1/*
2 * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
4 * Copyright (c) 2004 Intel Corporation.  All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
9 *
10 * This software is available to you under a choice of one of two
11 * licenses.  You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
15 *
16 *     Redistribution and use in source and binary forms, with or
17 *     without modification, are permitted provided that the following
18 *     conditions are met:
19 *
20 *      - Redistributions of source code must retain the above
21 *        copyright notice, this list of conditions and the following
22 *        disclaimer.
23 *
24 *      - Redistributions in binary form must reproduce the above
25 *        copyright notice, this list of conditions and the following
26 *        disclaimer in the documentation and/or other materials
27 *        provided with the distribution.
28 *
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 * SOFTWARE.
37 */
38
39#if !defined(IB_VERBS_H)
40#define IB_VERBS_H
41
42#include <linux/types.h>
43#include <linux/device.h>
44#include <linux/mm.h>
45#include <linux/dma-mapping.h>
46#include <linux/kref.h>
47#include <linux/list.h>
48#include <linux/rwsem.h>
49#include <linux/scatterlist.h>
50#include <linux/workqueue.h>
51
52#include <linux/atomic.h>
53#include <asm/uaccess.h>
54#include <linux/rbtree.h>
55#include <linux/mutex.h>
56
57extern struct workqueue_struct *ib_wq;
58
59union ib_gid {
60	u8	raw[16];
61	struct {
62		__be64	subnet_prefix;
63		__be64	interface_id;
64	} global;
65};
66
67enum rdma_node_type {
68	/* IB values map to NodeInfo:NodeType. */
69	RDMA_NODE_IB_CA 	= 1,
70	RDMA_NODE_IB_SWITCH,
71	RDMA_NODE_IB_ROUTER,
72	RDMA_NODE_RNIC
73};
74
75enum rdma_transport_type {
76	RDMA_TRANSPORT_IB,
77	RDMA_TRANSPORT_IWARP
78};
79
80enum rdma_transport_type
81rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__;
82
83enum rdma_link_layer {
84	IB_LINK_LAYER_UNSPECIFIED,
85	IB_LINK_LAYER_INFINIBAND,
86	IB_LINK_LAYER_ETHERNET,
87};
88
89enum ib_device_cap_flags {
90	IB_DEVICE_RESIZE_MAX_WR		= 1,
91	IB_DEVICE_BAD_PKEY_CNTR		= (1<<1),
92	IB_DEVICE_BAD_QKEY_CNTR		= (1<<2),
93	IB_DEVICE_RAW_MULTI		= (1<<3),
94	IB_DEVICE_AUTO_PATH_MIG		= (1<<4),
95	IB_DEVICE_CHANGE_PHY_PORT	= (1<<5),
96	IB_DEVICE_UD_AV_PORT_ENFORCE	= (1<<6),
97	IB_DEVICE_CURR_QP_STATE_MOD	= (1<<7),
98	IB_DEVICE_SHUTDOWN_PORT		= (1<<8),
99	IB_DEVICE_INIT_TYPE		= (1<<9),
100	IB_DEVICE_PORT_ACTIVE_EVENT	= (1<<10),
101	IB_DEVICE_SYS_IMAGE_GUID	= (1<<11),
102	IB_DEVICE_RC_RNR_NAK_GEN	= (1<<12),
103	IB_DEVICE_SRQ_RESIZE		= (1<<13),
104	IB_DEVICE_N_NOTIFY_CQ		= (1<<14),
105	IB_DEVICE_LOCAL_DMA_LKEY	= (1<<15),
106	IB_DEVICE_RESERVED		= (1<<16), /* old SEND_W_INV */
107	IB_DEVICE_MEM_WINDOW		= (1<<17),
108	/*
109	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
110	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
111	 * messages and can verify the validity of checksum for
112	 * incoming messages.  Setting this flag implies that the
113	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
114	 */
115	IB_DEVICE_UD_IP_CSUM		= (1<<18),
116	IB_DEVICE_UD_TSO		= (1<<19),
117	IB_DEVICE_XRC			= (1<<20),
118	IB_DEVICE_MEM_MGT_EXTENSIONS	= (1<<21),
119	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
120	IB_DEVICE_MR_ALLOCATE		= (1<<23),
121	IB_DEVICE_SHARED_MR             = (1<<24),
122	IB_DEVICE_QPG			= (1<<25),
123	IB_DEVICE_UD_RSS		= (1<<26),
124	IB_DEVICE_UD_TSS		= (1<<27)
125};
126
127enum ib_atomic_cap {
128	IB_ATOMIC_NONE,
129	IB_ATOMIC_HCA,
130	IB_ATOMIC_GLOB
131};
132
133struct ib_device_attr {
134	u64			fw_ver;
135	__be64			sys_image_guid;
136	u64			max_mr_size;
137	u64			page_size_cap;
138	u32			vendor_id;
139	u32			vendor_part_id;
140	u32			hw_ver;
141	int			max_qp;
142	int			max_qp_wr;
143	int			device_cap_flags;
