209 	int                        (*dealloc_mw)(struct ib_mw *mw);
1210 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1211 						int mr_access_flags,
1212 						struct ib_fmr_attr *fmr_attr);
1213 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1214 						   u64 *page_list, int list_len,
1215 						   u64 iova);
1216 	int		           (*unmap_fmr)(struct list_head *fmr_list);
1217 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1218 	int                        (*attach_mcast)(struct ib_qp *qp,
1219 						   union ib_gid *gid,
1220 						   u16 lid);
1221 	int                        (*detach_mcast)(struct ib_qp *qp,
1222 						   union ib_gid *gid,
1223 						   u16 lid);
1224 	int                        (*process_mad)(struct ib_device *device,
1225 						  int process_mad_flags,
1226 						  u8 port_num,
1227 						  struct ib_wc *in_wc,
1228 						  struct ib_grh *in_grh,
1229 						  struct ib_mad *in_mad,
1230 						  struct ib_mad *out_mad);
1231 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
1232 						 struct ib_ucontext *ucontext,
1233 						 struct ib_udata *udata);
1234 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1235 
1236 	struct ib_dma_mapping_ops   *dma_ops;
1237 
1238 	struct module               *owner;
1239 	struct device                dev;
1240 	struct kobject               *ports_parent;
1241 	struct list_head             port_list;
1242 
1243 	enum {
1244 		IB_DEV_UNINITIALIZED,
1245 		IB_DEV_REGISTERED,
1246 		IB_DEV_UNREGISTERED
1247 	}                            reg_state;
1248 
1249 	int			     uverbs_abi_ver;
1250 	u64			     uverbs_cmd_mask;
1251 
1252 	char			     node_desc[64];
1253 	__be64			     node_guid;
1254 	u32			     local_dma_lkey;
1255 	u8                           node_type;
1256 	u8                           phys_port_cnt;
1257 };
1258 
1259 struct ib_client {
1260 	char  *name;
1261 	void (*add)   (struct ib_device *);
1262 	void (*remove)(struct ib_device *);
1263 
1264 	struct list_head list;
1265 };
1266 
1267 struct ib_device *ib_alloc_device(size_t size);
1268 void ib_dealloc_device(struct ib_device *device);
1269 
1270 int ib_register_device(struct ib_device *device,
1271 		       int (*port_callback)(struct ib_device *,
1272 					    u8, struct kobject *));
1273 void ib_unregister_device(struct ib_device *device);
1274 
1275 int ib_register_client   (struct ib_client *client);
1276 void ib_unregister_client(struct ib_client *client);
1277 
1278 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1279 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
1280 			 void *data);
1281 
1282 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1283 {
1284 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1285 }
1286 
1287 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1288 {
1289 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1290 }
1291 
1292 /**
1293  * ib_modify_qp_is_ok - Check that the supplied attribute mask
1294  * contains all required attributes and no attributes not allowed for
1295  * the given QP state transition.
1296  * @cur_state: Current QP state
1297  * @next_state: Next QP state
1298  * @type: QP type
1299  * @mask: Mask of supplied QP attributes
1300  *
1301  * This function is a helper function that a low-level driver's
1302  * modify_qp method can use to validate the consumer's input.  It
1303  * checks that cur_state and next_state are valid QP states, that a
1304  * transition from cur_state to next_state is allowed by the IB spec,
1305  * and that the attribute mask supplied is allowed for the transition.
1306  */
1307 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1308 		       enum ib_qp_type type, enum ib_qp_attr_mask mask);
1309 
1310 int ib_register_event_handler  (struct ib_event_handler *event_handler);
1311 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1312 void ib_dispatch_event(struct ib_event *event);
1313 
1314 int ib_query_device(struct ib_device *device,
1315 		    struct ib_device_attr *device_attr);
1316 
1317 int ib_query_port(struct ib_device *device,
1318 		  u8 port_num, struct ib_port_attr *port_attr);
1319 
1320 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1321 					       u8 port_num);
1322 
1323 int ib_query_gid(struct ib_device *device,
1324 		 u8 port_num, int index, union ib_gid *gid);
1325 
1326 int ib_query_pkey(struct ib_device *device,
1327 		  u8 port_num, u16 index, u16 *pkey);
1328 
1329 int ib_modify_device(struct ib_device *device,
1330 		     int device_modify_mask,
1331 		     struct ib_device_modify *device_modify);
1332 
1333 int ib_modify_port(struct ib_device *device,
1334 		   u8 port_num, int port_modify_mask,
1335 		   struct ib_port_modify *port_modify);
1336 
1337 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1338 		u8 *port_num, u16 *index);
1339 
1340 int ib_find_pkey(struct ib_device *device,
1341 		 u8 port_num, u16 pkey, u16 *index);
1342 
1343 /**
1344  * ib_alloc_pd - Allocates an unused protection domain.
