xref: /freebsd/sys/ofed/include/rdma/ib_verbs.h (revision 7d99ab9fd0cc2c1ce2ecef0ed6d0672c2a50b0cb)
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 
51 #include <asm/atomic.h>
52 #include <asm/uaccess.h>
53 #include <linux/rbtree.h>
54 #include <linux/mutex.h>
55 
56 union ib_gid {
57 	u8	raw[16];
58 	struct {
59 		__be64	subnet_prefix;
60 		__be64	interface_id;
61 	} global;
62 };
63 
64 enum rdma_node_type {
65 	/* IB values map to NodeInfo:NodeType. */
66 	RDMA_NODE_IB_CA 	= 1,
67 	RDMA_NODE_IB_SWITCH,
68 	RDMA_NODE_IB_ROUTER,
69 	RDMA_NODE_RNIC
70 };
71 
72 enum rdma_transport_type {
73 	RDMA_TRANSPORT_IB,
74 	RDMA_TRANSPORT_IWARP
75 };
76 
77 enum rdma_transport_type
78 rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__;
79 
80 enum rdma_link_layer {
81 	IB_LINK_LAYER_UNSPECIFIED,
82 	IB_LINK_LAYER_INFINIBAND,
83 	IB_LINK_LAYER_ETHERNET,
84 };
85 
86 enum ib_device_cap_flags {
87 	IB_DEVICE_RESIZE_MAX_WR		= 1,
88 	IB_DEVICE_BAD_PKEY_CNTR		= (1<<1),
89 	IB_DEVICE_BAD_QKEY_CNTR		= (1<<2),
90 	IB_DEVICE_RAW_MULTI		= (1<<3),
91 	IB_DEVICE_AUTO_PATH_MIG		= (1<<4),
92 	IB_DEVICE_CHANGE_PHY_PORT	= (1<<5),
93 	IB_DEVICE_UD_AV_PORT_ENFORCE	= (1<<6),
94 	IB_DEVICE_CURR_QP_STATE_MOD	= (1<<7),
95 	IB_DEVICE_SHUTDOWN_PORT		= (1<<8),
96 	IB_DEVICE_INIT_TYPE		= (1<<9),
97 	IB_DEVICE_PORT_ACTIVE_EVENT	= (1<<10),
98 	IB_DEVICE_SYS_IMAGE_GUID	= (1<<11),
99 	IB_DEVICE_RC_RNR_NAK_GEN	= (1<<12),
100 	IB_DEVICE_SRQ_RESIZE		= (1<<13),
101 	IB_DEVICE_N_NOTIFY_CQ		= (1<<14),
102 	IB_DEVICE_LOCAL_DMA_LKEY	= (1<<15),
103 	IB_DEVICE_RESERVED		= (1<<16), /* old SEND_W_INV */
104 	IB_DEVICE_MEM_WINDOW		= (1<<17),
105 	/*
106 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
107 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
108 	 * messages and can verify the validity of checksum for
109 	 * incoming messages.  Setting this flag implies that the
110 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
111 	 */
112 	IB_DEVICE_UD_IP_CSUM		= (1<<18),
113 	IB_DEVICE_UD_TSO		= (1<<19),
114 	IB_DEVICE_XRC			= (1<<20),
115 	IB_DEVICE_MEM_MGT_EXTENSIONS	= (1<<21),
116 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
117 };
118 
119 enum ib_atomic_cap {
120 	IB_ATOMIC_NONE,
121 	IB_ATOMIC_HCA,
122 	IB_ATOMIC_GLOB
123 };
124 
125 struct ib_device_attr {
126 	u64			fw_ver;
127 	__be64			sys_image_guid;
128 	u64			max_mr_size;
129 	u64			page_size_cap;
130 	u32			vendor_id;
131 	u32			vendor_part_id;
132 	u32			hw_ver;
133 	int			max_qp;
134 	int			max_qp_wr;
135 	int			device_cap_flags;
136 	int			max_sge;
137 	int			max_sge_rd;
138 	int			max_cq;
139 	int			max_cqe;
140 	int			max_mr;
141 	int			max_pd;
142 	int			max_qp_rd_atom;
143 	int			max_ee_rd_atom;
144 	int			max_res_rd_atom;
145 	int			max_qp_init_rd_atom;
146 	int			max_ee_init_rd_atom;
147 	enum ib_atomic_cap	atomic_cap;
148 	enum ib_atomic_cap	masked_atomic_cap;
149 	int			max_ee;
150 	int			max_rdd;
151 	int			max_mw;
152 	int			max_raw_ipv6_qp;
153 	int			max_raw_ethy_qp;
154 	int			max_mcast_grp;
155 	int			max_mcast_qp_attach;
156 	int			max_total_mcast_qp_attach;
157 	int			max_ah;
158 	int			max_fmr;
159 	int			max_map_per_fmr;
160 	int			max_srq;
161 	int			max_srq_wr;
162 	int			max_srq_sge;
163 	unsigned int		max_fast_reg_page_list_len;
164 	u16			max_pkeys;
165 	u8			local_ca_ack_delay;
166 };
167 
168 enum ib_mtu {
169 	IB_MTU_256  = 1,
170 	IB_MTU_512  = 2,
171 	IB_MTU_1024 = 3,
172 	IB_MTU_2048 = 4,
173 	IB_MTU_4096 = 5
174 };
175 
176 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
177 {
178 	switch (mtu) {
179 	case IB_MTU_256:  return  256;
180 	case IB_MTU_512:  return  512;
181 	case IB_MTU_1024: return 1024;
182 	case IB_MTU_2048: return 2048;
183 	case IB_MTU_4096: return 4096;
184 	default: 	  return -1;
185 	}
186 }
187 
188 enum ib_port_state {
189 	IB_PORT_NOP		= 0,
190 	IB_PORT_DOWN		= 1,
191 	IB_PORT_INIT		= 2,
192 	IB_PORT_ARMED		= 3,
193 	IB_PORT_ACTIVE		= 4,
194 	IB_PORT_ACTIVE_DEFER	= 5
195 };
196 
197 enum ib_port_cap_flags {
198 	IB_PORT_SM				= 1 <<  1,
199 	IB_PORT_NOTICE_SUP			= 1 <<  2,
200 	IB_PORT_TRAP_SUP			= 1 <<  3,
201 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
202 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
203 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
204 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
205 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
206 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
207 	IB_PORT_SM_DISABLED			= 1 << 10,
208 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
209 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
210 	IB_PORT_CM_SUP				= 1 << 16,
211 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
212 	IB_PORT_REINIT_SUP			= 1 << 18,
213 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
214 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
215 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
216 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
217 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
218 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
219 	IB_PORT_CLIENT_REG_SUP			= 1 << 25
220 };
221 
222 enum ib_port_width {
223 	IB_WIDTH_1X	= 1,
224 	IB_WIDTH_4X	= 2,
225 	IB_WIDTH_8X	= 4,
226 	IB_WIDTH_12X	= 8
227 };
228 
229 static inline int ib_width_enum_to_int(enum ib_port_width width)
230 {
231 	switch (width) {
232 	case IB_WIDTH_1X:  return  1;
233 	case IB_WIDTH_4X:  return  4;
234 	case IB_WIDTH_8X:  return  8;
235 	case IB_WIDTH_12X: return 12;
236 	default: 	  return -1;
237 	}
238 }
239 
240 struct ib_protocol_stats {
241 	/* TBD... */
242 };
243 
244 struct iw_protocol_stats {
245 	u64	ipInReceives;
246 	u64	ipInHdrErrors;
247 	u64	ipInTooBigErrors;
248 	u64	ipInNoRoutes;
249 	u64	ipInAddrErrors;
250 	u64	ipInUnknownProtos;
251 	u64	ipInTruncatedPkts;
252 	u64	ipInDiscards;
253 	u64	ipInDelivers;
254 	u64	ipOutForwDatagrams;
255 	u64	ipOutRequests;
256 	u64	ipOutDiscards;
257 	u64	ipOutNoRoutes;
258 	u64	ipReasmTimeout;
259 	u64	ipReasmReqds;
260 	u64	ipReasmOKs;
261 	u64	ipReasmFails;
262 	u64	ipFragOKs;
263 	u64	ipFragFails;
264 	u64	ipFragCreates;
265 	u64	ipInMcastPkts;
266 	u64	ipOutMcastPkts;
267 	u64	ipInBcastPkts;
268 	u64	ipOutBcastPkts;
269 
270 	u64	tcpRtoAlgorithm;
271 	u64	tcpRtoMin;
272 	u64	tcpRtoMax;
273 	u64	tcpMaxConn;
274 	u64	tcpActiveOpens;
275 	u64	tcpPassiveOpens;
276 	u64	tcpAttemptFails;
277 	u64	tcpEstabResets;
278 	u64	tcpCurrEstab;
279 	u64	tcpInSegs;
280 	u64	tcpOutSegs;
281 	u64	tcpRetransSegs;
282 	u64	tcpInErrs;
283 	u64	tcpOutRsts;
284 };
285 
286 union rdma_protocol_stats {
