xref: /openbmc/linux/include/rdma/ib_verbs.h (revision ee8a99bd)
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 	IB_DEVICE_MEM_WINDOW_TYPE_2A	= (1<<23),
119 	IB_DEVICE_MEM_WINDOW_TYPE_2B	= (1<<24)
120 };
121 
122 enum ib_atomic_cap {
123 	IB_ATOMIC_NONE,
124 	IB_ATOMIC_HCA,
125 	IB_ATOMIC_GLOB
126 };
127 
128 struct ib_device_attr {
129 	u64			fw_ver;
130 	__be64			sys_image_guid;
131 	u64			max_mr_size;
132 	u64			page_size_cap;
133 	u32			vendor_id;
134 	u32			vendor_part_id;
135 	u32			hw_ver;
136 	int			max_qp;
137 	int			max_qp_wr;
138 	int			device_cap_flags;
139 	int			max_sge;
140 	int			max_sge_rd;
141 	int			max_cq;
142 	int			max_cqe;
143 	int			max_mr;
144 	int			max_pd;
145 	int			max_qp_rd_atom;
146 	int			max_ee_rd_atom;
147 	int			max_res_rd_atom;
148 	int			max_qp_init_rd_atom;
149 	int			max_ee_init_rd_atom;
150 	enum ib_atomic_cap	atomic_cap;
151 	enum ib_atomic_cap	masked_atomic_cap;
152 	int			max_ee;
153 	int			max_rdd;
154 	int			max_mw;
155 	int			max_raw_ipv6_qp;
156 	int			max_raw_ethy_qp;
157 	int			max_mcast_grp;
158 	int			max_mcast_qp_attach;
159 	int			max_total_mcast_qp_attach;
160 	int			max_ah;
161 	int			max_fmr;
162 	int			max_map_per_fmr;
163 	int			max_srq;
164 	int			max_srq_wr;
165 	int			max_srq_sge;
166 	unsigned int		max_fast_reg_page_list_len;
167 	u16			max_pkeys;
168 	u8			local_ca_ack_delay;
169 };
170 
171 enum ib_mtu {
172 	IB_MTU_256  = 1,
173 	IB_MTU_512  = 2,
174 	IB_MTU_1024 = 3,
175 	IB_MTU_2048 = 4,
176 	IB_MTU_4096 = 5
177 };
178 
179 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
180 {
181 	switch (mtu) {
182 	case IB_MTU_256:  return  256;
183 	case IB_MTU_512:  return  512;
184 	case IB_MTU_1024: return 1024;
185 	case IB_MTU_2048: return 2048;
186 	case IB_MTU_4096: return 4096;
187 	default: 	  return -1;
188 	}
189 }
190 
191 enum ib_port_state {
192 	IB_PORT_NOP		= 0,
193 	IB_PORT_DOWN		= 1,
194 	IB_PORT_INIT		= 2,
195 	IB_PORT_ARMED		= 3,
196 	IB_PORT_ACTIVE		= 4,
197 	IB_PORT_ACTIVE_DEFER	= 5
198 };
199 
200 enum ib_port_cap_flags {
201 	IB_PORT_SM				= 1 <<  1,
202 	IB_PORT_NOTICE_SUP			= 1 <<  2,
203 	IB_PORT_TRAP_SUP			= 1 <<  3,
204 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
205 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
206 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
207 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
208 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
209 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
210 	IB_PORT_SM_DISABLED			= 1 << 10,
211 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
212 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
213 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
214 	IB_PORT_CM_SUP				= 1 << 16,
215 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
216 	IB_PORT_REINIT_SUP			= 1 << 18,
217 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
218 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
219 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
220 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
221 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
222 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
223 	IB_PORT_CLIENT_REG_SUP			= 1 << 25
224 };
225 
226 enum ib_port_width {
227 	IB_WIDTH_1X	= 1,
228 	IB_WIDTH_4X	= 2,
229 	IB_WIDTH_8X	= 4,
230 	IB_WIDTH_12X	= 8
231 };
232 
233 static inline int ib_width_enum_to_int(enum ib_port_width width)
234 {
235 	switch (width) {
236 	case IB_WIDTH_1X:  return  1;
237 	case IB_WIDTH_4X:  return  4;
238 	case IB_WIDTH_8X:  return  8;
239 	case IB_WIDTH_12X: return 12;
240 	default: 	  return -1;
241 	}
242 }
243 
244 enum ib_port_speed {
245 	IB_SPEED_SDR	= 1,
246 	IB_SPEED_DDR	= 2,
247 	IB_SPEED_QDR	= 4,
248 	IB_SPEED_FDR10	= 8,
249 	IB_SPEED_FDR	= 16,
250 	IB_SPEED_EDR	= 32
251 };
252 
253 struct ib_protocol_stats {
254 	/* TBD... */
255 };
256 
257 struct iw_protocol_stats {
258 	u64	ipInReceives;
259 	u64	ipInHdrErrors;
260 	u64	ipInTooBigErrors;
261 	u64	ipInNoRoutes;
262 	u64	ipInAddrErrors;
263 	u64	ipInUnknownProtos;
264 	u64	ipInTruncatedPkts;
265 	u64	ipInDiscards;
266 	u64	ipInDelivers;
267 	u64	ipOutForwDatagrams;
268 	u64	ipOutRequests;
269 	u64	ipOutDiscards;
270 	u64	ipOutNoRoutes;
271 	u64	ipReasmTimeout;
272 	u64	ipReasmReqds;
273 	u64	ipReasmOKs;
274 	u64	ipReasmFails;
275 	u64	ipFragOKs;
276 	u64	ipFragFails;
277 	u64	ipFragCreates;
278 	u64	ipInMcastPkts;
279 	u64	ipOutMcastPkts;
280 	u64	ipInBcastPkts;
281 	u64	ipOutBcastPkts;
282 
283 	u64	tcpRtoAlgorithm;
284 	u64	tcpRtoMin;
285 	u64	tcpRtoMax;
286 	u64	tcpMaxConn;
287 	u64	tcpActiveOpens;
288 	u64	tcpPassiveOpens;
289 	u64	tcpAttemptFails;
290 	u64	tcpEstabResets;
291 	u64	tcpCurrEstab;
292 	u64	tcpInSegs;
293 	u64	tcpOutSegs;
294 	u64	tcpRetransSegs;
295 	u64	tcpInErrs;
296 	u64	tcpOutRsts;
297 };
298 
299 union rdma_protocol_stats {
300 	struct ib_protocol_stats	ib;
301 	struct iw_protocol_stats	iw;
302 };
303 
304 struct ib_port_attr {
305 	enum ib_port_state	state;
306 	enum ib_mtu		max_mtu;
307 	enum ib_mtu		active_mtu;
308 	int			gid_tbl_len;
309 	u32			port_cap_flags;
310 	u32			max_msg_sz;
311 	u32			bad_pkey_cntr;
312 	u32			qkey_viol_cntr;
313 	u16			pkey_tbl_len;
314 	u16			lid;
315 	u16			sm_lid;
316 	u8			lmc;
317 	u8			max_vl_num;
318 	u8			sm_sl;
319 	u8			subnet_timeout;
320 	u8			init_type_reply;
321 	u8			active_width;
322 	u8			active_speed;
323 	u8                      phys_state;
324 };
325 
326 enum ib_device_modify_flags {
327 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
328 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
329 };
330 
331 struct ib_device_modify {
332 	u64	sys_image_guid;
333 	char	node_desc[64];
334 };
335 
336 enum ib_port_modify_flags {
337 	IB_PORT_SHUTDOWN		= 1,
338 	IB_PORT_INIT_TYPE		= (1<<2),
339 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
340 };
341 
342 struct ib_port_modify {
343 	u32	set_port_cap_mask;
344 	u32	clr_port_cap_mask;
345 	u8	init_type;
346 };
347 
348 enum ib_event_type {
349 	IB_EVENT_CQ_ERR,
350 	IB_EVENT_QP_FATAL,
351 	IB_EVENT_QP_REQ_ERR,
352 	IB_EVENT_QP_ACCESS_ERR,
353 	IB_EVENT_COMM_EST,
354 	IB_EVENT_SQ_DRAINED,
355 	IB_EVENT_PATH_MIG,
356 	IB_EVENT_PATH_MIG_ERR,
357 	IB_EVENT_DEVICE_FATAL,
358 	IB_EVENT_PORT_ACTIVE,
359 	IB_EVENT_PORT_ERR,
360 	IB_EVENT_LID_CHANGE,
361 	IB_EVENT_PKEY_CHANGE,
362 	IB_EVENT_SM_CHANGE,
363 	IB_EVENT_SRQ_ERR,
364 	IB_EVENT_SRQ_LIMIT_REACHED,
365 	IB_EVENT_QP_LAST_WQE_REACHED,
366 	IB_EVENT_CLIENT_REREGISTER,
367 	IB_EVENT_GID_CHANGE,
368 };
369 
370 struct ib_event {
371 	struct ib_device	*device;
372 	union {
373 		struct ib_cq	*cq;
374 		struct ib_qp	*qp;
375 		struct ib_srq	*srq;
376 		u8		port_num;
377 	} element;
378 	enum ib_event_type	event;
379 };
380 
381 struct ib_event_handler {
382 	struct ib_device *device;
383 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
384 	struct list_head  list;
385 };
386 
387 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
388 	do {							\
389 		(_ptr)->device  = _device;			\
390 		(_ptr)->handler = _handler;			\
391 		INIT_LIST_HEAD(&(_ptr)->list);			\
392 	} while (0)
393 
394 struct ib_global_route {
395 	union ib_gid	dgid;
396 	u32		flow_label;
397 	u8		sgid_index;
398 	u8		hop_limit;
399 	u8		traffic_class;
400 };
401 
402 struct ib_grh {
403 	__be32		version_tclass_flow;
404 	__be16		paylen;
405 	u8		next_hdr;
406 	u8		hop_limit;
407 	union ib_gid	sgid;
408 	union ib_gid	dgid;
409 };
410 
411 enum {
412 	IB_MULTICAST_QPN = 0xffffff
413 };
414 
415 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
416 
417 enum ib_ah_flags {
418 	IB_AH_GRH	= 1
419 };
420 
421 enum ib_rate {
422 	IB_RATE_PORT_CURRENT = 0,
423 	IB_RATE_2_5_GBPS = 2,
424 	IB_RATE_5_GBPS   = 5,
425 	IB_RATE_10_GBPS  = 3,
426 	IB_RATE_20_GBPS  = 6,
427 	IB_RATE_30_GBPS  = 4,
428 	IB_RATE_40_GBPS  = 7,
429 	IB_RATE_60_GBPS  = 8,
430 	IB_RATE_80_GBPS  = 9,
431 	IB_RATE_120_GBPS = 10,
432 	IB_RATE_14_GBPS  = 11,
433 	IB_RATE_56_GBPS  = 12,
434 	IB_RATE_112_GBPS = 13,
435 	IB_RATE_168_GBPS = 14,
436 	IB_RATE_25_GBPS  = 15,
437 	IB_RATE_100_GBPS = 16,
438 	IB_RATE_200_GBPS = 17,
439 	IB_RATE_300_GBPS = 18
440 };
441 
442 /**
443  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
444  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
445  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
446  * @rate: rate to convert.