1345  * @device: The device on which to allocate the protection domain.
1346  *
1347  * A protection domain object provides an association between QPs, shared
1348  * receive queues, address handles, memory regions, and memory windows.
1349  */
1350 struct ib_pd *ib_alloc_pd(struct ib_device *device);
1351 
1352 /**
1353  * ib_dealloc_pd - Deallocates a protection domain.
1354  * @pd: The protection domain to deallocate.
1355  */
1356 int ib_dealloc_pd(struct ib_pd *pd);
1357 
1358 /**
1359  * ib_create_ah - Creates an address handle for the given address vector.
1360  * @pd: The protection domain associated with the address handle.
1361  * @ah_attr: The attributes of the address vector.
1362  *
1363  * The address handle is used to reference a local or global destination
1364  * in all UD QP post sends.
1365  */
1366 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
1367 
1368 /**
1369  * ib_init_ah_from_wc - Initializes address handle attributes from a
1370  *   work completion.
1371  * @device: Device on which the received message arrived.
1372  * @port_num: Port on which the received message arrived.
1373  * @wc: Work completion associated with the received message.
1374  * @grh: References the received global route header.  This parameter is
1375  *   ignored unless the work completion indicates that the GRH is valid.
1376  * @ah_attr: Returned attributes that can be used when creating an address
1377  *   handle for replying to the message.
1378  */
1379 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
1380 		       struct ib_grh *grh, struct ib_ah_attr *ah_attr);
1381 
1382 /**
1383  * ib_create_ah_from_wc - Creates an address handle associated with the
1384  *   sender of the specified work completion.
1385  * @pd: The protection domain associated with the address handle.
1386  * @wc: Work completion information associated with a received message.
1387  * @grh: References the received global route header.  This parameter is
1388  *   ignored unless the work completion indicates that the GRH is valid.
1389  * @port_num: The outbound port number to associate with the address.
1390  *
1391  * The address handle is used to reference a local or global destination
1392  * in all UD QP post sends.
1393  */
1394 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
1395 				   struct ib_grh *grh, u8 port_num);
1396 
1397 /**
1398  * ib_modify_ah - Modifies the address vector associated with an address
1399  *   handle.
1400  * @ah: The address handle to modify.
1401  * @ah_attr: The new address vector attributes to associate with the
1402  *   address handle.
1403  */
1404 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1405 
1406 /**
1407  * ib_query_ah - Queries the address vector associated with an address
1408  *   handle.
1409  * @ah: The address handle to query.
1410  * @ah_attr: The address vector attributes associated with the address
1411  *   handle.
1412  */
1413 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1414 
1415 /**
1416  * ib_destroy_ah - Destroys an address handle.
1417  * @ah: The address handle to destroy.
1418  */
1419 int ib_destroy_ah(struct ib_ah *ah);
1420 
1421 /**
1422  * ib_create_srq - Creates a SRQ associated with the specified protection
1423  *   domain.
1424  * @pd: The protection domain associated with the SRQ.
1425  * @srq_init_attr: A list of initial attributes required to create the
1426  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
1427  *   the actual capabilities of the created SRQ.
1428  *
1429  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1430  * requested size of the SRQ, and set to the actual values allocated
1431  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
1432  * will always be at least as large as the requested values.
1433  */
1434 struct ib_srq *ib_create_srq(struct ib_pd *pd,
1435 			     struct ib_srq_init_attr *srq_init_attr);
1436 
1437 /**
1438  * ib_modify_srq - Modifies the attributes for the specified SRQ.
1439  * @srq: The SRQ to modify.
1440  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
1441  *   the current values of selected SRQ attributes are returned.
1442  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1443  *   are being modified.
1444  *
1445  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1446  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1447  * the number of receives queued drops below the limit.