287 	struct ib_protocol_stats	ib;
288 	struct iw_protocol_stats	iw;
289 };
290 
291 struct ib_port_attr {
292 	enum ib_port_state	state;
293 	enum ib_mtu		max_mtu;
294 	enum ib_mtu		active_mtu;
295 	int			gid_tbl_len;
296 	u32			port_cap_flags;
297 	u32			max_msg_sz;
298 	u32			bad_pkey_cntr;
299 	u32			qkey_viol_cntr;
300 	u16			pkey_tbl_len;
301 	u16			lid;
302 	u16			sm_lid;
303 	u8			lmc;
304 	u8			max_vl_num;
305 	u8			sm_sl;
306 	u8			subnet_timeout;
307 	u8			init_type_reply;
308 	u8			active_width;
309 	u8			active_speed;
310 	u8                      phys_state;
311 	enum rdma_link_layer	link_layer;
312 };
313 
314 enum ib_device_modify_flags {
315 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
316 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
317 };
318 
319 struct ib_device_modify {
320 	u64	sys_image_guid;
321 	char	node_desc[64];
322 };
323 
324 enum ib_port_modify_flags {
325 	IB_PORT_SHUTDOWN		= 1,
326 	IB_PORT_INIT_TYPE		= (1<<2),
327 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
328 };
329 
330 struct ib_port_modify {
331 	u32	set_port_cap_mask;
332 	u32	clr_port_cap_mask;
333 	u8	init_type;
334 };
335 
336 enum ib_event_type {
337 	IB_EVENT_CQ_ERR,
338 	IB_EVENT_QP_FATAL,
339 	IB_EVENT_QP_REQ_ERR,
340 	IB_EVENT_QP_ACCESS_ERR,
341 	IB_EVENT_COMM_EST,
342 	IB_EVENT_SQ_DRAINED,
343 	IB_EVENT_PATH_MIG,
344 	IB_EVENT_PATH_MIG_ERR,
345 	IB_EVENT_DEVICE_FATAL,
346 	IB_EVENT_PORT_ACTIVE,
347 	IB_EVENT_PORT_ERR,
348 	IB_EVENT_LID_CHANGE,
349 	IB_EVENT_PKEY_CHANGE,
350 	IB_EVENT_SM_CHANGE,
351 	IB_EVENT_SRQ_ERR,
352 	IB_EVENT_SRQ_LIMIT_REACHED,
353 	IB_EVENT_QP_LAST_WQE_REACHED,
354 	IB_EVENT_CLIENT_REREGISTER,
355 	IB_EVENT_GID_CHANGE,
356 };
357 
358 enum ib_event_flags {
359 	IB_XRC_QP_EVENT_FLAG = 0x80000000,
360 };
361 
362 struct ib_event {
363 	struct ib_device	*device;
364 	union {
365 		struct ib_cq	*cq;
366 		struct ib_qp	*qp;
367 		struct ib_srq	*srq;
368 		u8		port_num;
369 		u32		xrc_qp_num;
370 	} element;
371 	enum ib_event_type	event;
372 };
373 
374 struct ib_event_handler {
375 	struct ib_device *device;
376 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
377 	struct list_head  list;
378 };
379 
380 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
381 	do {							\
382 		(_ptr)->device  = _device;			\
383 		(_ptr)->handler = _handler;			\
384 		INIT_LIST_HEAD(&(_ptr)->list);			\
385 	} while (0)
386 
387 struct ib_global_route {
388 	union ib_gid	dgid;
389 	u32		flow_label;
390 	u8		sgid_index;
391 	u8		hop_limit;
392 	u8		traffic_class;
393 };
394 
395 struct ib_grh {
396 	__be32		version_tclass_flow;
397 	__be16		paylen;
398 	u8		next_hdr;
399 	u8		hop_limit;
400 	union ib_gid	sgid;
401 	union ib_gid	dgid;
402 };
403 
404 enum {
405 	IB_MULTICAST_QPN = 0xffffff
406 };
407 
408 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
409 
410 enum ib_ah_flags {
411 	IB_AH_GRH	= 1
412 };
413 
414 enum ib_rate {
415 	IB_RATE_PORT_CURRENT = 0,
416 	IB_RATE_2_5_GBPS = 2,
417 	IB_RATE_5_GBPS   = 5,
418 	IB_RATE_10_GBPS  = 3,
419 	IB_RATE_20_GBPS  = 6,
420 	IB_RATE_30_GBPS  = 4,
421 	IB_RATE_40_GBPS  = 7,
422 	IB_RATE_60_GBPS  = 8,
423 	IB_RATE_80_GBPS  = 9,
424 	IB_RATE_120_GBPS = 10
425 };
426 
427 /**
428  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
429  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
430  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
431  * @rate: rate to convert.
432  */
433 int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
434 
435 /**
436  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
437  * enum.
438  * @mult: multiple to convert.
439  */
440 enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
441 
442 struct ib_ah_attr {
443 	struct ib_global_route	grh;
444 	u16			dlid;
445 	u8			sl;
446 	u8			src_path_bits;
447 	u8			static_rate;
448 	u8			ah_flags;
449 	u8			port_num;
450 };
451 
452 enum ib_wc_status {
453 	IB_WC_SUCCESS,
454 	IB_WC_LOC_LEN_ERR,
455 	IB_WC_LOC_QP_OP_ERR,
456 	IB_WC_LOC_EEC_OP_ERR,
457 	IB_WC_LOC_PROT_ERR,
458 	IB_WC_WR_FLUSH_ERR,
459 	IB_WC_MW_BIND_ERR,
460 	IB_WC_BAD_RESP_ERR,
461 	IB_WC_LOC_ACCESS_ERR,
462 	IB_WC_REM_INV_REQ_ERR,
463 	IB_WC_REM_ACCESS_ERR,
464 	IB_WC_REM_OP_ERR,
465 	IB_WC_RETRY_EXC_ERR,
466 	IB_WC_RNR_RETRY_EXC_ERR,
467 	IB_WC_LOC_RDD_VIOL_ERR,
468 	IB_WC_REM_INV_RD_REQ_ERR,
469 	IB_WC_REM_ABORT_ERR,
470 	IB_WC_INV_EECN_ERR,
471 	IB_WC_INV_EEC_STATE_ERR,
472 	IB_WC_FATAL_ERR,
473 	IB_WC_RESP_TIMEOUT_ERR,
474 	IB_WC_GENERAL_ERR
475 };
476 
477 enum ib_wc_opcode {
478 	IB_WC_SEND,
479 	IB_WC_RDMA_WRITE,
480 	IB_WC_RDMA_READ,
481 	IB_WC_COMP_SWAP,
482 	IB_WC_FETCH_ADD,
483 	IB_WC_BIND_MW,
484 	IB_WC_LSO,
485 	IB_WC_LOCAL_INV,
486 	IB_WC_FAST_REG_MR,
487 	IB_WC_MASKED_COMP_SWAP,
488 	IB_WC_MASKED_FETCH_ADD,
489 /*
490  * Set value of IB_WC_RECV so consumers can test if a completion is a
491  * receive by testing (opcode & IB_WC_RECV).
492  */
493 	IB_WC_RECV			= 1 << 7,
494 	IB_WC_RECV_RDMA_WITH_IMM
495 };
496 
497 enum ib_wc_flags {
498 	IB_WC_GRH		= 1,
499 	IB_WC_WITH_IMM		= (1<<1),
500 	IB_WC_WITH_INVALIDATE	= (1<<2),
501 };
502 
503 struct ib_wc {
504 	u64			wr_id;
505 	enum ib_wc_status	status;
506 	enum ib_wc_opcode	opcode;
507 	u32			vendor_err;
508 	u32			byte_len;
509 	struct ib_qp	       *qp;
510 	union {
511 		__be32		imm_data;
512 		u32		invalidate_rkey;
513 	} ex;
514 	u32			src_qp;
515 	int			wc_flags;
516 	u16			pkey_index;
517 	u16			slid;
518 	u8			sl;
519 	u8			dlid_path_bits;
520 	u8			port_num;	/* valid only for DR SMPs on switches */
521 	int			csum_ok;
522 };
523 
524 enum ib_cq_notify_flags {
525 	IB_CQ_SOLICITED			= 1 << 0,
526 	IB_CQ_NEXT_COMP			= 1 << 1,
527 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
528 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
529 };
530 
531 enum ib_srq_attr_mask {
532 	IB_SRQ_MAX_WR	= 1 << 0,
533 	IB_SRQ_LIMIT	= 1 << 1,
534 };
535 
536 struct ib_srq_attr {
537 	u32	max_wr;
538 	u32	max_sge;
539 	u32	srq_limit;
540 };
541 
542 struct ib_srq_init_attr {
543 	void		      (*event_handler)(struct ib_event *, void *);
544 	void		       *srq_context;
545 	struct ib_srq_attr	attr;
546 };
547 
548 struct ib_qp_cap {
549 	u32	max_send_wr;
550 	u32	max_recv_wr;
551 	u32	max_send_sge;
552 	u32	max_recv_sge;
553 	u32	max_inline_data;
554 };
555 
556 enum ib_sig_type {
557 	IB_SIGNAL_ALL_WR,
558 	IB_SIGNAL_REQ_WR
559 };
560 
561 enum ib_qp_type {
562 	/*
563 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
564 	 * here (and in that order) since the MAD layer uses them as
565 	 * indices into a 2-entry table.