144	int			max_sge;
145	int			max_sge_rd;
146	int			max_cq;
147	int			max_cqe;
148	int			max_mr;
149	int			max_pd;
150	int			max_qp_rd_atom;
151	int			max_ee_rd_atom;
152	int			max_res_rd_atom;
153	int			max_qp_init_rd_atom;
154	int			max_ee_init_rd_atom;
155	enum ib_atomic_cap	atomic_cap;
156	enum ib_atomic_cap	masked_atomic_cap;
157	int			max_ee;
158	int			max_rdd;
159	int			max_mw;
160	int			max_raw_ipv6_qp;
161	int			max_raw_ethy_qp;
162	int			max_mcast_grp;
163	int			max_mcast_qp_attach;
164	int			max_total_mcast_qp_attach;
165	int			max_ah;
166	int			max_fmr;
167	int			max_map_per_fmr;
168	int			max_srq;
169	int			max_srq_wr;
170	int			max_srq_sge;
171	unsigned int		max_fast_reg_page_list_len;
172	int			max_rss_tbl_sz;
173	u16			max_pkeys;
174	u8			local_ca_ack_delay;
175};
176
177enum ib_mtu {
178	IB_MTU_256  = 1,
179	IB_MTU_512  = 2,
180	IB_MTU_1024 = 3,
181	IB_MTU_2048 = 4,
182	IB_MTU_4096 = 5
183};
184
185static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
186{
187	switch (mtu) {
188	case IB_MTU_256:  return  256;
189	case IB_MTU_512:  return  512;
190	case IB_MTU_1024: return 1024;
191	case IB_MTU_2048: return 2048;
192	case IB_MTU_4096: return 4096;
193	default: 	  return -1;
194	}
195}
196
197enum ib_port_state {
198	IB_PORT_NOP		= 0,
199	IB_PORT_DOWN		= 1,
200	IB_PORT_INIT		= 2,
201	IB_PORT_ARMED		= 3,
202	IB_PORT_ACTIVE		= 4,
203	IB_PORT_ACTIVE_DEFER	= 5
204};
205
206enum ib_port_cap_flags {
207	IB_PORT_SM				= 1 <<  1,
208	IB_PORT_NOTICE_SUP			= 1 <<  2,
209	IB_PORT_TRAP_SUP			= 1 <<  3,
210	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
211	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
212	IB_PORT_SL_MAP_SUP			= 1 <<  6,
213	IB_PORT_MKEY_NVRAM			= 1 <<  7,
214	IB_PORT_PKEY_NVRAM			= 1 <<  8,
215	IB_PORT_LED_INFO_SUP			= 1 <<  9,
216	IB_PORT_SM_DISABLED			= 1 << 10,
217	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
218	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
219	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
220	IB_PORT_CM_SUP				= 1 << 16,
221	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
222	IB_PORT_REINIT_SUP			= 1 << 18,
223	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
224	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
225	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
226	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
227	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
228	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
229	IB_PORT_CLIENT_REG_SUP			= 1 << 25
230};
231
232enum ib_port_width {
233	IB_WIDTH_1X	= 1,
234	IB_WIDTH_4X	= 2,
235	IB_WIDTH_8X	= 4,
236	IB_WIDTH_12X	= 8
237};
238
239static inline int ib_width_enum_to_int(enum ib_port_width width)
240{
241	switch (width) {
242	case IB_WIDTH_1X:  return  1;
243	case IB_WIDTH_4X:  return  4;
244	case IB_WIDTH_8X:  return  8;
245	case IB_WIDTH_12X: return 12;
246	default: 	  return -1;
247	}
248}
249
250enum ib_port_speed {
251	IB_SPEED_SDR	= 1,
252	IB_SPEED_DDR	= 2,
253	IB_SPEED_QDR	= 4,
254	IB_SPEED_FDR10	= 8,
255	IB_SPEED_FDR	= 16,
256	IB_SPEED_EDR	= 32
257};
258
259struct ib_protocol_stats {
260	/* TBD... */
261};
262
263struct iw_protocol_stats {
264	u64	ipInReceives;
265	u64	ipInHdrErrors;
266	u64	ipInTooBigErrors;
267	u64	ipInNoRoutes;
268	u64	ipInAddrErrors;
269	u64	ipInUnknownProtos;
270	u64	ipInTruncatedPkts;
271	u64	ipInDiscards;
272	u64	ipInDelivers;
273	u64	ipOutForwDatagrams;
274	u64	ipOutRequests;
275	u64	ipOutDiscards;
276	u64	ipOutNoRoutes;
277	u64	ipReasmTimeout;
278	u64	ipReasmReqds;
279	u64	ipReasmOKs;
280	u64	ipReasmFails;
281	u64	ipFragOKs;
282	u64	ipFragFails;
283	u64	ipFragCreates;
284	u64	ipInMcastPkts;
285	u64	ipOutMcastPkts;
286	u64	ipInBcastPkts;
287	u64	ipOutBcastPkts;
288
289	u64	tcpRtoAlgorithm;
290	u64	tcpRtoMin;
291	u64	tcpRtoMax;
