ib_verbs.h revision 271127
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 <asm/uaccess.h>
53#include <linux/rbtree.h>
54#include <linux/mutex.h>
55
56extern struct workqueue_struct *ib_wq;
57
58union ib_gid {
59	u8	raw[16];
60	struct {
61		__be64	subnet_prefix;
62		__be64	interface_id;
63	} global;
64};
65
66enum rdma_node_type {
67	/* IB values map to NodeInfo:NodeType. */
68	RDMA_NODE_IB_CA 	= 1,
69	RDMA_NODE_IB_SWITCH,
70	RDMA_NODE_IB_ROUTER,
71	RDMA_NODE_RNIC
72};
73
74enum rdma_transport_type {
75	RDMA_TRANSPORT_IB,
76	RDMA_TRANSPORT_IWARP
77};
78
79enum rdma_transport_type
80rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__;
81
82enum rdma_link_layer {
83	IB_LINK_LAYER_UNSPECIFIED,
84	IB_LINK_LAYER_INFINIBAND,
85	IB_LINK_LAYER_ETHERNET,
86};
87
88enum ib_device_cap_flags {
89	IB_DEVICE_RESIZE_MAX_WR		= 1,
90	IB_DEVICE_BAD_PKEY_CNTR		= (1<<1),
91	IB_DEVICE_BAD_QKEY_CNTR		= (1<<2),
92	IB_DEVICE_RAW_MULTI		= (1<<3),
93	IB_DEVICE_AUTO_PATH_MIG		= (1<<4),
94	IB_DEVICE_CHANGE_PHY_PORT	= (1<<5),
95	IB_DEVICE_UD_AV_PORT_ENFORCE	= (1<<6),
96	IB_DEVICE_CURR_QP_STATE_MOD	= (1<<7),
97	IB_DEVICE_SHUTDOWN_PORT		= (1<<8),
98	IB_DEVICE_INIT_TYPE		= (1<<9),
99	IB_DEVICE_PORT_ACTIVE_EVENT	= (1<<10),
100	IB_DEVICE_SYS_IMAGE_GUID	= (1<<11),
101	IB_DEVICE_RC_RNR_NAK_GEN	= (1<<12),
102	IB_DEVICE_SRQ_RESIZE		= (1<<13),
103	IB_DEVICE_N_NOTIFY_CQ		= (1<<14),
104	IB_DEVICE_LOCAL_DMA_LKEY	= (1<<15),
105	IB_DEVICE_RESERVED		= (1<<16), /* old SEND_W_INV */
106	IB_DEVICE_MEM_WINDOW		= (1<<17),
107	/*
108	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
109	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
110	 * messages and can verify the validity of checksum for
111	 * incoming messages.  Setting this flag implies that the
112	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
113	 */
114	IB_DEVICE_UD_IP_CSUM		= (1<<18),
115	IB_DEVICE_UD_TSO		= (1<<19),
116	IB_DEVICE_XRC			= (1<<20),
117	IB_DEVICE_MEM_MGT_EXTENSIONS	= (1<<21),
118	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
119	IB_DEVICE_MR_ALLOCATE		= (1<<23),
120	IB_DEVICE_SHARED_MR             = (1<<24),
121	IB_DEVICE_QPG			= (1<<25),
122	IB_DEVICE_UD_RSS		= (1<<26),
123	IB_DEVICE_UD_TSS		= (1<<27)
124};
125
126enum ib_atomic_cap {
127	IB_ATOMIC_NONE,
128	IB_ATOMIC_HCA,
129	IB_ATOMIC_GLOB
130};
131
132struct ib_device_attr {
133	u64			fw_ver;
134	__be64			sys_image_guid;
135	u64			max_mr_size;
136	u64			page_size_cap;
137	u32			vendor_id;
138	u32			vendor_part_id;
139	u32			hw_ver;
140	int			max_qp;
141	int			max_qp_wr;
142	int			device_cap_flags;
143	int			max_sge;
144	int			max_sge_rd;
145	int			max_cq;
146	int			max_cqe;
147	int			max_mr;
148	int			max_pd;
149	int			max_qp_rd_atom;
150	int			max_ee_rd_atom;
151	int			max_res_rd_atom;
152	int			max_qp_init_rd_atom;
153	int			max_ee_init_rd_atom;
154	enum ib_atomic_cap	atomic_cap;
155	enum ib_atomic_cap	masked_atomic_cap;
156	int			max_ee;
157	int			max_rdd;
158	int			max_mw;
159	int			max_raw_ipv6_qp;
160	int			max_raw_ethy_qp;
161	int			max_mcast_grp;
162	int			max_mcast_qp_attach;
163	int			max_total_mcast_qp_attach;
164	int			max_ah;
165	int			max_fmr;
166	int			max_map_per_fmr;
167	int			max_srq;
168	int			max_srq_wr;
169	int			max_srq_sge;
170	unsigned int		max_fast_reg_page_list_len;
171	int			max_rss_tbl_sz;
172	u16			max_pkeys;
173	u8			local_ca_ack_delay;
174};
175
176enum ib_mtu {
177	IB_MTU_256  = 1,
178	IB_MTU_512  = 2,
179	IB_MTU_1024 = 3,
180	IB_MTU_2048 = 4,
181	IB_MTU_4096 = 5
182};
183
184static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
185{
186	switch (mtu) {
187	case IB_MTU_256:  return  256;
188	case IB_MTU_512:  return  512;
189	case IB_MTU_1024: return 1024;
190	case IB_MTU_2048: return 2048;
191	case IB_MTU_4096: return 4096;
192	default: 	  return -1;
193	}
194}
195
196enum ib_port_state {
197	IB_PORT_NOP		= 0,
198	IB_PORT_DOWN		= 1,
199	IB_PORT_INIT		= 2,
200	IB_PORT_ARMED		= 3,
201	IB_PORT_ACTIVE		= 4,
202	IB_PORT_ACTIVE_DEFER	= 5
203};
204
205enum ib_port_cap_flags {
206	IB_PORT_SM				= 1 <<  1,
207	IB_PORT_NOTICE_SUP			= 1 <<  2,
208	IB_PORT_TRAP_SUP			= 1 <<  3,
209	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
210	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
211	IB_PORT_SL_MAP_SUP			= 1 <<  6,
212	IB_PORT_MKEY_NVRAM			= 1 <<  7,
213	IB_PORT_PKEY_NVRAM			= 1 <<  8,
214	IB_PORT_LED_INFO_SUP			= 1 <<  9,
215	IB_PORT_SM_DISABLED			= 1 << 10,
216	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
217	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
218	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
219	IB_PORT_CM_SUP				= 1 << 16,
220	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
221	IB_PORT_REINIT_SUP			= 1 << 18,
222	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
223	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
224	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
225	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
226	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
227	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
228	IB_PORT_CLIENT_REG_SUP			= 1 << 25
229};
230
231enum ib_port_width {
232	IB_WIDTH_1X	= 1,
233	IB_WIDTH_4X	= 2,
234	IB_WIDTH_8X	= 4,
235	IB_WIDTH_12X	= 8
236};
237
238static inline int ib_width_enum_to_int(enum ib_port_width width)
239{
240	switch (width) {
241	case IB_WIDTH_1X:  return  1;
242	case IB_WIDTH_4X:  return  4;
243	case IB_WIDTH_8X:  return  8;
244	case IB_WIDTH_12X: return 12;
245	default: 	  return -1;
246	}
247}
248
249enum ib_port_speed {
250	IB_SPEED_SDR	= 1,
251	IB_SPEED_DDR	= 2,
252	IB_SPEED_QDR	= 4,
253	IB_SPEED_FDR10	= 8,
254	IB_SPEED_FDR	= 16,
255	IB_SPEED_EDR	= 32
256};
257
258struct ib_protocol_stats {
259	/* TBD... */
260};
261
262struct iw_protocol_stats {
263	u64	ipInReceives;
264	u64	ipInHdrErrors;
265	u64	ipInTooBigErrors;
266	u64	ipInNoRoutes;
267	u64	ipInAddrErrors;
268	u64	ipInUnknownProtos;
269	u64	ipInTruncatedPkts;
270	u64	ipInDiscards;
271	u64	ipInDelivers;
272	u64	ipOutForwDatagrams;
273	u64	ipOutRequests;
274	u64	ipOutDiscards;
275	u64	ipOutNoRoutes;
276	u64	ipReasmTimeout;
277	u64	ipReasmReqds;
278	u64	ipReasmOKs;
279	u64	ipReasmFails;
280	u64	ipFragOKs;
281	u64	ipFragFails;
282	u64	ipFragCreates;
283	u64	ipInMcastPkts;
284	u64	ipOutMcastPkts;
285	u64	ipInBcastPkts;
286	u64	ipOutBcastPkts;
287
288	u64	tcpRtoAlgorithm;
289	u64	tcpRtoMin;
290	u64	tcpRtoMax;
291	u64	tcpMaxConn;
292	u64	tcpActiveOpens;
293	u64	tcpPassiveOpens;
294	u64	tcpAttemptFails;
295	u64	tcpEstabResets;
296	u64	tcpCurrEstab;
297	u64	tcpInSegs;
298	u64	tcpOutSegs;
299	u64	tcpRetransSegs;
300	u64	tcpInErrs;
301	u64	tcpOutRsts;
302};
303
304union rdma_protocol_stats {
305	struct ib_protocol_stats	ib;
306	struct iw_protocol_stats	iw;
307};
308
309struct ib_port_attr {
310	enum ib_port_state	state;
311	enum ib_mtu		max_mtu;
312	enum ib_mtu		active_mtu;