566 	 */
567 	IB_QPT_SMI,
568 	IB_QPT_GSI,
569 
570 	IB_QPT_RC,
571 	IB_QPT_UC,
572 	IB_QPT_UD,
573 	IB_QPT_XRC,
574 	IB_QPT_RAW_IPV6,
575 	IB_QPT_RAW_ETY,
576 	IB_QPT_RAW_ETH
577 };
578 
579 enum ib_qp_create_flags {
580 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
581 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
582 };
583 
584 struct ib_qp_init_attr {
585 	void                  (*event_handler)(struct ib_event *, void *);
586 	void		       *qp_context;
587 	struct ib_cq	       *send_cq;
588 	struct ib_cq	       *recv_cq;
589 	struct ib_srq	       *srq;
590 	struct ib_qp_cap	cap;
591 	enum ib_sig_type	sq_sig_type;
592 	enum ib_qp_type		qp_type;
593 	enum ib_qp_create_flags	create_flags;
594 	struct ib_xrcd	       *xrc_domain; /* XRC qp's only */
595 	u8			port_num; /* special QP types only */
596 };
597 
598 enum ib_rnr_timeout {
599 	IB_RNR_TIMER_655_36 =  0,
600 	IB_RNR_TIMER_000_01 =  1,
601 	IB_RNR_TIMER_000_02 =  2,
602 	IB_RNR_TIMER_000_03 =  3,
603 	IB_RNR_TIMER_000_04 =  4,
604 	IB_RNR_TIMER_000_06 =  5,
605 	IB_RNR_TIMER_000_08 =  6,
606 	IB_RNR_TIMER_000_12 =  7,
607 	IB_RNR_TIMER_000_16 =  8,
608 	IB_RNR_TIMER_000_24 =  9,
609 	IB_RNR_TIMER_000_32 = 10,
610 	IB_RNR_TIMER_000_48 = 11,
611 	IB_RNR_TIMER_000_64 = 12,
612 	IB_RNR_TIMER_000_96 = 13,
613 	IB_RNR_TIMER_001_28 = 14,
614 	IB_RNR_TIMER_001_92 = 15,
615 	IB_RNR_TIMER_002_56 = 16,
616 	IB_RNR_TIMER_003_84 = 17,
617 	IB_RNR_TIMER_005_12 = 18,
618 	IB_RNR_TIMER_007_68 = 19,
619 	IB_RNR_TIMER_010_24 = 20,
620 	IB_RNR_TIMER_015_36 = 21,
621 	IB_RNR_TIMER_020_48 = 22,
622 	IB_RNR_TIMER_030_72 = 23,
623 	IB_RNR_TIMER_040_96 = 24,
624 	IB_RNR_TIMER_061_44 = 25,
625 	IB_RNR_TIMER_081_92 = 26,
626 	IB_RNR_TIMER_122_88 = 27,
627 	IB_RNR_TIMER_163_84 = 28,
628 	IB_RNR_TIMER_245_76 = 29,
629 	IB_RNR_TIMER_327_68 = 30,
630 	IB_RNR_TIMER_491_52 = 31
631 };
632 
633 enum ib_qp_attr_mask {
634 	IB_QP_STATE			= 1,
635 	IB_QP_CUR_STATE			= (1<<1),
636 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
637 	IB_QP_ACCESS_FLAGS		= (1<<3),
638 	IB_QP_PKEY_INDEX		= (1<<4),
639 	IB_QP_PORT			= (1<<5),
640 	IB_QP_QKEY			= (1<<6),
641 	IB_QP_AV			= (1<<7),
642 	IB_QP_PATH_MTU			= (1<<8),
643 	IB_QP_TIMEOUT			= (1<<9),
644 	IB_QP_RETRY_CNT			= (1<<10),
645 	IB_QP_RNR_RETRY			= (1<<11),
646 	IB_QP_RQ_PSN			= (1<<12),
647 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
648 	IB_QP_ALT_PATH			= (1<<14),
649 	IB_QP_MIN_RNR_TIMER		= (1<<15),
650 	IB_QP_SQ_PSN			= (1<<16),
651 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
652 	IB_QP_PATH_MIG_STATE		= (1<<18),
653 	IB_QP_CAP			= (1<<19),
654 	IB_QP_DEST_QPN			= (1<<20)
655 };
656 
657 enum ib_qp_state {
658 	IB_QPS_RESET,
659 	IB_QPS_INIT,
660 	IB_QPS_RTR,
661 	IB_QPS_RTS,
662 	IB_QPS_SQD,
663 	IB_QPS_SQE,
664 	IB_QPS_ERR
665 };
666 
667 enum ib_mig_state {
668 	IB_MIG_MIGRATED,
669 	IB_MIG_REARM,
670 	IB_MIG_ARMED
671 };
672 
673 struct ib_qp_attr {
674 	enum ib_qp_state	qp_state;
675 	enum ib_qp_state	cur_qp_state;
676 	enum ib_mtu		path_mtu;
677 	enum ib_mig_state	path_mig_state;
678 	u32			qkey;
679 	u32			rq_psn;
680 	u32			sq_psn;
681 	u32			dest_qp_num;
682 	int			qp_access_flags;
683 	struct ib_qp_cap	cap;
684 	struct ib_ah_attr	ah_attr;
685 	struct ib_ah_attr	alt_ah_attr;
686 	u16			pkey_index;
687 	u16			alt_pkey_index;
688 	u8			en_sqd_async_notify;
689 	u8			sq_draining;
690 	u8			max_rd_atomic;
691 	u8			max_dest_rd_atomic;
692 	u8			min_rnr_timer;
693 	u8			port_num;
694 	u8			timeout;
695 	u8			retry_cnt;
696 	u8			rnr_retry;
697 	u8			alt_port_num;
698 	u8			alt_timeout;
699 };
700 
701 enum ib_wr_opcode {
702 	IB_WR_RDMA_WRITE,
703 	IB_WR_RDMA_WRITE_WITH_IMM,
704 	IB_WR_SEND,
705 	IB_WR_SEND_WITH_IMM,
706 	IB_WR_RDMA_READ,
707 	IB_WR_ATOMIC_CMP_AND_SWP,
708 	IB_WR_ATOMIC_FETCH_AND_ADD,
709 	IB_WR_LSO,
710 	IB_WR_BIG_LSO,
711 	IB_WR_SEND_WITH_INV,
712 	IB_WR_RDMA_READ_WITH_INV,
713 	IB_WR_LOCAL_INV,
714 	IB_WR_FAST_REG_MR,
715 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
716 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
717 };
718 
719 enum ib_send_flags {
720 	IB_SEND_FENCE		= 1,
721 	IB_SEND_SIGNALED	= (1<<1),
722 	IB_SEND_SOLICITED	= (1<<2),
723 	IB_SEND_INLINE		= (1<<3),
724 	IB_SEND_IP_CSUM		= (1<<4)
725 };
726 
727 struct ib_sge {
728 	u64	addr;
729 	u32	length;
730 	u32	lkey;
731 };
732 
733 struct ib_fast_reg_page_list {
734 	struct ib_device       *device;
735 	u64		       *page_list;
736 	unsigned int		max_page_list_len;
737 };
738 
739 struct ib_send_wr {
740 	struct ib_send_wr      *next;
741 	u64			wr_id;
742 	struct ib_sge	       *sg_list;
743 	int			num_sge;
744 	enum ib_wr_opcode	opcode;
745 	int			send_flags;
746 	union {
747 		__be32		imm_data;
748 		u32		invalidate_rkey;
749 	} ex;
750 	union {
751 		struct {
752 			u64	remote_addr;
753 			u32	rkey;
754 		} rdma;
755 		struct {
756 			u64	remote_addr;
757 			u64	compare_add;
758 			u64	swap;
759 			u64	compare_add_mask;
760 			u64	swap_mask;
761 			u32	rkey;
762 		} atomic;
763 		struct {
764 			struct ib_ah *ah;
765 			void   *header;
766 			int     hlen;
767 			int     mss;
768 			u32	remote_qpn;
769 			u32	remote_qkey;
770 			u16	pkey_index; /* valid for GSI only */
771 			u8	port_num;   /* valid for DR SMPs on switch only */
772 		} ud;
773 		struct {
774 			u64				iova_start;
775 			struct ib_fast_reg_page_list   *page_list;
776 			unsigned int			page_shift;
777 			unsigned int			page_list_len;
778 			u32				length;
779 			int				access_flags;
780 			u32				rkey;
781 		} fast_reg;
782 		struct {
783 			struct ib_unpacked_lrh	*lrh;
784 			u32			eth_type;
785 			u8			static_rate;
786 		} raw_ety;
787 	} wr;
788 	u32			xrc_remote_srq_num; /* valid for XRC sends only */
789 };
790 
791 struct ib_recv_wr {
792 	struct ib_recv_wr      *next;
793 	u64			wr_id;
794 	struct ib_sge	       *sg_list;
795 	int			num_sge;
796 };
797 
798 enum ib_access_flags {
799 	IB_ACCESS_LOCAL_WRITE	= 1,
800 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
801 	IB_ACCESS_REMOTE_READ	= (1<<2),
802 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
803 	IB_ACCESS_MW_BIND	= (1<<4)
804 };
805 
806 struct ib_phys_buf {
807 	u64      addr;
808 	u64      size;
809 };
810 
811 struct ib_mr_attr {
812 	struct ib_pd	*pd;
813 	u64		device_virt_addr;
814 	u64		size;
815 	int		mr_access_flags;
816 	u32		lkey;
817 	u32		rkey;
818 };
819 
820 enum ib_mr_rereg_flags {
821 	IB_MR_REREG_TRANS	= 1,
822 	IB_MR_REREG_PD		= (1<<1),
823 	IB_MR_REREG_ACCESS	= (1<<2)
824 };
825 
826 struct ib_mw_bind {
827 	struct ib_mr   *mr;
828 	u64		wr_id;
829 	u64		addr;
830 	u32		length;
831 	int		send_flags;
832 	int		mw_access_flags;
833 };
834 
835 struct ib_fmr_attr {
836 	int	max_pages;
837 	int	max_maps;
838 	u8	page_shift;
839 };
840 
841 struct ib_ucontext {
842 	struct ib_device       *device;
843 	struct list_head	pd_list;
844 	struct list_head	mr_list;
845 	struct list_head	mw_list;
846 	struct list_head	cq_list;
847 	struct list_head	qp_list;
848 	struct list_head	srq_list;
849 	struct list_head	ah_list;
850 	struct list_head	xrc_domain_list;
851 	int			closing;
852 };
853 
854 struct ib_uobject {
855 	u64			user_handle;	/* handle given to us by userspace */
856 	struct ib_ucontext     *context;	/* associated user context */
857 	void		       *object;		/* containing object */
858 	struct list_head	list;		/* link to context's list */
859 	int			id;		/* index into kernel idr */
860 	struct kref		ref;
861 	struct rw_semaphore	mutex;		/* protects .live */
862 	int			live;
863 };
864 
865 struct ib_udata {
866 	void __user *inbuf;
867 	void __user *outbuf;
868 	size_t       inlen;
869 	size_t       outlen;
870 };
871 
872 struct ib_uxrc_rcv_object {
873 	struct list_head	list;		/* link to context's list */
