xref: /openbmc/linux/include/rdma/ib_verbs.h (revision d78c317f)
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 struct ib_protocol_stats {
243 	/* TBD... */
244 };
245 
246 struct iw_protocol_stats {
247 	u64	ipInReceives;
248 	u64	ipInHdrErrors;
249 	u64	ipInTooBigErrors;
250 	u64	ipInNoRoutes;
251 	u64	ipInAddrErrors;
252 	u64	ipInUnknownProtos;
253 	u64	ipInTruncatedPkts;
254 	u64	ipInDiscards;
255 	u64	ipInDelivers;
256 	u64	ipOutForwDatagrams;
257 	u64	ipOutRequests;
258 	u64	ipOutDiscards;
259 	u64	ipOutNoRoutes;
260 	u64	ipReasmTimeout;
261 	u64	ipReasmReqds;
262 	u64	ipReasmOKs;
263 	u64	ipReasmFails;
264 	u64	ipFragOKs;
265 	u64	ipFragFails;
266 	u64	ipFragCreates;
267 	u64	ipInMcastPkts;
268 	u64	ipOutMcastPkts;
269 	u64	ipInBcastPkts;
270 	u64	ipOutBcastPkts;
271 
272 	u64	tcpRtoAlgorithm;
273 	u64	tcpRtoMin;
274 	u64	tcpRtoMax;
275 	u64	tcpMaxConn;
276 	u64	tcpActiveOpens;
277 	u64	tcpPassiveOpens;
278 	u64	tcpAttemptFails;
279 	u64	tcpEstabResets;
280 	u64	tcpCurrEstab;
281 	u64	tcpInSegs;
282 	u64	tcpOutSegs;
283 	u64	tcpRetransSegs;
284 	u64	tcpInErrs;
285 	u64	tcpOutRsts;
286 };
287 
288 union rdma_protocol_stats {
289 	struct ib_protocol_stats	ib;
290 	struct iw_protocol_stats	iw;
291 };
292 
293 struct ib_port_attr {
294 	enum ib_port_state	state;
295 	enum ib_mtu		max_mtu;
296 	enum ib_mtu		active_mtu;
297 	int			gid_tbl_len;
298 	u32			port_cap_flags;
299 	u32			max_msg_sz;
300 	u32			bad_pkey_cntr;
301 	u32			qkey_viol_cntr;
302 	u16			pkey_tbl_len;
303 	u16			lid;
304 	u16			sm_lid;
305 	u8			lmc;
306 	u8			max_vl_num;
307 	u8			sm_sl;
308 	u8			subnet_timeout;
309 	u8			init_type_reply;
310 	u8			active_width;
311 	u8			active_speed;
312 	u8                      phys_state;
313 };
314 
315 enum ib_device_modify_flags {
316 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
317 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
318 };
319 
320 struct ib_device_modify {
321 	u64	sys_image_guid;
322 	char	node_desc[64];
323 };
324 
325 enum ib_port_modify_flags {
326 	IB_PORT_SHUTDOWN		= 1,
327 	IB_PORT_INIT_TYPE		= (1<<2),
328 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
329 };
330 
331 struct ib_port_modify {
332 	u32	set_port_cap_mask;
333 	u32	clr_port_cap_mask;
334 	u8	init_type;
335 };
336 
337 enum ib_event_type {
338 	IB_EVENT_CQ_ERR,
339 	IB_EVENT_QP_FATAL,
340 	IB_EVENT_QP_REQ_ERR,
341 	IB_EVENT_QP_ACCESS_ERR,
342 	IB_EVENT_COMM_EST,
343 	IB_EVENT_SQ_DRAINED,
344 	IB_EVENT_PATH_MIG,
345 	IB_EVENT_PATH_MIG_ERR,
346 	IB_EVENT_DEVICE_FATAL,
347 	IB_EVENT_PORT_ACTIVE,
348 	IB_EVENT_PORT_ERR,
349 	IB_EVENT_LID_CHANGE,
350 	IB_EVENT_PKEY_CHANGE,
351 	IB_EVENT_SM_CHANGE,
352 	IB_EVENT_SRQ_ERR,
353 	IB_EVENT_SRQ_LIMIT_REACHED,
354 	IB_EVENT_QP_LAST_WQE_REACHED,
355 	IB_EVENT_CLIENT_REREGISTER,
356 	IB_EVENT_GID_CHANGE,
357 };
358 
359 struct ib_event {
360 	struct ib_device	*device;
361 	union {
362 		struct ib_cq	*cq;
363 		struct ib_qp	*qp;
364 		struct ib_srq	*srq;
365 		u8		port_num;
366 	} element;
367 	enum ib_event_type	event;
368 };
369 
370 struct ib_event_handler {
371 	struct ib_device *device;
372 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
373 	struct list_head  list;
374 };
375 
376 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
377 	do {							\
378 		(_ptr)->device  = _device;			\
379 		(_ptr)->handler = _handler;			\
380 		INIT_LIST_HEAD(&(_ptr)->list);			\
381 	} while (0)
382 
383 struct ib_global_route {
384 	union ib_gid	dgid;
385 	u32		flow_label;
386 	u8		sgid_index;
387 	u8		hop_limit;
388 	u8		traffic_class;
389 };
390 
391 struct ib_grh {
392 	__be32		version_tclass_flow;
393 	__be16		paylen;
394 	u8		next_hdr;
395 	u8		hop_limit;
396 	union ib_gid	sgid;
397 	union ib_gid	dgid;
398 };
399 
400 enum {
401 	IB_MULTICAST_QPN = 0xffffff
402 };
403 
404 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
405 
406 enum ib_ah_flags {
407 	IB_AH_GRH	= 1
408 };
409 
410 enum ib_rate {
411 	IB_RATE_PORT_CURRENT = 0,
412 	IB_RATE_2_5_GBPS = 2,
413 	IB_RATE_5_GBPS   = 5,
414 	IB_RATE_10_GBPS  = 3,
415 	IB_RATE_20_GBPS  = 6,
416 	IB_RATE_30_GBPS  = 4,
417 	IB_RATE_40_GBPS  = 7,
418 	IB_RATE_60_GBPS  = 8,
419 	IB_RATE_80_GBPS  = 9,
420 	IB_RATE_120_GBPS = 10,
421 	IB_RATE_14_GBPS  = 11,
422 	IB_RATE_56_GBPS  = 12,
423 	IB_RATE_112_GBPS = 13,
424 	IB_RATE_168_GBPS = 14,
425 	IB_RATE_25_GBPS  = 15,
426 	IB_RATE_100_GBPS = 16,
427 	IB_RATE_200_GBPS = 17,
428 	IB_RATE_300_GBPS = 18
429 };
430 
431 /**
432  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
433  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
434  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
435  * @rate: rate to convert.
436  */
437 int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
438 
439 /**
440  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
441  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
442  * @rate: rate to convert.
443  */
444 int ib_rate_to_mbps(enum ib_rate rate) __attribute_const__;
445 
446 /**
447  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
448  * enum.
449  * @mult: multiple to convert.
450  */
451 enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
452 
453 struct ib_ah_attr {
454 	struct ib_global_route	grh;
455 	u16			dlid;
456 	u8			sl;
457 	u8			src_path_bits;
458 	u8			static_rate;
459 	u8			ah_flags;
460 	u8			port_num;
461 };
462 
463 enum ib_wc_status {
464 	IB_WC_SUCCESS,
465 	IB_WC_LOC_LEN_ERR,
466 	IB_WC_LOC_QP_OP_ERR,
467 	IB_WC_LOC_EEC_OP_ERR,
468 	IB_WC_LOC_PROT_ERR,
469 	IB_WC_WR_FLUSH_ERR,
470 	IB_WC_MW_BIND_ERR,
471 	IB_WC_BAD_RESP_ERR,
472 	IB_WC_LOC_ACCESS_ERR,
473 	IB_WC_REM_INV_REQ_ERR,
474 	IB_WC_REM_ACCESS_ERR,
475 	IB_WC_REM_OP_ERR,
476 	IB_WC_RETRY_EXC_ERR,
477 	IB_WC_RNR_RETRY_EXC_ERR,
478 	IB_WC_LOC_RDD_VIOL_ERR,
479 	IB_WC_REM_INV_RD_REQ_ERR,
480 	IB_WC_REM_ABORT_ERR,
481 	IB_WC_INV_EECN_ERR,
482 	IB_WC_INV_EEC_STATE_ERR,
483 	IB_WC_FATAL_ERR,
484 	IB_WC_RESP_TIMEOUT_ERR,
485 	IB_WC_GENERAL_ERR
486 };
487 
488 enum ib_wc_opcode {
489 	IB_WC_SEND,
490 	IB_WC_RDMA_WRITE,
491 	IB_WC_RDMA_READ,
492 	IB_WC_COMP_SWAP,
493 	IB_WC_FETCH_ADD,
494 	IB_WC_BIND_MW,
495 	IB_WC_LSO,
496 	IB_WC_LOCAL_INV,
497 	IB_WC_FAST_REG_MR,
498 	IB_WC_MASKED_COMP_SWAP,
499 	IB_WC_MASKED_FETCH_ADD,
500 /*
501  * Set value of IB_WC_RECV so consumers can test if a completion is a
502  * receive by testing (opcode & IB_WC_RECV).