313	int			gid_tbl_len;
314	u32			port_cap_flags;
315	u32			max_msg_sz;
316	u32			bad_pkey_cntr;
317	u32			qkey_viol_cntr;
318	u16			pkey_tbl_len;
319	u16			lid;
320	u16			sm_lid;
321	u8			lmc;
322	u8			max_vl_num;
323	u8			sm_sl;
324	u8			subnet_timeout;
325	u8			init_type_reply;
326	u8			active_width;
327	u8			active_speed;
328	u8                      phys_state;
329	enum rdma_link_layer	link_layer;
330};
331
332enum ib_device_modify_flags {
333	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
334	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
335};
336
337struct ib_device_modify {
338	u64	sys_image_guid;
339	char	node_desc[64];
340};
341
342enum ib_port_modify_flags {
343	IB_PORT_SHUTDOWN		= 1,
344	IB_PORT_INIT_TYPE		= (1<<2),
345	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
346};
347
348struct ib_port_modify {
349	u32	set_port_cap_mask;
350	u32	clr_port_cap_mask;
351	u8	init_type;
352};
353
354enum ib_event_type {
355	IB_EVENT_CQ_ERR,
356	IB_EVENT_QP_FATAL,
357	IB_EVENT_QP_REQ_ERR,
358	IB_EVENT_QP_ACCESS_ERR,
359	IB_EVENT_COMM_EST,
360	IB_EVENT_SQ_DRAINED,
361	IB_EVENT_PATH_MIG,
362	IB_EVENT_PATH_MIG_ERR,
363	IB_EVENT_DEVICE_FATAL,
364	IB_EVENT_PORT_ACTIVE,
365	IB_EVENT_PORT_ERR,
366	IB_EVENT_LID_CHANGE,
367	IB_EVENT_PKEY_CHANGE,
368	IB_EVENT_SM_CHANGE,
369	IB_EVENT_SRQ_ERR,
370	IB_EVENT_SRQ_LIMIT_REACHED,
371	IB_EVENT_QP_LAST_WQE_REACHED,
372	IB_EVENT_CLIENT_REREGISTER,
373	IB_EVENT_GID_CHANGE,
374};
375
376enum ib_event_flags {
377	IB_XRC_QP_EVENT_FLAG = 0x80000000,
378};
379
380struct ib_event {
381	struct ib_device	*device;
382	union {
383		struct ib_cq	*cq;
384		struct ib_qp	*qp;
385		struct ib_srq	*srq;
386		u8		port_num;
387		u32		xrc_qp_num;
388	} element;
389	enum ib_event_type	event;
390};
391
392struct ib_event_handler {
393	struct ib_device *device;
394	void            (*handler)(struct ib_event_handler *, struct ib_event *);
395	struct list_head  list;
396};
397
398#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
399	do {							\
400		(_ptr)->device  = _device;			\
401		(_ptr)->handler = _handler;			\
402		INIT_LIST_HEAD(&(_ptr)->list);			\
403	} while (0)
404
405struct ib_global_route {
406	union ib_gid	dgid;
407	u32		flow_label;
408	u8		sgid_index;
409	u8		hop_limit;
410	u8		traffic_class;
411};
412
413struct ib_grh {
414	__be32		version_tclass_flow;
415	__be16		paylen;
416	u8		next_hdr;
417	u8		hop_limit;
418	union ib_gid	sgid;
419	union ib_gid	dgid;
420};
421
422enum {
423	IB_MULTICAST_QPN = 0xffffff
424};
425
426#define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
427
428enum ib_ah_flags {
429	IB_AH_GRH	= 1
430};
431
432enum ib_rate {
433	IB_RATE_PORT_CURRENT = 0,
434	IB_RATE_2_5_GBPS = 2,
435	IB_RATE_5_GBPS   = 5,
436	IB_RATE_10_GBPS  = 3,
437	IB_RATE_20_GBPS  = 6,
438	IB_RATE_30_GBPS  = 4,
439	IB_RATE_40_GBPS  = 7,
440	IB_RATE_60_GBPS  = 8,
441	IB_RATE_80_GBPS  = 9,
442	IB_RATE_120_GBPS = 10,
443	IB_RATE_14_GBPS  = 11,
444	IB_RATE_56_GBPS  = 12,
445	IB_RATE_112_GBPS = 13,
446	IB_RATE_168_GBPS = 14,
447	IB_RATE_25_GBPS  = 15,
448	IB_RATE_100_GBPS = 16,
449	IB_RATE_200_GBPS = 17,
450	IB_RATE_300_GBPS = 18
451};
452
453/**
454 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
455 * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
456 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
457 * @rate: rate to convert.
458 */
459int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
460
461/**
462 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
463 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
464 * @rate: rate to convert.
465 */
466int ib_rate_to_mbps(enum ib_rate rate) __attribute_const__;
467
468/**
469 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
470 * enum.
471 * @mult: multiple to convert.
472 */
473enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
474
475struct ib_ah_attr {
476	struct ib_global_route	grh;
477	u16			dlid;
478	u8			sl;
479	u8			src_path_bits;
480	u8			static_rate;
481	u8			ah_flags;
482	u8			port_num;
483};
484
485enum ib_wc_status {
486	IB_WC_SUCCESS,
487	IB_WC_LOC_LEN_ERR,
488	IB_WC_LOC_QP_OP_ERR,
489	IB_WC_LOC_EEC_OP_ERR,
490	IB_WC_LOC_PROT_ERR,
491	IB_WC_WR_FLUSH_ERR,
492	IB_WC_MW_BIND_ERR,
493	IB_WC_BAD_RESP_ERR,
494	IB_WC_LOC_ACCESS_ERR,
495	IB_WC_REM_INV_REQ_ERR,
496	IB_WC_REM_ACCESS_ERR,
497	IB_WC_REM_OP_ERR,
498	IB_WC_RETRY_EXC_ERR,
499	IB_WC_RNR_RETRY_EXC_ERR,
500	IB_WC_LOC_RDD_VIOL_ERR,
501	IB_WC_REM_INV_RD_REQ_ERR,
502	IB_WC_REM_ABORT_ERR,
503	IB_WC_INV_EECN_ERR,
504	IB_WC_INV_EEC_STATE_ERR,
505	IB_WC_FATAL_ERR,
506	IB_WC_RESP_TIMEOUT_ERR,
507	IB_WC_GENERAL_ERR
508};
509
510enum ib_wc_opcode {
511	IB_WC_SEND,
512	IB_WC_RDMA_WRITE,
513	IB_WC_RDMA_READ,
514	IB_WC_COMP_SWAP,
515	IB_WC_FETCH_ADD,
516	IB_WC_BIND_MW,
517	IB_WC_LSO,
518	IB_WC_LOCAL_INV,
519	IB_WC_FAST_REG_MR,
520	IB_WC_MASKED_COMP_SWAP,
521	IB_WC_MASKED_FETCH_ADD,
522/*
523 * Set value of IB_WC_RECV so consumers can test if a completion is a
524 * receive by testing (opcode & IB_WC_RECV).
525 */
526	IB_WC_RECV			= 1 << 7,
527	IB_WC_RECV_RDMA_WITH_IMM
528};
529
530enum ib_wc_flags {
531	IB_WC_GRH		= 1,
532	IB_WC_WITH_IMM		= (1<<1),
533	IB_WC_WITH_INVALIDATE	= (1<<2),
534	IB_WC_IP_CSUM_OK	= (1<<3),
535};
536
537struct ib_wc {
538	u64			wr_id;
539	enum ib_wc_status	status;
540	enum ib_wc_opcode	opcode;
541	u32			vendor_err;
542	u32			byte_len;
543	struct ib_qp	       *qp;
544	union {
545		__be32		imm_data;
546		u32		invalidate_rkey;
547	} ex;
548	u32			src_qp;
549	int			wc_flags;
550	u16			pkey_index;
551	u16			slid;
552	u8			sl;
553	u8			dlid_path_bits;
554	u8			port_num;	/* valid only for DR SMPs on switches */
555	int			csum_ok;
556};
557
558enum ib_cq_notify_flags {
559	IB_CQ_SOLICITED			= 1 << 0,
560	IB_CQ_NEXT_COMP			= 1 << 1,
561	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
562	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
563};
564
565enum ib_srq_type {
566	IB_SRQT_BASIC,
567	IB_SRQT_XRC
568};
569
570enum ib_srq_attr_mask {
571	IB_SRQ_MAX_WR	= 1 << 0,
572	IB_SRQ_LIMIT	= 1 << 1,
573};
574
575struct ib_srq_attr {
576	u32	max_wr;
577	u32	max_sge;
578	u32	srq_limit;
579};
580
581struct ib_srq_init_attr {
582	void		      (*event_handler)(struct ib_event *, void *);
583	void		       *srq_context;
584	struct ib_srq_attr	attr;
585	enum ib_srq_type	srq_type;
586
587	union {
588		struct {
589			struct ib_xrcd *xrcd;
590			struct ib_cq   *cq;
591		} xrc;
592	} ext;
593};
594
595struct ib_qp_cap {
596	u32	max_send_wr;
597	u32	max_recv_wr;
598	u32	max_send_sge;
599	u32	max_recv_sge;
600	u32	max_inline_data;
601	u32	qpg_tss_mask_sz;
602};
603
604enum ib_sig_type {
605	IB_SIGNAL_ALL_WR,
606	IB_SIGNAL_REQ_WR
607};
608
609enum ib_qp_type {
610	/*
611	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
612	 * here (and in that order) since the MAD layer uses them as
613	 * indices into a 2-entry table.