874 	u32			qp_num;
875 	u32			domain_handle;
876 };
877 
878 struct ib_pd {
879 	struct ib_device       *device;
880 	struct ib_uobject      *uobject;
881 	atomic_t          	usecnt; /* count all resources */
882 };
883 
884 struct ib_xrcd {
885 	struct ib_device       *device;
886 	struct ib_uobject      *uobject;
887 	struct inode	       *inode;
888 	struct rb_node		node;
889 	atomic_t		usecnt; /* count all resources */
890 };
891 
892 
893 struct ib_ah {
894 	struct ib_device	*device;
895 	struct ib_pd		*pd;
896 	struct ib_uobject	*uobject;
897 };
898 
899 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
900 
901 struct ib_cq {
902 	struct ib_device       *device;
903 	struct ib_uobject      *uobject;
904 	ib_comp_handler   	comp_handler;
905 	void                  (*event_handler)(struct ib_event *, void *);
906 	void                   *cq_context;
907 	int               	cqe;
908 	atomic_t          	usecnt; /* count number of work queues */
909 };
910 
911 struct ib_srq {
912 	struct ib_device       *device;
913 	struct ib_pd	       *pd;
914 	struct ib_cq	       *xrc_cq;
915 	struct ib_xrcd	       *xrcd;
916 	struct ib_uobject      *uobject;
917 	void		      (*event_handler)(struct ib_event *, void *);
918 	void		       *srq_context;
919 	atomic_t		usecnt;
920 	u32			xrc_srq_num;
921 };
922 
923 struct ib_qp {
924 	struct ib_device       *device;
925 	struct ib_pd	       *pd;
926 	struct ib_cq	       *send_cq;
927 	struct ib_cq	       *recv_cq;
928 	struct ib_srq	       *srq;
929 	struct ib_uobject      *uobject;
930 	void                  (*event_handler)(struct ib_event *, void *);
931 	void		       *qp_context;
932 	u32			qp_num;
933 	enum ib_qp_type		qp_type;
934 	struct ib_xrcd	       *xrcd;  /* XRC QPs only */
935 };
936 
937 struct ib_mr {
938 	struct ib_device  *device;
939 	struct ib_pd	  *pd;
940 	struct ib_uobject *uobject;
941 	u32		   lkey;
942 	u32		   rkey;
943 	atomic_t	   usecnt; /* count number of MWs */
944 };
945 
946 struct ib_mw {
947 	struct ib_device	*device;
948 	struct ib_pd		*pd;
949 	struct ib_uobject	*uobject;
950 	u32			rkey;
951 };
952 
953 struct ib_fmr {
954 	struct ib_device	*device;
955 	struct ib_pd		*pd;
956 	struct list_head	list;
957 	u32			lkey;
958 	u32			rkey;
959 };
960 
961 struct ib_mad;
962 struct ib_grh;
963 
964 enum ib_process_mad_flags {
965 	IB_MAD_IGNORE_MKEY	= 1,
966 	IB_MAD_IGNORE_BKEY	= 2,
967 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
968 };
969 
970 enum ib_mad_result {
971 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
972 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
973 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
974 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
975 };
976 
977 #define IB_DEVICE_NAME_MAX 64
978 
979 struct ib_cache {
980 	rwlock_t                lock;
981 	struct ib_event_handler event_handler;
982 	struct ib_pkey_cache  **pkey_cache;
983 	struct ib_gid_cache   **gid_cache;
984 	u8                     *lmc_cache;
985 };
986 
987 struct ib_dma_mapping_ops {
988 	int		(*mapping_error)(struct ib_device *dev,
989 					 u64 dma_addr);
990 	u64		(*map_single)(struct ib_device *dev,
991 				      void *ptr, size_t size,
992 				      enum dma_data_direction direction);
993 	void		(*unmap_single)(struct ib_device *dev,
994 					u64 addr, size_t size,
995 					enum dma_data_direction direction);
996 	u64		(*map_page)(struct ib_device *dev,
997 				    struct page *page, unsigned long offset,
998 				    size_t size,
999 				    enum dma_data_direction direction);
1000 	void		(*unmap_page)(struct ib_device *dev,
1001 				      u64 addr, size_t size,
1002 				      enum dma_data_direction direction);
1003 	int		(*map_sg)(struct ib_device *dev,
1004 				  struct scatterlist *sg, int nents,
1005 				  enum dma_data_direction direction);
1006 	void		(*unmap_sg)(struct ib_device *dev,
1007 				    struct scatterlist *sg, int nents,
1008 				    enum dma_data_direction direction);
1009 	u64		(*dma_address)(struct ib_device *dev,
1010 				       struct scatterlist *sg);
1011 	unsigned int	(*dma_len)(struct ib_device *dev,
1012 				   struct scatterlist *sg);
1013 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1014 					       u64 dma_handle,
1015 					       size_t size,
1016 					       enum dma_data_direction dir);
1017 	void		(*sync_single_for_device)(struct ib_device *dev,
1018 						  u64 dma_handle,
1019 						  size_t size,
1020 						  enum dma_data_direction dir);
1021 	void		*(*alloc_coherent)(struct ib_device *dev,
1022 					   size_t size,
1023 					   u64 *dma_handle,
1024 					   gfp_t flag);
1025 	void		(*free_coherent)(struct ib_device *dev,
1026 					 size_t size, void *cpu_addr,
1027 					 u64 dma_handle);
1028 };
1029 
1030 struct iw_cm_verbs;
1031 
1032 struct ib_device {
1033 	struct device                *dma_device;
1034 
1035 	char                          name[IB_DEVICE_NAME_MAX];
1036 
1037 	struct list_head              event_handler_list;
1038 	spinlock_t                    event_handler_lock;
1039 
1040 	struct list_head              core_list;
1041 	struct list_head              client_data_list;
1042 	spinlock_t                    client_data_lock;
1043 
1044 	struct ib_cache               cache;
1045 	int                          *pkey_tbl_len;
1046 	int                          *gid_tbl_len;
1047 
1048 	int			      num_comp_vectors;
1049 
1050 	struct iw_cm_verbs	     *iwcm;
1051 
1052 	int		           (*get_protocol_stats)(struct ib_device *device,
1053 							 union rdma_protocol_stats *stats);
1054 	int		           (*query_device)(struct ib_device *device,
1055 						   struct ib_device_attr *device_attr);
1056 	int		           (*query_port)(struct ib_device *device,
1057 						 u8 port_num,
1058 						 struct ib_port_attr *port_attr);
1059 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1060 						     u8 port_num);
1061 	int		           (*query_gid)(struct ib_device *device,
1062 						u8 port_num, int index,
1063 						union ib_gid *gid);
1064 	int		           (*query_pkey)(struct ib_device *device,
1065 						 u8 port_num, u16 index, u16 *pkey);
1066 	int		           (*modify_device)(struct ib_device *device,
1067 						    int device_modify_mask,
1068 						    struct ib_device_modify *device_modify);
1069 	int		           (*modify_port)(struct ib_device *device,
1070 						  u8 port_num, int port_modify_mask,
1071 						  struct ib_port_modify *port_modify);
1072 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1073 						     struct ib_udata *udata);
1074 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1075 	int                        (*mmap)(struct ib_ucontext *context,
1076 					   struct vm_area_struct *vma);
1077 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1078 					       struct ib_ucontext *context,
1079 					       struct ib_udata *udata);
1080 	int                        (*dealloc_pd)(struct ib_pd *pd);
1081 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1082 						struct ib_ah_attr *ah_attr);
1083 	int                        (*modify_ah)(struct ib_ah *ah,
1084 						struct ib_ah_attr *ah_attr);
1085 	int                        (*query_ah)(struct ib_ah *ah,
1086 					       struct ib_ah_attr *ah_attr);
1087 	int                        (*destroy_ah)(struct ib_ah *ah);
1088 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1089 						 struct ib_srq_init_attr *srq_init_attr,
1090 						 struct ib_udata *udata);
1091 	int                        (*modify_srq)(struct ib_srq *srq,
1092 						 struct ib_srq_attr *srq_attr,
1093 						 enum ib_srq_attr_mask srq_attr_mask,
1094 						 struct ib_udata *udata);
1095 	int                        (*query_srq)(struct ib_srq *srq,