292	u64	tcpMaxConn;
293	u64	tcpActiveOpens;
294	u64	tcpPassiveOpens;
295	u64	tcpAttemptFails;
296	u64	tcpEstabResets;
297	u64	tcpCurrEstab;
298	u64	tcpInSegs;
299	u64	tcpOutSegs;
300	u64	tcpRetransSegs;
301	u64	tcpInErrs;
302	u64	tcpOutRsts;
303};
304
305union rdma_protocol_stats {
306	struct ib_protocol_stats	ib;
307	struct iw_protocol_stats	iw;
308};
309
310struct ib_port_attr {
311	enum ib_port_state	state;
312	enum ib_mtu		max_mtu;
313	enum ib_mtu		active_mtu;
314	int			gid_tbl_len;
315	u32			port_cap_flags;
316	u32			max_msg_sz;
317	u32			bad_pkey_cntr;
318	u32			qkey_viol_cntr;
319	u16			pkey_tbl_len;
320	u16			lid;
321	u16			sm_lid;
322	u8			lmc;
323	u8			max_vl_num;
324	u8			sm_sl;
325	u8			subnet_timeout;
326	u8			init_type_reply;
327	u8			active_width;
328	u8			active_speed;
329	u8                      phys_state;
330	enum rdma_link_layer	link_layer;
331};
332
333enum ib_device_modify_flags {
334	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
335	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
336};
337
338struct ib_device_modify {
339	u64	sys_image_guid;
340	char	node_desc[64];
341};
342
343enum ib_port_modify_flags {
344	IB_PORT_SHUTDOWN		= 1,
345	IB_PORT_INIT_TYPE		= (1<<2),
346	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
347};
348
349struct ib_port_modify {
350	u32	set_port_cap_mask;
351	u32	clr_port_cap_mask;
352	u8	init_type;
353};
354
355enum ib_event_type {
356	IB_EVENT_CQ_ERR,
357	IB_EVENT_QP_FATAL,
358	IB_EVENT_QP_REQ_ERR,
359	IB_EVENT_QP_ACCESS_ERR,
360	IB_EVENT_COMM_EST,
361	IB_EVENT_SQ_DRAINED,
362	IB_EVENT_PATH_MIG,
363	IB_EVENT_PATH_MIG_ERR,
364	IB_EVENT_DEVICE_FATAL,
365	IB_EVENT_PORT_ACTIVE,
366	IB_EVENT_PORT_ERR,
367	IB_EVENT_LID_CHANGE,
368	IB_EVENT_PKEY_CHANGE,
369	IB_EVENT_SM_CHANGE,
370	IB_EVENT_SRQ_ERR,
371	IB_EVENT_SRQ_LIMIT_REACHED,
372	IB_EVENT_QP_LAST_WQE_REACHED,
373	IB_EVENT_CLIENT_REREGISTER,
374	IB_EVENT_GID_CHANGE,
375};
376
377enum ib_event_flags {
378	IB_XRC_QP_EVENT_FLAG = 0x80000000,
379};
380
381struct ib_event {
382	struct ib_device	*device;
383	union {
384		struct ib_cq	*cq;
385		struct ib_qp	*qp;
386		struct ib_srq	*srq;
387		u8		port_num;
388		u32		xrc_qp_num;
389	} element;
390	enum ib_event_type	event;
391};
392
393struct ib_event_handler {
394	struct ib_device *device;
395	void            (*handler)(struct ib_event_handler *, struct ib_event *);
396	struct list_head  list;
397};
398
399#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
400	do {							\
401		(_ptr)->device  = _device;			\
402		(_ptr)->handler = _handler;			\
403		INIT_LIST_HEAD(&(_ptr)->list);			\
404	} while (0)
405
406struct ib_global_route {
407	union ib_gid	dgid;
408	u32		flow_label;
409	u8		sgid_index;
410	u8		hop_limit;
411	u8		traffic_class;
412};
413
414struct ib_grh {
415	__be32		version_tclass_flow;
416	__be16		paylen;
417	u8		next_hdr;
418	u8		hop_limit;
419	union ib_gid	sgid;
420	union ib_gid	dgid;
421};
422
423enum {
424	IB_MULTICAST_QPN = 0xffffff
425};
426
427#define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
428
429enum ib_ah_flags {
430	IB_AH_GRH	= 1
431};
432
433enum ib_rate {
434	IB_RATE_PORT_CURRENT = 0,
435	IB_RATE_2_5_GBPS = 2,
436	IB_RATE_5_GBPS   = 5,
437	IB_RATE_10_GBPS  = 3,
438	IB_RATE_20_GBPS  = 6,
439	IB_RATE_30_GBPS  = 4,
440	IB_RATE_40_GBPS  = 7,
441	IB_RATE_60_GBPS  = 8,
442	IB_RATE_80_GBPS  = 9,
443	IB_RATE_120_GBPS = 10,
444	IB_RATE_14_GBPS  = 11,
445	IB_RATE_56_GBPS  = 12,
446	IB_RATE_112_GBPS = 13,
447	IB_RATE_168_GBPS = 14,
448	IB_RATE_25_GBPS  = 15,
449	IB_RATE_100_GBPS = 16,
450	IB_RATE_200_GBPS = 17,
451	IB_RATE_300_GBPS = 18
452};
453
454/**
455 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
456 * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
457 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
458 * @rate: rate to convert.