1448  */
1449 int ib_modify_srq(struct ib_srq *srq,
1450 		  struct ib_srq_attr *srq_attr,
1451 		  enum ib_srq_attr_mask srq_attr_mask);
1452 
1453 /**
1454  * ib_query_srq - Returns the attribute list and current values for the
1455  *   specified SRQ.
1456  * @srq: The SRQ to query.
1457  * @srq_attr: The attributes of the specified SRQ.
1458  */
1459 int ib_query_srq(struct ib_srq *srq,
1460 		 struct ib_srq_attr *srq_attr);
1461 
1462 /**
1463  * ib_destroy_srq - Destroys the specified SRQ.
1464  * @srq: The SRQ to destroy.
1465  */
1466 int ib_destroy_srq(struct ib_srq *srq);
1467 
1468 /**
1469  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1470  * @srq: The SRQ to post the work request on.
1471  * @recv_wr: A list of work requests to post on the receive queue.
1472  * @bad_recv_wr: On an immediate failure, this parameter will reference
1473  *   the work request that failed to be posted on the QP.
1474  */
1475 static inline int ib_post_srq_recv(struct ib_srq *srq,
1476 				   struct ib_recv_wr *recv_wr,
1477 				   struct ib_recv_wr **bad_recv_wr)
1478 {
1479 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1480 }
1481 
1482 /**
1483  * ib_create_qp - Creates a QP associated with the specified protection
1484  *   domain.
1485  * @pd: The protection domain associated with the QP.
1486  * @qp_init_attr: A list of initial attributes required to create the
1487  *   QP.  If QP creation succeeds, then the attributes are updated to
1488  *   the actual capabilities of the created QP.
1489  */
1490 struct ib_qp *ib_create_qp(struct ib_pd *pd,
1491 			   struct ib_qp_init_attr *qp_init_attr);
1492 
1493 /**
1494  * ib_modify_qp - Modifies the attributes for the specified QP and then
1495  *   transitions the QP to the given state.
1496  * @qp: The QP to modify.
1497  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
1498  *   the current values of selected QP attributes are returned.
1499  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1500  *   are being modified.
1501  */
1502 int ib_modify_qp(struct ib_qp *qp,
1503 		 struct ib_qp_attr *qp_attr,
1504 		 int qp_attr_mask);
1505 
1506 /**
1507  * ib_query_qp - Returns the attribute list and current values for the
1508  *   specified QP.
1509  * @qp: The QP to query.
1510  * @qp_attr: The attributes of the specified QP.
1511  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1512  * @qp_init_attr: Additional attributes of the selected QP.
1513  *
1514  * The qp_attr_mask may be used to limit the query to gathering only the
1515  * selected attributes.
1516  */
1517 int ib_query_qp(struct ib_qp *qp,
1518 		struct ib_qp_attr *qp_attr,
1519 		int qp_attr_mask,
1520 		struct ib_qp_init_attr *qp_init_attr);
1521 
1522 /**
1523  * ib_destroy_qp - Destroys the specified QP.
1524  * @qp: The QP to destroy.
1525  */
1526 int ib_destroy_qp(struct ib_qp *qp);
1527 
1528 /**
1529  * ib_open_qp - Obtain a reference to an existing sharable QP.
1530  * @xrcd - XRC domain
1531  * @qp_open_attr: Attributes identifying the QP to open.
1532  *
1533  * Returns a reference to a sharable QP.
1534  */
1535 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
1536 			 struct ib_qp_open_attr *qp_open_attr);
1537 
1538 /**
1539  * ib_close_qp - Release an external reference to a QP.
1540  * @qp: The QP handle to release
1541  *
1542  * The opened QP handle is released by the caller.  The underlying
1543  * shared QP is not destroyed until all internal references are released.
1544  */
1545 int ib_close_qp(struct ib_qp *qp);
1546 
1547 /**
1548  * ib_post_send - Posts a list of work requests to the send queue of
1549  *   the specified QP.
1550  * @qp: The QP to post the work request on.
1551  * @send_wr: A list of work requests to post on the send queue.
1552  * @bad_send_wr: On an immediate failure, this parameter will reference
1553  *   the work request that failed to be posted on the QP.
1554  *
1555  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
1556  * error is returned, the QP state shall not be affected,
1557  * ib_post_send() will return an immediate error after queueing any
1558  * earlier work requests in the list.