1096 						struct ib_srq_attr *srq_attr);
1097 	int                        (*destroy_srq)(struct ib_srq *srq);
1098 	int                        (*post_srq_recv)(struct ib_srq *srq,
1099 						    struct ib_recv_wr *recv_wr,
1100 						    struct ib_recv_wr **bad_recv_wr);
1101 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1102 						struct ib_qp_init_attr *qp_init_attr,
1103 						struct ib_udata *udata);
1104 	int                        (*modify_qp)(struct ib_qp *qp,
1105 						struct ib_qp_attr *qp_attr,
1106 						int qp_attr_mask,
1107 						struct ib_udata *udata);
1108 	int                        (*query_qp)(struct ib_qp *qp,
1109 					       struct ib_qp_attr *qp_attr,
1110 					       int qp_attr_mask,
1111 					       struct ib_qp_init_attr *qp_init_attr);
1112 	int                        (*destroy_qp)(struct ib_qp *qp);
1113 	int                        (*post_send)(struct ib_qp *qp,
1114 						struct ib_send_wr *send_wr,
1115 						struct ib_send_wr **bad_send_wr);
1116 	int                        (*post_recv)(struct ib_qp *qp,
1117 						struct ib_recv_wr *recv_wr,
1118 						struct ib_recv_wr **bad_recv_wr);
1119 	struct ib_cq *             (*create_cq)(struct ib_device *device, int cqe,
1120 						int comp_vector,
1121 						struct ib_ucontext *context,
1122 						struct ib_udata *udata);
1123 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1124 						u16 cq_period);
1125 	int                        (*destroy_cq)(struct ib_cq *cq);
1126 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1127 						struct ib_udata *udata);
1128 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1129 					      struct ib_wc *wc);
1130 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1131 	int                        (*req_notify_cq)(struct ib_cq *cq,
1132 						    enum ib_cq_notify_flags flags);
1133 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1134 						      int wc_cnt);
1135 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1136 						 int mr_access_flags);
1137 	struct ib_mr *             (*reg_phys_mr)(struct ib_pd *pd,
1138 						  struct ib_phys_buf *phys_buf_array,
1139 						  int num_phys_buf,
1140 						  int mr_access_flags,
1141 						  u64 *iova_start);
1142 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1143 						  u64 start, u64 length,
1144 						  u64 virt_addr,
1145 						  int mr_access_flags,
1146 						  struct ib_udata *udata);
1147 	int                        (*query_mr)(struct ib_mr *mr,
1148 					       struct ib_mr_attr *mr_attr);
1149 	int                        (*dereg_mr)(struct ib_mr *mr);
1150 	struct ib_mr *		   (*alloc_fast_reg_mr)(struct ib_pd *pd,
1151 					       int max_page_list_len);
1152 	struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1153 								   int page_list_len);
1154 	void			   (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1155 	int                        (*rereg_phys_mr)(struct ib_mr *mr,
1156 						    int mr_rereg_mask,
1157 						    struct ib_pd *pd,
1158 						    struct ib_phys_buf *phys_buf_array,
1159 						    int num_phys_buf,
1160 						    int mr_access_flags,
1161 						    u64 *iova_start);
1162 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd);
1163 	int                        (*bind_mw)(struct ib_qp *qp,
1164 					      struct ib_mw *mw,
1165 					      struct ib_mw_bind *mw_bind);
1166 	int                        (*dealloc_mw)(struct ib_mw *mw);
1167 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1168 						int mr_access_flags,
1169 						struct ib_fmr_attr *fmr_attr);
1170 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1171 						   u64 *page_list, int list_len,
1172 						   u64 iova);
1173 	int		           (*unmap_fmr)(struct list_head *fmr_list);
1174 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1175 	int                        (*attach_mcast)(struct ib_qp *qp,
1176 						   union ib_gid *gid,
1177 						   u16 lid);
1178 	int                        (*detach_mcast)(struct ib_qp *qp,
1179 						   union ib_gid *gid,
1180 						   u16 lid);
1181 	int                        (*process_mad)(struct ib_device *device,
1182 						  int process_mad_flags,
1183 						  u8 port_num,
1184 						  struct ib_wc *in_wc,
1185 						  struct ib_grh *in_grh,
1186 						  struct ib_mad *in_mad,
1187 						  struct ib_mad *out_mad);
1188 	struct ib_srq *		   (*create_xrc_srq)(struct ib_pd *pd,
1189 						     struct ib_cq *xrc_cq,
1190 						     struct ib_xrcd *xrcd,
1191 						     struct ib_srq_init_attr *srq_init_attr,
1192 						     struct ib_udata *udata);
1193 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
1194 						 struct ib_ucontext *context,
1195 						 struct ib_udata *udata);
1196 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1197 	int			   (*create_xrc_rcv_qp)(struct ib_qp_init_attr *init_attr,
1198 							u32 *qp_num);
1199 	int			   (*modify_xrc_rcv_qp)(struct ib_xrcd *xrcd,
1200 							u32 qp_num,
1201 							struct ib_qp_attr *attr,
1202 							int attr_mask);
1203 	int			   (*query_xrc_rcv_qp)(struct ib_xrcd *xrcd,
1204 						       u32 qp_num,
1205 						       struct ib_qp_attr *attr,
1206 						       int attr_mask,
1207 						       struct ib_qp_init_attr *init_attr);
1208 	int 			   (*reg_xrc_rcv_qp)(struct ib_xrcd *xrcd,
1209 						     void *context,
1210 						     u32 qp_num);
1211 	int 			   (*unreg_xrc_rcv_qp)(struct ib_xrcd *xrcd,
1212 						       void *context,
1213 						       u32 qp_num);
1214 
1215 	struct ib_dma_mapping_ops   *dma_ops;
1216 
1217 	struct module               *owner;
1218 	struct device                dev;
1219 	struct kobject               *ports_parent;
1220 	struct list_head             port_list;
1221 
1222 	enum {
1223 		IB_DEV_UNINITIALIZED,
1224 		IB_DEV_REGISTERED,
1225 		IB_DEV_UNREGISTERED
1226 	}                            reg_state;
1227 
1228 	u64			     uverbs_cmd_mask;
1229 	int			     uverbs_abi_ver;
1230 
1231 	char			     node_desc[64];
1232 	__be64			     node_guid;
1233 	u32			     local_dma_lkey;
1234 	u8                           node_type;
1235 	u8                           phys_port_cnt;
1236 	struct rb_root		     ib_uverbs_xrcd_table;
1237 	struct mutex		     xrcd_table_mutex;
1238 };
1239 
1240 struct ib_client {
1241 	char  *name;
1242 	void (*add)   (struct ib_device *);
1243 	void (*remove)(struct ib_device *);
1244 
1245 	struct list_head list;
1246 };
1247 
1248 struct ib_device *ib_alloc_device(size_t size);
1249 void ib_dealloc_device(struct ib_device *device);
1250 
1251 int ib_register_device   (struct ib_device *device);
1252 void ib_unregister_device(struct ib_device *device);
1253 
1254 int ib_register_client   (struct ib_client *client);
1255 void ib_unregister_client(struct ib_client *client);
1256 
1257 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1258 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
1259 			 void *data);
1260 
1261 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1262 {
1263 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1264 }
1265 
1266 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1267 {
1268 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1269 }
1270 
1271 /**
1272  * ib_sysfs_create_port_files - iterate over port sysfs directories
1273  * @device: the IB device
1274  * @create: a function to create sysfs files in each port directory
1275  */
1276 int ib_sysfs_create_port_files(struct ib_device *device,
1277 			       int (*create)(struct ib_device *dev, u8 port_num,
1278 					     struct kobject *kobj));
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_xrc_srq - Creates an XRC SRQ associated with the specified
1411  *   protection domain, cq, and xrc domain.