459 */
460int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
461
462/**
463 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
464 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
465 * @rate: rate to convert.
466 */
467int ib_rate_to_mbps(enum ib_rate rate) __attribute_const__;
468
469/**
470 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
471 * enum.
472 * @mult: multiple to convert.
473 */
474enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
475
476struct ib_ah_attr {
477	struct ib_global_route	grh;
478	u16			dlid;
479	u8			sl;
480	u8			src_path_bits;
481	u8			static_rate;
482	u8			ah_flags;
483	u8			port_num;
484};
485
486enum ib_wc_status {
487	IB_WC_SUCCESS,
488	IB_WC_LOC_LEN_ERR,
489	IB_WC_LOC_QP_OP_ERR,
490	IB_WC_LOC_EEC_OP_ERR,
491	IB_WC_LOC_PROT_ERR,
492	IB_WC_WR_FLUSH_ERR,
493	IB_WC_MW_BIND_ERR,
494	IB_WC_BAD_RESP_ERR,
495	IB_WC_LOC_ACCESS_ERR,
496	IB_WC_REM_INV_REQ_ERR,
497	IB_WC_REM_ACCESS_ERR,
498	IB_WC_REM_OP_ERR,
499	IB_WC_RETRY_EXC_ERR,
500	IB_WC_RNR_RETRY_EXC_ERR,
501	IB_WC_LOC_RDD_VIOL_ERR,
502	IB_WC_REM_INV_RD_REQ_ERR,
503	IB_WC_REM_ABORT_ERR,
504	IB_WC_INV_EECN_ERR,
505	IB_WC_INV_EEC_STATE_ERR,
506	IB_WC_FATAL_ERR,
507	IB_WC_RESP_TIMEOUT_ERR,
508	IB_WC_GENERAL_ERR
509};
510
511enum ib_wc_opcode {
512	IB_WC_SEND,
513	IB_WC_RDMA_WRITE,
514	IB_WC_RDMA_READ,
515	IB_WC_COMP_SWAP,
516	IB_WC_FETCH_ADD,
517	IB_WC_BIND_MW,
518	IB_WC_LSO,
519	IB_WC_LOCAL_INV,
520	IB_WC_FAST_REG_MR,
521	IB_WC_MASKED_COMP_SWAP,
522	IB_WC_MASKED_FETCH_ADD,
523/*
524 * Set value of IB_WC_RECV so consumers can test if a completion is a
525 * receive by testing (opcode & IB_WC_RECV).