1559  */
1560 static inline int ib_post_send(struct ib_qp *qp,
1561 			       struct ib_send_wr *send_wr,
1562 			       struct ib_send_wr **bad_send_wr)
1563 {
1564 	return qp->device->post_send(qp, send_wr, bad_send_wr);
1565 }
1566 
1567 /**
1568  * ib_post_recv - Posts a list of work requests to the receive queue of
1569  *   the specified QP.
1570  * @qp: The QP to post the work request on.
1571  * @recv_wr: A list of work requests to post on the receive queue.
1572  * @bad_recv_wr: On an immediate failure, this parameter will reference
1573  *   the work request that failed to be posted on the QP.
1574  */
1575 static inline int ib_post_recv(struct ib_qp *qp,
1576 			       struct ib_recv_wr *recv_wr,
1577 			       struct ib_recv_wr **bad_recv_wr)
1578 {
1579 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1580 }
1581 
1582 /**
1583  * ib_create_cq - Creates a CQ on the specified device.
1584  * @device: The device on which to create the CQ.
1585  * @comp_handler: A user-specified callback that is invoked when a
1586  *   completion event occurs on the CQ.
1587  * @event_handler: A user-specified callback that is invoked when an
1588  *   asynchronous event not associated with a completion occurs on the CQ.
1589  * @cq_context: Context associated with the CQ returned to the user via
1590  *   the associated completion and event handlers.
1591  * @cqe: The minimum size of the CQ.
1592  * @comp_vector - Completion vector used to signal completion events.
1593  *     Must be >= 0 and < context->num_comp_vectors.
1594  *
1595  * Users can examine the cq structure to determine the actual CQ size.
1596  */
1597 struct ib_cq *ib_create_cq(struct ib_device *device,
1598 			   ib_comp_handler comp_handler,
1599 			   void (*event_handler)(struct ib_event *, void *),
1600 			   void *cq_context, int cqe, int comp_vector);
1601 
1602 /**
1603  * ib_resize_cq - Modifies the capacity of the CQ.
1604  * @cq: The CQ to resize.
1605  * @cqe: The minimum size of the CQ.
1606  *
1607  * Users can examine the cq structure to determine the actual CQ size.
1608  */
1609 int ib_resize_cq(struct ib_cq *cq, int cqe);
1610 
1611 /**
1612  * ib_modify_cq - Modifies moderation params of the CQ
1613  * @cq: The CQ to modify.
1614  * @cq_count: number of CQEs that will trigger an event
1615  * @cq_period: max period of time in usec before triggering an event
1616  *
1617  */
1618 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1619 
1620 /**
1621  * ib_destroy_cq - Destroys the specified CQ.
1622  * @cq: The CQ to destroy.
1623  */
1624 int ib_destroy_cq(struct ib_cq *cq);
1625 
1626 /**
1627  * ib_poll_cq - poll a CQ for completion(s)
1628  * @cq:the CQ being polled
1629  * @num_entries:maximum number of completions to return
1630  * @wc:array of at least @num_entries &struct ib_wc where completions
1631  *   will be returned
1632  *
1633  * Poll a CQ for (possibly multiple) completions.  If the return value
1634  * is < 0, an error occurred.  If the return value is >= 0, it is the
1635  * number of completions returned.  If the return value is
1636  * non-negative and < num_entries, then the CQ was emptied.
1637  */
1638 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1639 			     struct ib_wc *wc)
1640 {
1641 	return cq->device->poll_cq(cq, num_entries, wc);
1642 }
1643 
1644 /**
1645  * ib_peek_cq - Returns the number of unreaped completions currently
1646  *   on the specified CQ.
1647  * @cq: The CQ to peek.
1648  * @wc_cnt: A minimum number of unreaped completions to check for.
1649  *
1650  * If the number of unreaped completions is greater than or equal to wc_cnt,
1651  * this function returns wc_cnt, otherwise, it returns the actual number of
1652  * unreaped completions.
1653  */
1654 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1655 
1656 /**
1657  * ib_req_notify_cq - Request completion notification on a CQ.
1658  * @cq: The CQ to generate an event for.
1659  * @flags:
1660  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1661  *   to request an event on the next solicited event or next work
1662  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1663  *   may also be |ed in to request a hint about missed events, as
1664  *   described below.