1412  * @pd: The protection domain associated with the SRQ.
1413  * @xrc_cq: The cq to be associated with the XRC SRQ.
1414  * @xrcd: The XRC domain to be associated with the XRC SRQ.
1415  * @srq_init_attr: A list of initial attributes required to create the
1416  *   XRC SRQ.  If XRC SRQ creation succeeds, then the attributes are updated
1417  *   to the actual capabilities of the created XRC SRQ.
1418  *
1419  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1420  * requested size of the XRC SRQ, and set to the actual values allocated
1421  * on return.  If ib_create_xrc_srq() succeeds, then max_wr and max_sge
1422  * will always be at least as large as the requested values.
1423  */
1424 struct ib_srq *ib_create_xrc_srq(struct ib_pd *pd,
1425 				 struct ib_cq *xrc_cq,
1426 				 struct ib_xrcd *xrcd,
1427 				 struct ib_srq_init_attr *srq_init_attr);
1428 
1429 /**
1430  * ib_create_srq - Creates an SRQ associated with the specified
1431  *   protection domain.
1432  * @pd: The protection domain associated with the SRQ.
1433  * @srq_init_attr: A list of initial attributes required to create the
1434  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
1435  *   the actual capabilities of the created SRQ.
1436  *
1437  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1438  * requested size of the SRQ, and set to the actual values allocated
1439  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
1440  * will always be at least as large as the requested values.
1441  */
1442 struct ib_srq *ib_create_srq(struct ib_pd *pd,
1443 			     struct ib_srq_init_attr *srq_init_attr);
1444 
1445 /**
1446  * ib_modify_srq - Modifies the attributes for the specified SRQ.
1447  * @srq: The SRQ to modify.
1448  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
1449  *   the current values of selected SRQ attributes are returned.
1450  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1451  *   are being modified.
1452  *
1453  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1454  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1455  * the number of receives queued drops below the limit.
1456  */
1457 int ib_modify_srq(struct ib_srq *srq,
1458 		  struct ib_srq_attr *srq_attr,
1459 		  enum ib_srq_attr_mask srq_attr_mask);
1460 
1461 /**
1462  * ib_query_srq - Returns the attribute list and current values for the
1463  *   specified SRQ.
1464  * @srq: The SRQ to query.
1465  * @srq_attr: The attributes of the specified SRQ.
1466  */
1467 int ib_query_srq(struct ib_srq *srq,
1468 		 struct ib_srq_attr *srq_attr);
1469 
1470 /**
1471  * ib_destroy_srq - Destroys the specified SRQ.
1472  * @srq: The SRQ to destroy.
1473  */
1474 int ib_destroy_srq(struct ib_srq *srq);
1475 
1476 /**
1477  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1478  * @srq: The SRQ to post the work request on.
1479  * @recv_wr: A list of work requests to post on the receive queue.
1480  * @bad_recv_wr: On an immediate failure, this parameter will reference
1481  *   the work request that failed to be posted on the QP.
1482  */
1483 static inline int ib_post_srq_recv(struct ib_srq *srq,
1484 				   struct ib_recv_wr *recv_wr,
1485 				   struct ib_recv_wr **bad_recv_wr)
1486 {
1487 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1488 }
1489 
1490 /**
1491  * ib_create_qp - Creates a QP associated with the specified protection
1492  *   domain.
1493  * @pd: The protection domain associated with the QP.
1494  * @qp_init_attr: A list of initial attributes required to create the
1495  *   QP.  If QP creation succeeds, then the attributes are updated to
1496  *   the actual capabilities of the created QP.
1497  */
1498 struct ib_qp *ib_create_qp(struct ib_pd *pd,
1499 			   struct ib_qp_init_attr *qp_init_attr);
1500 
1501 /**
1502  * ib_modify_qp - Modifies the attributes for the specified QP and then
1503  *   transitions the QP to the given state.
1504  * @qp: The QP to modify.
1505  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
1506  *   the current values of selected QP attributes are returned.
1507  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1508  *   are being modified.
1509  */
1510 int ib_modify_qp(struct ib_qp *qp,
1511 		 struct ib_qp_attr *qp_attr,
1512 		 int qp_attr_mask);
1513 
1514 /**
1515  * ib_query_qp - Returns the attribute list and current values for the
1516  *   specified QP.
1517  * @qp: The QP to query.
1518  * @qp_attr: The attributes of the specified QP.
1519  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1520  * @qp_init_attr: Additional attributes of the selected QP.
1521  *
1522  * The qp_attr_mask may be used to limit the query to gathering only the
1523  * selected attributes.
1524  */
1525 int ib_query_qp(struct ib_qp *qp,
1526 		struct ib_qp_attr *qp_attr,
1527 		int qp_attr_mask,
1528 		struct ib_qp_init_attr *qp_init_attr);
1529 
1530 /**
1531  * ib_destroy_qp - Destroys the specified QP.
1532  * @qp: The QP to destroy.
1533  */
1534 int ib_destroy_qp(struct ib_qp *qp);
1535 
1536 /**
1537  * ib_post_send - Posts a list of work requests to the send queue of
1538  *   the specified QP.
1539  * @qp: The QP to post the work request on.
1540  * @send_wr: A list of work requests to post on the send queue.
1541  * @bad_send_wr: On an immediate failure, this parameter will reference
1542  *   the work request that failed to be posted on the QP.
1543  */
1544 static inline int ib_post_send(struct ib_qp *qp,
1545 			       struct ib_send_wr *send_wr,
1546 			       struct ib_send_wr **bad_send_wr)
1547 {
1548 	return qp->device->post_send(qp, send_wr, bad_send_wr);
1549 }
1550 
1551 /**
1552  * ib_post_recv - Posts a list of work requests to the receive queue of
1553  *   the specified QP.
1554  * @qp: The QP to post the work request on.
1555  * @recv_wr: A list of work requests to post on the receive queue.
1556  * @bad_recv_wr: On an immediate failure, this parameter will reference
1557  *   the work request that failed to be posted on the QP.
1558  */
1559 static inline int ib_post_recv(struct ib_qp *qp,
1560 			       struct ib_recv_wr *recv_wr,
1561 			       struct ib_recv_wr **bad_recv_wr)
1562 {
1563 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1564 }
1565 
1566 /*
1567  * IB_CQ_VECTOR_LEAST_ATTACHED: The constant specifies that
1568  *	the CQ will be attached to the completion vector that has
1569  *	the least number of CQs already attached to it.
1570  */
1571 #define IB_CQ_VECTOR_LEAST_ATTACHED	0xffffffff
1572 
1573 /**
1574  * ib_create_cq - Creates a CQ on the specified device.
1575  * @device: The device on which to create the CQ.
1576  * @comp_handler: A user-specified callback that is invoked when a
1577  *   completion event occurs on the CQ.
1578  * @event_handler: A user-specified callback that is invoked when an
1579  *   asynchronous event not associated with a completion occurs on the CQ.
1580  * @cq_context: Context associated with the CQ returned to the user via
1581  *   the associated completion and event handlers.
1582  * @cqe: The minimum size of the CQ.
1583  * @comp_vector - Completion vector used to signal completion events.
1584  *     Must be >= 0 and < context->num_comp_vectors
1585  *     or IB_CQ_VECTOR_LEAST_ATTACHED.
1586  *
1587  * Users can examine the cq structure to determine the actual CQ size.
1588  */
1589 struct ib_cq *ib_create_cq(struct ib_device *device,
1590 			   ib_comp_handler comp_handler,
1591 			   void (*event_handler)(struct ib_event *, void *),
1592 			   void *cq_context, int cqe, int comp_vector);
1593 
1594 /**
1595  * ib_resize_cq - Modifies the capacity of the CQ.
1596  * @cq: The CQ to resize.
1597  * @cqe: The minimum size of the CQ.
1598  *
1599  * Users can examine the cq structure to determine the actual CQ size.
1600  */
1601 int ib_resize_cq(struct ib_cq *cq, int cqe);
1602 
1603 /**
1604  * ib_modify_cq - Modifies moderation params of the CQ
1605  * @cq: The CQ to modify.
1606  * @cq_count: number of CQEs that will trigger an event
1607  * @cq_period: max period of time in usec before triggering an event
1608  *
1609  */
1610 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1611 
1612 /**
1613  * ib_destroy_cq - Destroys the specified CQ.
1614  * @cq: The CQ to destroy.
1615  */
1616 int ib_destroy_cq(struct ib_cq *cq);
1617 
1618 /**
1619  * ib_poll_cq - poll a CQ for completion(s)
1620  * @cq:the CQ being polled
1621  * @num_entries:maximum number of completions to return
1622  * @wc:array of at least @num_entries &struct ib_wc where completions
1623  *   will be returned
1624  *
1625  * Poll a CQ for (possibly multiple) completions.  If the return value
1626  * is < 0, an error occurred.  If the return value is >= 0, it is the
1627  * number of completions returned.  If the return value is
1628  * non-negative and < num_entries, then the CQ was emptied.