503  */
504 	IB_WC_RECV			= 1 << 7,
505 	IB_WC_RECV_RDMA_WITH_IMM
506 };
507 
508 enum ib_wc_flags {
509 	IB_WC_GRH		= 1,
510 	IB_WC_WITH_IMM		= (1<<1),
511 	IB_WC_WITH_INVALIDATE	= (1<<2),
512 };
513 
514 struct ib_wc {
515 	u64			wr_id;
516 	enum ib_wc_status	status;
517 	enum ib_wc_opcode	opcode;
518 	u32			vendor_err;
519 	u32			byte_len;
520 	struct ib_qp	       *qp;
521 	union {
522 		__be32		imm_data;
523 		u32		invalidate_rkey;
524 	} ex;
525 	u32			src_qp;
526 	int			wc_flags;
527 	u16			pkey_index;
528 	u16			slid;
529 	u8			sl;
530 	u8			dlid_path_bits;
531 	u8			port_num;	/* valid only for DR SMPs on switches */
532 	int			csum_ok;
533 };
534 
535 enum ib_cq_notify_flags {
536 	IB_CQ_SOLICITED			= 1 << 0,
537 	IB_CQ_NEXT_COMP			= 1 << 1,
538 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
539 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
540 };
541 
542 enum ib_srq_type {
543 	IB_SRQT_BASIC,
544 	IB_SRQT_XRC
545 };
546 
547 enum ib_srq_attr_mask {
548 	IB_SRQ_MAX_WR	= 1 << 0,
549 	IB_SRQ_LIMIT	= 1 << 1,
550 };
551 
552 struct ib_srq_attr {
553 	u32	max_wr;
554 	u32	max_sge;
555 	u32	srq_limit;
556 };
557 
558 struct ib_srq_init_attr {
559 	void		      (*event_handler)(struct ib_event *, void *);
560 	void		       *srq_context;
561 	struct ib_srq_attr	attr;
562 	enum ib_srq_type	srq_type;
563 
564 	union {
565 		struct {
566 			struct ib_xrcd *xrcd;
567 			struct ib_cq   *cq;
568 		} xrc;
569 	} ext;
570 };
571 
572 struct ib_qp_cap {
573 	u32	max_send_wr;
574 	u32	max_recv_wr;
575 	u32	max_send_sge;
576 	u32	max_recv_sge;
577 	u32	max_inline_data;
578 };
579 
580 enum ib_sig_type {
581 	IB_SIGNAL_ALL_WR,
582 	IB_SIGNAL_REQ_WR
583 };
584 
585 enum ib_qp_type {
586 	/*
587 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
588 	 * here (and in that order) since the MAD layer uses them as
589 	 * indices into a 2-entry table.
590 	 */
591 	IB_QPT_SMI,
592 	IB_QPT_GSI,
593 
594 	IB_QPT_RC,
595 	IB_QPT_UC,
596 	IB_QPT_UD,
597 	IB_QPT_RAW_IPV6,
598 	IB_QPT_RAW_ETHERTYPE,
599 	/* Save 8 for RAW_PACKET */
600 	IB_QPT_XRC_INI = 9,
601 	IB_QPT_XRC_TGT,
602 	IB_QPT_MAX
603 };
604 
605 enum ib_qp_create_flags {
606 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
607 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
608 };
609 
610 struct ib_qp_init_attr {
611 	void                  (*event_handler)(struct ib_event *, void *);
612 	void		       *qp_context;
613 	struct ib_cq	       *send_cq;
614 	struct ib_cq	       *recv_cq;
615 	struct ib_srq	       *srq;
616 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
617 	struct ib_qp_cap	cap;
618 	enum ib_sig_type	sq_sig_type;
619 	enum ib_qp_type		qp_type;
620 	enum ib_qp_create_flags	create_flags;
621 	u8			port_num; /* special QP types only */
622 };
623 
624 struct ib_qp_open_attr {
625 	void                  (*event_handler)(struct ib_event *, void *);
626 	void		       *qp_context;
627 	u32			qp_num;
628 	enum ib_qp_type		qp_type;
629 };
630 
631 enum ib_rnr_timeout {
632 	IB_RNR_TIMER_655_36 =  0,
633 	IB_RNR_TIMER_000_01 =  1,
634 	IB_RNR_TIMER_000_02 =  2,
635 	IB_RNR_TIMER_000_03 =  3,
636 	IB_RNR_TIMER_000_04 =  4,
637 	IB_RNR_TIMER_000_06 =  5,
638 	IB_RNR_TIMER_000_08 =  6,
639 	IB_RNR_TIMER_000_12 =  7,
640 	IB_RNR_TIMER_000_16 =  8,
641 	IB_RNR_TIMER_000_24 =  9,
642 	IB_RNR_TIMER_000_32 = 10,
643 	IB_RNR_TIMER_000_48 = 11,
644 	IB_RNR_TIMER_000_64 = 12,
645 	IB_RNR_TIMER_000_96 = 13,
646 	IB_RNR_TIMER_001_28 = 14,
647 	IB_RNR_TIMER_001_92 = 15,
648 	IB_RNR_TIMER_002_56 = 16,
649 	IB_RNR_TIMER_003_84 = 17,
650 	IB_RNR_TIMER_005_12 = 18,
651 	IB_RNR_TIMER_007_68 = 19,
652 	IB_RNR_TIMER_010_24 = 20,
653 	IB_RNR_TIMER_015_36 = 21,
654 	IB_RNR_TIMER_020_48 = 22,
655 	IB_RNR_TIMER_030_72 = 23,
656 	IB_RNR_TIMER_040_96 = 24,
657 	IB_RNR_TIMER_061_44 = 25,
658 	IB_RNR_TIMER_081_92 = 26,
659 	IB_RNR_TIMER_122_88 = 27,
660 	IB_RNR_TIMER_163_84 = 28,
661 	IB_RNR_TIMER_245_76 = 29,
662 	IB_RNR_TIMER_327_68 = 30,
663 	IB_RNR_TIMER_491_52 = 31
664 };
665 
666 enum ib_qp_attr_mask {
667 	IB_QP_STATE			= 1,
668 	IB_QP_CUR_STATE			= (1<<1),
669 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
670 	IB_QP_ACCESS_FLAGS		= (1<<3),
671 	IB_QP_PKEY_INDEX		= (1<<4),
672 	IB_QP_PORT			= (1<<5),
673 	IB_QP_QKEY			= (1<<6),
674 	IB_QP_AV			= (1<<7),
675 	IB_QP_PATH_MTU			= (1<<8),
676 	IB_QP_TIMEOUT			= (1<<9),
677 	IB_QP_RETRY_CNT			= (1<<10),
678 	IB_QP_RNR_RETRY			= (1<<11),
679 	IB_QP_RQ_PSN			= (1<<12),
680 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
681 	IB_QP_ALT_PATH			= (1<<14),
682 	IB_QP_MIN_RNR_TIMER		= (1<<15),
683 	IB_QP_SQ_PSN			= (1<<16),
684 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
685 	IB_QP_PATH_MIG_STATE		= (1<<18),
686 	IB_QP_CAP			= (1<<19),
687 	IB_QP_DEST_QPN			= (1<<20)
688 };
689 
690 enum ib_qp_state {
691 	IB_QPS_RESET,
692 	IB_QPS_INIT,
693 	IB_QPS_RTR,
694 	IB_QPS_RTS,
695 	IB_QPS_SQD,
696 	IB_QPS_SQE,
697 	IB_QPS_ERR
698 };
699 
700 enum ib_mig_state {
701 	IB_MIG_MIGRATED,
702 	IB_MIG_REARM,
703 	IB_MIG_ARMED
704 };
705 
706 struct ib_qp_attr {
707 	enum ib_qp_state	qp_state;
708 	enum ib_qp_state	cur_qp_state;
709 	enum ib_mtu		path_mtu;
710 	enum ib_mig_state	path_mig_state;
711 	u32			qkey;
712 	u32			rq_psn;
713 	u32			sq_psn;
714 	u32			dest_qp_num;
715 	int			qp_access_flags;
716 	struct ib_qp_cap	cap;
717 	struct ib_ah_attr	ah_attr;
718 	struct ib_ah_attr	alt_ah_attr;
719 	u16			pkey_index;
720 	u16			alt_pkey_index;
721 	u8			en_sqd_async_notify;
722 	u8			sq_draining;
723 	u8			max_rd_atomic;
724 	u8			max_dest_rd_atomic;
725 	u8			min_rnr_timer;
726 	u8			port_num;
727 	u8			timeout;
728 	u8			retry_cnt;
729 	u8			rnr_retry;
730 	u8			alt_port_num;
731 	u8			alt_timeout;
732 };
733 
734 enum ib_wr_opcode {
735 	IB_WR_RDMA_WRITE,
736 	IB_WR_RDMA_WRITE_WITH_IMM,
737 	IB_WR_SEND,
738 	IB_WR_SEND_WITH_IMM,
739 	IB_WR_RDMA_READ,
740 	IB_WR_ATOMIC_CMP_AND_SWP,
741 	IB_WR_ATOMIC_FETCH_AND_ADD,
742 	IB_WR_LSO,
743 	IB_WR_SEND_WITH_INV,
744 	IB_WR_RDMA_READ_WITH_INV,
745 	IB_WR_LOCAL_INV,
746 	IB_WR_FAST_REG_MR,
747 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
748 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
749 };
750 
751 enum ib_send_flags {
752 	IB_SEND_FENCE		= 1,
753 	IB_SEND_SIGNALED	= (1<<1),
754 	IB_SEND_SOLICITED	= (1<<2),
755 	IB_SEND_INLINE		= (1<<3),
756 	IB_SEND_IP_CSUM		= (1<<4)
757 };
758 
759 struct ib_sge {
760 	u64	addr;
761 	u32	length;
762 	u32	lkey;
763 };
764 
765 struct ib_fast_reg_page_list {
766 	struct ib_device       *device;
767 	u64		       *page_list;
768 	unsigned int		max_page_list_len;
769 };
770 
771 struct ib_send_wr {
772 	struct ib_send_wr      *next;
773 	u64			wr_id;
774 	struct ib_sge	       *sg_list;
775 	int			num_sge;
776 	enum ib_wr_opcode	opcode;
777 	int			send_flags;
778 	union {
779 		__be32		imm_data;
780 		u32		invalidate_rkey;
781 	} ex;
782 	union {
783 		struct {
784 			u64	remote_addr;
785 			u32	rkey;
786 		} rdma;
787 		struct {
788 			u64	remote_addr;
789 			u64	compare_add;
790 			u64	swap;
791 			u64	compare_add_mask;
792 			u64	swap_mask;
793 			u32	rkey;
794 		} atomic;
795 		struct {
796 			struct ib_ah *ah;
797 			void   *header;
798 			int     hlen;
799 			int     mss;
800 			u32	remote_qpn;
801 			u32	remote_qkey;
802 			u16	pkey_index; /* valid for GSI only */
803 			u8	port_num;   /* valid for DR SMPs on switch only */
804 		} ud;
805 		struct {
806 			u64				iova_start;
807 			struct ib_fast_reg_page_list   *page_list;