614	 */
615	IB_QPT_SMI,
616	IB_QPT_GSI,
617
618	IB_QPT_RC,
619	IB_QPT_UC,
620	IB_QPT_UD,
621	IB_QPT_XRC,
622	IB_QPT_RAW_IPV6,
623	IB_QPT_RAW_ETHERTYPE,
624	IB_QPT_RAW_PACKET = 8,
625	IB_QPT_XRC_INI = 9,
626	IB_QPT_XRC_TGT,
627	IB_QPT_MAX,
628};
629
630enum ib_qp_create_flags {
631	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
632	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
633	IB_QP_CREATE_NETIF_QP			= 1 << 2,
634	/* reserve bits 26-31 for low level drivers' internal use */
635	IB_QP_CREATE_RESERVED_START		= 1 << 26,
636	IB_QP_CREATE_RESERVED_END		= 1 << 31,
637};
638
639enum ib_qpg_type {
640	IB_QPG_NONE	= 0,
641	IB_QPG_PARENT	= (1<<0),
642	IB_QPG_CHILD_RX = (1<<1),
643	IB_QPG_CHILD_TX = (1<<2)
644};
645
646struct ib_qpg_init_attrib {
647	u32 tss_child_count;
648	u32 rss_child_count;
649};
650
651struct ib_qp_init_attr {
652	void                  (*event_handler)(struct ib_event *, void *);
653	void		       *qp_context;
654	struct ib_cq	       *send_cq;
655	struct ib_cq	       *recv_cq;
656	struct ib_srq	       *srq;
657	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
658	struct ib_qp_cap	cap;
659	union {
660		struct ib_qp *qpg_parent; /* see qpg_type */
661		struct ib_qpg_init_attrib parent_attrib;
662	} pp;
663	enum ib_sig_type	sq_sig_type;
664	enum ib_qp_type		qp_type;
665	enum ib_qp_create_flags	create_flags;
666	enum ib_qpg_type	qpg_type;
667	u8			port_num; /* special QP types only */
668};
669
670struct ib_qp_open_attr {
671	void                  (*event_handler)(struct ib_event *, void *);
672	void		       *qp_context;
673	u32			qp_num;
674	enum ib_qp_type		qp_type;
675};
676
677enum ib_rnr_timeout {
678	IB_RNR_TIMER_655_36 =  0,
679	IB_RNR_TIMER_000_01 =  1,
680	IB_RNR_TIMER_000_02 =  2,
681	IB_RNR_TIMER_000_03 =  3,
682	IB_RNR_TIMER_000_04 =  4,
683	IB_RNR_TIMER_000_06 =  5,
684	IB_RNR_TIMER_000_08 =  6,
685	IB_RNR_TIMER_000_12 =  7,
686	IB_RNR_TIMER_000_16 =  8,
687	IB_RNR_TIMER_000_24 =  9,
688	IB_RNR_TIMER_000_32 = 10,
689	IB_RNR_TIMER_000_48 = 11,
690	IB_RNR_TIMER_000_64 = 12,
691	IB_RNR_TIMER_000_96 = 13,
692	IB_RNR_TIMER_001_28 = 14,
693	IB_RNR_TIMER_001_92 = 15,
694	IB_RNR_TIMER_002_56 = 16,
695	IB_RNR_TIMER_003_84 = 17,
696	IB_RNR_TIMER_005_12 = 18,
697	IB_RNR_TIMER_007_68 = 19,
698	IB_RNR_TIMER_010_24 = 20,
699	IB_RNR_TIMER_015_36 = 21,
700	IB_RNR_TIMER_020_48 = 22,
701	IB_RNR_TIMER_030_72 = 23,
702	IB_RNR_TIMER_040_96 = 24,
703	IB_RNR_TIMER_061_44 = 25,
704	IB_RNR_TIMER_081_92 = 26,
705	IB_RNR_TIMER_122_88 = 27,
706	IB_RNR_TIMER_163_84 = 28,
707	IB_RNR_TIMER_245_76 = 29,
708	IB_RNR_TIMER_327_68 = 30,
709	IB_RNR_TIMER_491_52 = 31
710};
711
712enum ib_qp_attr_mask {
713	IB_QP_STATE			= 1,
714	IB_QP_CUR_STATE			= (1<<1),
715	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
716	IB_QP_ACCESS_FLAGS		= (1<<3),
717	IB_QP_PKEY_INDEX		= (1<<4),
718	IB_QP_PORT			= (1<<5),
719	IB_QP_QKEY			= (1<<6),
720	IB_QP_AV			= (1<<7),
721	IB_QP_PATH_MTU			= (1<<8),
722	IB_QP_TIMEOUT			= (1<<9),
723	IB_QP_RETRY_CNT			= (1<<10),
724	IB_QP_RNR_RETRY			= (1<<11),
725	IB_QP_RQ_PSN			= (1<<12),
726	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
727	IB_QP_ALT_PATH			= (1<<14),
728	IB_QP_MIN_RNR_TIMER		= (1<<15),
729	IB_QP_SQ_PSN			= (1<<16),
730	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
731	IB_QP_PATH_MIG_STATE		= (1<<18),
732	IB_QP_CAP			= (1<<19),
733	IB_QP_DEST_QPN			= (1<<20),
734	IB_QP_GROUP_RSS			= (1<<21)
735};
736
737enum ib_qp_state {
738	IB_QPS_RESET,
739	IB_QPS_INIT,
740	IB_QPS_RTR,
741	IB_QPS_RTS,
742	IB_QPS_SQD,
743	IB_QPS_SQE,
744	IB_QPS_ERR
745};
746
747enum ib_mig_state {
748	IB_MIG_MIGRATED,
749	IB_MIG_REARM,
750	IB_MIG_ARMED
751};
752
753struct ib_qp_attr {
754	enum ib_qp_state	qp_state;
755	enum ib_qp_state	cur_qp_state;
756	enum ib_mtu		path_mtu;
757	enum ib_mig_state	path_mig_state;
758	u32			qkey;
759	u32			rq_psn;
760	u32			sq_psn;
761	u32			dest_qp_num;
762	int			qp_access_flags;
763	struct ib_qp_cap	cap;
764	struct ib_ah_attr	ah_attr;
765	struct ib_ah_attr	alt_ah_attr;
766	u16			pkey_index;
767	u16			alt_pkey_index;
768	u8			en_sqd_async_notify;
769	u8			sq_draining;
770	u8			max_rd_atomic;
771	u8			max_dest_rd_atomic;
772	u8			min_rnr_timer;
773	u8			port_num;
774	u8			timeout;
775	u8			retry_cnt;
776	u8			rnr_retry;
777	u8			alt_port_num;
778	u8			alt_timeout;
779};
780
781enum ib_wr_opcode {
782	IB_WR_RDMA_WRITE,
783	IB_WR_RDMA_WRITE_WITH_IMM,
784	IB_WR_SEND,
785	IB_WR_SEND_WITH_IMM,
786	IB_WR_RDMA_READ,
787	IB_WR_ATOMIC_CMP_AND_SWP,
788	IB_WR_ATOMIC_FETCH_AND_ADD,
789	IB_WR_LSO,
790	IB_WR_BIG_LSO,
791	IB_WR_SEND_WITH_INV,
792	IB_WR_RDMA_READ_WITH_INV,
793	IB_WR_LOCAL_INV,
794	IB_WR_FAST_REG_MR,
795	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
796	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
797};
798
799enum ib_send_flags {
800	IB_SEND_FENCE		= 1,
801	IB_SEND_SIGNALED	= (1<<1),
802	IB_SEND_SOLICITED	= (1<<2),
803	IB_SEND_INLINE		= (1<<3),
804	IB_SEND_IP_CSUM		= (1<<4)
805};
806
807enum ib_flow_types {
808	IB_FLOW_ETH = 0,
809	IB_FLOW_IB_UC = 1,
810	IB_FLOW_IB_MC_IPV4 = 2,
811	IB_FLOW_IB_MC_IPV6 = 3
812};
813
814enum {
815	IB_FLOW_L4_NONE = 0,
816	IB_FLOW_L4_OTHER = 3,
817	IB_FLOW_L4_UDP = 5,
818	IB_FLOW_L4_TCP = 6
819};
820
821struct ib_sge {
822	u64	addr;
823	u32	length;
824	u32	lkey;
825};
826
827struct ib_fast_reg_page_list {
828	struct ib_device       *device;
829	u64		       *page_list;
830	unsigned int		max_page_list_len;
831};
832
833struct ib_send_wr {
834	struct ib_send_wr      *next;
835	u64			wr_id;
836	struct ib_sge	       *sg_list;
837	int			num_sge;
838	enum ib_wr_opcode	opcode;
839	int			send_flags;
840	union {
841		__be32		imm_data;
842		u32		invalidate_rkey;
843	} ex;
844	union {
845		struct {
846			u64	remote_addr;
847			u32	rkey;
848		} rdma;
849		struct {
850			u64	remote_addr;
851			u64	compare_add;
852			u64	swap;
853			u64	compare_add_mask;
854			u64	swap_mask;
855			u32	rkey;
856		} atomic;
857		struct {
858			struct ib_ah *ah;
859			void   *header;
860			int     hlen;
861			int     mss;
862			u32	remote_qpn;
863			u32	remote_qkey;
864			u16	pkey_index; /* valid for GSI only */
865			u8	port_num;   /* valid for DR SMPs on switch only */
866		} ud;
867		struct {
868			u64				iova_start;
869			struct ib_fast_reg_page_list   *page_list;
870			unsigned int			page_shift;
871			unsigned int			page_list_len;
872			u32				length;
873			int				access_flags;
874			u32				rkey;
875		} fast_reg;
876		struct {
877			struct ib_unpacked_lrh	*lrh;
878			u32			eth_type;
879			u8			static_rate;
880		} raw_ety;
881	} wr;
882	u32			xrc_remote_srq_num;	/* XRC TGT QPs only */
883};
884
885struct ib_recv_wr {
886	struct ib_recv_wr      *next;
887	u64			wr_id;
888	struct ib_sge	       *sg_list;
889	int			num_sge;
890};
891
892enum ib_access_flags {
893	IB_ACCESS_LOCAL_WRITE	= 1,
894	IB_ACCESS_REMOTE_WRITE	= (1<<1),
895	IB_ACCESS_REMOTE_READ	= (1<<2),
896	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
897	IB_ACCESS_MW_BIND	= (1<<4),
898	IB_ACCESS_ALLOCATE_MR	= (1<<5),
899	IB_ACCESS_SHARED_MR_USER_READ   = (1<<6),
900	IB_ACCESS_SHARED_MR_USER_WRITE  = (1<<7),
901	IB_ACCESS_SHARED_MR_GROUP_READ  = (1<<8),
902	IB_ACCESS_SHARED_MR_GROUP_WRITE = (1<<9),
903	IB_ACCESS_SHARED_MR_OTHER_READ  = (1<<10),
904	IB_ACCESS_SHARED_MR_OTHER_WRITE = (1<<11)
905
906};
907
908struct ib_phys_buf {
909	u64      addr;
910	u64      size;
911};
912
913struct 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
922enum 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
928struct ib_mw_bind {
929	struct ib_mr   *mr;
930	u64		wr_id;
931	u64		addr;
932	u32		length;
933	int		send_flags;
934	int		mw_access_flags;
935};
936
937struct ib_fmr_attr {
938	int	max_pages;
939	int	max_maps;
940	u8	page_shift;
941};
942
943struct ib_ucontext {
944	struct ib_device       *device;
945	struct list_head	pd_list;
946	struct list_head	mr_list;
947	struct list_head	mw_list;
948	struct list_head	cq_list;
949	struct list_head	qp_list;
950	struct list_head	srq_list;
951	struct list_head	ah_list;
952	struct list_head	xrcd_list;
953	int			closing;
954};
955
956struct ib_uobject {
957	u64			user_handle;	/* handle given to us by userspace */
958	struct ib_ucontext     *context;	/* associated user context */
959	void		       *object;		/* containing object */
960	struct list_head	list;		/* link to context's list */
961	int			id;		/* index into kernel idr */
962	struct kref		ref;
963	struct rw_semaphore	mutex;		/* protects .live */
964	int			live;
965};
966
967struct ib_udata {
968	void __user *inbuf;
969	void __user *outbuf;
970	size_t       inlen;
971	size_t       outlen;
972};
973
974struct ib_uxrc_rcv_object {
975	struct list_head	list;		/* link to context's list */
976	u32			qp_num;
977	u32			domain_handle;
978};
979
980struct ib_pd {
981	struct ib_device       *device;
982	struct ib_uobject      *uobject;
983	atomic_t          	usecnt; /* count all resources */
984};
985
986struct ib_xrcd {
987	struct ib_device       *device;
988	struct ib_uobject      *uobject;