447  */
448 int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
449 
450 /**
451  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
452  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
453  * @rate: rate to convert.
454  */
455 int ib_rate_to_mbps(enum ib_rate rate) __attribute_const__;
456 
457 /**
458  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
459  * enum.
460  * @mult: multiple to convert.
461  */
462 enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
463 
464 struct ib_ah_attr {
465 	struct ib_global_route	grh;
466 	u16			dlid;
467 	u8			sl;
468 	u8			src_path_bits;
469 	u8			static_rate;
470 	u8			ah_flags;
471 	u8			port_num;
472 };
473 
474 enum ib_wc_status {
475 	IB_WC_SUCCESS,
476 	IB_WC_LOC_LEN_ERR,
477 	IB_WC_LOC_QP_OP_ERR,
478 	IB_WC_LOC_EEC_OP_ERR,
479 	IB_WC_LOC_PROT_ERR,
480 	IB_WC_WR_FLUSH_ERR,
481 	IB_WC_MW_BIND_ERR,
482 	IB_WC_BAD_RESP_ERR,
483 	IB_WC_LOC_ACCESS_ERR,
484 	IB_WC_REM_INV_REQ_ERR,
485 	IB_WC_REM_ACCESS_ERR,
486 	IB_WC_REM_OP_ERR,
487 	IB_WC_RETRY_EXC_ERR,
488 	IB_WC_RNR_RETRY_EXC_ERR,
489 	IB_WC_LOC_RDD_VIOL_ERR,
490 	IB_WC_REM_INV_RD_REQ_ERR,
491 	IB_WC_REM_ABORT_ERR,
492 	IB_WC_INV_EECN_ERR,
493 	IB_WC_INV_EEC_STATE_ERR,
494 	IB_WC_FATAL_ERR,
495 	IB_WC_RESP_TIMEOUT_ERR,
496 	IB_WC_GENERAL_ERR
497 };
498 
499 enum ib_wc_opcode {
500 	IB_WC_SEND,
501 	IB_WC_RDMA_WRITE,
502 	IB_WC_RDMA_READ,
503 	IB_WC_COMP_SWAP,
504 	IB_WC_FETCH_ADD,
505 	IB_WC_BIND_MW,
506 	IB_WC_LSO,
507 	IB_WC_LOCAL_INV,
508 	IB_WC_FAST_REG_MR,
509 	IB_WC_MASKED_COMP_SWAP,
510 	IB_WC_MASKED_FETCH_ADD,
511 /*
512  * Set value of IB_WC_RECV so consumers can test if a completion is a
513  * receive by testing (opcode & IB_WC_RECV).
514  */
515 	IB_WC_RECV			= 1 << 7,
516 	IB_WC_RECV_RDMA_WITH_IMM
517 };
518 
519 enum ib_wc_flags {
520 	IB_WC_GRH		= 1,
521 	IB_WC_WITH_IMM		= (1<<1),
522 	IB_WC_WITH_INVALIDATE	= (1<<2),
523 	IB_WC_IP_CSUM_OK	= (1<<3),
524 };
525 
526 struct ib_wc {
527 	u64			wr_id;
528 	enum ib_wc_status	status;
529 	enum ib_wc_opcode	opcode;
530 	u32			vendor_err;
531 	u32			byte_len;
532 	struct ib_qp	       *qp;
533 	union {
534 		__be32		imm_data;
535 		u32		invalidate_rkey;
536 	} ex;
537 	u32			src_qp;
538 	int			wc_flags;
539 	u16			pkey_index;
540 	u16			slid;
541 	u8			sl;
542 	u8			dlid_path_bits;
543 	u8			port_num;	/* valid only for DR SMPs on switches */
544 };
545 
546 enum ib_cq_notify_flags {
547 	IB_CQ_SOLICITED			= 1 << 0,
548 	IB_CQ_NEXT_COMP			= 1 << 1,
549 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
550 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
551 };
552 
553 enum ib_srq_type {
554 	IB_SRQT_BASIC,
555 	IB_SRQT_XRC
556 };
557 
558 enum ib_srq_attr_mask {
559 	IB_SRQ_MAX_WR	= 1 << 0,
560 	IB_SRQ_LIMIT	= 1 << 1,
561 };
562 
563 struct ib_srq_attr {
564 	u32	max_wr;
565 	u32	max_sge;
566 	u32	srq_limit;
567 };
568 
569 struct ib_srq_init_attr {
570 	void		      (*event_handler)(struct ib_event *, void *);
571 	void		       *srq_context;
572 	struct ib_srq_attr	attr;
573 	enum ib_srq_type	srq_type;
574 
575 	union {
576 		struct {
577 			struct ib_xrcd *xrcd;
578 			struct ib_cq   *cq;
579 		} xrc;
580 	} ext;
581 };
582 
583 struct ib_qp_cap {
584 	u32	max_send_wr;
585 	u32	max_recv_wr;
586 	u32	max_send_sge;
587 	u32	max_recv_sge;
588 	u32	max_inline_data;
589 };
590 
591 enum ib_sig_type {
592 	IB_SIGNAL_ALL_WR,
593 	IB_SIGNAL_REQ_WR
594 };
595 
596 enum ib_qp_type {
597 	/*
598 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
599 	 * here (and in that order) since the MAD layer uses them as
600 	 * indices into a 2-entry table.
601 	 */
602 	IB_QPT_SMI,
603 	IB_QPT_GSI,
604 
605 	IB_QPT_RC,
606 	IB_QPT_UC,
607 	IB_QPT_UD,
608 	IB_QPT_RAW_IPV6,
609 	IB_QPT_RAW_ETHERTYPE,
610 	IB_QPT_RAW_PACKET = 8,
611 	IB_QPT_XRC_INI = 9,
612 	IB_QPT_XRC_TGT,
613 	IB_QPT_MAX,
614 	/* Reserve a range for qp types internal to the low level driver.