526 */
527	IB_WC_RECV			= 1 << 7,
528	IB_WC_RECV_RDMA_WITH_IMM
529};
530
531enum ib_wc_flags {
532	IB_WC_GRH		= 1,
533	IB_WC_WITH_IMM		= (1<<1),
534	IB_WC_WITH_INVALIDATE	= (1<<2),
535	IB_WC_IP_CSUM_OK	= (1<<3),
536};
537
538struct ib_wc {
539	u64			wr_id;
540	enum ib_wc_status	status;
541	enum ib_wc_opcode	opcode;
542	u32			vendor_err;
543	u32			byte_len;
544	struct ib_qp	       *qp;
545	union {
546		__be32		imm_data;
547		u32		invalidate_rkey;
548	} ex;
549	u32			src_qp;
550	int			wc_flags;
551	u16			pkey_index;
552	u16			slid;
553	u8			sl;
554	u8			dlid_path_bits;
555	u8			port_num;	/* valid only for DR SMPs on switches */
556	int			csum_ok;
557};
558
559enum ib_cq_notify_flags {
560	IB_CQ_SOLICITED			= 1 << 0,
561	IB_CQ_NEXT_COMP			= 1 << 1,
562	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
563	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
564};
565
566enum ib_srq_type {
567	IB_SRQT_BASIC,
568	IB_SRQT_XRC
569};
570
571enum ib_srq_attr_mask {
572	IB_SRQ_MAX_WR	= 1 << 0,
573	IB_SRQ_LIMIT	= 1 << 1,
574};
575
576struct ib_srq_attr {
577	u32	max_wr;
578	u32	max_sge;
579	u32	srq_limit;
580};
581
582struct ib_srq_init_attr {
583	void		      (*event_handler)(struct ib_event *, void *);
584	void		       *srq_context;
585	struct ib_srq_attr	attr;
586	enum ib_srq_type	srq_type;
587
588	union {
589		struct {
590			struct ib_xrcd *xrcd;
591			struct ib_cq   *cq;
592		} xrc;
593	} ext;
594};
595
596struct ib_qp_cap {
597	u32	max_send_wr;
598	u32	max_recv_wr;
599	u32	max_send_sge;
600	u32	max_recv_sge;
601	u32	max_inline_data;
602	u32	qpg_tss_mask_sz;
603};
604
605enum ib_sig_type {
606	IB_SIGNAL_ALL_WR,
607	IB_SIGNAL_REQ_WR
608};
609
610enum ib_qp_type {
611	/*
612	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
613	 * here (and in that order) since the MAD layer uses them as
614	 * indices into a 2-entry table.
615	 */
616	IB_QPT_SMI,
617	IB_QPT_GSI,
618
619	IB_QPT_RC,
620	IB_QPT_UC,
621	IB_QPT_UD,
622	IB_QPT_XRC,
623	IB_QPT_RAW_IPV6,
624	IB_QPT_RAW_ETHERTYPE,
625	IB_QPT_RAW_PACKET = 8,
626	IB_QPT_XRC_INI = 9,
627	IB_QPT_XRC_TGT,
628	IB_QPT_MAX,
629};
630
631enum ib_qp_create_flags {
632	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
633	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
634	IB_QP_CREATE_NETIF_QP			= 1 << 2,
635	/* reserve bits 26-31 for low level drivers' internal use */
636	IB_QP_CREATE_RESERVED_START		= 1 << 26,
637	IB_QP_CREATE_RESERVED_END		= 1 << 31,
638};
639
640enum ib_qpg_type {
641	IB_QPG_NONE	= 0,
642	IB_QPG_PARENT	= (1<<0),
643	IB_QPG_CHILD_RX = (1<<1),
644	IB_QPG_CHILD_TX = (1<<2)
645};
646
647struct ib_qpg_init_attrib {
648	u32 tss_child_count;
649	u32 rss_child_count;
650};
651
652struct ib_qp_init_attr {
653	void                  (*event_handler)(struct ib_event *, void *);
654	void		       *qp_context;
655	struct ib_cq	       *send_cq;
656	struct ib_cq	       *recv_cq;
657	struct ib_srq	       *srq;
658	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
659	struct ib_qp_cap	cap;
660	union {
661		struct ib_qp *qpg_parent; /* see qpg_type */
662		struct ib_qpg_init_attrib parent_attrib;
663	} pp;
664	enum ib_sig_type	sq_sig_type;
665	enum ib_qp_type		qp_type;
666	enum ib_qp_create_flags	create_flags;
667	enum ib_qpg_type	qpg_type;
668	u8			port_num; /* special QP types only */
669};
670
671struct ib_qp_open_attr {
672	void                  (*event_handler)(struct ib_event *, void *);
673	void		       *qp_context;
674	u32			qp_num;
675	enum ib_qp_type		qp_type;
676};
677
678enum ib_rnr_timeout {
679	IB_RNR_TIMER_655_36 =  0,
680	IB_RNR_TIMER_000_01 =  1,
681	IB_RNR_TIMER_000_02 =  2,