989	atomic_t		usecnt; /* count all exposed resources */
990	struct inode	       *inode;
991	struct rb_node		node;
992
993	struct mutex		tgt_qp_mutex;
994	struct list_head	tgt_qp_list;
995};
996
997struct ib_ah {
998	struct ib_device	*device;
999	struct ib_pd		*pd;
1000	struct ib_uobject	*uobject;
1001};
1002
1003typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1004
1005struct ib_cq {
1006	struct ib_device       *device;
1007	struct ib_uobject      *uobject;
1008	ib_comp_handler   	comp_handler;
1009	void                  (*event_handler)(struct ib_event *, void *);
1010	void                   *cq_context;
1011	int               	cqe;
1012	atomic_t          	usecnt; /* count number of work queues */
1013};
1014
1015struct ib_srq {
1016	struct ib_device       *device;
1017	struct ib_pd	       *pd;
1018	struct ib_uobject      *uobject;
1019	void		      (*event_handler)(struct ib_event *, void *);
1020	void		       *srq_context;
1021	enum ib_srq_type	srq_type;
1022	atomic_t		usecnt;
1023
1024	union {
1025		struct {
1026			struct ib_xrcd *xrcd;
1027			struct ib_cq   *cq;
1028			u32		srq_num;
1029		} xrc;
1030	} ext;
1031};
1032
1033struct ib_qp {
1034	struct ib_device       *device;
1035	struct ib_pd	       *pd;
1036	struct ib_cq	       *send_cq;
1037	struct ib_cq	       *recv_cq;
1038	struct ib_srq	       *srq;
1039	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1040	struct list_head	xrcd_list;
1041	atomic_t		usecnt; /* count times opened, mcast attaches */
1042	struct list_head	open_list;
1043	struct ib_qp           *real_qp;
1044	struct ib_uobject      *uobject;
1045	void                  (*event_handler)(struct ib_event *, void *);
1046	void		       *qp_context;
1047	u32			qp_num;
1048	enum ib_qp_type		qp_type;
1049	enum ib_qpg_type	qpg_type;
1050};
1051
1052struct ib_mr {
1053	struct ib_device  *device;
1054	struct ib_pd	  *pd;
1055	struct ib_uobject *uobject;
1056	u32		   lkey;
1057	u32		   rkey;
1058	atomic_t	   usecnt; /* count number of MWs */
1059};
1060
1061struct ib_mw {
1062	struct ib_device	*device;
1063	struct ib_pd		*pd;
1064	struct ib_uobject	*uobject;
1065	u32			rkey;
1066};
1067
1068struct ib_fmr {
1069	struct ib_device	*device;
1070	struct ib_pd		*pd;
1071	struct list_head	list;
1072	u32			lkey;
1073	u32			rkey;
1074};
1075
1076struct ib_flow_spec {
1077	enum ib_flow_types type;
1078	union {
1079		struct {
1080			__be16 ethertype;
1081			__be16 vlan;
1082			u8 vlan_present;
1083			u8  mac[6];
1084			u8  port;
1085		} eth;
1086		struct {
1087			__be32 qpn;
1088		} ib_uc;
1089		struct {
1090			u8  mgid[16];
1091		} ib_mc;
1092	} l2_id;
1093	__be32 src_ip;
1094	__be32 dst_ip;
1095	__be16 src_port;
1096	__be16 dst_port;
1097	u8 l4_protocol;
1098	u8 block_mc_loopback;
1099	u8 rule_type;
1100};
1101
1102struct ib_mad;
1103struct ib_grh;
1104
1105enum ib_process_mad_flags {
1106	IB_MAD_IGNORE_MKEY	= 1,
1107	IB_MAD_IGNORE_BKEY	= 2,
1108	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1109};
1110
1111enum ib_mad_result {
1112	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1113	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1114	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1115	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1116};
1117
1118#define IB_DEVICE_NAME_MAX 64
1119
1120struct ib_cache {
1121	rwlock_t                lock;
1122	struct ib_event_handler event_handler;
1123	struct ib_pkey_cache  **pkey_cache;
1124	struct ib_gid_cache   **gid_cache;
1125	u8                     *lmc_cache;
1126};
1127
1128struct ib_dma_mapping_ops {
1129	int		(*mapping_error)(struct ib_device *dev,
1130					 u64 dma_addr);
1131	u64		(*map_single)(struct ib_device *dev,
1132				      void *ptr, size_t size,
1133				      enum dma_data_direction direction);
1134	void		(*unmap_single)(struct ib_device *dev,
1135					u64 addr, size_t size,
1136					enum dma_data_direction direction);
1137	u64		(*map_page)(struct ib_device *dev,
1138				    struct page *page, unsigned long offset,
1139				    size_t size,
1140				    enum dma_data_direction direction);
1141	void		(*unmap_page)(struct ib_device *dev,
1142				      u64 addr, size_t size,
1143				      enum dma_data_direction direction);
1144	int		(*map_sg)(struct ib_device *dev,
1145				  struct scatterlist *sg, int nents,
1146				  enum dma_data_direction direction);
1147	void		(*unmap_sg)(struct ib_device *dev,
1148				    struct scatterlist *sg, int nents,
1149				    enum dma_data_direction direction);
1150	u64		(*dma_address)(struct ib_device *dev,
1151				       struct scatterlist *sg);
1152	unsigned int	(*dma_len)(struct ib_device *dev,
1153				   struct scatterlist *sg);
1154	void		(*sync_single_for_cpu)(struct ib_device *dev,
1155					       u64 dma_handle,
1156					       size_t size,
1157					       enum dma_data_direction dir);
1158	void		(*sync_single_for_device)(struct ib_device *dev,
1159						  u64 dma_handle,
1160						  size_t size,
1161						  enum dma_data_direction dir);
1162	void		*(*alloc_coherent)(struct ib_device *dev,
1163					   size_t size,
1164					   u64 *dma_handle,
1165					   gfp_t flag);
1166	void		(*free_coherent)(struct ib_device *dev,
1167					 size_t size, void *cpu_addr,
1168					 u64 dma_handle);
1169};
1170
1171struct iw_cm_verbs;
1172
1173struct ib_device {
1174	struct device                *dma_device;
1175
1176	char                          name[IB_DEVICE_NAME_MAX];
1177
1178	struct list_head              event_handler_list;
1179	spinlock_t                    event_handler_lock;
1180
1181	spinlock_t                    client_data_lock;
1182	struct list_head              core_list;
1183	struct list_head              client_data_list;
1184
1185	struct ib_cache               cache;
1186	int                          *pkey_tbl_len;
1187	int                          *gid_tbl_len;
1188
1189	int			      num_comp_vectors;
1190
1191	struct iw_cm_verbs	     *iwcm;
1192
1193	int		           (*get_protocol_stats)(struct ib_device *device,
1194							 union rdma_protocol_stats *stats);
1195	int		           (*query_device)(struct ib_device *device,
1196						   struct ib_device_attr *device_attr);
1197	int		           (*query_port)(struct ib_device *device,
1198						 u8 port_num,
1199						 struct ib_port_attr *port_attr);
1200	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1201						     u8 port_num);
1202	int		           (*query_gid)(struct ib_device *device,
1203						u8 port_num, int index,
1204						union ib_gid *gid);
1205	int		           (*query_pkey)(struct ib_device *device,
1206						 u8 port_num, u16 index, u16 *pkey);
1207	int		           (*modify_device)(struct ib_device *device,
1208						    int device_modify_mask,
1209						    struct ib_device_modify *device_modify);
1210	int		           (*modify_port)(struct ib_device *device,
1211						  u8 port_num, int port_modify_mask,
1212						  struct ib_port_modify *port_modify);
1213	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1214						     struct ib_udata *udata);
1215	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1216	int                        (*mmap)(struct ib_ucontext *context,
1217					   struct vm_area_struct *vma);
1218	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1219					       struct ib_ucontext *context,
1220					       struct ib_udata *udata);
1221	int                        (*dealloc_pd)(struct ib_pd *pd);
1222	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1223						struct ib_ah_attr *ah_attr);
1224	int                        (*modify_ah)(struct ib_ah *ah,
1225						struct ib_ah_attr *ah_attr);
1226	int                        (*query_ah)(struct ib_ah *ah,
1227					       struct ib_ah_attr *ah_attr);
1228	int                        (*destroy_ah)(struct ib_ah *ah);
1229	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1230						 struct ib_srq_init_attr *srq_init_attr,
1231						 struct ib_udata *udata);
1232	int                        (*modify_srq)(struct ib_srq *srq,
1233						 struct ib_srq_attr *srq_attr,
1234						 enum ib_srq_attr_mask srq_attr_mask,
1235						 struct ib_udata *udata);
1236	int                        (*query_srq)(struct ib_srq *srq,
1237						struct ib_srq_attr *srq_attr);
1238	int                        (*destroy_srq)(struct ib_srq *srq);
1239	int                        (*post_srq_recv)(struct ib_srq *srq,
1240						    struct ib_recv_wr *recv_wr,
1241						    struct ib_recv_wr **bad_recv_wr);
1242	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1243						struct ib_qp_init_attr *qp_init_attr,
1244						struct ib_udata *udata);
1245	int                        (*modify_qp)(struct ib_qp *qp,
1246						struct ib_qp_attr *qp_attr,
1247						int qp_attr_mask,
1248						struct ib_udata *udata);
1249	int                        (*query_qp)(struct ib_qp *qp,
1250					       struct ib_qp_attr *qp_attr,
1251					       int qp_attr_mask,
1252					       struct ib_qp_init_attr *qp_init_attr);
1253	int                        (*destroy_qp)(struct ib_qp *qp);
1254	int                        (*post_send)(struct ib_qp *qp,