615 	 * These qp types will not be visible at the IB core layer, so the
616 	 * IB_QPT_MAX usages should not be affected in the core layer
617 	 */
618 	IB_QPT_RESERVED1 = 0x1000,
619 	IB_QPT_RESERVED2,
620 	IB_QPT_RESERVED3,
621 	IB_QPT_RESERVED4,
622 	IB_QPT_RESERVED5,
623 	IB_QPT_RESERVED6,
624 	IB_QPT_RESERVED7,
625 	IB_QPT_RESERVED8,
626 	IB_QPT_RESERVED9,
627 	IB_QPT_RESERVED10,
628 };
629 
630 enum ib_qp_create_flags {
631 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
632 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
633 	/* reserve bits 26-31 for low level drivers' internal use */
634 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
635 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
636 };
637 
638 struct ib_qp_init_attr {
639 	void                  (*event_handler)(struct ib_event *, void *);
640 	void		       *qp_context;
641 	struct ib_cq	       *send_cq;
642 	struct ib_cq	       *recv_cq;
643 	struct ib_srq	       *srq;
644 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
645 	struct ib_qp_cap	cap;
646 	enum ib_sig_type	sq_sig_type;
647 	enum ib_qp_type		qp_type;
648 	enum ib_qp_create_flags	create_flags;
649 	u8			port_num; /* special QP types only */
650 };
651 
652 struct ib_qp_open_attr {
653 	void                  (*event_handler)(struct ib_event *, void *);
654 	void		       *qp_context;
655 	u32			qp_num;
656 	enum ib_qp_type		qp_type;
657 };
658 
659 enum ib_rnr_timeout {
660 	IB_RNR_TIMER_655_36 =  0,
661 	IB_RNR_TIMER_000_01 =  1,
662 	IB_RNR_TIMER_000_02 =  2,
663 	IB_RNR_TIMER_000_03 =  3,
664 	IB_RNR_TIMER_000_04 =  4,
665 	IB_RNR_TIMER_000_06 =  5,
666 	IB_RNR_TIMER_000_08 =  6,
667 	IB_RNR_TIMER_000_12 =  7,
668 	IB_RNR_TIMER_000_16 =  8,
669 	IB_RNR_TIMER_000_24 =  9,
670 	IB_RNR_TIMER_000_32 = 10,
671 	IB_RNR_TIMER_000_48 = 11,
672 	IB_RNR_TIMER_000_64 = 12,
673 	IB_RNR_TIMER_000_96 = 13,
674 	IB_RNR_TIMER_001_28 = 14,
675 	IB_RNR_TIMER_001_92 = 15,
676 	IB_RNR_TIMER_002_56 = 16,
677 	IB_RNR_TIMER_003_84 = 17,
678 	IB_RNR_TIMER_005_12 = 18,
679 	IB_RNR_TIMER_007_68 = 19,
680 	IB_RNR_TIMER_010_24 = 20,
681 	IB_RNR_TIMER_015_36 = 21,
682 	IB_RNR_TIMER_020_48 = 22,
683 	IB_RNR_TIMER_030_72 = 23,
684 	IB_RNR_TIMER_040_96 = 24,
685 	IB_RNR_TIMER_061_44 = 25,
686 	IB_RNR_TIMER_081_92 = 26,
687 	IB_RNR_TIMER_122_88 = 27,
688 	IB_RNR_TIMER_163_84 = 28,
689 	IB_RNR_TIMER_245_76 = 29,
690 	IB_RNR_TIMER_327_68 = 30,
691 	IB_RNR_TIMER_491_52 = 31
692 };
693 
694 enum ib_qp_attr_mask {
695 	IB_QP_STATE			= 1,
696 	IB_QP_CUR_STATE			= (1<<1),
697 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
698 	IB_QP_ACCESS_FLAGS		= (1<<3),
699 	IB_QP_PKEY_INDEX		= (1<<4),
700 	IB_QP_PORT			= (1<<5),
701 	IB_QP_QKEY			= (1<<6),
702 	IB_QP_AV			= (1<<7),
703 	IB_QP_PATH_MTU			= (1<<8),
704 	IB_QP_TIMEOUT			= (1<<9),
705 	IB_QP_RETRY_CNT			= (1<<10),
706 	IB_QP_RNR_RETRY			= (1<<11),
707 	IB_QP_RQ_PSN			= (1<<12),
708 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
709 	IB_QP_ALT_PATH			= (1<<14),
710 	IB_QP_MIN_RNR_TIMER		= (1<<15),
711 	IB_QP_SQ_PSN			= (1<<16),
712 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
713 	IB_QP_PATH_MIG_STATE		= (1<<18),
714 	IB_QP_CAP			= (1<<19),
715 	IB_QP_DEST_QPN			= (1<<20)
716 };
717 
718 enum ib_qp_state {
719 	IB_QPS_RESET,
720 	IB_QPS_INIT,
721 	IB_QPS_RTR,
722 	IB_QPS_RTS,
723 	IB_QPS_SQD,
724 	IB_QPS_SQE,
725 	IB_QPS_ERR
726 };
727 
728 enum ib_mig_state {
729 	IB_MIG_MIGRATED,
730 	IB_MIG_REARM,
731 	IB_MIG_ARMED
732 };
733 
734 enum ib_mw_type {
735 	IB_MW_TYPE_1 = 1,
736 	IB_MW_TYPE_2 = 2
737 };
738 
739 struct ib_qp_attr {
740 	enum ib_qp_state	qp_state;
741 	enum ib_qp_state	cur_qp_state;
742 	enum ib_mtu		path_mtu;
743 	enum ib_mig_state	path_mig_state;
744 	u32			qkey;
745 	u32			rq_psn;
746 	u32			sq_psn;
747 	u32			dest_qp_num;
748 	int			qp_access_flags;
749 	struct ib_qp_cap	cap;
750 	struct ib_ah_attr	ah_attr;
751 	struct ib_ah_attr	alt_ah_attr;
752 	u16			pkey_index;
753 	u16			alt_pkey_index;
754 	u8			en_sqd_async_notify;
755 	u8			sq_draining;
756 	u8			max_rd_atomic;
757 	u8			max_dest_rd_atomic;
758 	u8			min_rnr_timer;
759 	u8			port_num;
760 	u8			timeout;
761 	u8			retry_cnt;
762 	u8			rnr_retry;
763 	u8			alt_port_num;
764 	u8			alt_timeout;
765 };
766 
767 enum ib_wr_opcode {
768 	IB_WR_RDMA_WRITE,
769 	IB_WR_RDMA_WRITE_WITH_IMM,
770 	IB_WR_SEND,
771 	IB_WR_SEND_WITH_IMM,
772 	IB_WR_RDMA_READ,
773 	IB_WR_ATOMIC_CMP_AND_SWP,
774 	IB_WR_ATOMIC_FETCH_AND_ADD,
775 	IB_WR_LSO,
776 	IB_WR_SEND_WITH_INV,
777 	IB_WR_RDMA_READ_WITH_INV,
778 	IB_WR_LOCAL_INV,
779 	IB_WR_FAST_REG_MR,
780 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
781 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
782 	IB_WR_BIND_MW,
783 	/* reserve values for low level drivers' internal use.
784 	 * These values will not be used at all in the ib core layer.
785 	 */
786 	IB_WR_RESERVED1 = 0xf0,
787 	IB_WR_RESERVED2,
788 	IB_WR_RESERVED3,
789 	IB_WR_RESERVED4,
790 	IB_WR_RESERVED5,
791 	IB_WR_RESERVED6,
792 	IB_WR_RESERVED7,
793 	IB_WR_RESERVED8,
794 	IB_WR_RESERVED9,
795 	IB_WR_RESERVED10,
796 };
797 
798 enum ib_send_flags {
799 	IB_SEND_FENCE		= 1,
800 	IB_SEND_SIGNALED	= (1<<1),
801 	IB_SEND_SOLICITED	= (1<<2),
802 	IB_SEND_INLINE		= (1<<3),
803 	IB_SEND_IP_CSUM		= (1<<4),
804 
805 	/* reserve bits 26-31 for low level drivers' internal use */
806 	IB_SEND_RESERVED_START	= (1 << 26),
807 	IB_SEND_RESERVED_END	= (1 << 31),
808 };
809 
810 struct ib_sge {
811 	u64	addr;
812 	u32	length;
813 	u32	lkey;
814 };
815 
816 struct ib_fast_reg_page_list {
817 	struct ib_device       *device;
818 	u64		       *page_list;
819 	unsigned int		max_page_list_len;
820 };
821 
822 /**
823  * struct ib_mw_bind_info - Parameters for a memory window bind operation.
824  * @mr: A memory region to bind the memory window to.
825  * @addr: The address where the memory window should begin.
826  * @length: The length of the memory window, in bytes.
827  * @mw_access_flags: Access flags from enum ib_access_flags for the window.
828  *
829  * This struct contains the shared parameters for type 1 and type 2
830  * memory window bind operations.
831  */
832 struct ib_mw_bind_info {
833 	struct ib_mr   *mr;
834 	u64		addr;
835 	u64		length;
836 	int		mw_access_flags;
837 };
838 
839 struct ib_send_wr {
840 	struct ib_send_wr      *next;
841 	u64			wr_id;
842 	struct ib_sge	       *sg_list;
843 	int			num_sge;
844 	enum ib_wr_opcode	opcode;
845 	int			send_flags;
846 	union {
847 		__be32		imm_data;
848 		u32		invalidate_rkey;
849 	} ex;
850 	union {
851 		struct {
852 			u64	remote_addr;
853 			u32	rkey;
854 		} rdma;
855 		struct {
856 			u64	remote_addr;
857 			u64	compare_add;
858 			u64	swap;
859 			u64	compare_add_mask;
860 			u64	swap_mask;
861 			u32	rkey;
862 		} atomic;
863 		struct {
864 			struct ib_ah *ah;
865 			void   *header;
866 			int     hlen;
867 			int     mss;
868 			u32	remote_qpn;
869 			u32	remote_qkey;
870 			u16	pkey_index; /* valid for GSI only */
871 			u8	port_num;   /* valid for DR SMPs on switch only */
872 		} ud;
873 		struct {
874 			u64				iova_start;
875 			struct ib_fast_reg_page_list   *page_list;
876 			unsigned int			page_shift;
877 			unsigned int			page_list_len;
878 			u32				length;
879 			int				access_flags;
880 			u32				rkey;
881 		} fast_reg;
882 		struct {
883 			struct ib_mw            *mw;
884 			/* The new rkey for the memory window. */
885 			u32                      rkey;
886 			struct ib_mw_bind_info   bind_info;
887 		} bind_mw;
888 	} wr;
889 	u32			xrc_remote_srq_num;	/* XRC TGT QPs only */
890 };
891 
892 struct ib_recv_wr {
893 	struct ib_recv_wr      *next;
894 	u64			wr_id;
895 	struct ib_sge	       *sg_list;
896 	int			num_sge;
897 };
898 
899 enum ib_access_flags {
900 	IB_ACCESS_LOCAL_WRITE	= 1,
901 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
902 	IB_ACCESS_REMOTE_READ	= (1<<2),
903 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
904 	IB_ACCESS_MW_BIND	= (1<<4),
905 	IB_ZERO_BASED		= (1<<5)
906 };
907 
908 struct ib_phys_buf {
909 	u64      addr;
910 	u64      size;
911 };
912 
913 struct ib_mr_attr {
914 	struct ib_pd	*pd;
915 	u64		device_virt_addr;
916 	u64		size;
917 	int		mr_access_flags;
918 	u32		lkey;
919 	u32		rkey;
920 };
921 
922 enum ib_mr_rereg_flags {
923 	IB_MR_REREG_TRANS	= 1,
924 	IB_MR_REREG_PD		= (1<<1),
925 	IB_MR_REREG_ACCESS	= (1<<2)
926 };
927 
928 /**
929  * struct ib_mw_bind - Parameters for a type 1 memory window bind operation.