808 			unsigned int			page_shift;
809 			unsigned int			page_list_len;
810 			u32				length;
811 			int				access_flags;
812 			u32				rkey;
813 		} fast_reg;
814 	} wr;
815 	u32			xrc_remote_srq_num;	/* XRC TGT QPs only */
816 };
817 
818 struct ib_recv_wr {
819 	struct ib_recv_wr      *next;
820 	u64			wr_id;
821 	struct ib_sge	       *sg_list;
822 	int			num_sge;
823 };
824 
825 enum ib_access_flags {
826 	IB_ACCESS_LOCAL_WRITE	= 1,
827 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
828 	IB_ACCESS_REMOTE_READ	= (1<<2),
829 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
830 	IB_ACCESS_MW_BIND	= (1<<4)
831 };
832 
833 struct ib_phys_buf {
834 	u64      addr;
835 	u64      size;
836 };
837 
838 struct ib_mr_attr {
839 	struct ib_pd	*pd;
840 	u64		device_virt_addr;
841 	u64		size;
842 	int		mr_access_flags;
843 	u32		lkey;
844 	u32		rkey;
845 };
846 
847 enum ib_mr_rereg_flags {
848 	IB_MR_REREG_TRANS	= 1,
849 	IB_MR_REREG_PD		= (1<<1),
850 	IB_MR_REREG_ACCESS	= (1<<2)
851 };
852 
853 struct ib_mw_bind {
854 	struct ib_mr   *mr;
855 	u64		wr_id;
856 	u64		addr;
857 	u32		length;
858 	int		send_flags;
859 	int		mw_access_flags;
860 };
861 
862 struct ib_fmr_attr {
863 	int	max_pages;
864 	int	max_maps;
865 	u8	page_shift;
866 };
867 
868 struct ib_ucontext {
869 	struct ib_device       *device;
870 	struct list_head	pd_list;
871 	struct list_head	mr_list;
872 	struct list_head	mw_list;
873 	struct list_head	cq_list;
874 	struct list_head	qp_list;
875 	struct list_head	srq_list;
876 	struct list_head	ah_list;
877 	struct list_head	xrcd_list;
878 	int			closing;
879 };
880 
881 struct ib_uobject {
882 	u64			user_handle;	/* handle given to us by userspace */
883 	struct ib_ucontext     *context;	/* associated user context */
884 	void		       *object;		/* containing object */
885 	struct list_head	list;		/* link to context's list */
886 	int			id;		/* index into kernel idr */
887 	struct kref		ref;
888 	struct rw_semaphore	mutex;		/* protects .live */
889 	int			live;
890 };
891 
892 struct ib_udata {
893 	void __user *inbuf;
894 	void __user *outbuf;
895 	size_t       inlen;
896 	size_t       outlen;
897 };
898 
899 struct ib_pd {
900 	struct ib_device       *device;
901 	struct ib_uobject      *uobject;
902 	atomic_t          	usecnt; /* count all resources */
903 };
904 
905 struct ib_xrcd {
906 	struct ib_device       *device;
907 	atomic_t		usecnt; /* count all exposed resources */
908 	struct inode	       *inode;
909 
910 	struct mutex		tgt_qp_mutex;
911 	struct list_head	tgt_qp_list;
912 };
913 
914 struct ib_ah {
915 	struct ib_device	*device;
916 	struct ib_pd		*pd;
917 	struct ib_uobject	*uobject;
918 };
919 
920 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
921 
922 struct ib_cq {
923 	struct ib_device       *device;
924 	struct ib_uobject      *uobject;
925 	ib_comp_handler   	comp_handler;
926 	void                  (*event_handler)(struct ib_event *, void *);
927 	void                   *cq_context;
928 	int               	cqe;
929 	atomic_t          	usecnt; /* count number of work queues */
930 };
931 
932 struct ib_srq {
933 	struct ib_device       *device;
934 	struct ib_pd	       *pd;
935 	struct ib_uobject      *uobject;
936 	void		      (*event_handler)(struct ib_event *, void *);
937 	void		       *srq_context;
938 	enum ib_srq_type	srq_type;
939 	atomic_t		usecnt;
940 
941 	union {
942 		struct {
943 			struct ib_xrcd *xrcd;
944 			struct ib_cq   *cq;
945 			u32		srq_num;
946 		} xrc;
947 	} ext;
948 };
949 
950 struct ib_qp {
951 	struct ib_device       *device;
952 	struct ib_pd	       *pd;
953 	struct ib_cq	       *send_cq;
954 	struct ib_cq	       *recv_cq;
955 	struct ib_srq	       *srq;
956 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
957 	struct list_head	xrcd_list;
958 	atomic_t		usecnt; /* count times opened */
959 	struct list_head	open_list;
960 	struct ib_qp           *real_qp;
961 	struct ib_uobject      *uobject;
962 	void                  (*event_handler)(struct ib_event *, void *);
963 	void		       *qp_context;
964 	u32			qp_num;
965 	enum ib_qp_type		qp_type;
966 };
967 
968 struct ib_mr {
969 	struct ib_device  *device;
970 	struct ib_pd	  *pd;
971 	struct ib_uobject *uobject;
972 	u32		   lkey;
973 	u32		   rkey;
974 	atomic_t	   usecnt; /* count number of MWs */
975 };
976 
977 struct ib_mw {
978 	struct ib_device	*device;
979 	struct ib_pd		*pd;
980 	struct ib_uobject	*uobject;
981 	u32			rkey;
982 };
983 
984 struct ib_fmr {
985 	struct ib_device	*device;
986 	struct ib_pd		*pd;
987 	struct list_head	list;
988 	u32			lkey;
989 	u32			rkey;
990 };
991 
992 struct ib_mad;
993 struct ib_grh;
994 
995 enum ib_process_mad_flags {
996 	IB_MAD_IGNORE_MKEY	= 1,
997 	IB_MAD_IGNORE_BKEY	= 2,
998 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
999 };
1000 
1001 enum ib_mad_result {
1002 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1003 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1004 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1005 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1006 };
1007 
1008 #define IB_DEVICE_NAME_MAX 64
1009 
1010 struct ib_cache {
1011 	rwlock_t                lock;
1012 	struct ib_event_handler event_handler;
1013 	struct ib_pkey_cache  **pkey_cache;
1014 	struct ib_gid_cache   **gid_cache;
1015 	u8                     *lmc_cache;
1016 };
1017 
1018 struct ib_dma_mapping_ops {
1019 	int		(*mapping_error)(struct ib_device *dev,
1020 					 u64 dma_addr);
1021 	u64		(*map_single)(struct ib_device *dev,
1022 				      void *ptr, size_t size,
1023 				      enum dma_data_direction direction);
1024 	void		(*unmap_single)(struct ib_device *dev,
1025 					u64 addr, size_t size,
1026 					enum dma_data_direction direction);
1027 	u64		(*map_page)(struct ib_device *dev,
1028 				    struct page *page, unsigned long offset,
1029 				    size_t size,
1030 				    enum dma_data_direction direction);
1031 	void		(*unmap_page)(struct ib_device *dev,
1032 				      u64 addr, size_t size,
1033 				      enum dma_data_direction direction);
1034 	int		(*map_sg)(struct ib_device *dev,
1035 				  struct scatterlist *sg, int nents,
1036 				  enum dma_data_direction direction);
1037 	void		(*unmap_sg)(struct ib_device *dev,
1038 				    struct scatterlist *sg, int nents,
1039 				    enum dma_data_direction direction);
1040 	u64		(*dma_address)(struct ib_device *dev,
1041 				       struct scatterlist *sg);
1042 	unsigned int	(*dma_len)(struct ib_device *dev,
1043 				   struct scatterlist *sg);
1044 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1045 					       u64 dma_handle,
1046 					       size_t size,
1047 					       enum dma_data_direction dir);
1048 	void		(*sync_single_for_device)(struct ib_device *dev,
1049 						  u64 dma_handle,
1050 						  size_t size,
1051 						  enum dma_data_direction dir);
1052 	void		*(*alloc_coherent)(struct ib_device *dev,
1053 					   size_t size,
1054 					   u64 *dma_handle,
1055 					   gfp_t flag);
1056 	void		(*free_coherent)(struct ib_device *dev,
1057 					 size_t size, void *cpu_addr,
1058 					 u64 dma_handle);
1059 };
1060 
1061 struct iw_cm_verbs;
1062 
1063 struct ib_device {
1064 	struct device                *dma_device;
1065 
1066 	char                          name[IB_DEVICE_NAME_MAX];
1067 
1068 	struct list_head              event_handler_list;
1069 	spinlock_t                    event_handler_lock;
1070 
1071 	spinlock_t                    client_data_lock;
1072 	struct list_head              core_list;