1665  *
1666  * Return Value:
1667  *    < 0 means an error occurred while requesting notification
1668  *   == 0 means notification was requested successfully, and if
1669  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1670  *        were missed and it is safe to wait for another event.  In
1671  *        this case is it guaranteed that any work completions added
1672  *        to the CQ since the last CQ poll will trigger a completion
1673  *        notification event.
1674  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1675  *        in.  It means that the consumer must poll the CQ again to
1676  *        make sure it is empty to avoid missing an event because of a
1677  *        race between requesting notification and an entry being
1678  *        added to the CQ.  This return value means it is possible
1679  *        (but not guaranteed) that a work completion has been added
1680  *        to the CQ since the last poll without triggering a
1681  *        completion notification event.
1682  */
1683 static inline int ib_req_notify_cq(struct ib_cq *cq,
1684 				   enum ib_cq_notify_flags flags)
1685 {
1686 	return cq->device->req_notify_cq(cq, flags);
1687 }
1688 
1689 /**
1690  * ib_req_ncomp_notif - Request completion notification when there are
1691  *   at least the specified number of unreaped completions on the CQ.
1692  * @cq: The CQ to generate an event for.
1693  * @wc_cnt: The number of unreaped completions that should be on the
1694  *   CQ before an event is generated.
1695  */
1696 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1697 {
1698 	return cq->device->req_ncomp_notif ?
1699 		cq->device->req_ncomp_notif(cq, wc_cnt) :
1700 		-ENOSYS;
1701 }
1702 
1703 /**
1704  * ib_get_dma_mr - Returns a memory region for system memory that is
1705  *   usable for DMA.
1706  * @pd: The protection domain associated with the memory region.
1707  * @mr_access_flags: Specifies the memory access rights.
1708  *
1709  * Note that the ib_dma_*() functions defined below must be used
1710  * to create/destroy addresses used with the Lkey or Rkey returned
1711  * by ib_get_dma_mr().
1712  */
1713 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1714 
1715 /**
1716  * ib_dma_mapping_error - check a DMA addr for error
1717  * @dev: The device for which the dma_addr was created
1718  * @dma_addr: The DMA address to check
1719  */
1720 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1721 {
1722 	if (dev->dma_ops)
1723 		return dev->dma_ops->mapping_error(dev, dma_addr);
1724 	return dma_mapping_error(dev->dma_device, dma_addr);
1725 }
1726 
1727 /**
1728  * ib_dma_map_single - Map a kernel virtual address to DMA address
1729  * @dev: The device for which the dma_addr is to be created
1730  * @cpu_addr: The kernel virtual address
1731  * @size: The size of the region in bytes
1732  * @direction: The direction of the DMA
1733  */
1734 static inline u64 ib_dma_map_single(struct ib_device *dev,
1735 				    void *cpu_addr, size_t size,
1736 				    enum dma_data_direction direction)
1737 {
1738 	if (dev->dma_ops)
1739 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1740 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1741 }
1742 
1743 /**
1744  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1745  * @dev: The device for which the DMA address was created
1746  * @addr: The DMA address
1747  * @size: The size of the region in bytes
1748  * @direction: The direction of the DMA
1749  */
1750 static inline void ib_dma_unmap_single(struct ib_device *dev,
1751 				       u64 addr, size_t size,
1752 				       enum dma_data_direction direction)
1753 {
1754 	if (dev->dma_ops)
1755 		dev->dma_ops->unmap_single(dev, addr, size, direction);
1756 	else
1757 		dma_unmap_single(dev->dma_device, addr, size, direction);
1758 }
1759 
1760 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
1761 					  void *cpu_addr, size_t size,
1762 					  enum dma_data_direction direction,
1763 					  struct dma_attrs *attrs)
1764 {
1765 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
1766 				    direction, attrs);
1767 }
1768 
1769 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
1770 					     u64 addr, size_t size,
1771 					     enum dma_data_direction direction,
1772 					     struct dma_attrs *attrs)
1773 {
1774 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
1775 				      direction, attrs);
1776 }
1777 
1778 /**
1779  * ib_dma_map_page - Map a physical page to DMA address
1780  * @dev: The device for which the dma_addr is to be created
1781  * @page: The page to be mapped
1782  * @offset: The offset within the page
1783  * @size: The size of the region in bytes
1784  * @direction: The direction of the DMA
1785  */
1786 static inline u64 ib_dma_map_page(struct ib_device *dev,
1787 				  struct page *page,
1788 				  unsigned long offset,
1789 				  size_t size,
1790 					 enum dma_data_direction direction)
1791 {
1792 	if (dev->dma_ops)
1793 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
1794 	return dma_map_page(dev->dma_device, page, offset, size, direction);
1795 }
1796 
1797 /**
1798  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1799  * @dev: The device for which the DMA address was created
1800  * @addr: The DMA address
1801  * @size: The size of the region in bytes
1802  * @direction: The direction of the DMA
1803  */
1804 static inline void ib_dma_unmap_page(struct ib_device *dev,