930  * @wr_id:      Work request id.
931  * @send_flags: Flags from ib_send_flags enum.
932  * @bind_info:  More parameters of the bind operation.
933  */
934 struct ib_mw_bind {
935 	u64                    wr_id;
936 	int                    send_flags;
937 	struct ib_mw_bind_info bind_info;
938 };
939 
940 struct ib_fmr_attr {
941 	int	max_pages;
942 	int	max_maps;
943 	u8	page_shift;
944 };
945 
946 struct ib_ucontext {
947 	struct ib_device       *device;
948 	struct list_head	pd_list;
949 	struct list_head	mr_list;
950 	struct list_head	mw_list;
951 	struct list_head	cq_list;
952 	struct list_head	qp_list;
953 	struct list_head	srq_list;
954 	struct list_head	ah_list;
955 	struct list_head	xrcd_list;
956 	int			closing;
957 };
958 
959 struct ib_uobject {
960 	u64			user_handle;	/* handle given to us by userspace */
961 	struct ib_ucontext     *context;	/* associated user context */
962 	void		       *object;		/* containing object */
963 	struct list_head	list;		/* link to context's list */
964 	int			id;		/* index into kernel idr */
965 	struct kref		ref;
966 	struct rw_semaphore	mutex;		/* protects .live */
967 	int			live;
968 };
969 
970 struct ib_udata {
971 	void __user *inbuf;
972 	void __user *outbuf;
973 	size_t       inlen;
974 	size_t       outlen;
975 };
976 
977 struct ib_pd {
978 	struct ib_device       *device;
979 	struct ib_uobject      *uobject;
980 	atomic_t          	usecnt; /* count all resources */
981 };
982 
983 struct ib_xrcd {
984 	struct ib_device       *device;
985 	atomic_t		usecnt; /* count all exposed resources */
986 	struct inode	       *inode;
987 
988 	struct mutex		tgt_qp_mutex;
989 	struct list_head	tgt_qp_list;
990 };
991 
992 struct ib_ah {
993 	struct ib_device	*device;
994 	struct ib_pd		*pd;
995 	struct ib_uobject	*uobject;
996 };
997 
998 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
999 
1000 struct ib_cq {
1001 	struct ib_device       *device;
1002 	struct ib_uobject      *uobject;
1003 	ib_comp_handler   	comp_handler;
1004 	void                  (*event_handler)(struct ib_event *, void *);
1005 	void                   *cq_context;
1006 	int               	cqe;
1007 	atomic_t          	usecnt; /* count number of work queues */
1008 };
1009 
1010 struct ib_srq {
1011 	struct ib_device       *device;
1012 	struct ib_pd	       *pd;
1013 	struct ib_uobject      *uobject;
1014 	void		      (*event_handler)(struct ib_event *, void *);
1015 	void		       *srq_context;
1016 	enum ib_srq_type	srq_type;
1017 	atomic_t		usecnt;
1018 
1019 	union {
1020 		struct {
1021 			struct ib_xrcd *xrcd;
1022 			struct ib_cq   *cq;
1023 			u32		srq_num;
1024 		} xrc;
1025 	} ext;
1026 };
1027 
1028 struct ib_qp {
1029 	struct ib_device       *device;
1030 	struct ib_pd	       *pd;
1031 	struct ib_cq	       *send_cq;
1032 	struct ib_cq	       *recv_cq;
1033 	struct ib_srq	       *srq;
1034 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1035 	struct list_head	xrcd_list;
1036 	atomic_t		usecnt; /* count times opened, mcast attaches */
1037 	struct list_head	open_list;
1038 	struct ib_qp           *real_qp;
1039 	struct ib_uobject      *uobject;
1040 	void                  (*event_handler)(struct ib_event *, void *);
1041 	void		       *qp_context;
1042 	u32			qp_num;
1043 	enum ib_qp_type		qp_type;
1044 };
1045 
1046 struct ib_mr {
1047 	struct ib_device  *device;
1048 	struct ib_pd	  *pd;
1049 	struct ib_uobject *uobject;
1050 	u32		   lkey;
1051 	u32		   rkey;
1052 	atomic_t	   usecnt; /* count number of MWs */
1053 };
1054 
1055 struct ib_mw {
1056 	struct ib_device	*device;
1057 	struct ib_pd		*pd;
1058 	struct ib_uobject	*uobject;
1059 	u32			rkey;
1060 	enum ib_mw_type         type;
1061 };
1062 
1063 struct ib_fmr {
1064 	struct ib_device	*device;
1065 	struct ib_pd		*pd;
1066 	struct list_head	list;
1067 	u32			lkey;
1068 	u32			rkey;
1069 };
1070 
1071 struct ib_mad;
1072 struct ib_grh;
1073 
1074 enum ib_process_mad_flags {
1075 	IB_MAD_IGNORE_MKEY	= 1,
1076 	IB_MAD_IGNORE_BKEY	= 2,
1077 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1078 };
1079 
1080 enum ib_mad_result {
1081 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1082 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1083 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1084 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1085 };
1086 
1087 #define IB_DEVICE_NAME_MAX 64
1088 
1089 struct ib_cache {
1090 	rwlock_t                lock;
1091 	struct ib_event_handler event_handler;
1092 	struct ib_pkey_cache  **pkey_cache;
1093 	struct ib_gid_cache   **gid_cache;
1094 	u8                     *lmc_cache;
1095 };
1096 
1097 struct ib_dma_mapping_ops {
1098 	int		(*mapping_error)(struct ib_device *dev,
1099 					 u64 dma_addr);
1100 	u64		(*map_single)(struct ib_device *dev,
1101 				      void *ptr, size_t size,
1102 				      enum dma_data_direction direction);
1103 	void		(*unmap_single)(struct ib_device *dev,
1104 					u64 addr, size_t size,
1105 					enum dma_data_direction direction);
1106 	u64		(*map_page)(struct ib_device *dev,
1107 				    struct page *page, unsigned long offset,
1108 				    size_t size,
1109 				    enum dma_data_direction direction);
1110 	void		(*unmap_page)(struct ib_device *dev,
1111 				      u64 addr, size_t size,
1112 				      enum dma_data_direction direction);
1113 	int		(*map_sg)(struct ib_device *dev,
1114 				  struct scatterlist *sg, int nents,
1115 				  enum dma_data_direction direction);
1116 	void		(*unmap_sg)(struct ib_device *dev,
1117 				    struct scatterlist *sg, int nents,
1118 				    enum dma_data_direction direction);
1119 	u64		(*dma_address)(struct ib_device *dev,
1120 				       struct scatterlist *sg);
1121 	unsigned int	(*dma_len)(struct ib_device *dev,
1122 				   struct scatterlist *sg);
1123 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1124 					       u64 dma_handle,
1125 					       size_t size,
1126 					       enum dma_data_direction dir);
1127 	void		(*sync_single_for_device)(struct ib_device *dev,
1128 						  u64 dma_handle,
1129 						  size_t size,
1130 						  enum dma_data_direction dir);
1131 	void		*(*alloc_coherent)(struct ib_device *dev,
1132 					   size_t size,
1133 					   u64 *dma_handle,
1134 					   gfp_t flag);
1135 	void		(*free_coherent)(struct ib_device *dev,
1136 					 size_t size, void *cpu_addr,
1137 					 u64 dma_handle);
1138 };
1139 
1140 struct iw_cm_verbs;
1141 
1142 struct ib_device {
1143 	struct device                *dma_device;
1144 
1145 	char                          name[IB_DEVICE_NAME_MAX];
1146 
1147 	struct list_head              event_handler_list;
1148 	spinlock_t                    event_handler_lock;
1149 
1150 	spinlock_t                    client_data_lock;
1151 	struct list_head              core_list;
1152 	struct list_head              client_data_list;
1153 
1154 	struct ib_cache               cache;
1155 	int                          *pkey_tbl_len;
1156 	int                          *gid_tbl_len;
1157 
1158 	int			      num_comp_vectors;
1159 
1160 	struct iw_cm_verbs	     *iwcm;
1161 
1162 	int		           (*get_protocol_stats)(struct ib_device *device,
1163 							 union rdma_protocol_stats *stats);
1164 	int		           (*query_device)(struct ib_device *device,
1165 						   struct ib_device_attr *device_attr);
1166 	int		           (*query_port)(struct ib_device *device,
1167 						 u8 port_num,
1168 						 struct ib_port_attr *port_attr);
1169 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1170 						     u8 port_num);
1171 	int		           (*query_gid)(struct ib_device *device,
1172 						u8 port_num, int index,
1173 						union ib_gid *gid);
1174 	int		           (*query_pkey)(struct ib_device *device,
1175 						 u8 port_num, u16 index, u16 *pkey);
1176 	int		           (*modify_device)(struct ib_device *device,
1177 						    int device_modify_mask,
1178 						    struct ib_device_modify *device_modify);
1179 	int		           (*modify_port)(struct ib_device *device,
1180 						  u8 port_num, int port_modify_mask,
1181 						  struct ib_port_modify *port_modify);
1182 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1183 						     struct ib_udata *udata);
1184 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1185 	int                        (*mmap)(struct ib_ucontext *context,
1186 					   struct vm_area_struct *vma);
1187 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1188 					       struct ib_ucontext *context,
1189 					       struct ib_udata *udata);
1190 	int                        (*dealloc_pd)(struct ib_pd *pd);
1191 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1192 						struct ib_ah_attr *ah_attr);
1193 	int                        (*modify_ah)(struct ib_ah *ah,
1194 						struct ib_ah_attr *ah_attr);
1195 	int                        (*query_ah)(struct ib_ah *ah,
1196 					       struct ib_ah_attr *ah_attr);
1197 	int                        (*destroy_ah)(struct ib_ah *ah);
1198 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1199 						 struct ib_srq_init_attr *srq_init_attr,
1200 						 struct ib_udata *udata);
1201 	int                        (*modify_srq)(struct ib_srq *srq,
1202 						 struct ib_srq_attr *srq_attr,
1203 						 enum ib_srq_attr_mask srq_attr_mask,
1204 						 struct ib_udata *udata);
1205 	int                        (*query_srq)(struct ib_srq *srq,
1206 						struct ib_srq_attr *srq_attr);