1073 	struct list_head              client_data_list;
1074 
1075 	struct ib_cache               cache;
1076 	int                          *pkey_tbl_len;
1077 	int                          *gid_tbl_len;
1078 
1079 	int			      num_comp_vectors;
1080 
1081 	struct iw_cm_verbs	     *iwcm;
1082 
1083 	int		           (*get_protocol_stats)(struct ib_device *device,
1084 							 union rdma_protocol_stats *stats);
1085 	int		           (*query_device)(struct ib_device *device,
1086 						   struct ib_device_attr *device_attr);
1087 	int		           (*query_port)(struct ib_device *device,
1088 						 u8 port_num,
1089 						 struct ib_port_attr *port_attr);
1090 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1091 						     u8 port_num);
1092 	int		           (*query_gid)(struct ib_device *device,
1093 						u8 port_num, int index,
1094 						union ib_gid *gid);
1095 	int		           (*query_pkey)(struct ib_device *device,
1096 						 u8 port_num, u16 index, u16 *pkey);
1097 	int		           (*modify_device)(struct ib_device *device,
1098 						    int device_modify_mask,
1099 						    struct ib_device_modify *device_modify);
1100 	int		           (*modify_port)(struct ib_device *device,
1101 						  u8 port_num, int port_modify_mask,
1102 						  struct ib_port_modify *port_modify);
1103 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1104 						     struct ib_udata *udata);
1105 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1106 	int                        (*mmap)(struct ib_ucontext *context,
1107 					   struct vm_area_struct *vma);
1108 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1109 					       struct ib_ucontext *context,
1110 					       struct ib_udata *udata);
1111 	int                        (*dealloc_pd)(struct ib_pd *pd);
1112 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1113 						struct ib_ah_attr *ah_attr);
1114 	int                        (*modify_ah)(struct ib_ah *ah,
1115 						struct ib_ah_attr *ah_attr);
1116 	int                        (*query_ah)(struct ib_ah *ah,
1117 					       struct ib_ah_attr *ah_attr);
1118 	int                        (*destroy_ah)(struct ib_ah *ah);
1119 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1120 						 struct ib_srq_init_attr *srq_init_attr,
1121 						 struct ib_udata *udata);
1122 	int                        (*modify_srq)(struct ib_srq *srq,
1123 						 struct ib_srq_attr *srq_attr,
1124 						 enum ib_srq_attr_mask srq_attr_mask,
1125 						 struct ib_udata *udata);
1126 	int                        (*query_srq)(struct ib_srq *srq,
1127 						struct ib_srq_attr *srq_attr);
1128 	int                        (*destroy_srq)(struct ib_srq *srq);
1129 	int                        (*post_srq_recv)(struct ib_srq *srq,
1130 						    struct ib_recv_wr *recv_wr,
1131 						    struct ib_recv_wr **bad_recv_wr);
1132 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1133 						struct ib_qp_init_attr *qp_init_attr,
1134 						struct ib_udata *udata);
1135 	int                        (*modify_qp)(struct ib_qp *qp,
1136 						struct ib_qp_attr *qp_attr,
1137 						int qp_attr_mask,
1138 						struct ib_udata *udata);
1139 	int                        (*query_qp)(struct ib_qp *qp,
1140 					       struct ib_qp_attr *qp_attr,
1141 					       int qp_attr_mask,
1142 					       struct ib_qp_init_attr *qp_init_attr);
1143 	int                        (*destroy_qp)(struct ib_qp *qp);
1144 	int                        (*post_send)(struct ib_qp *qp,
1145 						struct ib_send_wr *send_wr,
1146 						struct ib_send_wr **bad_send_wr);
1147 	int                        (*post_recv)(struct ib_qp *qp,
1148 						struct ib_recv_wr *recv_wr,
1149 						struct ib_recv_wr **bad_recv_wr);
1150 	struct ib_cq *             (*create_cq)(struct ib_device *device, int cqe,
1151 						int comp_vector,
1152 						struct ib_ucontext *context,
1153 						struct ib_udata *udata);
1154 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1155 						u16 cq_period);
1156 	int                        (*destroy_cq)(struct ib_cq *cq);
1157 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1158 						struct ib_udata *udata);
1159 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1160 					      struct ib_wc *wc);
1161 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1162 	int                        (*req_notify_cq)(struct ib_cq *cq,
1163 						    enum ib_cq_notify_flags flags);
1164 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1165 						      int wc_cnt);
1166 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1167 						 int mr_access_flags);
1168 	struct ib_mr *             (*reg_phys_mr)(struct ib_pd *pd,
1169 						  struct ib_phys_buf *phys_buf_array,
1170 						  int num_phys_buf,
1171 						  int mr_access_flags,
1172 						  u64 *iova_start);
1173 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1174 						  u64 start, u64 length,
1175 						  u64 virt_addr,
1176 						  int mr_access_flags,
1177 						  struct ib_udata *udata);
1178 	int                        (*query_mr)(struct ib_mr *mr,
1179 					       struct ib_mr_attr *mr_attr);
1180 	int                        (*dereg_mr)(struct ib_mr *mr);
1181 	struct ib_mr *		   (*alloc_fast_reg_mr)(struct ib_pd *pd,
1182 					       int max_page_list_len);
1183 	struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1184 								   int page_list_len);
1185 	void			   (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1186 	int                        (*rereg_phys_mr)(struct ib_mr *mr,
1187 						    int mr_rereg_mask,
1188 						    struct ib_pd *pd,
1189 						    struct ib_phys_buf *phys_buf_array,
1190 						    int num_phys_buf,
1191 						    int mr_access_flags,
1192 						    u64 *iova_start);
1193 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd);
1194 	int                        (*bind_mw)(struct ib_qp *qp,
1195 					      struct ib_mw *mw,
1196 					      struct ib_mw_bind *mw_bind);
1197 	int                        (*dealloc_mw)(struct ib_mw *mw);
1198 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1199 						int mr_access_flags,
1200 						struct ib_fmr_attr *fmr_attr);
1201 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1202 						   u64 *page_list, int list_len,
1203 						   u64 iova);
1204 	int		           (*unmap_fmr)(struct list_head *fmr_list);
1205 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1206 	int                        (*attach_mcast)(struct ib_qp *qp,
1207 						   union ib_gid *gid,
1208 						   u16 lid);
1209 	int                        (*detach_mcast)(struct ib_qp *qp,
1210 						   union ib_gid *gid,
1211 						   u16 lid);
1212 	int                        (*process_mad)(struct ib_device *device,
1213 						  int process_mad_flags,
1214 						  u8 port_num,
1215 						  struct ib_wc *in_wc,
1216 						  struct ib_grh *in_grh,
1217 						  struct ib_mad *in_mad,
1218 						  struct ib_mad *out_mad);
1219 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
1220 						 struct ib_ucontext *ucontext,
1221 						 struct ib_udata *udata);
1222 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1223 
1224 	struct ib_dma_mapping_ops   *dma_ops;
1225 
1226 	struct module               *owner;
1227 	struct device                dev;
1228 	struct kobject               *ports_parent;
1229 	struct list_head             port_list;
1230 
1231 	enum {
1232 		IB_DEV_UNINITIALIZED,
1233 		IB_DEV_REGISTERED,
1234 		IB_DEV_UNREGISTERED