1805 				     u64 addr, size_t size,
1806 				     enum dma_data_direction direction)
1807 {
1808 	if (dev->dma_ops)
1809 		dev->dma_ops->unmap_page(dev, addr, size, direction);
1810 	else
1811 		dma_unmap_page(dev->dma_device, addr, size, direction);
1812 }
1813 
1814 /**
1815  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
1816  * @dev: The device for which the DMA addresses are to be created
1817  * @sg: The array of scatter/gather entries
1818  * @nents: The number of scatter/gather entries
1819  * @direction: The direction of the DMA
1820  */
1821 static inline int ib_dma_map_sg(struct ib_device *dev,
1822 				struct scatterlist *sg, int nents,
1823 				enum dma_data_direction direction)
1824 {
1825 	if (dev->dma_ops)
1826 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
1827 	return dma_map_sg(dev->dma_device, sg, nents, direction);
1828 }
1829 
1830 /**
1831  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
1832  * @dev: The device for which the DMA addresses were created
1833  * @sg: The array of scatter/gather entries
1834  * @nents: The number of scatter/gather entries
1835  * @direction: The direction of the DMA
1836  */
1837 static inline void ib_dma_unmap_sg(struct ib_device *dev,
1838 				   struct scatterlist *sg, int nents,
1839 				   enum dma_data_direction direction)
1840 {
1841 	if (dev->dma_ops)
1842 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
1843 	else
1844 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
1845 }
1846 
1847 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
1848 				      struct scatterlist *sg, int nents,
1849 				      enum dma_data_direction direction,
1850 				      struct dma_attrs *attrs)
1851 {
1852 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1853 }
1854 
1855 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
1856 					 struct scatterlist *sg, int nents,
1857 					 enum dma_data_direction direction,
1858 					 struct dma_attrs *attrs)
1859 {
1860 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1861 }
1862 /**
1863  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
1864  * @dev: The device for which the DMA addresses were created
1865  * @sg: The scatter/gather entry
1866  */
1867 static inline u64 ib_sg_dma_address(struct ib_device *dev,
1868 				    struct scatterlist *sg)
1869 {
1870 	if (dev->dma_ops)
1871 		return dev->dma_ops->dma_address(dev, sg);
1872 	return sg_dma_address(sg);
1873 }
1874 
1875 /**
1876  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
1877  * @dev: The device for which the DMA addresses were created
1878  * @sg: The scatter/gather entry
1879  */
1880 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
1881 					 struct scatterlist *sg)
1882 {
1883 	if (dev->dma_ops)
1884 		return dev->dma_ops->dma_len(dev, sg);
1885 	return sg_dma_len(sg);
1886 }
1887 
1888 /**
1889  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
1890  * @dev: The device for which the DMA address was created
1891  * @addr: The DMA address
1892  * @size: The size of the region in bytes
1893  * @dir: The direction of the DMA
1894  */
1895 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
1896 					      u64 addr,
1897 					      size_t size,
1898 					      enum dma_data_direction dir)
1899 {
1900 	if (dev->dma_ops)
1901 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
1902 	else
1903 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
1904 }
1905 
1906 /**
1907  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
1908  * @dev: The device for which the DMA address was created
1909  * @addr: The DMA address
1910  * @size: The size of the region in bytes
1911  * @dir: The direction of the DMA
1912  */
1913 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
1914 						 u64 addr,
1915 						 size_t size,
1916 						 enum dma_data_direction dir)
1917 {
1918 	if (dev->dma_ops)
1919 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
1920 	else
1921 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
1922 }
1923 
1924 /**
1925  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
1926  * @dev: The device for which the DMA address is requested
1927  * @size: The size of the region to allocate in bytes
1928  * @dma_handle: A pointer for returning the DMA address of the region
1929  * @flag: memory allocator flags
1930  */
1931 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
1932 					   size_t size,
1933 					   u64 *dma_handle,
1934 					   gfp_t flag)
1935 {
1936 	if (dev->dma_ops)
1937 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
1938 	else {
1939 		dma_addr_t handle;
1940 		void *ret;
1941 
1942 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
1943 		*dma_handle = handle;
1944 		return ret;
1945 	}
1946 }
1947 
1948 /**
1949  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
1950  * @dev: The device for which the DMA addresses were allocated
1951  * @size: The size of the region
1952  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
1953  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
1954  */
1955 static inline void ib_dma_free_coherent(struct ib_device *dev,
1956 					size_t size, void *cpu_addr,
1957 					u64 dma_handle)
1958 {
1959 	if (dev->dma_ops)
1960 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
1961 	else
1962 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
1963 }
1964 
1965 /**
1966  * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
1967  *   by an HCA.