682	IB_RNR_TIMER_000_03 =  3,
683	IB_RNR_TIMER_000_04 =  4,
684	IB_RNR_TIMER_000_06 =  5,
685	IB_RNR_TIMER_000_08 =  6,
686	IB_RNR_TIMER_000_12 =  7,
687	IB_RNR_TIMER_000_16 =  8,
688	IB_RNR_TIMER_000_24 =  9,
689	IB_RNR_TIMER_000_32 = 10,
690	IB_RNR_TIMER_000_48 = 11,
691	IB_RNR_TIMER_000_64 = 12,
692	IB_RNR_TIMER_000_96 = 13,
693	IB_RNR_TIMER_001_28 = 14,
694	IB_RNR_TIMER_001_92 = 15,
695	IB_RNR_TIMER_002_56 = 16,
696	IB_RNR_TIMER_003_84 = 17,
697	IB_RNR_TIMER_005_12 = 18,
698	IB_RNR_TIMER_007_68 = 19,
699	IB_RNR_TIMER_010_24 = 20,
700	IB_RNR_TIMER_015_36 = 21,
701	IB_RNR_TIMER_020_48 = 22,
702	IB_RNR_TIMER_030_72 = 23,
703	IB_RNR_TIMER_040_96 = 24,
704	IB_RNR_TIMER_061_44 = 25,
705	IB_RNR_TIMER_081_92 = 26,
706	IB_RNR_TIMER_122_88 = 27,
707	IB_RNR_TIMER_163_84 = 28,
708	IB_RNR_TIMER_245_76 = 29,
709	IB_RNR_TIMER_327_68 = 30,
710	IB_RNR_TIMER_491_52 = 31
711};
712
713enum ib_qp_attr_mask {
714	IB_QP_STATE			= 1,
715	IB_QP_CUR_STATE			= (1<<1),
716	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
717	IB_QP_ACCESS_FLAGS		= (1<<3),
718	IB_QP_PKEY_INDEX		= (1<<4),
719	IB_QP_PORT			= (1<<5),
720	IB_QP_QKEY			= (1<<6),
721	IB_QP_AV			= (1<<7),
722	IB_QP_PATH_MTU			= (1<<8),
723	IB_QP_TIMEOUT			= (1<<9),
724	IB_QP_RETRY_CNT			= (1<<10),
725	IB_QP_RNR_RETRY			= (1<<11),
726	IB_QP_RQ_PSN			= (1<<12),
727	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
728	IB_QP_ALT_PATH			= (1<<14),
729	IB_QP_MIN_RNR_TIMER		= (1<<15),
730	IB_QP_SQ_PSN			= (1<<16),
731	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
732	IB_QP_PATH_MIG_STATE		= (1<<18),
733	IB_QP_CAP			= (1<<19),
734	IB_QP_DEST_QPN			= (1<<20),
735	IB_QP_GROUP_RSS			= (1<<21)
736};
737
738enum ib_qp_state {
739	IB_QPS_RESET,
740	IB_QPS_INIT,
741	IB_QPS_RTR,
742	IB_QPS_RTS,
743	IB_QPS_SQD,
744	IB_QPS_SQE,
745	IB_QPS_ERR
746};
747
748enum ib_mig_state {
749	IB_MIG_MIGRATED,
750	IB_MIG_REARM,
751	IB_MIG_ARMED
752};
753
754struct ib_qp_attr {
755	enum ib_qp_state	qp_state;
756	enum ib_qp_state	cur_qp_state;
757	enum ib_mtu		path_mtu;
758	enum ib_mig_state	path_mig_state;
759	u32			qkey;
760	u32			rq_psn;
761	u32			sq_psn;
762	u32			dest_qp_num;
763	int			qp_access_flags;
764	struct ib_qp_cap	cap;
765	struct ib_ah_attr	ah_attr;
766	struct ib_ah_attr	alt_ah_attr;
767	u16			pkey_index;
768	u16			alt_pkey_index;
769	u8			en_sqd_async_notify;
770	u8			sq_draining;
771	u8			max_rd_atomic;
772	u8			max_dest_rd_atomic;
773	u8			min_rnr_timer;
774	u8			port_num;
775	u8			timeout;
776	u8			retry_cnt;
777	u8			rnr_retry;
778	u8			alt_port_num;
779	u8			alt_timeout;
780};
781
782enum ib_wr_opcode {
783	IB_WR_RDMA_WRITE,
784	IB_WR_RDMA_WRITE_WITH_IMM,
785	IB_WR_SEND,
786	IB_WR_SEND_WITH_IMM,
787	IB_WR_RDMA_READ,
788	IB_WR_ATOMIC_CMP_AND_SWP,
789	IB_WR_ATOMIC_FETCH_AND_ADD,
790	IB_WR_LSO,
791	IB_WR_BIG_LSO,
792	IB_WR_SEND_WITH_INV,
793	IB_WR_RDMA_READ_WITH_INV,
794	IB_WR_LOCAL_INV,
795	IB_WR_FAST_REG_MR,
796	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
797	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
798};
799
800enum ib_send_flags {
801	IB_SEND_FENCE		= 1,
802	IB_SEND_SIGNALED	= (1<<1),
803	IB_SEND_SOLICITED	= (1<<2),
804	IB_SEND_INLINE		= (1<<3),
805	IB_SEND_IP_CSUM		= (1<<4)
806};
807
808enum ib_flow_types {
809	IB_FLOW_ETH = 0,
810	IB_FLOW_IB_UC = 1,
811	IB_FLOW_IB_MC_IPV4 = 2,
812	IB_FLOW_IB_MC_IPV6 = 3
813};
814
815enum {
816	IB_FLOW_L4_NONE = 0,
817	IB_FLOW_L4_OTHER = 3,
818	IB_FLOW_L4_UDP = 5,
819	IB_FLOW_L4_TCP = 6
820};
821
822struct ib_sge {
823	u64	addr;
824	u32	length;
825	u32	lkey;
826};
827
828struct ib_fast_reg_page_list {
829	struct ib_device       *device;