1235 	}                            reg_state;
1236 
1237 	int			     uverbs_abi_ver;
1238 	u64			     uverbs_cmd_mask;
1239 
1240 	char			     node_desc[64];
1241 	__be64			     node_guid;
1242 	u32			     local_dma_lkey;
1243 	u8                           node_type;
1244 	u8                           phys_port_cnt;
1245 };
1246 
1247 struct ib_client {
1248 	char  *name;
1249 	void (*add)   (struct ib_device *);
1250 	void (*remove)(struct ib_device *);
1251 
1252 	struct list_head list;
1253 };
1254 
1255 struct ib_device *ib_alloc_device(size_t size);
1256 void ib_dealloc_device(struct ib_device *device);
1257 
1258 int ib_register_device(struct ib_device *device,
1259 		       int (*port_callback)(struct ib_device *,
1260 					    u8, struct kobject *));
1261 void ib_unregister_device(struct ib_device *device);
1262 
1263 int ib_register_client   (struct ib_client *client);
1264 void ib_unregister_client(struct ib_client *client);
1265 
1266 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1267 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
1268 			 void *data);
1269 
1270 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1271 {
1272 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1273 }
1274 
1275 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1276 {
1277 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1278 }
1279 
1280 /**
1281  * ib_modify_qp_is_ok - Check that the supplied attribute mask
1282  * contains all required attributes and no attributes not allowed for
1283  * the given QP state transition.
1284  * @cur_state: Current QP state
1285  * @next_state: Next QP state
1286  * @type: QP type
1287  * @mask: Mask of supplied QP attributes
1288  *
1289  * This function is a helper function that a low-level driver's
1290  * modify_qp method can use to validate the consumer's input.  It
1291  * checks that cur_state and next_state are valid QP states, that a
1292  * transition from cur_state to next_state is allowed by the IB spec,
1293  * and that the attribute mask supplied is allowed for the transition.
1294  */
1295 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1296 		       enum ib_qp_type type, enum ib_qp_attr_mask mask);
1297 
1298 int ib_register_event_handler  (struct ib_event_handler *event_handler);
1299 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1300 void ib_dispatch_event(struct ib_event *event);
1301 
1302 int ib_query_device(struct ib_device *device,
1303 		    struct ib_device_attr *device_attr);
1304 
1305 int ib_query_port(struct ib_device *device,
1306 		  u8 port_num, struct ib_port_attr *port_attr);
1307 
1308 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1309 					       u8 port_num);
1310 
1311 int ib_query_gid(struct ib_device *device,
1312 		 u8 port_num, int index, union ib_gid *gid);
1313 
1314 int ib_query_pkey(struct ib_device *device,
1315 		  u8 port_num, u16 index, u16 *pkey);
1316 
1317 int ib_modify_device(struct ib_device *device,
1318 		     int device_modify_mask,
1319 		     struct ib_device_modify *device_modify);
1320 
1321 int ib_modify_port(struct ib_device *device,
1322 		   u8 port_num, int port_modify_mask,
1323 		   struct ib_port_modify *port_modify);
1324 
1325 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1326 		u8 *port_num, u16 *index);
1327 
1328 int ib_find_pkey(struct ib_device *device,
1329 		 u8 port_num, u16 pkey, u16 *index);
1330 
1331 /**
1332  * ib_alloc_pd - Allocates an unused protection domain.
1333  * @device: The device on which to allocate the protection domain.
1334  *
1335  * A protection domain object provides an association between QPs, shared
1336  * receive queues, address handles, memory regions, and memory windows.
1337  */
1338 struct ib_pd *ib_alloc_pd(struct ib_device *device);
1339 
1340 /**
1341  * ib_dealloc_pd - Deallocates a protection domain.
1342  * @pd: The protection domain to deallocate.
1343  */
1344 int ib_dealloc_pd(struct ib_pd *pd);
1345 
1346 /**
1347  * ib_create_ah - Creates an address handle for the given address vector.
1348  * @pd: The protection domain associated with the address handle.
1349  * @ah_attr: The attributes of the address vector.
1350  *
1351  * The address handle is used to reference a local or global destination
1352  * in all UD QP post sends.
1353  */
1354 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
1355 
1356 /**
1357  * ib_init_ah_from_wc - Initializes address handle attributes from a
1358  *   work completion.
1359  * @device: Device on which the received message arrived.
1360  * @port_num: Port on which the received message arrived.
1361  * @wc: Work completion associated with the received message.
1362  * @grh: References the received global route header.  This parameter is
1363  *   ignored unless the work completion indicates that the GRH is valid.
1364  * @ah_attr: Returned attributes that can be used when creating an address
1365  *   handle for replying to the message.
1366  */
1367 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
1368 		       struct ib_grh *grh, struct ib_ah_attr *ah_attr);
1369 
1370 /**
1371  * ib_create_ah_from_wc - Creates an address handle associated with the
1372  *   sender of the specified work completion.
1373  * @pd: The protection domain associated with the address handle.
1374  * @wc: Work completion information associated with a received message.
1375  * @grh: References the received global route header.  This parameter is
1376  *   ignored unless the work completion indicates that the GRH is valid.
1377  * @port_num: The outbound port number to associate with the address.
1378  *
1379  * The address handle is used to reference a local or global destination
1380  * in all UD QP post sends.
1381  */
1382 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
1383 				   struct ib_grh *grh, u8 port_num);
1384 
1385 /**
1386  * ib_modify_ah - Modifies the address vector associated with an address
1387  *   handle.
1388  * @ah: The address handle to modify.
1389  * @ah_attr: The new address vector attributes to associate with the
1390  *   address handle.
1391  */
1392 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1393 
1394 /**
1395  * ib_query_ah - Queries the address vector associated with an address
1396  *   handle.
1397  * @ah: The address handle to query.
1398  * @ah_attr: The address vector attributes associated with the address
1399  *   handle.
1400  */
1401 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1402 
1403 /**
1404  * ib_destroy_ah - Destroys an address handle.
1405  * @ah: The address handle to destroy.
1406  */
1407 int ib_destroy_ah(struct ib_ah *ah);
1408 
1409 /**
1410  * ib_create_srq - Creates a SRQ associated with the specified protection
1411  *   domain.
1412  * @pd: The protection domain associated with the SRQ.
1413  * @srq_init_attr: A list of initial attributes required to create the
1414  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
1415  *   the actual capabilities of the created SRQ.
1416  *
1417  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1418  * requested size of the SRQ, and set to the actual values allocated
1419  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
1420  * will always be at least as large as the requested values.
1421  */
1422 struct ib_srq *ib_create_srq(struct ib_pd *pd,
1423 			     struct ib_srq_init_attr *srq_init_attr);
1424 
1425 /**
1426  * ib_modify_srq - Modifies the attributes for the specified SRQ.
1427  * @srq: The SRQ to modify.