1255						struct ib_send_wr *send_wr,
1256						struct ib_send_wr **bad_send_wr);
1257	int                        (*post_recv)(struct ib_qp *qp,
1258						struct ib_recv_wr *recv_wr,
1259						struct ib_recv_wr **bad_recv_wr);
1260	struct ib_cq *             (*create_cq)(struct ib_device *device, int cqe,
1261						int comp_vector,
1262						struct ib_ucontext *context,
1263						struct ib_udata *udata);
1264	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1265						u16 cq_period);
1266	int                        (*destroy_cq)(struct ib_cq *cq);
1267	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1268						struct ib_udata *udata);
1269	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1270					      struct ib_wc *wc);
1271	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1272	int                        (*req_notify_cq)(struct ib_cq *cq,
1273						    enum ib_cq_notify_flags flags);
1274	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1275						      int wc_cnt);
1276	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1277						 int mr_access_flags);
1278	struct ib_mr *             (*reg_phys_mr)(struct ib_pd *pd,
1279						  struct ib_phys_buf *phys_buf_array,
1280						  int num_phys_buf,
1281						  int mr_access_flags,
1282						  u64 *iova_start);
1283	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1284						  u64 start, u64 length,
1285						  u64 virt_addr,
1286						  int mr_access_flags,
1287						  struct ib_udata *udata,
1288							int mr_id);
1289	int                        (*query_mr)(struct ib_mr *mr,
1290					       struct ib_mr_attr *mr_attr);
1291	int                        (*dereg_mr)(struct ib_mr *mr);
1292	struct ib_mr *		   (*alloc_fast_reg_mr)(struct ib_pd *pd,
1293					       int max_page_list_len);
1294	struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1295								   int page_list_len);
1296	void			   (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1297	int                        (*rereg_phys_mr)(struct ib_mr *mr,
1298						    int mr_rereg_mask,
1299						    struct ib_pd *pd,
1300						    struct ib_phys_buf *phys_buf_array,
1301						    int num_phys_buf,
1302						    int mr_access_flags,
1303						    u64 *iova_start);
1304	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd);
1305	int                        (*bind_mw)(struct ib_qp *qp,
1306					      struct ib_mw *mw,
1307					      struct ib_mw_bind *mw_bind);
1308	int                        (*dealloc_mw)(struct ib_mw *mw);
1309	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1310						int mr_access_flags,
1311						struct ib_fmr_attr *fmr_attr);
1312	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1313						   u64 *page_list, int list_len,
1314						   u64 iova);
1315	int		           (*unmap_fmr)(struct list_head *fmr_list);
1316	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1317	int                        (*attach_mcast)(struct ib_qp *qp,
1318						   union ib_gid *gid,
1319						   u16 lid);
1320	int                        (*detach_mcast)(struct ib_qp *qp,
1321						   union ib_gid *gid,
1322						   u16 lid);
1323	int                        (*process_mad)(struct ib_device *device,
1324						  int process_mad_flags,
1325						  u8 port_num,
1326						  struct ib_wc *in_wc,
1327						  struct ib_grh *in_grh,
1328						  struct ib_mad *in_mad,
1329						  struct ib_mad *out_mad);
1330	struct ib_srq *		   (*create_xrc_srq)(struct ib_pd *pd,
1331						     struct ib_cq *xrc_cq,
1332						     struct ib_xrcd *xrcd,
1333						     struct ib_srq_init_attr *srq_init_attr,
1334						     struct ib_udata *udata);
1335	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
1336						 struct ib_ucontext *ucontext,
1337						 struct ib_udata *udata);
1338	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1339	int			   (*create_xrc_rcv_qp)(struct ib_qp_init_attr *init_attr,
1340							u32 *qp_num);
1341	int			   (*modify_xrc_rcv_qp)(struct ib_xrcd *xrcd,
1342							u32 qp_num,
1343							struct ib_qp_attr *attr,
1344							int attr_mask);
1345	int			   (*query_xrc_rcv_qp)(struct ib_xrcd *xrcd,
1346						       u32 qp_num,
1347						       struct ib_qp_attr *attr,
1348						       int attr_mask,
1349						       struct ib_qp_init_attr *init_attr);
1350	int 			   (*reg_xrc_rcv_qp)(struct ib_xrcd *xrcd,
1351						     void *context,
1352						     u32 qp_num);
1353	int 			   (*unreg_xrc_rcv_qp)(struct ib_xrcd *xrcd,
1354						       void *context,
1355						       u32 qp_num);
1356	int                        (*attach_flow)(struct ib_qp *qp,
1357						  struct ib_flow_spec *spec,
1358						  int priority);
1359	int                        (*detach_flow)(struct ib_qp *qp,
1360						  struct ib_flow_spec *spec,
1361						  int priority);
1362
1363	unsigned long		   (*get_unmapped_area)(struct file *file,
1364					unsigned long addr,
1365					unsigned long len, unsigned long pgoff,
1366					unsigned long flags);
1367	struct ib_dma_mapping_ops   *dma_ops;
1368
1369	struct module               *owner;
1370	struct device                dev;
1371	struct kobject               *ports_parent;
1372	struct list_head             port_list;
1373
1374	enum {
1375		IB_DEV_UNINITIALIZED,
1376		IB_DEV_REGISTERED,
1377		IB_DEV_UNREGISTERED
1378	}                            reg_state;
1379
1380	int			     uverbs_abi_ver;
1381	u64			     uverbs_cmd_mask;
1382
1383	char			     node_desc[64];
1384	__be64			     node_guid;
1385	u32			     local_dma_lkey;
1386	u8                           node_type;
1387	u8                           phys_port_cnt;
1388	struct rb_root		     ib_uverbs_xrcd_table;
1389	struct mutex		     xrcd_table_mutex;
1390};
1391
1392struct ib_client {
1393	char  *name;
1394	void (*add)   (struct ib_device *);
1395	void (*remove)(struct ib_device *);
1396
1397	struct list_head list;
1398};
1399
1400struct ib_device *ib_alloc_device(size_t size);
1401void ib_dealloc_device(struct ib_device *device);
1402
1403int ib_register_device(struct ib_device *device,
1404		       int (*port_callback)(struct ib_device *,
1405					    u8, struct kobject *));
1406void ib_unregister_device(struct ib_device *device);
1407
1408int ib_register_client   (struct ib_client *client);
1409void ib_unregister_client(struct ib_client *client);
1410
1411void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1412void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
1413			 void *data);
1414
1415static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1416{
1417	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1418}
1419
1420static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1421{
1422	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1423}
1424
1425/**
1426 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1427 * contains all required attributes and no attributes not allowed for
1428 * the given QP state transition.
1429 * @cur_state: Current QP state
1430 * @next_state: Next QP state
1431 * @type: QP type
1432 * @mask: Mask of supplied QP attributes
1433 *
1434 * This function is a helper function that a low-level driver's
1435 * modify_qp method can use to validate the consumer's input.  It
1436 * checks that cur_state and next_state are valid QP states, that a
1437 * transition from cur_state to next_state is allowed by the IB spec,
1438 * and that the attribute mask supplied is allowed for the transition.
1439 */
1440int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1441		       enum ib_qp_type type, enum ib_qp_attr_mask mask);
1442
1443int ib_register_event_handler  (struct ib_event_handler *event_handler);
1444int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1445void ib_dispatch_event(struct ib_event *event);
1446
1447int ib_query_device(struct ib_device *device,
1448		    struct ib_device_attr *device_attr);
1449
1450int ib_query_port(struct ib_device *device,
1451		  u8 port_num, struct ib_port_attr *port_attr);
1452
1453enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1454					       u8 port_num);
1455
1456int ib_query_gid(struct ib_device *device,
1457		 u8 port_num, int index, union ib_gid *gid);
1458
1459int ib_query_pkey(struct ib_device *device,
1460		  u8 port_num, u16 index, u16 *pkey);
1461
1462int ib_modify_device(struct ib_device *device,
1463		     int device_modify_mask,
1464		     struct ib_device_modify *device_modify);
1465
1466int ib_modify_port(struct ib_device *device,
1467		   u8 port_num, int port_modify_mask,
1468		   struct ib_port_modify *port_modify);
1469
1470int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1471		u8 *port_num, u16 *index);
1472
1473int ib_find_pkey(struct ib_device *device,
1474		 u8 port_num, u16 pkey, u16 *index);
1475
1476/**
1477 * ib_alloc_pd - Allocates an unused protection domain.
1478 * @device: The device on which to allocate the protection domain.
1479 *
1480 * A protection domain object provides an association between QPs, shared
1481 * receive queues, address handles, memory regions, and memory windows.