1207 	int                        (*destroy_srq)(struct ib_srq *srq);
1208 	int                        (*post_srq_recv)(struct ib_srq *srq,
1209 						    struct ib_recv_wr *recv_wr,
1210 						    struct ib_recv_wr **bad_recv_wr);
1211 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1212 						struct ib_qp_init_attr *qp_init_attr,
1213 						struct ib_udata *udata);
1214 	int                        (*modify_qp)(struct ib_qp *qp,
1215 						struct ib_qp_attr *qp_attr,
1216 						int qp_attr_mask,
1217 						struct ib_udata *udata);
1218 	int                        (*query_qp)(struct ib_qp *qp,
1219 					       struct ib_qp_attr *qp_attr,
1220 					       int qp_attr_mask,
1221 					       struct ib_qp_init_attr *qp_init_attr);
1222 	int                        (*destroy_qp)(struct ib_qp *qp);
1223 	int                        (*post_send)(struct ib_qp *qp,
1224 						struct ib_send_wr *send_wr,
1225 						struct ib_send_wr **bad_send_wr);
1226 	int                        (*post_recv)(struct ib_qp *qp,
1227 						struct ib_recv_wr *recv_wr,
1228 						struct ib_recv_wr **bad_recv_wr);
1229 	struct ib_cq *             (*create_cq)(struct ib_device *device, int cqe,
1230 						int comp_vector,
1231 						struct ib_ucontext *context,
1232 						struct ib_udata *udata);
1233 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1234 						u16 cq_period);
1235 	int                        (*destroy_cq)(struct ib_cq *cq);
1236 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1237 						struct ib_udata *udata);
1238 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1239 					      struct ib_wc *wc);
1240 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1241 	int                        (*req_notify_cq)(struct ib_cq *cq,
1242 						    enum ib_cq_notify_flags flags);
1243 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1244 						      int wc_cnt);
1245 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1246 						 int mr_access_flags);
1247 	struct ib_mr *             (*reg_phys_mr)(struct ib_pd *pd,
1248 						  struct ib_phys_buf *phys_buf_array,
1249 						  int num_phys_buf,
1250 						  int mr_access_flags,
1251 						  u64 *iova_start);
1252 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1253 						  u64 start, u64 length,
1254 						  u64 virt_addr,
1255 						  int mr_access_flags,
1256 						  struct ib_udata *udata);
1257 	int                        (*query_mr)(struct ib_mr *mr,
1258 					       struct ib_mr_attr *mr_attr);
1259 	int                        (*dereg_mr)(struct ib_mr *mr);
1260 	struct ib_mr *		   (*alloc_fast_reg_mr)(struct ib_pd *pd,
1261 					       int max_page_list_len);
1262 	struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1263 								   int page_list_len);
1264 	void			   (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1265 	int                        (*rereg_phys_mr)(struct ib_mr *mr,
1266 						    int mr_rereg_mask,
1267 						    struct ib_pd *pd,
1268 						    struct ib_phys_buf *phys_buf_array,
1269 						    int num_phys_buf,
1270 						    int mr_access_flags,
1271 						    u64 *iova_start);
1272 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
1273 					       enum ib_mw_type type);
1274 	int                        (*bind_mw)(struct ib_qp *qp,
1275 					      struct ib_mw *mw,
1276 					      struct ib_mw_bind *mw_bind);
1277 	int                        (*dealloc_mw)(struct ib_mw *mw);
1278 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1279 						int mr_access_flags,
1280 						struct ib_fmr_attr *fmr_attr);
1281 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1282 						   u64 *page_list, int list_len,
1283 						   u64 iova);
1284 	int		           (*unmap_fmr)(struct list_head *fmr_list);
1285 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1286 	int                        (*attach_mcast)(struct ib_qp *qp,
1287 						   union ib_gid *gid,
1288 						   u16 lid);
1289 	int                        (*detach_mcast)(struct ib_qp *qp,
1290 						   union ib_gid *gid,
1291 						   u16 lid);
1292 	int                        (*process_mad)(struct ib_device *device,
1293 						  int process_mad_flags,
1294 						  u8 port_num,
1295 						  struct ib_wc *in_wc,
1296 						  struct ib_grh *in_grh,
1297 						  struct ib_mad *in_mad,
1298 						  struct ib_mad *out_mad);
1299 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
1300 						 struct ib_ucontext *ucontext,
1301 						 struct ib_udata *udata);
1302 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1303 
1304 	struct ib_dma_mapping_ops   *dma_ops;
1305 
1306 	struct module               *owner;
1307 	struct device                dev;
1308 	struct kobject               *ports_parent;
1309 	struct list_head             port_list;
1310 
1311 	enum {
1312 		IB_DEV_UNINITIALIZED,
1313 		IB_DEV_REGISTERED,
1314 		IB_DEV_UNREGISTERED
1315 	}                            reg_state;
1316 
1317 	int			     uverbs_abi_ver;
1318 	u64			     uverbs_cmd_mask;
1319 
1320 	char			     node_desc[64];
1321 	__be64			     node_guid;
1322 	u32			     local_dma_lkey;
1323 	u8                           node_type;
1324 	u8                           phys_port_cnt;
1325 };
1326 
1327 struct ib_client {
1328 	char  *name;
1329 	void (*add)   (struct ib_device *);
1330 	void (*remove)(struct ib_device *);
1331 
1332 	struct list_head list;
1333 };
1334 
1335 struct ib_device *ib_alloc_device(size_t size);
1336 void ib_dealloc_device(struct ib_device *device);
1337 
1338 int ib_register_device(struct ib_device *device,
1339 		       int (*port_callback)(struct ib_device *,
1340 					    u8, struct kobject *));
1341 void ib_unregister_device(struct ib_device *device);
1342 
1343 int ib_register_client   (struct ib_client *client);
1344 void ib_unregister_client(struct ib_client *client);
1345 
1346 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1347 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
1348 			 void *data);
1349 
1350 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1351 {
1352 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1353 }
1354 
1355 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1356 {
1357 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1358 }
1359 
1360 /**
1361  * ib_modify_qp_is_ok - Check that the supplied attribute mask
1362  * contains all required attributes and no attributes not allowed for
1363  * the given QP state transition.
1364  * @cur_state: Current QP state
1365  * @next_state: Next QP state
1366  * @type: QP type
1367  * @mask: Mask of supplied QP attributes
1368  *
1369  * This function is a helper function that a low-level driver's
1370  * modify_qp method can use to validate the consumer's input.  It
1371  * checks that cur_state and next_state are valid QP states, that a
1372  * transition from cur_state to next_state is allowed by the IB spec,
1373  * and that the attribute mask supplied is allowed for the transition.
1374  */
1375 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1376 		       enum ib_qp_type type, enum ib_qp_attr_mask mask);
1377 
1378 int ib_register_event_handler  (struct ib_event_handler *event_handler);
1379 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1380 void ib_dispatch_event(struct ib_event *event);
1381 
1382 int ib_query_device(struct ib_device *device,
1383 		    struct ib_device_attr *device_attr);
1384 
1385 int ib_query_port(struct ib_device *device,
1386 		  u8 port_num, struct ib_port_attr *port_attr);
1387 
1388 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1389 					       u8 port_num);
1390 
1391 int ib_query_gid(struct ib_device *device,
1392 		 u8 port_num, int index, union ib_gid *gid);
1393 
1394 int ib_query_pkey(struct ib_device *device,
1395 		  u8 port_num, u16 index, u16 *pkey);
1396 
1397 int ib_modify_device(struct ib_device *device,
1398 		     int device_modify_mask,
1399 		     struct ib_device_modify *device_modify);
1400 
1401 int ib_modify_port(struct ib_device *device,
1402 		   u8 port_num, int port_modify_mask,
1403 		   struct ib_port_modify *port_modify);
1404 
1405 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1406 		u8 *port_num, u16 *index);
1407 
1408 int ib_find_pkey(struct ib_device *device,
1409 		 u8 port_num, u16 pkey, u16 *index);
1410 
1411 /**
1412  * ib_alloc_pd - Allocates an unused protection domain.
1413  * @device: The device on which to allocate the protection domain.
1414  *
1415  * A protection domain object provides an association between QPs, shared
1416  * receive queues, address handles, memory regions, and memory windows.
1417  */
1418 struct ib_pd *ib_alloc_pd(struct ib_device *device);
1419 
1420 /**
1421  * ib_dealloc_pd - Deallocates a protection domain.
1422  * @pd: The protection domain to deallocate.
1423  */
1424 int ib_dealloc_pd(struct ib_pd *pd);
1425 
1426 /**
1427  * ib_create_ah - Creates an address handle for the given address vector.
1428  * @pd: The protection domain associated with the address handle.
1429  * @ah_attr: The attributes of the address vector.
1430  *
1431  * The address handle is used to reference a local or global destination
1432  * in all UD QP post sends.
1433  */
1434 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
1435 
1436 /**
1437  * ib_init_ah_from_wc - Initializes address handle attributes from a
1438  *   work completion.
1439  * @device: Device on which the received message arrived.