1968  * @pd: The protection domain associated assigned to the registered region.
1969  * @phys_buf_array: Specifies a list of physical buffers to use in the
1970  *   memory region.
1971  * @num_phys_buf: Specifies the size of the phys_buf_array.
1972  * @mr_access_flags: Specifies the memory access rights.
1973  * @iova_start: The offset of the region's starting I/O virtual address.
1974  */
1975 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
1976 			     struct ib_phys_buf *phys_buf_array,
1977 			     int num_phys_buf,
1978 			     int mr_access_flags,
1979 			     u64 *iova_start);
1980 
1981 /**
1982  * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
1983  *   Conceptually, this call performs the functions deregister memory region
1984  *   followed by register physical memory region.  Where possible,
1985  *   resources are reused instead of deallocated and reallocated.
1986  * @mr: The memory region to modify.
1987  * @mr_rereg_mask: A bit-mask used to indicate which of the following
1988  *   properties of the memory region are being modified.
1989  * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
1990  *   the new protection domain to associated with the memory region,
1991  *   otherwise, this parameter is ignored.
1992  * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1993  *   field specifies a list of physical buffers to use in the new
1994  *   translation, otherwise, this parameter is ignored.
1995  * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1996  *   field specifies the size of the phys_buf_array, otherwise, this
1997  *   parameter is ignored.
1998  * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
1999  *   field specifies the new memory access rights, otherwise, this
2000  *   parameter is ignored.
2001  * @iova_start: The offset of the region's starting I/O virtual address.
2002  */
2003 int ib_rereg_phys_mr(struct ib_mr *mr,
2004 		     int mr_rereg_mask,
2005 		     struct ib_pd *pd,
2006 		     struct ib_phys_buf *phys_buf_array,
2007 		     int num_phys_buf,
2008 		     int mr_access_flags,
2009 		     u64 *iova_start);
2010 
2011 /**
2012  * ib_query_mr - Retrieves information about a specific memory region.
2013  * @mr: The memory region to retrieve information about.
2014  * @mr_attr: The attributes of the specified memory region.
2015  */
2016 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2017 
2018 /**
2019  * ib_dereg_mr - Deregisters a memory region and removes it from the
2020  *   HCA translation table.
2021  * @mr: The memory region to deregister.
2022  */
2023 int ib_dereg_mr(struct ib_mr *mr);
2024 
2025 /**
2026  * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2027  *   IB_WR_FAST_REG_MR send work request.
2028  * @pd: The protection domain associated with the region.
2029  * @max_page_list_len: requested max physical buffer list length to be
2030  *   used with fast register work requests for this MR.
2031  */
2032 struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
2033 
2034 /**
2035  * ib_alloc_fast_reg_page_list - Allocates a page list array
2036  * @device - ib device pointer.
2037  * @page_list_len - size of the page list array to be allocated.
2038  *
2039  * This allocates and returns a struct ib_fast_reg_page_list * and a
2040  * page_list array that is at least page_list_len in size.  The actual
2041  * size is returned in max_page_list_len.  The caller is responsible
2042  * for initializing the contents of the page_list array before posting
2043  * a send work request with the IB_WC_FAST_REG_MR opcode.
2044  *
2045  * The page_list array entries must be translated using one of the
2046  * ib_dma_*() functions just like the addresses passed to
2047  * ib_map_phys_fmr().  Once the ib_post_send() is issued, the struct
2048  * ib_fast_reg_page_list must not be modified by the caller until the
2049  * IB_WC_FAST_REG_MR work request completes.