1482 */
1483struct ib_pd *ib_alloc_pd(struct ib_device *device);
1484
1485/**
1486 * ib_dealloc_pd - Deallocates a protection domain.
1487 * @pd: The protection domain to deallocate.
1488 */
1489int ib_dealloc_pd(struct ib_pd *pd);
1490
1491/**
1492 * ib_create_ah - Creates an address handle for the given address vector.
1493 * @pd: The protection domain associated with the address handle.
1494 * @ah_attr: The attributes of the address vector.
1495 *
1496 * The address handle is used to reference a local or global destination
1497 * in all UD QP post sends.
1498 */
1499struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
1500
1501/**
1502 * ib_init_ah_from_wc - Initializes address handle attributes from a
1503 *   work completion.
1504 * @device: Device on which the received message arrived.
1505 * @port_num: Port on which the received message arrived.
1506 * @wc: Work completion associated with the received message.
1507 * @grh: References the received global route header.  This parameter is
1508 *   ignored unless the work completion indicates that the GRH is valid.
1509 * @ah_attr: Returned attributes that can be used when creating an address
1510 *   handle for replying to the message.
1511 */
1512int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
1513		       struct ib_grh *grh, struct ib_ah_attr *ah_attr);
1514
1515/**
1516 * ib_create_ah_from_wc - Creates an address handle associated with the
1517 *   sender of the specified work completion.
1518 * @pd: The protection domain associated with the address handle.
1519 * @wc: Work completion information associated with a received message.
1520 * @grh: References the received global route header.  This parameter is
1521 *   ignored unless the work completion indicates that the GRH is valid.
1522 * @port_num: The outbound port number to associate with the address.
1523 *
1524 * The address handle is used to reference a local or global destination
1525 * in all UD QP post sends.
1526 */
1527struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
1528				   struct ib_grh *grh, u8 port_num);
1529
1530/**
1531 * ib_modify_ah - Modifies the address vector associated with an address
1532 *   handle.
1533 * @ah: The address handle to modify.
1534 * @ah_attr: The new address vector attributes to associate with the
1535 *   address handle.
1536 */
1537int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1538
1539/**
1540 * ib_query_ah - Queries the address vector associated with an address
1541 *   handle.
1542 * @ah: The address handle to query.
1543 * @ah_attr: The address vector attributes associated with the address
1544 *   handle.
1545 */
1546int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1547
1548/**
1549 * ib_destroy_ah - Destroys an address handle.
1550 * @ah: The address handle to destroy.
1551 */
1552int ib_destroy_ah(struct ib_ah *ah);
1553
1554/**
1555 * ib_create_xrc_srq - Creates an XRC SRQ associated with the specified
1556 *   protection domain, cq, and xrc domain.
1557 * @pd: The protection domain associated with the SRQ.
1558 * @xrc_cq: The cq to be associated with the XRC SRQ.
1559 * @xrcd: The XRC domain to be associated with the XRC SRQ.
1560 * @srq_init_attr: A list of initial attributes required to create the
1561 *   XRC SRQ.  If XRC SRQ creation succeeds, then the attributes are updated
1562 *   to the actual capabilities of the created XRC SRQ.
1563 *
1564 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1565 * requested size of the XRC SRQ, and set to the actual values allocated
1566 * on return.  If ib_create_xrc_srq() succeeds, then max_wr and max_sge
1567 * will always be at least as large as the requested values.
1568 */
1569struct ib_srq *ib_create_xrc_srq(struct ib_pd *pd,
1570				 struct ib_cq *xrc_cq,
1571				 struct ib_xrcd *xrcd,
1572				 struct ib_srq_init_attr *srq_init_attr);
1573
1574/**
1575 * ib_create_srq - Creates a SRQ associated with the specified protection
1576 *   domain.
1577 * @pd: The protection domain associated with the SRQ.
1578 * @srq_init_attr: A list of initial attributes required to create the
1579 *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
1580 *   the actual capabilities of the created SRQ.
1581 *
1582 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1583 * requested size of the SRQ, and set to the actual values allocated
1584 * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
1585 * will always be at least as large as the requested values.
1586 */
1587struct ib_srq *ib_create_srq(struct ib_pd *pd,
1588			     struct ib_srq_init_attr *srq_init_attr);
1589
1590/**
1591 * ib_modify_srq - Modifies the attributes for the specified SRQ.
1592 * @srq: The SRQ to modify.
1593 * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
1594 *   the current values of selected SRQ attributes are returned.
1595 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1596 *   are being modified.
1597 *
1598 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1599 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1600 * the number of receives queued drops below the limit.
1601 */
1602int ib_modify_srq(struct ib_srq *srq,
1603		  struct ib_srq_attr *srq_attr,
1604		  enum ib_srq_attr_mask srq_attr_mask);
1605
1606/**
1607 * ib_query_srq - Returns the attribute list and current values for the
1608 *   specified SRQ.
1609 * @srq: The SRQ to query.
1610 * @srq_attr: The attributes of the specified SRQ.
1611 */
1612int ib_query_srq(struct ib_srq *srq,
1613		 struct ib_srq_attr *srq_attr);
1614
1615/**
1616 * ib_destroy_srq - Destroys the specified SRQ.
1617 * @srq: The SRQ to destroy.
1618 */
1619int ib_destroy_srq(struct ib_srq *srq);
1620
1621/**
1622 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1623 * @srq: The SRQ to post the work request on.
1624 * @recv_wr: A list of work requests to post on the receive queue.
1625 * @bad_recv_wr: On an immediate failure, this parameter will reference
1626 *   the work request that failed to be posted on the QP.
1627 */
1628static inline int ib_post_srq_recv(struct ib_srq *srq,
1629				   struct ib_recv_wr *recv_wr,
1630				   struct ib_recv_wr **bad_recv_wr)
1631{
1632	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1633}
1634
1635/**
1636 * ib_create_qp - Creates a QP associated with the specified protection
1637 *   domain.
1638 * @pd: The protection domain associated with the QP.
1639 * @qp_init_attr: A list of initial attributes required to create the
1640 *   QP.  If QP creation succeeds, then the attributes are updated to
1641 *   the actual capabilities of the created QP.
1642 */
1643struct ib_qp *ib_create_qp(struct ib_pd *pd,
1644			   struct ib_qp_init_attr *qp_init_attr);
1645
1646/**
1647 * ib_modify_qp - Modifies the attributes for the specified QP and then
1648 *   transitions the QP to the given state.
1649 * @qp: The QP to modify.
1650 * @qp_attr: On input, specifies the QP attributes to modify.  On output,
1651 *   the current values of selected QP attributes are returned.
1652 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1653 *   are being modified.
1654 */
1655int ib_modify_qp(struct ib_qp *qp,
1656		 struct ib_qp_attr *qp_attr,
1657		 int qp_attr_mask);
1658
1659/**
1660 * ib_query_qp - Returns the attribute list and current values for the
1661 *   specified QP.
1662 * @qp: The QP to query.
1663 * @qp_attr: The attributes of the specified QP.
1664 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1665 * @qp_init_attr: Additional attributes of the selected QP.
1666 *
1667 * The qp_attr_mask may be used to limit the query to gathering only the
1668 * selected attributes.
1669 */
1670int ib_query_qp(struct ib_qp *qp,
1671		struct ib_qp_attr *qp_attr,
1672		int qp_attr_mask,
1673		struct ib_qp_init_attr *qp_init_attr);
1674
1675/**
1676 * ib_destroy_qp - Destroys the specified QP.
1677 * @qp: The QP to destroy.
1678 */
1679int ib_destroy_qp(struct ib_qp *qp);
1680
1681/**
1682 * ib_open_qp - Obtain a reference to an existing sharable QP.
1683 * @xrcd - XRC domain
1684 * @qp_open_attr: Attributes identifying the QP to open.
1685 *
1686 * Returns a reference to a sharable QP.
1687 */
1688struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
1689			 struct ib_qp_open_attr *qp_open_attr);
1690
1691/**
1692 * ib_close_qp - Release an external reference to a QP.
1693 * @qp: The QP handle to release
1694 *
1695 * The opened QP handle is released by the caller.  The underlying
1696 * shared QP is not destroyed until all internal references are released.
1697 */
1698int ib_close_qp(struct ib_qp *qp);
1699
1700/**
1701 * ib_post_send - Posts a list of work requests to the send queue of
1702 *   the specified QP.
1703 * @qp: The QP to post the work request on.
1704 * @send_wr: A list of work requests to post on the send queue.
1705 * @bad_send_wr: On an immediate failure, this parameter will reference
1706 *   the work request that failed to be posted on the QP.
1707 *
1708 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
1709 * error is returned, the QP state shall not be affected,
1710 * ib_post_send() will return an immediate error after queueing any
1711 * earlier work requests in the list.
1712 */
1713static inline int ib_post_send(struct ib_qp *qp,
1714			       struct ib_send_wr *send_wr,
1715			       struct ib_send_wr **bad_send_wr)
1716{
1717	return qp->device->post_send(qp, send_wr, bad_send_wr);
1718}
1719
1720/**
1721 * ib_post_recv - Posts a list of work requests to the receive queue of
1722 *   the specified QP.
1723 * @qp: The QP to post the work request on.
1724 * @recv_wr: A list of work requests to post on the receive queue.
1725 * @bad_recv_wr: On an immediate failure, this parameter will reference
1726 *   the work request that failed to be posted on the QP.
1727 */
1728static inline int ib_post_recv(struct ib_qp *qp,
1729			       struct ib_recv_wr *recv_wr,
1730			       struct ib_recv_wr **bad_recv_wr)
1731{
1732	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1733}
1734
1735/*
1736 * IB_CQ_VECTOR_LEAST_ATTACHED: The constant specifies that
1737 *	the CQ will be attached to the completion vector that has
1738 *	the least number of CQs already attached to it.
1739 */
1740#define IB_CQ_VECTOR_LEAST_ATTACHED	0xffffffff
1741
1742/**
1743 * ib_create_cq - Creates a CQ on the specified device.