1440  * @port_num: Port on which the received message arrived.
1441  * @wc: Work completion associated with the received message.
1442  * @grh: References the received global route header.  This parameter is
1443  *   ignored unless the work completion indicates that the GRH is valid.
1444  * @ah_attr: Returned attributes that can be used when creating an address
1445  *   handle for replying to the message.
1446  */
1447 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
1448 		       struct ib_grh *grh, struct ib_ah_attr *ah_attr);
1449 
1450 /**
1451  * ib_create_ah_from_wc - Creates an address handle associated with the
1452  *   sender of the specified work completion.
1453  * @pd: The protection domain associated with the address handle.
1454  * @wc: Work completion information associated with a received message.
1455  * @grh: References the received global route header.  This parameter is
1456  *   ignored unless the work completion indicates that the GRH is valid.
1457  * @port_num: The outbound port number to associate with the address.
1458  *
1459  * The address handle is used to reference a local or global destination
1460  * in all UD QP post sends.
1461  */
1462 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
1463 				   struct ib_grh *grh, u8 port_num);
1464 
1465 /**
1466  * ib_modify_ah - Modifies the address vector associated with an address
1467  *   handle.
1468  * @ah: The address handle to modify.
1469  * @ah_attr: The new address vector attributes to associate with the
1470  *   address handle.
1471  */
1472 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1473 
1474 /**
1475  * ib_query_ah - Queries the address vector associated with an address
1476  *   handle.
1477  * @ah: The address handle to query.
1478  * @ah_attr: The address vector attributes associated with the address
1479  *   handle.
1480  */
1481 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1482 
1483 /**
1484  * ib_destroy_ah - Destroys an address handle.
1485  * @ah: The address handle to destroy.
1486  */
1487 int ib_destroy_ah(struct ib_ah *ah);
1488 
1489 /**
1490  * ib_create_srq - Creates a SRQ associated with the specified protection
1491  *   domain.
1492  * @pd: The protection domain associated with the SRQ.
1493  * @srq_init_attr: A list of initial attributes required to create the
1494  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
1495  *   the actual capabilities of the created SRQ.
1496  *
1497  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1498  * requested size of the SRQ, and set to the actual values allocated
1499  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
1500  * will always be at least as large as the requested values.
1501  */
1502 struct ib_srq *ib_create_srq(struct ib_pd *pd,
1503 			     struct ib_srq_init_attr *srq_init_attr);
1504 
1505 /**
1506  * ib_modify_srq - Modifies the attributes for the specified SRQ.
1507  * @srq: The SRQ to modify.
1508  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
1509  *   the current values of selected SRQ attributes are returned.
1510  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1511  *   are being modified.
1512  *
1513  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1514  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1515  * the number of receives queued drops below the limit.
1516  */
1517 int ib_modify_srq(struct ib_srq *srq,
1518 		  struct ib_srq_attr *srq_attr,
1519 		  enum ib_srq_attr_mask srq_attr_mask);
1520 
1521 /**
1522  * ib_query_srq - Returns the attribute list and current values for the
1523  *   specified SRQ.
1524  * @srq: The SRQ to query.
1525  * @srq_attr: The attributes of the specified SRQ.
1526  */
1527 int ib_query_srq(struct ib_srq *srq,
1528 		 struct ib_srq_attr *srq_attr);
1529 
1530 /**
1531  * ib_destroy_srq - Destroys the specified SRQ.
1532  * @srq: The SRQ to destroy.
1533  */
1534 int ib_destroy_srq(struct ib_srq *srq);
1535 
1536 /**
1537  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1538  * @srq: The SRQ to post the work request on.
1539  * @recv_wr: A list of work requests to post on the receive queue.
1540  * @bad_recv_wr: On an immediate failure, this parameter will reference
1541  *   the work request that failed to be posted on the QP.
1542  */
1543 static inline int ib_post_srq_recv(struct ib_srq *srq,
1544 				   struct ib_recv_wr *recv_wr,
1545 				   struct ib_recv_wr **bad_recv_wr)
1546 {
1547 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1548 }
1549 
1550 /**
1551  * ib_create_qp - Creates a QP associated with the specified protection
1552  *   domain.
1553  * @pd: The protection domain associated with the QP.
1554  * @qp_init_attr: A list of initial attributes required to create the
1555  *   QP.  If QP creation succeeds, then the attributes are updated to
1556  *   the actual capabilities of the created QP.
1557  */
1558 struct ib_qp *ib_create_qp(struct ib_pd *pd,
1559 			   struct ib_qp_init_attr *qp_init_attr);
1560 
1561 /**
1562  * ib_modify_qp - Modifies the attributes for the specified QP and then
1563  *   transitions the QP to the given state.
1564  * @qp: The QP to modify.
1565  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
1566  *   the current values of selected QP attributes are returned.
1567  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1568  *   are being modified.
1569  */
1570 int ib_modify_qp(struct ib_qp *qp,
1571 		 struct ib_qp_attr *qp_attr,
1572 		 int qp_attr_mask);
1573 
1574 /**
1575  * ib_query_qp - Returns the attribute list and current values for the
1576  *   specified QP.
1577  * @qp: The QP to query.
1578  * @qp_attr: The attributes of the specified QP.
1579  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1580  * @qp_init_attr: Additional attributes of the selected QP.
1581  *
1582  * The qp_attr_mask may be used to limit the query to gathering only the
1583  * selected attributes.
1584  */
1585 int ib_query_qp(struct ib_qp *qp,
1586 		struct ib_qp_attr *qp_attr,
1587 		int qp_attr_mask,
1588 		struct ib_qp_init_attr *qp_init_attr);
1589 
1590 /**
1591  * ib_destroy_qp - Destroys the specified QP.
1592  * @qp: The QP to destroy.
1593  */
1594 int ib_destroy_qp(struct ib_qp *qp);
1595 
1596 /**
1597  * ib_open_qp - Obtain a reference to an existing sharable QP.
1598  * @xrcd - XRC domain
1599  * @qp_open_attr: Attributes identifying the QP to open.
1600  *
1601  * Returns a reference to a sharable QP.
1602  */
1603 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
1604 			 struct ib_qp_open_attr *qp_open_attr);
1605 
1606 /**
1607  * ib_close_qp - Release an external reference to a QP.
1608  * @qp: The QP handle to release
1609  *
1610  * The opened QP handle is released by the caller.  The underlying
1611  * shared QP is not destroyed until all internal references are released.
1612  */
1613 int ib_close_qp(struct ib_qp *qp);
1614 
1615 /**
1616  * ib_post_send - Posts a list of work requests to the send queue of
1617  *   the specified QP.
1618  * @qp: The QP to post the work request on.
1619  * @send_wr: A list of work requests to post on the send queue.
1620  * @bad_send_wr: On an immediate failure, this parameter will reference
1621  *   the work request that failed to be posted on the QP.
1622  *
1623  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
1624  * error is returned, the QP state shall not be affected,
1625  * ib_post_send() will return an immediate error after queueing any
1626  * earlier work requests in the list.
1627  */
1628 static inline int ib_post_send(struct ib_qp *qp,
1629 			       struct ib_send_wr *send_wr,
1630 			       struct ib_send_wr **bad_send_wr)
1631 {
1632 	return qp->device->post_send(qp, send_wr, bad_send_wr);
1633 }
1634 
1635 /**
1636  * ib_post_recv - Posts a list of work requests to the receive queue of
1637  *   the specified QP.
1638  * @qp: The QP to post the work request on.
1639  * @recv_wr: A list of work requests to post on the receive queue.
1640  * @bad_recv_wr: On an immediate failure, this parameter will reference
1641  *   the work request that failed to be posted on the QP.
1642  */
1643 static inline int ib_post_recv(struct ib_qp *qp,
1644 			       struct ib_recv_wr *recv_wr,
1645 			       struct ib_recv_wr **bad_recv_wr)
1646 {
1647 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1648 }
1649 
1650 /**
1651  * ib_create_cq - Creates a CQ on the specified device.
1652  * @device: The device on which to create the CQ.
1653  * @comp_handler: A user-specified callback that is invoked when a
1654  *   completion event occurs on the CQ.
1655  * @event_handler: A user-specified callback that is invoked when an
1656  *   asynchronous event not associated with a completion occurs on the CQ.
1657  * @cq_context: Context associated with the CQ returned to the user via
1658  *   the associated completion and event handlers.
1659  * @cqe: The minimum size of the CQ.
1660  * @comp_vector - Completion vector used to signal completion events.
1661  *     Must be >= 0 and < context->num_comp_vectors.
1662  *
1663  * Users can examine the cq structure to determine the actual CQ size.
1664  */
1665 struct ib_cq *ib_create_cq(struct ib_device *device,
1666 			   ib_comp_handler comp_handler,
1667 			   void (*event_handler)(struct ib_event *, void *),
1668 			   void *cq_context, int cqe, int comp_vector);
1669 
1670 /**
1671  * ib_resize_cq - Modifies the capacity of the CQ.
1672  * @cq: The CQ to resize.
1673  * @cqe: The minimum size of the CQ.
1674  *
1675  * Users can examine the cq structure to determine the actual CQ size.
1676  */
1677 int ib_resize_cq(struct ib_cq *cq, int cqe);
1678 
1679 /**
1680  * ib_modify_cq - Modifies moderation params of the CQ
1681  * @cq: The CQ to modify.
1682  * @cq_count: number of CQEs that will trigger an event
1683  * @cq_period: max period of time in usec before triggering an event
1684  *
1685  */
1686 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1687 
1688 /**
1689  * ib_destroy_cq - Destroys the specified CQ.
1690  * @cq: The CQ to destroy.