2050  */
2051 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2052 				struct ib_device *device, int page_list_len);
2053 
2054 /**
2055  * ib_free_fast_reg_page_list - Deallocates a previously allocated
2056  *   page list array.
2057  * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2058  */
2059 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2060 
2061 /**
2062  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2063  *   R_Key and L_Key.
2064  * @mr - struct ib_mr pointer to be updated.
2065  * @newkey - new key to be used.
2066  */
2067 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2068 {
2069 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2070 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2071 }
2072 
2073 /**
2074  * ib_alloc_mw - Allocates a memory window.
2075  * @pd: The protection domain associated with the memory window.
2076  */
2077 struct ib_mw *ib_alloc_mw(struct ib_pd *pd);
2078 
2079 /**
2080  * ib_bind_mw - Posts a work request to the send queue of the specified
2081  *   QP, which binds the memory window to the given address range and
2082  *   remote access attributes.
2083  * @qp: QP to post the bind work request on.
2084  * @mw: The memory window to bind.
2085  * @mw_bind: Specifies information about the memory window, including
2086  *   its address range, remote access rights, and associated memory region.
2087  */
2088 static inline int ib_bind_mw(struct ib_qp *qp,
2089 			     struct ib_mw *mw,
2090 			     struct ib_mw_bind *mw_bind)
2091 {
2092 	/* XXX reference counting in corresponding MR? */
2093 	return mw->device->bind_mw ?
2094 		mw->device->bind_mw(qp, mw, mw_bind) :
2095 		-ENOSYS;
2096 }
2097 
2098 /**
2099  * ib_dealloc_mw - Deallocates a memory window.
2100  * @mw: The memory window to deallocate.
2101  */
2102 int ib_dealloc_mw(struct ib_mw *mw);
2103 
2104 /**
2105  * ib_alloc_fmr - Allocates a unmapped fast memory region.
2106  * @pd: The protection domain associated with the unmapped region.
2107  * @mr_access_flags: Specifies the memory access rights.
2108  * @fmr_attr: Attributes of the unmapped region.
2109  *
2110  * A fast memory region must be mapped before it can be used as part of
2111  * a work request.
2112  */
2113 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2114 			    int mr_access_flags,
2115 			    struct ib_fmr_attr *fmr_attr);
2116 
2117 /**
2118  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2119  * @fmr: The fast memory region to associate with the pages.
2120  * @page_list: An array of physical pages to map to the fast memory region.
2121  * @list_len: The number of pages in page_list.
2122  * @iova: The I/O virtual address to use with the mapped region.
2123  */
2124 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2125 				  u64 *page_list, int list_len,
2126 				  u64 iova)
2127 {
2128 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2129 }
2130 
2131 /**
2132  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2133  * @fmr_list: A linked list of fast memory regions to unmap.
2134  */
2135 int ib_unmap_fmr(struct list_head *fmr_list);
2136 
2137 /**
2138  * ib_dealloc_fmr - Deallocates a fast memory region.
2139  * @fmr: The fast memory region to deallocate.
2140  */
2141 int ib_dealloc_fmr(struct ib_fmr *fmr);
2142 
2143 /**
2144  * ib_attach_mcast - Attaches the specified QP to a multicast group.
2145  * @qp: QP to attach to the multicast group.  The QP must be type
2146  *   IB_QPT_UD.
2147  * @gid: Multicast group GID.
2148  * @lid: Multicast group LID in host byte order.
2149  *
2150  * In order to send and receive multicast packets, subnet
2151  * administration must have created the multicast group and configured
2152  * the fabric appropriately.  The port associated with the specified
2153  * QP must also be a member of the multicast group.
2154  */
2155 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2156 
2157 /**
2158  * ib_detach_mcast - Detaches the specified QP from a multicast group.
2159  * @qp: QP to detach from the multicast group.
2160  * @gid: Multicast group GID.
2161  * @lid: Multicast group LID in host byte order.
2162  */
2163 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2164 
2165 /**
2166  * ib_alloc_xrcd - Allocates an XRC domain.
2167  * @device: The device on which to allocate the XRC domain.
2168  */
2169 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2170 
2171 /**
2172  * ib_dealloc_xrcd - Deallocates an XRC domain.
2173  * @xrcd: The XRC domain to deallocate.
2174  */
2175 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2176 
2177 #endif /* IB_VERBS_H */
2178