1744 * @device: The device on which to create the CQ.
1745 * @comp_handler: A user-specified callback that is invoked when a
1746 *   completion event occurs on the CQ.
1747 * @event_handler: A user-specified callback that is invoked when an
1748 *   asynchronous event not associated with a completion occurs on the CQ.
1749 * @cq_context: Context associated with the CQ returned to the user via
1750 *   the associated completion and event handlers.
1751 * @cqe: The minimum size of the CQ.
1752 * @comp_vector - Completion vector used to signal completion events.
1753 *     Must be >= 0 and < context->num_comp_vectors.
1754 *
1755 * Users can examine the cq structure to determine the actual CQ size.
1756 */
1757struct ib_cq *ib_create_cq(struct ib_device *device,
1758			   ib_comp_handler comp_handler,
1759			   void (*event_handler)(struct ib_event *, void *),
1760			   void *cq_context, int cqe, int comp_vector);
1761
1762/**
1763 * ib_resize_cq - Modifies the capacity of the CQ.
1764 * @cq: The CQ to resize.
1765 * @cqe: The minimum size of the CQ.
1766 *
1767 * Users can examine the cq structure to determine the actual CQ size.
1768 */
1769int ib_resize_cq(struct ib_cq *cq, int cqe);
1770
1771/**
1772 * ib_modify_cq - Modifies moderation params of the CQ
1773 * @cq: The CQ to modify.
1774 * @cq_count: number of CQEs that will trigger an event
1775 * @cq_period: max period of time in usec before triggering an event
1776 *
1777 */
1778int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1779
1780/**
1781 * ib_destroy_cq - Destroys the specified CQ.
1782 * @cq: The CQ to destroy.
1783 */
1784int ib_destroy_cq(struct ib_cq *cq);
1785
1786/**
1787 * ib_poll_cq - poll a CQ for completion(s)
1788 * @cq:the CQ being polled
1789 * @num_entries:maximum number of completions to return
1790 * @wc:array of at least @num_entries &struct ib_wc where completions
1791 *   will be returned
1792 *
1793 * Poll a CQ for (possibly multiple) completions.  If the return value
1794 * is < 0, an error occurred.  If the return value is >= 0, it is the
1795 * number of completions returned.  If the return value is
1796 * non-negative and < num_entries, then the CQ was emptied.
1797 */
1798static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1799			     struct ib_wc *wc)
1800{
1801	return cq->device->poll_cq(cq, num_entries, wc);
1802}
1803
1804/**
1805 * ib_peek_cq - Returns the number of unreaped completions currently
1806 *   on the specified CQ.
1807 * @cq: The CQ to peek.
1808 * @wc_cnt: A minimum number of unreaped completions to check for.
1809 *
1810 * If the number of unreaped completions is greater than or equal to wc_cnt,
1811 * this function returns wc_cnt, otherwise, it returns the actual number of
1812 * unreaped completions.
1813 */
1814int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1815
1816/**
1817 * ib_req_notify_cq - Request completion notification on a CQ.
1818 * @cq: The CQ to generate an event for.
1819 * @flags:
1820 *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1821 *   to request an event on the next solicited event or next work
1822 *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1823 *   may also be |ed in to request a hint about missed events, as
1824 *   described below.
1825 *
1826 * Return Value:
1827 *    < 0 means an error occurred while requesting notification
1828 *   == 0 means notification was requested successfully, and if
1829 *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1830 *        were missed and it is safe to wait for another event.  In
1831 *        this case is it guaranteed that any work completions added
1832 *        to the CQ since the last CQ poll will trigger a completion
1833 *        notification event.
1834 *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1835 *        in.  It means that the consumer must poll the CQ again to
1836 *        make sure it is empty to avoid missing an event because of a
1837 *        race between requesting notification and an entry being
1838 *        added to the CQ.  This return value means it is possible
1839 *        (but not guaranteed) that a work completion has been added
1840 *        to the CQ since the last poll without triggering a
1841 *        completion notification event.
1842 */
1843static inline int ib_req_notify_cq(struct ib_cq *cq,
1844				   enum ib_cq_notify_flags flags)
1845{
1846	return cq->device->req_notify_cq(cq, flags);
1847}
1848
1849/**
1850 * ib_req_ncomp_notif - Request completion notification when there are
1851 *   at least the specified number of unreaped completions on the CQ.
1852 * @cq: The CQ to generate an event for.
1853 * @wc_cnt: The number of unreaped completions that should be on the
1854 *   CQ before an event is generated.
1855 */
1856static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1857{
1858	return cq->device->req_ncomp_notif ?
1859		cq->device->req_ncomp_notif(cq, wc_cnt) :
1860		-ENOSYS;
1861}
1862
1863/**
1864 * ib_get_dma_mr - Returns a memory region for system memory that is
1865 *   usable for DMA.
1866 * @pd: The protection domain associated with the memory region.
1867 * @mr_access_flags: Specifies the memory access rights.
1868 *
1869 * Note that the ib_dma_*() functions defined below must be used
1870 * to create/destroy addresses used with the Lkey or Rkey returned
1871 * by ib_get_dma_mr().
1872 */
1873struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1874
1875/**
1876 * ib_dma_mapping_error - check a DMA addr for error
1877 * @dev: The device for which the dma_addr was created
1878 * @dma_addr: The DMA address to check
1879 */
1880static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1881{
1882	if (dev->dma_ops)
1883		return dev->dma_ops->mapping_error(dev, dma_addr);
1884	return dma_mapping_error(dev->dma_device, dma_addr);
1885}
1886
1887/**
1888 * ib_dma_map_single - Map a kernel virtual address to DMA address
1889 * @dev: The device for which the dma_addr is to be created
1890 * @cpu_addr: The kernel virtual address
1891 * @size: The size of the region in bytes
1892 * @direction: The direction of the DMA
1893 */
1894static inline u64 ib_dma_map_single(struct ib_device *dev,
1895				    void *cpu_addr, size_t size,
1896				    enum dma_data_direction direction)
1897{
1898	if (dev->dma_ops)
1899		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1900	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1901}
1902
1903/**
1904 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1905 * @dev: The device for which the DMA address was created
1906 * @addr: The DMA address
1907 * @size: The size of the region in bytes
1908 * @direction: The direction of the DMA
1909 */
1910static inline void ib_dma_unmap_single(struct ib_device *dev,
1911				       u64 addr, size_t size,
1912				       enum dma_data_direction direction)
1913{
1914	if (dev->dma_ops)
1915		dev->dma_ops->unmap_single(dev, addr, size, direction);
1916	else
1917		dma_unmap_single(dev->dma_device, addr, size, direction);
1918}
1919
1920static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
1921					  void *cpu_addr, size_t size,
1922					  enum dma_data_direction direction,
1923					  struct dma_attrs *attrs)
1924{
1925	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
1926				    direction, attrs);
1927}
1928
1929static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
1930					     u64 addr, size_t size,
1931					     enum dma_data_direction direction,
1932					     struct dma_attrs *attrs)
1933{
1934	return dma_unmap_single_attrs(dev->dma_device, addr, size,
1935				      direction, attrs);
1936}
1937
1938/**
1939 * ib_dma_map_page - Map a physical page to DMA address
1940 * @dev: The device for which the dma_addr is to be created
1941 * @page: The page to be mapped
1942 * @offset: The offset within the page
1943 * @size: The size of the region in bytes
1944 * @direction: The direction of the DMA
1945 */
1946static inline u64 ib_dma_map_page(struct ib_device *dev,
1947				  struct page *page,
1948				  unsigned long offset,
1949				  size_t size,
1950					 enum dma_data_direction direction)
1951{
1952	if (dev->dma_ops)
1953		return dev->dma_ops->map_page(dev, page, offset, size, direction);
1954	return dma_map_page(dev->dma_device, page, offset, size, direction);
1955}
1956
1957/**
1958 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1959 * @dev: The device for which the DMA address was created
1960 * @addr: The DMA address
1961 * @size: The size of the region in bytes
1962 * @direction: The direction of the DMA
1963 */
1964static inline void ib_dma_unmap_page(struct ib_device *dev,
1965				     u64 addr, size_t size,
1966				     enum dma_data_direction direction)
1967{
1968	if (dev->dma_ops)
1969		dev->dma_ops->unmap_page(dev, addr, size, direction);
1970	else
1971		dma_unmap_page(dev->dma_device, addr, size, direction);
1972}
1973
1974/**
1975 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
1976 * @dev: The device for which the DMA addresses are to be created
1977 * @sg: The array of scatter/gather entries
1978 * @nents: The number of scatter/gather entries
1979 * @direction: The direction of the DMA
1980 */
1981static inline int ib_dma_map_sg(struct ib_device *dev,
1982				struct scatterlist *sg, int nents,
1983				enum dma_data_direction direction)
1984{
1985	if (dev->dma_ops)
1986		return dev->dma_ops->map_sg(dev, sg, nents, direction);
1987	return dma_map_sg(dev->dma_device, sg, nents, direction);
1988}
1989
1990/**
1991 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
1992 * @dev: The device for which the DMA addresses were created
1993 * @sg: The array of scatter/gather entries
1994 * @nents: The number of scatter/gather entries
1995 * @direction: The direction of the DMA
1996 */
1997static inline void ib_dma_unmap_sg(struct ib_device *dev,
1998				   struct scatterlist *sg, int nents,
1999				   enum dma_data_direction direction)
2000{
2001	if (dev->dma_ops)
2002		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2003	else
2004		dma_unmap_sg(dev->dma_device, sg, nents, direction);
2005}
2006
2007static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2008				      struct scatterlist *sg, int nents,
2009				      enum dma_data_direction direction,
2010				      struct dma_attrs *attrs)
2011{
2012	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2013}
2014
2015static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2016					 struct scatterlist *sg, int nents,
2017					 enum dma_data_direction direction,
2018					 struct dma_attrs *attrs)
2019{
2020	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2021}
2022/**
2023 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2024 * @dev: The device for which the DMA addresses were created
2025 * @sg: The scatter/gather entry
2026 */
2027static inline u64 ib_sg_dma_address(struct ib_device *dev,
2028				    struct scatterlist *sg)
2029{
2030	if (dev->dma_ops)
2031		return dev->dma_ops->dma_address(dev, sg);
2032	return sg_dma_address(sg);
2033}
2034
2035/**
2036 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2037 * @dev: The device for which the DMA addresses were created
2038 * @sg: The scatter/gather entry
2039 */
2040static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2041					 struct scatterlist *sg)
2042{
2043	if (dev->dma_ops)
2044		return dev->dma_ops->dma_len(dev, sg);
2045	return sg_dma_len(sg);
2046}
2047
2048/**
2049 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2050 * @dev: The device for which the DMA address was created
2051 * @addr: The DMA address
2052 * @size: The size of the region in bytes
2053 * @dir: The direction of the DMA
2054 */
2055static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2056					      u64 addr,
2057					      size_t size,
2058					      enum dma_data_direction dir)
2059{
2060	if (dev->dma_ops)
2061		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2062	else
2063		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2064}
2065
2066/**
2067 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2068 * @dev: The device for which the DMA address was created
2069 * @addr: The DMA address
2070 * @size: The size of the region in bytes
2071 * @dir: The direction of the DMA
2072 */
2073static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2074						 u64 addr,
2075						 size_t size,
2076						 enum dma_data_direction dir)
2077{
2078	if (dev->dma_ops)
2079		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2080	else
2081		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2082}
2083
2084/**
2085 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2086 * @dev: The device for which the DMA address is requested
2087 * @size: The size of the region to allocate in bytes
2088 * @dma_handle: A pointer for returning the DMA address of the region
2089 * @flag: memory allocator flags
2090 */
2091static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2092					   size_t size,
2093					   u64 *dma_handle,
2094					   gfp_t flag)
2095{
2096	if (dev->dma_ops)
2097		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2098	else {
2099		dma_addr_t handle;
2100		void *ret;
2101
2102		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2103		*dma_handle = handle;
2104		return ret;
2105	}
2106}
2107
2108/**
2109 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2110 * @dev: The device for which the DMA addresses were allocated
2111 * @size: The size of the region
2112 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2113 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2114 */
2115static inline void ib_dma_free_coherent(struct ib_device *dev,
2116					size_t size, void *cpu_addr,
2117					u64 dma_handle)
2118{
2119	if (dev->dma_ops)
2120		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2121	else
2122		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2123}
2124
2125/**
2126 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
2127 *   by an HCA.