830	u64		       *page_list;
831	unsigned int		max_page_list_len;
832};
833
834struct ib_send_wr {
835	struct ib_send_wr      *next;
836	u64			wr_id;
837	struct ib_sge	       *sg_list;
838	int			num_sge;
839	enum ib_wr_opcode	opcode;
840	int			send_flags;
841	union {
842		__be32		imm_data;
843		u32		invalidate_rkey;
844	} ex;
845	union {
846		struct {
847			u64	remote_addr;
848			u32	rkey;
849		} rdma;
850		struct {
851			u64	remote_addr;
852			u64	compare_add;
853			u64	swap;
854			u64	compare_add_mask;
855			u64	swap_mask;
856			u32	rkey;
857		} atomic;
858		struct {
859			struct ib_ah *ah;
860			void   *header;
861			int     hlen;
862			int     mss;
863			u32	remote_qpn;
864			u32	remote_qkey;
865			u16	pkey_index; /* valid for GSI only */
866			u8	port_num;   /* valid for DR SMPs on switch only */
867		} ud;
868		struct {
869			u64				iova_start;
870			struct ib_fast_reg_page_list   *page_list;
871			unsigned int			page_shift;
872			unsigned int			page_list_len;
873			u32				length;
874			int				access_flags;
875			u32				rkey;
876		} fast_reg;
877		struct {
878			struct ib_unpacked_lrh	*lrh;
879			u32			eth_type;
880			u8			static_rate;
881		} raw_ety;
882	} wr;
883	u32			xrc_remote_srq_num;	/* XRC TGT QPs only */
884};
885
886struct ib_recv_wr {
887	struct ib_recv_wr      *next;
888	u64			wr_id;
889	struct ib_sge	       *sg_list;
890	int			num_sge;
891};
892
893enum ib_access_flags {
894	IB_ACCESS_LOCAL_WRITE	= 1,
895	IB_ACCESS_REMOTE_WRITE	= (1<<1),
896	IB_ACCESS_REMOTE_READ	= (1<<2),
897	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
898	IB_ACCESS_MW_BIND	= (1<<4),
899	IB_ACCESS_ALLOCATE_MR	= (1<<5),
900	IB_ACCESS_SHARED_MR_USER_READ   = (1<<6),
901	IB_ACCESS_SHARED_MR_USER_WRITE  = (1<<7),
902	IB_ACCESS_SHARED_MR_GROUP_READ  = (1<<8),
903	IB_ACCESS_SHARED_MR_GROUP_WRITE = (1<<9),
904	IB_ACCESS_SHARED_MR_OTHER_READ  = (1<<10),
905	IB_ACCESS_SHARED_MR_OTHER_WRITE = (1<<11)
906
907};
908
909struct ib_phys_buf {
910	u64      addr;
911	u64      size;
912};
913
914struct ib_mr_attr {
915	struct ib_pd	*pd;
916	u64		device_virt_addr;
917	u64		size;
918	int		mr_access_flags;
919	u32		lkey;
920	u32		rkey;
921};
922
923enum ib_mr_rereg_flags {
924	IB_MR_REREG_TRANS	= 1,
925	IB_MR_REREG_PD		= (1<<1),
926	IB_MR_REREG_ACCESS	= (1<<2)
927};
928
929struct ib_mw_bind {
930	struct ib_mr   *mr;
931	u64		wr_id;
932	u64		addr;
933	u32		length;
934	int		send_flags;
935	int		mw_access_flags;
936};
937
938struct ib_fmr_attr {
939	int	max_pages;
940	int	max_maps;
941	u8	page_shift;
942};
943
944struct ib_ucontext {
945	struct ib_device       *device;
946	struct list_head	pd_list;
947	struct list_head	mr_list;
948	struct list_head	mw_list;
949	struct list_head	cq_list;
950	struct list_head	qp_list;
951	struct list_head	srq_list;
952	struct list_head	ah_list;
953	struct list_head	xrcd_list;
954	int			closing;
955};
956
957struct ib_uobject {
958	u64			user_handle;	/* handle given to us by userspace */
959	struct ib_ucontext     *context;	/* associated user context */
960	void		       *object;		/* containing object */
961	struct list_head	list;		/* link to context's list */
962	int			id;		/* index into kernel idr */
963	struct kref		ref;
964	struct rw_semaphore	mutex;		/* protects .live */
965	int			live;
966};
967
968struct ib_udata {
969	void __user *inbuf;
970	void __user *outbuf;
971	size_t       inlen;
972	size_t       outlen;
973};
974
975struct ib_uxrc_rcv_object {
976	struct list_head	list;		/* link to context's list */
977	u32			qp_num;
978	u32			domain_handle;
979};
980
981struct ib_pd {
982	struct ib_device       *device;