1428  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
1429  *   the current values of selected SRQ attributes are returned.
1430  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1431  *   are being modified.
1432  *
1433  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1434  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1435  * the number of receives queued drops below the limit.
1436  */
1437 int ib_modify_srq(struct ib_srq *srq,
1438 		  struct ib_srq_attr *srq_attr,
1439 		  enum ib_srq_attr_mask srq_attr_mask);
1440 
1441 /**
1442  * ib_query_srq - Returns the attribute list and current values for the
1443  *   specified SRQ.
1444  * @srq: The SRQ to query.
1445  * @srq_attr: The attributes of the specified SRQ.
1446  */
1447 int ib_query_srq(struct ib_srq *srq,
1448 		 struct ib_srq_attr *srq_attr);
1449 
1450 /**
1451  * ib_destroy_srq - Destroys the specified SRQ.
1452  * @srq: The SRQ to destroy.
1453  */
1454 int ib_destroy_srq(struct ib_srq *srq);
1455 
1456 /**
1457  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1458  * @srq: The SRQ to post the work request on.
1459  * @recv_wr: A list of work requests to post on the receive queue.
1460  * @bad_recv_wr: On an immediate failure, this parameter will reference
1461  *   the work request that failed to be posted on the QP.
1462  */
1463 static inline int ib_post_srq_recv(struct ib_srq *srq,
1464 				   struct ib_recv_wr *recv_wr,
1465 				   struct ib_recv_wr **bad_recv_wr)
1466 {
1467 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1468 }
1469 
1470 /**
1471  * ib_create_qp - Creates a QP associated with the specified protection
1472  *   domain.
1473  * @pd: The protection domain associated with the QP.
1474  * @qp_init_attr: A list of initial attributes required to create the
1475  *   QP.  If QP creation succeeds, then the attributes are updated to
1476  *   the actual capabilities of the created QP.
1477  */
1478 struct ib_qp *ib_create_qp(struct ib_pd *pd,
1479 			   struct ib_qp_init_attr *qp_init_attr);
1480 
1481 /**
1482  * ib_modify_qp - Modifies the attributes for the specified QP and then
1483  *   transitions the QP to the given state.
1484  * @qp: The QP to modify.
1485  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
1486  *   the current values of selected QP attributes are returned.
1487  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1488  *   are being modified.
1489  */
1490 int ib_modify_qp(struct ib_qp *qp,
1491 		 struct ib_qp_attr *qp_attr,
1492 		 int qp_attr_mask);
1493 
1494 /**
1495  * ib_query_qp - Returns the attribute list and current values for the
1496  *   specified QP.
1497  * @qp: The QP to query.
1498  * @qp_attr: The attributes of the specified QP.
1499  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1500  * @qp_init_attr: Additional attributes of the selected QP.
1501  *
1502  * The qp_attr_mask may be used to limit the query to gathering only the
1503  * selected attributes.
1504  */
1505 int ib_query_qp(struct ib_qp *qp,
1506 		struct ib_qp_attr *qp_attr,
1507 		int qp_attr_mask,
1508 		struct ib_qp_init_attr *qp_init_attr);
1509 
1510 /**
1511  * ib_destroy_qp - Destroys the specified QP.
1512  * @qp: The QP to destroy.
1513  */
1514 int ib_destroy_qp(struct ib_qp *qp);
1515 
1516 /**
1517  * ib_open_qp - Obtain a reference to an existing sharable QP.
1518  * @xrcd - XRC domain
1519  * @qp_open_attr: Attributes identifying the QP to open.
1520  *
1521  * Returns a reference to a sharable QP.
1522  */
1523 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
1524 			 struct ib_qp_open_attr *qp_open_attr);
1525 
1526 /**
1527  * ib_close_qp - Release an external reference to a QP.
1528  * @qp: The QP handle to release
1529  *
1530  * The opened QP handle is released by the caller.  The underlying
1531  * shared QP is not destroyed until all internal references are released.
1532  */
1533 int ib_close_qp(struct ib_qp *qp);
1534 
1535 /**
1536  * ib_post_send - Posts a list of work requests to the send queue of
1537  *   the specified QP.
1538  * @qp: The QP to post the work request on.
1539  * @send_wr: A list of work requests to post on the send queue.
1540  * @bad_send_wr: On an immediate failure, this parameter will reference
1541  *   the work request that failed to be posted on the QP.
1542  *
1543  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
1544  * error is returned, the QP state shall not be affected,
1545  * ib_post_send() will return an immediate error after queueing any
1546  * earlier work requests in the list.
1547  */
1548 static inline int ib_post_send(struct ib_qp *qp,
1549 			       struct ib_send_wr *send_wr,
1550 			       struct ib_send_wr **bad_send_wr)
1551 {
1552 	return qp->device->post_send(qp, send_wr, bad_send_wr);
1553 }
1554 
1555 /**
1556  * ib_post_recv - Posts a list of work requests to the receive queue of
1557  *   the specified QP.
1558  * @qp: The QP to post the work request on.
1559  * @recv_wr: A list of work requests to post on the receive queue.
1560  * @bad_recv_wr: On an immediate failure, this parameter will reference
1561  *   the work request that failed to be posted on the QP.
1562  */
1563 static inline int ib_post_recv(struct ib_qp *qp,
1564 			       struct ib_recv_wr *recv_wr,
1565 			       struct ib_recv_wr **bad_recv_wr)
1566 {
1567 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1568 }
1569 
1570 /**
1571  * ib_create_cq - Creates a CQ on the specified device.
1572  * @device: The device on which to create the CQ.
1573  * @comp_handler: A user-specified callback that is invoked when a
1574  *   completion event occurs on the CQ.
1575  * @event_handler: A user-specified callback that is invoked when an
1576  *   asynchronous event not associated with a completion occurs on the CQ.
1577  * @cq_context: Context associated with the CQ returned to the user via
1578  *   the associated completion and event handlers.
1579  * @cqe: The minimum size of the CQ.
1580  * @comp_vector - Completion vector used to signal completion events.
1581  *     Must be >= 0 and < context->num_comp_vectors.
1582  *
1583  * Users can examine the cq structure to determine the actual CQ size.
1584  */
1585 struct ib_cq *ib_create_cq(struct ib_device *device,
1586 			   ib_comp_handler comp_handler,
1587 			   void (*event_handler)(struct ib_event *, void *),
1588 			   void *cq_context, int cqe, int comp_vector);
1589 
1590 /**
1591  * ib_resize_cq - Modifies the capacity of the CQ.
1592  * @cq: The CQ to resize.
1593  * @cqe: The minimum size of the CQ.
1594  *
1595  * Users can examine the cq structure to determine the actual CQ size.
1596  */
1597 int ib_resize_cq(struct ib_cq *cq, int cqe);
1598 
1599 /**
1600  * ib_modify_cq - Modifies moderation params of the CQ
1601  * @cq: The CQ to modify.
1602  * @cq_count: number of CQEs that will trigger an event
1603  * @cq_period: max period of time in usec before triggering an event
1604  *
1605  */
1606 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1607 
1608 /**
1609  * ib_destroy_cq - Destroys the specified CQ.
1610  * @cq: The CQ to destroy.
1611  */
1612 int ib_destroy_cq(struct ib_cq *cq);
1613 
1614 /**
1615  * ib_poll_cq - poll a CQ for completion(s)
1616  * @cq:the CQ being polled
1617  * @num_entries:maximum number of completions to return
1618  * @wc:array of at least @num_entries &struct ib_wc where completions
1619  *   will be returned
1620  *
1621  * Poll a CQ for (possibly multiple) completions.  If the return value
1622  * is < 0, an error occurred.  If the return value is >= 0, it is the
1623  * number of completions returned.  If the return value is
1624  * non-negative and < num_entries, then the CQ was emptied.
1625  */
1626 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1627 			     struct ib_wc *wc)
1628 {
1629 	return cq->device->poll_cq(cq, num_entries, wc);
1630 }
1631 
1632 /**
1633  * ib_peek_cq - Returns the number of unreaped completions currently
1634  *   on the specified CQ.
1635  * @cq: The CQ to peek.
1636  * @wc_cnt: A minimum number of unreaped completions to check for.
1637  *
1638  * If the number of unreaped completions is greater than or equal to wc_cnt,
1639  * this function returns wc_cnt, otherwise, it returns the actual number of
1640  * unreaped completions.
1641  */
1642 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1643 
1644 /**
1645  * ib_req_notify_cq - Request completion notification on a CQ.
1646  * @cq: The CQ to generate an event for.
1647  * @flags:
1648  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1649  *   to request an event on the next solicited event or next work
1650  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1651  *   may also be |ed in to request a hint about missed events, as
1652  *   described below.