1629  */
1630 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1631 			     struct ib_wc *wc)
1632 {
1633 	return cq->device->poll_cq(cq, num_entries, wc);
1634 }
1635 
1636 /**
1637  * ib_peek_cq - Returns the number of unreaped completions currently
1638  *   on the specified CQ.
1639  * @cq: The CQ to peek.
1640  * @wc_cnt: A minimum number of unreaped completions to check for.
1641  *
1642  * If the number of unreaped completions is greater than or equal to wc_cnt,
1643  * this function returns wc_cnt, otherwise, it returns the actual number of
1644  * unreaped completions.
1645  */
1646 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1647 
1648 /**
1649  * ib_req_notify_cq - Request completion notification on a CQ.
1650  * @cq: The CQ to generate an event for.
1651  * @flags:
1652  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1653  *   to request an event on the next solicited event or next work
1654  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1655  *   may also be |ed in to request a hint about missed events, as
1656  *   described below.
1657  *
1658  * Return Value:
1659  *    < 0 means an error occurred while requesting notification
1660  *   == 0 means notification was requested successfully, and if
1661  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1662  *        were missed and it is safe to wait for another event.  In
1663  *        this case is it guaranteed that any work completions added
1664  *        to the CQ since the last CQ poll will trigger a completion
1665  *        notification event.
1666  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1667  *        in.  It means that the consumer must poll the CQ again to
1668  *        make sure it is empty to avoid missing an event because of a
1669  *        race between requesting notification and an entry being
1670  *        added to the CQ.  This return value means it is possible
1671  *        (but not guaranteed) that a work completion has been added
1672  *        to the CQ since the last poll without triggering a
1673  *        completion notification event.
1674  */
1675 static inline int ib_req_notify_cq(struct ib_cq *cq,
1676 				   enum ib_cq_notify_flags flags)
1677 {
1678 	return cq->device->req_notify_cq(cq, flags);
1679 }
1680 
1681 /**
1682  * ib_req_ncomp_notif - Request completion notification when there are
1683  *   at least the specified number of unreaped completions on the CQ.
1684  * @cq: The CQ to generate an event for.
1685  * @wc_cnt: The number of unreaped completions that should be on the
1686  *   CQ before an event is generated.
1687  */
1688 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1689 {
1690 	return cq->device->req_ncomp_notif ?
1691 		cq->device->req_ncomp_notif(cq, wc_cnt) :
1692 		-ENOSYS;
1693 }
1694 
1695 /**
1696  * ib_get_dma_mr - Returns a memory region for system memory that is
1697  *   usable for DMA.
1698  * @pd: The protection domain associated with the memory region.
1699  * @mr_access_flags: Specifies the memory access rights.
1700  *
1701  * Note that the ib_dma_*() functions defined below must be used
1702  * to create/destroy addresses used with the Lkey or Rkey returned
1703  * by ib_get_dma_mr().
1704  */
1705 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1706 
1707 /**
1708  * ib_dma_mapping_error - check a DMA addr for error
1709  * @dev: The device for which the dma_addr was created
1710  * @dma_addr: The DMA address to check
1711  */
1712 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1713 {
1714 	if (dev->dma_ops)
1715 		return dev->dma_ops->mapping_error(dev, dma_addr);
1716 	return dma_mapping_error(dev->dma_device, dma_addr);
1717 }
1718 
1719 /**
1720  * ib_dma_map_single - Map a kernel virtual address to DMA address
1721  * @dev: The device for which the dma_addr is to be created
1722  * @cpu_addr: The kernel virtual address
1723  * @size: The size of the region in bytes
1724  * @direction: The direction of the DMA
1725  */
1726 static inline u64 ib_dma_map_single(struct ib_device *dev,
1727 				    void *cpu_addr, size_t size,
1728 				    enum dma_data_direction direction)
1729 {
1730 	if (dev->dma_ops)
1731 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1732 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1733 }
1734 
1735 /**
1736  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1737  * @dev: The device for which the DMA address was created
1738  * @addr: The DMA address
1739  * @size: The size of the region in bytes
1740  * @direction: The direction of the DMA
1741  */
1742 static inline void ib_dma_unmap_single(struct ib_device *dev,
1743 				       u64 addr, size_t size,
1744 				       enum dma_data_direction direction)
1745 {
1746 	if (dev->dma_ops)
1747 		dev->dma_ops->unmap_single(dev, addr, size, direction);
1748 	else
1749 		dma_unmap_single(dev->dma_device, addr, size, direction);
1750 }
1751 
1752 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
1753 					  void *cpu_addr, size_t size,
1754 					  enum dma_data_direction direction,
1755 					  struct dma_attrs *attrs)
1756 {
1757 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
1758 				    direction, attrs);
1759 }
1760 
1761 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
1762 					     u64 addr, size_t size,
1763 					     enum dma_data_direction direction,
1764 					     struct dma_attrs *attrs)
1765 {
1766 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
1767 				      direction, attrs);
1768 }
1769 
1770 /**
1771  * ib_dma_map_page - Map a physical page to DMA address
1772  * @dev: The device for which the dma_addr is to be created
1773  * @page: The page to be mapped
1774  * @offset: The offset within the page
1775  * @size: The size of the region in bytes
1776  * @direction: The direction of the DMA
1777  */
1778 static inline u64 ib_dma_map_page(struct ib_device *dev,
1779 				  struct page *page,
1780 				  unsigned long offset,
1781 				  size_t size,
1782 					 enum dma_data_direction direction)
1783 {
1784 	if (dev->dma_ops)
1785 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
1786 	return dma_map_page(dev->dma_device, page, offset, size, direction);
1787 }
1788 
1789 /**
1790  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1791  * @dev: The device for which the DMA address was created
1792  * @addr: The DMA address
1793  * @size: The size of the region in bytes
1794  * @direction: The direction of the DMA
1795  */
1796 static inline void ib_dma_unmap_page(struct ib_device *dev,
1797 				     u64 addr, size_t size,
1798 				     enum dma_data_direction direction)
1799 {
1800 	if (dev->dma_ops)
1801 		dev->dma_ops->unmap_page(dev, addr, size, direction);
1802 	else
1803 		dma_unmap_page(dev->dma_device, addr, size, direction);
1804 }
1805 
1806 /**
1807  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
1808  * @dev: The device for which the DMA addresses are to be created
1809  * @sg: The array of scatter/gather entries
1810  * @nents: The number of scatter/gather entries
1811  * @direction: The direction of the DMA
1812  */
1813 static inline int ib_dma_map_sg(struct ib_device *dev,
1814 				struct scatterlist *sg, int nents,
1815 				enum dma_data_direction direction)
1816 {
1817 	if (dev->dma_ops)
1818 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
1819 	return dma_map_sg(dev->dma_device, sg, nents, direction);
1820 }
1821 
1822 /**
1823  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
1824  * @dev: The device for which the DMA addresses were created
1825  * @sg: The array of scatter/gather entries
1826  * @nents: The number of scatter/gather entries
1827  * @direction: The direction of the DMA
1828  */
1829 static inline void ib_dma_unmap_sg(struct ib_device *dev,
1830 				   struct scatterlist *sg, int nents,
1831 				   enum dma_data_direction direction)
1832 {
1833 	if (dev->dma_ops)
1834 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
1835 	else
1836 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
1837 }
1838 
1839 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
1840 				      struct scatterlist *sg, int nents,
1841 				      enum dma_data_direction direction,
1842 				      struct dma_attrs *attrs)
1843 {
1844 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1845 }
1846 
1847 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
1848 					 struct scatterlist *sg, int nents,
1849 					 enum dma_data_direction direction,
1850 					 struct dma_attrs *attrs)
1851 {
1852 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1853 }
1854 /**
1855  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
1856  * @dev: The device for which the DMA addresses were created
1857  * @sg: The scatter/gather entry
1858  */
1859 static inline u64 ib_sg_dma_address(struct ib_device *dev,
1860 				    struct scatterlist *sg)
1861 {
1862 	if (dev->dma_ops)
1863 		return dev->dma_ops->dma_address(dev, sg);
1864 	return sg_dma_address(sg);
1865 }
1866 
1867 /**
1868  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
1869  * @dev: The device for which the DMA addresses were created
1870  * @sg: The scatter/gather entry
1871  */
1872 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
1873 					 struct scatterlist *sg)
1874 {
1875 	if (dev->dma_ops)
1876 		return dev->dma_ops->dma_len(dev, sg);
1877 	return sg_dma_len(sg);
1878 }
1879 
1880 /**
1881  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
1882  * @dev: The device for which the DMA address was created
1883  * @addr: The DMA address
1884  * @size: The size of the region in bytes
1885  * @dir: The direction of the DMA
1886  */
1887 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
1888 					      u64 addr,
1889 					      size_t size,
1890 					      enum dma_data_direction dir)
1891 {
1892 	if (dev->dma_ops)
1893 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
1894 	else
1895 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
1896 }
1897 
1898 /**
1899  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
1900  * @dev: The device for which the DMA address was created
1901  * @addr: The DMA address
1902  * @size: The size of the region in bytes
1903  * @dir: The direction of the DMA
1904  */
1905 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
1906 						 u64 addr,
1907 						 size_t size,
1908 						 enum dma_data_direction dir)
1909 {
1910 	if (dev->dma_ops)
1911 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
1912 	else
1913 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
1914 }
1915 
1916 /**
1917  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
1918  * @dev: The device for which the DMA address is requested
1919  * @size: The size of the region to allocate in bytes
1920  * @dma_handle: A pointer for returning the DMA address of the region
1921  * @flag: memory allocator flags
1922  */
1923 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
1924 					   size_t size,
1925 					   u64 *dma_handle,
1926 					   gfp_t flag)
1927 {
1928 	if (dev->dma_ops)
1929 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
1930 	else {
1931 		dma_addr_t handle;
1932 		void *ret;
1933 
1934 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
1935 		*dma_handle = handle;
1936 		return ret;
1937 	}
1938 }
1939 
1940 /**
1941  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
1942  * @dev: The device for which the DMA addresses were allocated
1943  * @size: The size of the region
1944  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
1945  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
1946  */
1947 static inline void ib_dma_free_coherent(struct ib_device *dev,
1948 					size_t size, void *cpu_addr,
1949 					u64 dma_handle)
1950 {
1951 	if (dev->dma_ops)
1952 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
1953 	else
1954 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
1955 }
1956 
1957 /**
1958  * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
1959  *   by an HCA.