1691  */
1692 int ib_destroy_cq(struct ib_cq *cq);
1693 
1694 /**
1695  * ib_poll_cq - poll a CQ for completion(s)
1696  * @cq:the CQ being polled
1697  * @num_entries:maximum number of completions to return
1698  * @wc:array of at least @num_entries &struct ib_wc where completions
1699  *   will be returned
1700  *
1701  * Poll a CQ for (possibly multiple) completions.  If the return value
1702  * is < 0, an error occurred.  If the return value is >= 0, it is the
1703  * number of completions returned.  If the return value is
1704  * non-negative and < num_entries, then the CQ was emptied.
1705  */
1706 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1707 			     struct ib_wc *wc)
1708 {
1709 	return cq->device->poll_cq(cq, num_entries, wc);
1710 }
1711 
1712 /**
1713  * ib_peek_cq - Returns the number of unreaped completions currently
1714  *   on the specified CQ.
1715  * @cq: The CQ to peek.
1716  * @wc_cnt: A minimum number of unreaped completions to check for.
1717  *
1718  * If the number of unreaped completions is greater than or equal to wc_cnt,
1719  * this function returns wc_cnt, otherwise, it returns the actual number of
1720  * unreaped completions.
1721  */
1722 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1723 
1724 /**
1725  * ib_req_notify_cq - Request completion notification on a CQ.
1726  * @cq: The CQ to generate an event for.
1727  * @flags:
1728  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1729  *   to request an event on the next solicited event or next work
1730  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1731  *   may also be |ed in to request a hint about missed events, as
1732  *   described below.
1733  *
1734  * Return Value:
1735  *    < 0 means an error occurred while requesting notification
1736  *   == 0 means notification was requested successfully, and if
1737  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1738  *        were missed and it is safe to wait for another event.  In
1739  *        this case is it guaranteed that any work completions added
1740  *        to the CQ since the last CQ poll will trigger a completion
1741  *        notification event.
1742  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1743  *        in.  It means that the consumer must poll the CQ again to
1744  *        make sure it is empty to avoid missing an event because of a
1745  *        race between requesting notification and an entry being
1746  *        added to the CQ.  This return value means it is possible
1747  *        (but not guaranteed) that a work completion has been added
1748  *        to the CQ since the last poll without triggering a
1749  *        completion notification event.
1750  */
1751 static inline int ib_req_notify_cq(struct ib_cq *cq,
1752 				   enum ib_cq_notify_flags flags)
1753 {
1754 	return cq->device->req_notify_cq(cq, flags);
1755 }
1756 
1757 /**
1758  * ib_req_ncomp_notif - Request completion notification when there are
1759  *   at least the specified number of unreaped completions on the CQ.
1760  * @cq: The CQ to generate an event for.
1761  * @wc_cnt: The number of unreaped completions that should be on the
1762  *   CQ before an event is generated.
1763  */
1764 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1765 {
1766 	return cq->device->req_ncomp_notif ?
1767 		cq->device->req_ncomp_notif(cq, wc_cnt) :
1768 		-ENOSYS;
1769 }
1770 
1771 /**
1772  * ib_get_dma_mr - Returns a memory region for system memory that is
1773  *   usable for DMA.
1774  * @pd: The protection domain associated with the memory region.
1775  * @mr_access_flags: Specifies the memory access rights.
1776  *
1777  * Note that the ib_dma_*() functions defined below must be used
1778  * to create/destroy addresses used with the Lkey or Rkey returned
1779  * by ib_get_dma_mr().
1780  */
1781 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1782 
1783 /**
1784  * ib_dma_mapping_error - check a DMA addr for error
1785  * @dev: The device for which the dma_addr was created
1786  * @dma_addr: The DMA address to check
1787  */
1788 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1789 {
1790 	if (dev->dma_ops)
1791 		return dev->dma_ops->mapping_error(dev, dma_addr);
1792 	return dma_mapping_error(dev->dma_device, dma_addr);
1793 }
1794 
1795 /**
1796  * ib_dma_map_single - Map a kernel virtual address to DMA address
1797  * @dev: The device for which the dma_addr is to be created
1798  * @cpu_addr: The kernel virtual address
1799  * @size: The size of the region in bytes
1800  * @direction: The direction of the DMA
1801  */
1802 static inline u64 ib_dma_map_single(struct ib_device *dev,
1803 				    void *cpu_addr, size_t size,
1804 				    enum dma_data_direction direction)
1805 {
1806 	if (dev->dma_ops)
1807 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1808 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1809 }
1810 
1811 /**
1812  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1813  * @dev: The device for which the DMA address was created
1814  * @addr: The DMA address
1815  * @size: The size of the region in bytes
1816  * @direction: The direction of the DMA
1817  */
1818 static inline void ib_dma_unmap_single(struct ib_device *dev,
1819 				       u64 addr, size_t size,
1820 				       enum dma_data_direction direction)
1821 {
1822 	if (dev->dma_ops)
1823 		dev->dma_ops->unmap_single(dev, addr, size, direction);
1824 	else
1825 		dma_unmap_single(dev->dma_device, addr, size, direction);
1826 }
1827 
1828 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
1829 					  void *cpu_addr, size_t size,
1830 					  enum dma_data_direction direction,
1831 					  struct dma_attrs *attrs)
1832 {
1833 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
1834 				    direction, attrs);
1835 }
1836 
1837 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
1838 					     u64 addr, size_t size,
1839 					     enum dma_data_direction direction,
1840 					     struct dma_attrs *attrs)
1841 {
1842 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
1843 				      direction, attrs);
1844 }
1845 
1846 /**
1847  * ib_dma_map_page - Map a physical page to DMA address
1848  * @dev: The device for which the dma_addr is to be created
1849  * @page: The page to be mapped
1850  * @offset: The offset within the page
1851  * @size: The size of the region in bytes
1852  * @direction: The direction of the DMA
1853  */
1854 static inline u64 ib_dma_map_page(struct ib_device *dev,
1855 				  struct page *page,
1856 				  unsigned long offset,
1857 				  size_t size,
1858 					 enum dma_data_direction direction)
1859 {
1860 	if (dev->dma_ops)
1861 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
1862 	return dma_map_page(dev->dma_device, page, offset, size, direction);
1863 }
1864 
1865 /**
1866  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1867  * @dev: The device for which the DMA address was created
1868  * @addr: The DMA address
1869  * @size: The size of the region in bytes
1870  * @direction: The direction of the DMA
1871  */
1872 static inline void ib_dma_unmap_page(struct ib_device *dev,
1873 				     u64 addr, size_t size,
1874 				     enum dma_data_direction direction)
1875 {
1876 	if (dev->dma_ops)
1877 		dev->dma_ops->unmap_page(dev, addr, size, direction);
1878 	else
1879 		dma_unmap_page(dev->dma_device, addr, size, direction);
1880 }
1881 
1882 /**
1883  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
1884  * @dev: The device for which the DMA addresses are to be created
1885  * @sg: The array of scatter/gather entries
1886  * @nents: The number of scatter/gather entries
1887  * @direction: The direction of the DMA
1888  */
1889 static inline int ib_dma_map_sg(struct ib_device *dev,
1890 				struct scatterlist *sg, int nents,
1891 				enum dma_data_direction direction)
1892 {
1893 	if (dev->dma_ops)
1894 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
1895 	return dma_map_sg(dev->dma_device, sg, nents, direction);
1896 }
1897 
1898 /**
1899  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
1900  * @dev: The device for which the DMA addresses were created
1901  * @sg: The array of scatter/gather entries
1902  * @nents: The number of scatter/gather entries
1903  * @direction: The direction of the DMA
1904  */
1905 static inline void ib_dma_unmap_sg(struct ib_device *dev,
1906 				   struct scatterlist *sg, int nents,
1907 				   enum dma_data_direction direction)
1908 {
1909 	if (dev->dma_ops)
1910 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
1911 	else
1912 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
1913 }
1914 
1915 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
1916 				      struct scatterlist *sg, int nents,
1917 				      enum dma_data_direction direction,
1918 				      struct dma_attrs *attrs)
1919 {
1920 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1921 }
1922 
1923 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
1924 					 struct scatterlist *sg, int nents,
1925 					 enum dma_data_direction direction,
1926 					 struct dma_attrs *attrs)
1927 {
1928 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1929 }
1930 /**
1931  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
1932  * @dev: The device for which the DMA addresses were created
1933  * @sg: The scatter/gather entry
1934  */
1935 static inline u64 ib_sg_dma_address(struct ib_device *dev,
1936 				    struct scatterlist *sg)
1937 {
1938 	if (dev->dma_ops)
1939 		return dev->dma_ops->dma_address(dev, sg);
1940 	return sg_dma_address(sg);
1941 }
1942 
1943 /**
1944  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
1945  * @dev: The device for which the DMA addresses were created
1946  * @sg: The scatter/gather entry
1947  */
1948 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
1949 					 struct scatterlist *sg)
1950 {
1951 	if (dev->dma_ops)
1952 		return dev->dma_ops->dma_len(dev, sg);
1953 	return sg_dma_len(sg);
1954 }
1955 
1956 /**
1957  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
1958  * @dev: The device for which the DMA address was created
1959  * @addr: The DMA address
1960  * @size: The size of the region in bytes
1961  * @dir: The direction of the DMA
1962  */
1963 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
1964 					      u64 addr,
1965 					      size_t size,
1966 					      enum dma_data_direction dir)
1967 {
1968 	if (dev->dma_ops)
1969 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
1970 	else
1971 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
1972 }
1973 
1974 /**
1975  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
1976  * @dev: The device for which the DMA address was created
1977  * @addr: The DMA address
1978  * @size: The size of the region in bytes
1979  * @dir: The direction of the DMA
1980  */
1981 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
1982 						 u64 addr,
1983 						 size_t size,
1984 						 enum dma_data_direction dir)
1985 {
1986 	if (dev->dma_ops)
1987 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
1988 	else
1989 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
1990 }
1991 
1992 /**
1993  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
1994  * @dev: The device for which the DMA address is requested
1995  * @size: The size of the region to allocate in bytes
1996  * @dma_handle: A pointer for returning the DMA address of the region
1997  * @flag: memory allocator flags
1998  */
1999 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2000 					   size_t size,
2001 					   u64 *dma_handle,
2002 					   gfp_t flag)
2003 {
2004 	if (dev->dma_ops)
2005 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2006 	else {
2007 		dma_addr_t handle;
2008 		void *ret;
2009 
2010 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2011 		*dma_handle = handle;
2012 		return ret;
2013 	}
2014 }
2015 
2016 /**
2017  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2018  * @dev: The device for which the DMA addresses were allocated
2019  * @size: The size of the region
2020  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2021  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2022  */
2023 static inline void ib_dma_free_coherent(struct ib_device *dev,
2024 					size_t size, void *cpu_addr,
2025 					u64 dma_handle)
2026 {
2027 	if (dev->dma_ops)
2028 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2029 	else
2030 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2031 }
2032 
2033 /**
2034  * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
2035  *   by an HCA.