1653  *
1654  * Return Value:
1655  *    < 0 means an error occurred while requesting notification
1656  *   == 0 means notification was requested successfully, and if
1657  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1658  *        were missed and it is safe to wait for another event.  In
1659  *        this case is it guaranteed that any work completions added
1660  *        to the CQ since the last CQ poll will trigger a completion
1661  *        notification event.
1662  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1663  *        in.  It means that the consumer must poll the CQ again to
1664  *        make sure it is empty to avoid missing an event because of a
1665  *        race between requesting notification and an entry being
1666  *        added to the CQ.  This return value means it is possible
1667  *        (but not guaranteed) that a work completion has been added
1668  *        to the CQ since the last poll without triggering a
1669  *        completion notification event.
1670  */
1671 static inline int ib_req_notify_cq(struct ib_cq *cq,
1672 				   enum ib_cq_notify_flags flags)
1673 {
1674 	return cq->device->req_notify_cq(cq, flags);
1675 }
1676 
1677 /**
1678  * ib_req_ncomp_notif - Request completion notification when there are
1679  *   at least the specified number of unreaped completions on the CQ.
1680  * @cq: The CQ to generate an event for.
1681  * @wc_cnt: The number of unreaped completions that should be on the
1682  *   CQ before an event is generated.
1683  */
1684 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1685 {
1686 	return cq->device->req_ncomp_notif ?
1687 		cq->device->req_ncomp_notif(cq, wc_cnt) :
1688 		-ENOSYS;
1689 }
1690 
1691 /**
1692  * ib_get_dma_mr - Returns a memory region for system memory that is
1693  *   usable for DMA.
1694  * @pd: The protection domain associated with the memory region.
1695  * @mr_access_flags: Specifies the memory access rights.
1696  *
1697  * Note that the ib_dma_*() functions defined below must be used
1698  * to create/destroy addresses used with the Lkey or Rkey returned
1699  * by ib_get_dma_mr().
1700  */
1701 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1702 
1703 /**
1704  * ib_dma_mapping_error - check a DMA addr for error
1705  * @dev: The device for which the dma_addr was created
1706  * @dma_addr: The DMA address to check
1707  */
1708 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1709 {
1710 	if (dev->dma_ops)
1711 		return dev->dma_ops->mapping_error(dev, dma_addr);
1712 	return dma_mapping_error(dev->dma_device, dma_addr);
1713 }
1714 
1715 /**
1716  * ib_dma_map_single - Map a kernel virtual address to DMA address
1717  * @dev: The device for which the dma_addr is to be created
1718  * @cpu_addr: The kernel virtual address
1719  * @size: The size of the region in bytes
1720  * @direction: The direction of the DMA
1721  */
1722 static inline u64 ib_dma_map_single(struct ib_device *dev,
1723 				    void *cpu_addr, size_t size,
1724 				    enum dma_data_direction direction)
1725 {
1726 	if (dev->dma_ops)
1727 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1728 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1729 }
1730 
1731 /**
1732  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1733  * @dev: The device for which the DMA address was created
1734  * @addr: The DMA address
1735  * @size: The size of the region in bytes
1736  * @direction: The direction of the DMA
1737  */
1738 static inline void ib_dma_unmap_single(struct ib_device *dev,
1739 				       u64 addr, size_t size,
1740 				       enum dma_data_direction direction)
1741 {
1742 	if (dev->dma_ops)
1743 		dev->dma_ops->unmap_single(dev, addr, size, direction);
1744 	else
1745 		dma_unmap_single(dev->dma_device, addr, size, direction);
1746 }
1747 
1748 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
1749 					  void *cpu_addr, size_t size,
1750 					  enum dma_data_direction direction,
1751 					  struct dma_attrs *attrs)
1752 {
1753 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
1754 				    direction, attrs);
1755 }
1756 
1757 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
1758 					     u64 addr, size_t size,
1759 					     enum dma_data_direction direction,
1760 					     struct dma_attrs *attrs)
1761 {
1762 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
1763 				      direction, attrs);
1764 }
1765 
1766 /**
1767  * ib_dma_map_page - Map a physical page to DMA address
1768  * @dev: The device for which the dma_addr is to be created
1769  * @page: The page to be mapped
1770  * @offset: The offset within the page
1771  * @size: The size of the region in bytes
1772  * @direction: The direction of the DMA
1773  */
1774 static inline u64 ib_dma_map_page(struct ib_device *dev,
1775 				  struct page *page,
1776 				  unsigned long offset,
1777 				  size_t size,
1778 					 enum dma_data_direction direction)
1779 {
1780 	if (dev->dma_ops)
1781 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
1782 	return dma_map_page(dev->dma_device, page, offset, size, direction);
1783 }
1784 
1785 /**
1786  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1787  * @dev: The device for which the DMA address was created
1788  * @addr: The DMA address
1789  * @size: The size of the region in bytes
1790  * @direction: The direction of the DMA
1791  */
1792 static inline void ib_dma_unmap_page(struct ib_device *dev,
1793 				     u64 addr, size_t size,
1794 				     enum dma_data_direction direction)
1795 {
1796 	if (dev->dma_ops)
1797 		dev->dma_ops->unmap_page(dev, addr, size, direction);
1798 	else
1799 		dma_unmap_page(dev->dma_device, addr, size, direction);
1800 }
1801 
1802 /**
1803  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
1804  * @dev: The device for which the DMA addresses are to be created
1805  * @sg: The array of scatter/gather entries
1806  * @nents: The number of scatter/gather entries
1807  * @direction: The direction of the DMA
1808  */
1809 static inline int ib_dma_map_sg(struct ib_device *dev,
1810 				struct scatterlist *sg, int nents,
1811 				enum dma_data_direction direction)
1812 {
1813 	if (dev->dma_ops)
1814 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
1815 	return dma_map_sg(dev->dma_device, sg, nents, direction);
1816 }
1817 
1818 /**
1819  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
1820  * @dev: The device for which the DMA addresses were created
1821  * @sg: The array of scatter/gather entries
1822  * @nents: The number of scatter/gather entries
1823  * @direction: The direction of the DMA
1824  */
1825 static inline void ib_dma_unmap_sg(struct ib_device *dev,
1826 				   struct scatterlist *sg, int nents,
1827 				   enum dma_data_direction direction)
1828 {
1829 	if (dev->dma_ops)
1830 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
1831 	else
1832 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
1833 }
1834 
1835 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
1836 				      struct scatterlist *sg, int nents,
1837 				      enum dma_data_direction direction,
1838 				      struct dma_attrs *attrs)
1839 {
1840 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1841 }
1842 
1843 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
1844 					 struct scatterlist *sg, int nents,
1845 					 enum dma_data_direction direction,
1846 					 struct dma_attrs *attrs)
1847 {
1848 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1849 }
1850 /**
1851  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
1852  * @dev: The device for which the DMA addresses were created
1853  * @sg: The scatter/gather entry
1854  */
1855 static inline u64 ib_sg_dma_address(struct ib_device *dev,
1856 				    struct scatterlist *sg)
1857 {
1858 	if (dev->dma_ops)
1859 		return dev->dma_ops->dma_address(dev, sg);
1860 	return sg_dma_address(sg);
1861 }
1862 
1863 /**
1864  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
1865  * @dev: The device for which the DMA addresses were created
1866  * @sg: The scatter/gather entry
1867  */
1868 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
1869 					 struct scatterlist *sg)
1870 {
1871 	if (dev->dma_ops)
1872 		return dev->dma_ops->dma_len(dev, sg);
1873 	return sg_dma_len(sg);
1874 }
1875 
1876 /**
1877  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
1878  * @dev: The device for which the DMA address was created
1879  * @addr: The DMA address
1880  * @size: The size of the region in bytes
1881  * @dir: The direction of the DMA
1882  */
1883 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
1884 					      u64 addr,
1885 					      size_t size,
1886 					      enum dma_data_direction dir)
1887 {
1888 	if (dev->dma_ops)
1889 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
1890 	else
1891 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
1892 }
1893 
1894 /**
1895  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
1896  * @dev: The device for which the DMA address was created
1897  * @addr: The DMA address
1898  * @size: The size of the region in bytes
1899  * @dir: The direction of the DMA
1900  */
1901 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
1902 						 u64 addr,
1903 						 size_t size,
1904 						 enum dma_data_direction dir)
1905 {
1906 	if (dev->dma_ops)
1907 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
1908 	else
1909 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
1910 }
1911 
1912 /**
1913  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
1914  * @dev: The device for which the DMA address is requested
1915  * @size: The size of the region to allocate in bytes
1916  * @dma_handle: A pointer for returning the DMA address of the region
1917  * @flag: memory allocator flags
1918  */
1919 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
1920 					   size_t size,
1921 					   u64 *dma_handle,
1922 					   gfp_t flag)
1923 {
1924 	if (dev->dma_ops)
1925 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
1926 	else {
1927 		dma_addr_t handle;
1928 		void *ret;
1929 
1930 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
1931 		*dma_handle = handle;
1932 		return ret;
1933 	}
1934 }
1935 
1936 /**
1937  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
1938  * @dev: The device for which the DMA addresses were allocated
1939  * @size: The size of the region
1940  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
1941  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
1942  */
1943 static inline void ib_dma_free_coherent(struct ib_device *dev,
1944 					size_t size, void *cpu_addr,
1945 					u64 dma_handle)
1946 {
1947 	if (dev->dma_ops)
1948 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
1949 	else
1950 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
1951 }
1952 
1953 /**
1954  * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
1955  *   by an HCA.