2128 * @pd: The protection domain associated assigned to the registered region.
2129 * @phys_buf_array: Specifies a list of physical buffers to use in the
2130 *   memory region.
2131 * @num_phys_buf: Specifies the size of the phys_buf_array.
2132 * @mr_access_flags: Specifies the memory access rights.
2133 * @iova_start: The offset of the region's starting I/O virtual address.
2134 */
2135struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
2136			     struct ib_phys_buf *phys_buf_array,
2137			     int num_phys_buf,
2138			     int mr_access_flags,
2139			     u64 *iova_start);
2140
2141/**
2142 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
2143 *   Conceptually, this call performs the functions deregister memory region
2144 *   followed by register physical memory region.  Where possible,
2145 *   resources are reused instead of deallocated and reallocated.
2146 * @mr: The memory region to modify.
2147 * @mr_rereg_mask: A bit-mask used to indicate which of the following
2148 *   properties of the memory region are being modified.
2149 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
2150 *   the new protection domain to associated with the memory region,
2151 *   otherwise, this parameter is ignored.
2152 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2153 *   field specifies a list of physical buffers to use in the new
2154 *   translation, otherwise, this parameter is ignored.
2155 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2156 *   field specifies the size of the phys_buf_array, otherwise, this
2157 *   parameter is ignored.
2158 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
2159 *   field specifies the new memory access rights, otherwise, this
2160 *   parameter is ignored.
2161 * @iova_start: The offset of the region's starting I/O virtual address.
2162 */
2163int ib_rereg_phys_mr(struct ib_mr *mr,
2164		     int mr_rereg_mask,
2165		     struct ib_pd *pd,
2166		     struct ib_phys_buf *phys_buf_array,
2167		     int num_phys_buf,
2168		     int mr_access_flags,
2169		     u64 *iova_start);
2170
2171/**
2172 * ib_query_mr - Retrieves information about a specific memory region.
2173 * @mr: The memory region to retrieve information about.
2174 * @mr_attr: The attributes of the specified memory region.
2175 */
2176int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2177
2178/**
2179 * ib_dereg_mr - Deregisters a memory region and removes it from the
2180 *   HCA translation table.
2181 * @mr: The memory region to deregister.
2182 */
2183int ib_dereg_mr(struct ib_mr *mr);
2184
2185/**
2186 * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2187 *   IB_WR_FAST_REG_MR send work request.
2188 * @pd: The protection domain associated with the region.
2189 * @max_page_list_len: requested max physical buffer list length to be
2190 *   used with fast register work requests for this MR.
2191 */
2192struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
2193
2194/**
2195 * ib_alloc_fast_reg_page_list - Allocates a page list array
2196 * @device - ib device pointer.
2197 * @page_list_len - size of the page list array to be allocated.
2198 *
2199 * This allocates and returns a struct ib_fast_reg_page_list * and a
2200 * page_list array that is at least page_list_len in size.  The actual
2201 * size is returned in max_page_list_len.  The caller is responsible
2202 * for initializing the contents of the page_list array before posting
2203 * a send work request with the IB_WC_FAST_REG_MR opcode.
2204 *
2205 * The page_list array entries must be translated using one of the
2206 * ib_dma_*() functions just like the addresses passed to
2207 * ib_map_phys_fmr().  Once the ib_post_send() is issued, the struct
2208 * ib_fast_reg_page_list must not be modified by the caller until the
2209 * IB_WC_FAST_REG_MR work request completes.
2210 */
2211struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2212				struct ib_device *device, int page_list_len);
2213
2214/**
2215 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2216 *   page list array.
2217 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2218 */
2219void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2220
2221/**
2222 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2223 *   R_Key and L_Key.
2224 * @mr - struct ib_mr pointer to be updated.
2225 * @newkey - new key to be used.
2226 */
2227static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2228{
2229	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2230	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2231}
2232
2233/**
2234 * ib_alloc_mw - Allocates a memory window.
2235 * @pd: The protection domain associated with the memory window.
2236 */
2237struct ib_mw *ib_alloc_mw(struct ib_pd *pd);
2238
2239/**
2240 * ib_bind_mw - Posts a work request to the send queue of the specified
2241 *   QP, which binds the memory window to the given address range and
2242 *   remote access attributes.
2243 * @qp: QP to post the bind work request on.
2244 * @mw: The memory window to bind.
2245 * @mw_bind: Specifies information about the memory window, including
2246 *   its address range, remote access rights, and associated memory region.
2247 */
2248static inline int ib_bind_mw(struct ib_qp *qp,
2249			     struct ib_mw *mw,
2250			     struct ib_mw_bind *mw_bind)
2251{
2252	/* XXX reference counting in corresponding MR? */
2253	return mw->device->bind_mw ?
2254		mw->device->bind_mw(qp, mw, mw_bind) :
2255		-ENOSYS;
2256}
2257
2258/**
2259 * ib_dealloc_mw - Deallocates a memory window.
2260 * @mw: The memory window to deallocate.
2261 */
2262int ib_dealloc_mw(struct ib_mw *mw);
2263
2264/**
2265 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2266 * @pd: The protection domain associated with the unmapped region.
2267 * @mr_access_flags: Specifies the memory access rights.
2268 * @fmr_attr: Attributes of the unmapped region.
2269 *
2270 * A fast memory region must be mapped before it can be used as part of
2271 * a work request.
2272 */
2273struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2274			    int mr_access_flags,
2275			    struct ib_fmr_attr *fmr_attr);
2276
2277/**
2278 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2279 * @fmr: The fast memory region to associate with the pages.
2280 * @page_list: An array of physical pages to map to the fast memory region.
2281 * @list_len: The number of pages in page_list.
2282 * @iova: The I/O virtual address to use with the mapped region.
2283 */
2284static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2285				  u64 *page_list, int list_len,
2286				  u64 iova)
2287{
2288	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2289}
2290
2291/**
2292 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2293 * @fmr_list: A linked list of fast memory regions to unmap.
2294 */
2295int ib_unmap_fmr(struct list_head *fmr_list);
2296
2297/**
2298 * ib_dealloc_fmr - Deallocates a fast memory region.
2299 * @fmr: The fast memory region to deallocate.
2300 */
2301int ib_dealloc_fmr(struct ib_fmr *fmr);
2302
2303/**
2304 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2305 * @qp: QP to attach to the multicast group.  The QP must be type
2306 *   IB_QPT_UD.
2307 * @gid: Multicast group GID.
2308 * @lid: Multicast group LID in host byte order.
2309 *
2310 * In order to send and receive multicast packets, subnet
2311 * administration must have created the multicast group and configured
2312 * the fabric appropriately.  The port associated with the specified
2313 * QP must also be a member of the multicast group.
2314 */
2315int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2316
2317/**
2318 * ib_detach_mcast - Detaches the specified QP from a multicast group.
2319 * @qp: QP to detach from the multicast group.
2320 * @gid: Multicast group GID.
2321 * @lid: Multicast group LID in host byte order.
2322 */
2323int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2324
2325/**
2326 * ib_alloc_xrcd - Allocates an XRC domain.
2327 * @device: The device on which to allocate the XRC domain.
2328 */
2329struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2330
2331/**
2332 * ib_dealloc_xrcd - Deallocates an XRC domain.
2333 * @xrcd: The XRC domain to deallocate.
2334 */
2335int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2336
2337int ib_attach_flow(struct ib_qp *qp, struct ib_flow_spec *spec, int priority);
2338int ib_detach_flow(struct ib_qp *qp, struct ib_flow_spec *spec, int priority);
2339
2340#endif /* IB_VERBS_H */
2341