2036  * @pd: The protection domain associated assigned to the registered region.
2037  * @phys_buf_array: Specifies a list of physical buffers to use in the
2038  *   memory region.
2039  * @num_phys_buf: Specifies the size of the phys_buf_array.
2040  * @mr_access_flags: Specifies the memory access rights.
2041  * @iova_start: The offset of the region's starting I/O virtual address.
2042  */
2043 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
2044 			     struct ib_phys_buf *phys_buf_array,
2045 			     int num_phys_buf,
2046 			     int mr_access_flags,
2047 			     u64 *iova_start);
2048 
2049 /**
2050  * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
2051  *   Conceptually, this call performs the functions deregister memory region
2052  *   followed by register physical memory region.  Where possible,
2053  *   resources are reused instead of deallocated and reallocated.
2054  * @mr: The memory region to modify.
2055  * @mr_rereg_mask: A bit-mask used to indicate which of the following
2056  *   properties of the memory region are being modified.
2057  * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
2058  *   the new protection domain to associated with the memory region,
2059  *   otherwise, this parameter is ignored.
2060  * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2061  *   field specifies a list of physical buffers to use in the new
2062  *   translation, otherwise, this parameter is ignored.
2063  * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2064  *   field specifies the size of the phys_buf_array, otherwise, this
2065  *   parameter is ignored.
2066  * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
2067  *   field specifies the new memory access rights, otherwise, this
2068  *   parameter is ignored.
2069  * @iova_start: The offset of the region's starting I/O virtual address.
2070  */
2071 int ib_rereg_phys_mr(struct ib_mr *mr,
2072 		     int mr_rereg_mask,
2073 		     struct ib_pd *pd,
2074 		     struct ib_phys_buf *phys_buf_array,
2075 		     int num_phys_buf,
2076 		     int mr_access_flags,
2077 		     u64 *iova_start);
2078 
2079 /**
2080  * ib_query_mr - Retrieves information about a specific memory region.
2081  * @mr: The memory region to retrieve information about.
2082  * @mr_attr: The attributes of the specified memory region.
2083  */
2084 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2085 
2086 /**
2087  * ib_dereg_mr - Deregisters a memory region and removes it from the
2088  *   HCA translation table.
2089  * @mr: The memory region to deregister.
2090  *
2091  * This function can fail, if the memory region has memory windows bound to it.
2092  */
2093 int ib_dereg_mr(struct ib_mr *mr);
2094 
2095 /**
2096  * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2097  *   IB_WR_FAST_REG_MR send work request.
2098  * @pd: The protection domain associated with the region.
2099  * @max_page_list_len: requested max physical buffer list length to be
2100  *   used with fast register work requests for this MR.
2101  */
2102 struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
2103 
2104 /**
2105  * ib_alloc_fast_reg_page_list - Allocates a page list array
2106  * @device - ib device pointer.
2107  * @page_list_len - size of the page list array to be allocated.
2108  *
2109  * This allocates and returns a struct ib_fast_reg_page_list * and a
2110  * page_list array that is at least page_list_len in size.  The actual
2111  * size is returned in max_page_list_len.  The caller is responsible
2112  * for initializing the contents of the page_list array before posting
2113  * a send work request with the IB_WC_FAST_REG_MR opcode.
2114  *
2115  * The page_list array entries must be translated using one of the
2116  * ib_dma_*() functions just like the addresses passed to
2117  * ib_map_phys_fmr().  Once the ib_post_send() is issued, the struct
2118  * ib_fast_reg_page_list must not be modified by the caller until the
2119  * IB_WC_FAST_REG_MR work request completes.
2120  */
2121 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2122 				struct ib_device *device, int page_list_len);
2123 
2124 /**
2125  * ib_free_fast_reg_page_list - Deallocates a previously allocated
2126  *   page list array.
2127  * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2128  */
2129 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2130 
2131 /**
2132  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2133  *   R_Key and L_Key.
2134  * @mr - struct ib_mr pointer to be updated.
2135  * @newkey - new key to be used.
2136  */
2137 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2138 {
2139 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2140 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2141 }
2142 
2143 /**
2144  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2145  * for calculating a new rkey for type 2 memory windows.
2146  * @rkey - the rkey to increment.
2147  */
2148 static inline u32 ib_inc_rkey(u32 rkey)
2149 {
2150 	const u32 mask = 0x000000ff;
2151 	return ((rkey + 1) & mask) | (rkey & ~mask);
2152 }
2153 
2154 /**
2155  * ib_alloc_mw - Allocates a memory window.
2156  * @pd: The protection domain associated with the memory window.
2157  * @type: The type of the memory window (1 or 2).
2158  */
2159 struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
2160 
2161 /**
2162  * ib_bind_mw - Posts a work request to the send queue of the specified
2163  *   QP, which binds the memory window to the given address range and
2164  *   remote access attributes.
2165  * @qp: QP to post the bind work request on.
2166  * @mw: The memory window to bind.
2167  * @mw_bind: Specifies information about the memory window, including
2168  *   its address range, remote access rights, and associated memory region.
2169  *
2170  * If there is no immediate error, the function will update the rkey member
2171  * of the mw parameter to its new value. The bind operation can still fail
2172  * asynchronously.
2173  */
2174 static inline int ib_bind_mw(struct ib_qp *qp,
2175 			     struct ib_mw *mw,
2176 			     struct ib_mw_bind *mw_bind)
2177 {
2178 	/* XXX reference counting in corresponding MR? */
2179 	return mw->device->bind_mw ?
2180 		mw->device->bind_mw(qp, mw, mw_bind) :
2181 		-ENOSYS;
2182 }
2183 
2184 /**
2185  * ib_dealloc_mw - Deallocates a memory window.
2186  * @mw: The memory window to deallocate.
2187  */
2188 int ib_dealloc_mw(struct ib_mw *mw);
2189 
2190 /**
2191  * ib_alloc_fmr - Allocates a unmapped fast memory region.
2192  * @pd: The protection domain associated with the unmapped region.
2193  * @mr_access_flags: Specifies the memory access rights.
2194  * @fmr_attr: Attributes of the unmapped region.
2195  *
2196  * A fast memory region must be mapped before it can be used as part of
2197  * a work request.
2198  */
2199 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2200 			    int mr_access_flags,
2201 			    struct ib_fmr_attr *fmr_attr);
2202 
2203 /**
2204  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2205  * @fmr: The fast memory region to associate with the pages.
2206  * @page_list: An array of physical pages to map to the fast memory region.
2207  * @list_len: The number of pages in page_list.
2208  * @iova: The I/O virtual address to use with the mapped region.
2209  */
2210 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2211 				  u64 *page_list, int list_len,
2212 				  u64 iova)
2213 {
2214 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2215 }
2216 
2217 /**
2218  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2219  * @fmr_list: A linked list of fast memory regions to unmap.
2220  */
2221 int ib_unmap_fmr(struct list_head *fmr_list);
2222 
2223 /**
2224  * ib_dealloc_fmr - Deallocates a fast memory region.
2225  * @fmr: The fast memory region to deallocate.
2226  */
2227 int ib_dealloc_fmr(struct ib_fmr *fmr);
2228 
2229 /**
2230  * ib_attach_mcast - Attaches the specified QP to a multicast group.
2231  * @qp: QP to attach to the multicast group.  The QP must be type
2232  *   IB_QPT_UD.
2233  * @gid: Multicast group GID.
2234  * @lid: Multicast group LID in host byte order.
2235  *
2236  * In order to send and receive multicast packets, subnet
2237  * administration must have created the multicast group and configured
2238  * the fabric appropriately.  The port associated with the specified
2239  * QP must also be a member of the multicast group.
2240  */
2241 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2242 
2243 /**
2244  * ib_detach_mcast - Detaches the specified QP from a multicast group.
2245  * @qp: QP to detach from the multicast group.
2246  * @gid: Multicast group GID.
2247  * @lid: Multicast group LID in host byte order.
2248  */
2249 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2250 
2251 /**
2252  * ib_alloc_xrcd - Allocates an XRC domain.
2253  * @device: The device on which to allocate the XRC domain.
2254  */
2255 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2256 
2257 /**
2258  * ib_dealloc_xrcd - Deallocates an XRC domain.
2259  * @xrcd: The XRC domain to deallocate.
2260  */
2261 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2262 
2263 #endif /* IB_VERBS_H */
2264