1960  * @pd: The protection domain associated assigned to the registered region.
1961  * @phys_buf_array: Specifies a list of physical buffers to use in the
1962  *   memory region.
1963  * @num_phys_buf: Specifies the size of the phys_buf_array.
1964  * @mr_access_flags: Specifies the memory access rights.
1965  * @iova_start: The offset of the region's starting I/O virtual address.
1966  */
1967 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
1968 			     struct ib_phys_buf *phys_buf_array,
1969 			     int num_phys_buf,
1970 			     int mr_access_flags,
1971 			     u64 *iova_start);
1972 
1973 /**
1974  * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
1975  *   Conceptually, this call performs the functions deregister memory region
1976  *   followed by register physical memory region.  Where possible,
1977  *   resources are reused instead of deallocated and reallocated.
1978  * @mr: The memory region to modify.
1979  * @mr_rereg_mask: A bit-mask used to indicate which of the following
1980  *   properties of the memory region are being modified.
1981  * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
1982  *   the new protection domain to associated with the memory region,
1983  *   otherwise, this parameter is ignored.
1984  * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1985  *   field specifies a list of physical buffers to use in the new
1986  *   translation, otherwise, this parameter is ignored.
1987  * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1988  *   field specifies the size of the phys_buf_array, otherwise, this
1989  *   parameter is ignored.
1990  * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
1991  *   field specifies the new memory access rights, otherwise, this
1992  *   parameter is ignored.
1993  * @iova_start: The offset of the region's starting I/O virtual address.
1994  */
1995 int ib_rereg_phys_mr(struct ib_mr *mr,
1996 		     int mr_rereg_mask,
1997 		     struct ib_pd *pd,
1998 		     struct ib_phys_buf *phys_buf_array,
1999 		     int num_phys_buf,
2000 		     int mr_access_flags,
2001 		     u64 *iova_start);
2002 
2003 /**
2004  * ib_query_mr - Retrieves information about a specific memory region.
2005  * @mr: The memory region to retrieve information about.
2006  * @mr_attr: The attributes of the specified memory region.
2007  */
2008 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2009 
2010 /**
2011  * ib_dereg_mr - Deregisters a memory region and removes it from the
2012  *   HCA translation table.
2013  * @mr: The memory region to deregister.
2014  */
2015 int ib_dereg_mr(struct ib_mr *mr);
2016 
2017 /**
2018  * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2019  *   IB_WR_FAST_REG_MR send work request.
2020  * @pd: The protection domain associated with the region.
2021  * @max_page_list_len: requested max physical buffer list length to be
2022  *   used with fast register work requests for this MR.
2023  */
2024 struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
2025 
2026 /**
2027  * ib_alloc_fast_reg_page_list - Allocates a page list array
2028  * @device - ib device pointer.
2029  * @page_list_len - size of the page list array to be allocated.
2030  *
2031  * This allocates and returns a struct ib_fast_reg_page_list * and a
2032  * page_list array that is at least page_list_len in size.  The actual
2033  * size is returned in max_page_list_len.  The caller is responsible
2034  * for initializing the contents of the page_list array before posting
2035  * a send work request with the IB_WC_FAST_REG_MR opcode.
2036  *
2037  * The page_list array entries must be translated using one of the
2038  * ib_dma_*() functions just like the addresses passed to
2039  * ib_map_phys_fmr().  Once the ib_post_send() is issued, the struct
2040  * ib_fast_reg_page_list must not be modified by the caller until the
2041  * IB_WC_FAST_REG_MR work request completes.
2042  */
2043 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2044 				struct ib_device *device, int page_list_len);
2045 
2046 /**
2047  * ib_free_fast_reg_page_list - Deallocates a previously allocated
2048  *   page list array.
2049  * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2050  */
2051 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2052 
2053 /**
2054  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2055  *   R_Key and L_Key.
2056  * @mr - struct ib_mr pointer to be updated.
2057  * @newkey - new key to be used.
2058  */
2059 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2060 {
2061 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2062 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2063 }
2064 
2065 /**
2066  * ib_alloc_mw - Allocates a memory window.
2067  * @pd: The protection domain associated with the memory window.
2068  */
2069 struct ib_mw *ib_alloc_mw(struct ib_pd *pd);
2070 
2071 /**
2072  * ib_bind_mw - Posts a work request to the send queue of the specified
2073  *   QP, which binds the memory window to the given address range and
2074  *   remote access attributes.
2075  * @qp: QP to post the bind work request on.
2076  * @mw: The memory window to bind.
2077  * @mw_bind: Specifies information about the memory window, including
2078  *   its address range, remote access rights, and associated memory region.
2079  */
2080 static inline int ib_bind_mw(struct ib_qp *qp,
2081 			     struct ib_mw *mw,
2082 			     struct ib_mw_bind *mw_bind)
2083 {
2084 	/* XXX reference counting in corresponding MR? */
2085 	return mw->device->bind_mw ?
2086 		mw->device->bind_mw(qp, mw, mw_bind) :
2087 		-ENOSYS;
2088 }
2089 
2090 /**
2091  * ib_dealloc_mw - Deallocates a memory window.
2092  * @mw: The memory window to deallocate.
2093  */
2094 int ib_dealloc_mw(struct ib_mw *mw);
2095 
2096 /**
2097  * ib_alloc_fmr - Allocates a unmapped fast memory region.
2098  * @pd: The protection domain associated with the unmapped region.
2099  * @mr_access_flags: Specifies the memory access rights.
2100  * @fmr_attr: Attributes of the unmapped region.
2101  *
2102  * A fast memory region must be mapped before it can be used as part of
2103  * a work request.
2104  */
2105 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2106 			    int mr_access_flags,
2107 			    struct ib_fmr_attr *fmr_attr);
2108 
2109 /**
2110  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2111  * @fmr: The fast memory region to associate with the pages.
2112  * @page_list: An array of physical pages to map to the fast memory region.
2113  * @list_len: The number of pages in page_list.
2114  * @iova: The I/O virtual address to use with the mapped region.
2115  */
2116 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2117 				  u64 *page_list, int list_len,
2118 				  u64 iova)
2119 {
2120 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2121 }
2122 
2123 /**
2124  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2125  * @fmr_list: A linked list of fast memory regions to unmap.
2126  */
2127 int ib_unmap_fmr(struct list_head *fmr_list);
2128 
2129 /**
2130  * ib_dealloc_fmr - Deallocates a fast memory region.
2131  * @fmr: The fast memory region to deallocate.
2132  */
2133 int ib_dealloc_fmr(struct ib_fmr *fmr);
2134 
2135 /**
2136  * ib_attach_mcast - Attaches the specified QP to a multicast group.
2137  * @qp: QP to attach to the multicast group.  The QP must be type
2138  *   IB_QPT_UD.
2139  * @gid: Multicast group GID.
2140  * @lid: Multicast group LID in host byte order.
2141  *
2142  * In order to send and receive multicast packets, subnet
2143  * administration must have created the multicast group and configured
2144  * the fabric appropriately.  The port associated with the specified
2145  * QP must also be a member of the multicast group.
2146  */
2147 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2148 
2149 /**
2150  * ib_detach_mcast - Detaches the specified QP from a multicast group.
2151  * @qp: QP to detach from the multicast group.
2152  * @gid: Multicast group GID.
2153  * @lid: Multicast group LID in host byte order.
2154  */
2155 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2156 
2157 
2158 /**
2159  * ib_dealloc_xrcd - Deallocates an extended reliably connected domain.
2160  * @xrcd: The xrc domain to deallocate.
2161  */
2162 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2163 
2164 /**
2165  * ib_alloc_xrcd - Allocates an extended reliably connected domain.
2166  * @device: The device on which to allocate the xrcd.
2167  */
2168 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2169 
2170 #endif /* IB_VERBS_H */
2171