1956  * @pd: The protection domain associated assigned to the registered region.
1957  * @phys_buf_array: Specifies a list of physical buffers to use in the
1958  *   memory region.
1959  * @num_phys_buf: Specifies the size of the phys_buf_array.
1960  * @mr_access_flags: Specifies the memory access rights.
1961  * @iova_start: The offset of the region's starting I/O virtual address.
1962  */
1963 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
1964 			     struct ib_phys_buf *phys_buf_array,
1965 			     int num_phys_buf,
1966 			     int mr_access_flags,
1967 			     u64 *iova_start);
1968 
1969 /**
1970  * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
1971  *   Conceptually, this call performs the functions deregister memory region
1972  *   followed by register physical memory region.  Where possible,
1973  *   resources are reused instead of deallocated and reallocated.
1974  * @mr: The memory region to modify.
1975  * @mr_rereg_mask: A bit-mask used to indicate which of the following
1976  *   properties of the memory region are being modified.
1977  * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
1978  *   the new protection domain to associated with the memory region,
1979  *   otherwise, this parameter is ignored.
1980  * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1981  *   field specifies a list of physical buffers to use in the new
1982  *   translation, otherwise, this parameter is ignored.
1983  * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1984  *   field specifies the size of the phys_buf_array, otherwise, this
1985  *   parameter is ignored.
1986  * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
1987  *   field specifies the new memory access rights, otherwise, this
1988  *   parameter is ignored.
1989  * @iova_start: The offset of the region's starting I/O virtual address.
1990  */
1991 int ib_rereg_phys_mr(struct ib_mr *mr,
1992 		     int mr_rereg_mask,
1993 		     struct ib_pd *pd,
1994 		     struct ib_phys_buf *phys_buf_array,
1995 		     int num_phys_buf,
1996 		     int mr_access_flags,
1997 		     u64 *iova_start);
1998 
1999 /**
2000  * ib_query_mr - Retrieves information about a specific memory region.
2001  * @mr: The memory region to retrieve information about.
2002  * @mr_attr: The attributes of the specified memory region.
2003  */
2004 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2005 
2006 /**
2007  * ib_dereg_mr - Deregisters a memory region and removes it from the
2008  *   HCA translation table.
2009  * @mr: The memory region to deregister.
2010  */
2011 int ib_dereg_mr(struct ib_mr *mr);
2012 
2013 /**
2014  * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2015  *   IB_WR_FAST_REG_MR send work request.
2016  * @pd: The protection domain associated with the region.
2017  * @max_page_list_len: requested max physical buffer list length to be
2018  *   used with fast register work requests for this MR.
2019  */
2020 struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
2021 
2022 /**
2023  * ib_alloc_fast_reg_page_list - Allocates a page list array
2024  * @device - ib device pointer.
2025  * @page_list_len - size of the page list array to be allocated.
2026  *
2027  * This allocates and returns a struct ib_fast_reg_page_list * and a
2028  * page_list array that is at least page_list_len in size.  The actual
2029  * size is returned in max_page_list_len.  The caller is responsible
2030  * for initializing the contents of the page_list array before posting
2031  * a send work request with the IB_WC_FAST_REG_MR opcode.
2032  *
2033  * The page_list array entries must be translated using one of the
2034  * ib_dma_*() functions just like the addresses passed to
2035  * ib_map_phys_fmr().  Once the ib_post_send() is issued, the struct
2036  * ib_fast_reg_page_list must not be modified by the caller until the
2037  * IB_WC_FAST_REG_MR work request completes.
2038  */
2039 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2040 				struct ib_device *device, int page_list_len);
2041 
2042 /**
2043  * ib_free_fast_reg_page_list - Deallocates a previously allocated
2044  *   page list array.
2045  * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2046  */
2047 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2048 
2049 /**
2050  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2051  *   R_Key and L_Key.
2052  * @mr - struct ib_mr pointer to be updated.
2053  * @newkey - new key to be used.
2054  */
2055 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2056 {
2057 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2058 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2059 }
2060 
2061 /**
2062  * ib_alloc_mw - Allocates a memory window.
2063  * @pd: The protection domain associated with the memory window.
2064  */
2065 struct ib_mw *ib_alloc_mw(struct ib_pd *pd);
2066 
2067 /**
2068  * ib_bind_mw - Posts a work request to the send queue of the specified
2069  *   QP, which binds the memory window to the given address range and
2070  *   remote access attributes.
2071  * @qp: QP to post the bind work request on.
2072  * @mw: The memory window to bind.
2073  * @mw_bind: Specifies information about the memory window, including
2074  *   its address range, remote access rights, and associated memory region.
2075  */
2076 static inline int ib_bind_mw(struct ib_qp *qp,
2077 			     struct ib_mw *mw,
2078 			     struct ib_mw_bind *mw_bind)
2079 {
2080 	/* XXX reference counting in corresponding MR? */
2081 	return mw->device->bind_mw ?
2082 		mw->device->bind_mw(qp, mw, mw_bind) :
2083 		-ENOSYS;
2084 }
2085 
2086 /**
2087  * ib_dealloc_mw - Deallocates a memory window.
2088  * @mw: The memory window to deallocate.
2089  */
2090 int ib_dealloc_mw(struct ib_mw *mw);
2091 
2092 /**
2093  * ib_alloc_fmr - Allocates a unmapped fast memory region.
2094  * @pd: The protection domain associated with the unmapped region.
2095  * @mr_access_flags: Specifies the memory access rights.
2096  * @fmr_attr: Attributes of the unmapped region.
2097  *
2098  * A fast memory region must be mapped before it can be used as part of
2099  * a work request.
2100  */
2101 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2102 			    int mr_access_flags,
2103 			    struct ib_fmr_attr *fmr_attr);
2104 
2105 /**
2106  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2107  * @fmr: The fast memory region to associate with the pages.
2108  * @page_list: An array of physical pages to map to the fast memory region.
2109  * @list_len: The number of pages in page_list.
2110  * @iova: The I/O virtual address to use with the mapped region.
2111  */
2112 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2113 				  u64 *page_list, int list_len,
2114 				  u64 iova)
2115 {
2116 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2117 }
2118 
2119 /**
2120  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2121  * @fmr_list: A linked list of fast memory regions to unmap.
2122  */
2123 int ib_unmap_fmr(struct list_head *fmr_list);
2124 
2125 /**
2126  * ib_dealloc_fmr - Deallocates a fast memory region.
2127  * @fmr: The fast memory region to deallocate.
2128  */
2129 int ib_dealloc_fmr(struct ib_fmr *fmr);
2130 
2131 /**
2132  * ib_attach_mcast - Attaches the specified QP to a multicast group.
2133  * @qp: QP to attach to the multicast group.  The QP must be type
2134  *   IB_QPT_UD.
2135  * @gid: Multicast group GID.
2136  * @lid: Multicast group LID in host byte order.
2137  *
2138  * In order to send and receive multicast packets, subnet
2139  * administration must have created the multicast group and configured
2140  * the fabric appropriately.  The port associated with the specified
2141  * QP must also be a member of the multicast group.
2142  */
2143 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2144 
2145 /**
2146  * ib_detach_mcast - Detaches the specified QP from a multicast group.
2147  * @qp: QP to detach from the multicast group.
2148  * @gid: Multicast group GID.
2149  * @lid: Multicast group LID in host byte order.
2150  */
2151 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2152 
2153 /**
2154  * ib_alloc_xrcd - Allocates an XRC domain.
2155  * @device: The device on which to allocate the XRC domain.
2156  */
2157 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2158 
2159 /**
2160  * ib_dealloc_xrcd - Deallocates an XRC domain.
2161  * @xrcd: The XRC domain to deallocate.
2162  */
2163 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2164 
2165 #endif /* IB_VERBS_H */
2166