983	struct ib_uobject      *uobject;
984	atomic_t          	usecnt; /* count all resources */
985};
986
987struct ib_xrcd {
988	struct ib_device       *device;
989	struct ib_uobject      *uobject;
990	atomic_t		usecnt; /* count all exposed resources */
991	struct inode	       *inode;
992	struct rb_node		node;
993
994	struct mutex		tgt_qp_mutex;
995	struct list_head	tgt_qp_list;
996};
997
998struct ib_ah {
999	struct ib_device	*device;
1000	struct ib_pd		*pd;
1001	struct ib_uobject	*uobject;
1002};
1003
1004typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1005
1006struct ib_cq {
1007	struct ib_device       *device;
1008	struct ib_uobject      *uobject;
1009	ib_comp_handler   	comp_handler;
1010	void                  (*event_handler)(struct ib_event *, void *);
1011	void                   *cq_context;
1012	int               	cqe;
1013	atomic_t          	usecnt; /* count number of work queues */
1014};
1015
1016struct ib_srq {
1017	struct ib_device       *device;
1018	struct ib_pd	       *pd;
1019	struct ib_uobject      *uobject;
1020	void		      (*event_handler)(struct ib_event *, void *);
1021	void		       *srq_context;
1022	enum ib_srq_type	srq_type;
1023	atomic_t		usecnt;
1024
1025	union {
1026		struct {
1027			struct ib_xrcd *xrcd;
1028			struct ib_cq   *cq;
1029			u32		srq_num;
1030		} xrc;
1031	} ext;
1032};
1033
1034struct ib_qp {
1035	struct ib_device       *device;
1036	struct ib_pd	       *pd;
1037	struct ib_cq	       *send_cq;
1038	struct ib_cq	       *recv_cq;
1039	struct ib_srq	       *srq;
1040	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1041	struct list_head	xrcd_list;
1042	atomic_t		usecnt; /* count times opened, mcast attaches */
1043	struct list_head	open_list;
1044	struct ib_qp           *real_qp;
1045	struct ib_uobject      *uobject;
1046	void                  (*event_handler)(struct ib_event *, void *);
1047	void		       *qp_context;
1048	u32			qp_num;
1049	enum ib_qp_type		qp_type;
1050	enum ib_qpg_type	qpg_type;
1051};
1052
1053struct ib_mr {
1054	struct ib_device  *device;
1055	struct ib_pd	  *pd;
1056	struct ib_uobject *uobject;
1057	u32		   lkey;
1058	u32		   rkey;
1059	atomic_t	   usecnt; /* count number of MWs */
1060};
1061
1062struct ib_mw {
1063	struct ib_device	*device;
1064	struct ib_pd		*pd;
1065	struct ib_uobject	*uobject;
1066	u32			rkey;
1067};
1068
1069struct ib_fmr {
1070	struct ib_device	*device;
1071	struct ib_pd		*pd;
1072	struct list_head	list;
1073	u32			lkey;
1074	u32			rkey;
1075};
1076
1077struct ib_flow_spec {
1078	enum ib_flow_types type;
1079	union {
1080		struct {
1081			__be16 ethertype;
1082			__be16 vlan;
1083			u8 vlan_present;
1084			u8  mac[6];
1085			u8  port;
1086		} eth;
1087		struct {
1088			__be32 qpn;
1089		} ib_uc;
1090		struct {
1091			u8  mgid[16];
1092		} ib_mc;
1093	} l2_id;
1094	__be32 src_ip;
1095	__be32 dst_ip;
1096	__be16 src_port;
1097	__be16 dst_port;
1098	u8 l4_protocol;
1099	u8 block_mc_loopback;
1100	u8 rule_type;
1101};
1102
1103struct ib_mad;
1104struct ib_grh;
1105
1106enum ib_process_mad_flags {
1107	IB_MAD_IGNORE_MKEY	= 1,
1108	IB_MAD_IGNORE_BKEY	= 2,
1109	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1110};
1111
1112enum ib_mad_result {
1113	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1114	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1115	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1116	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1117};
1118
1119#define IB_DEVICE_NAME_MAX 64
1120
1121struct ib_cache {
1122	rwlock_t                lock;
1123	struct ib_event_handler event_handler;
1124	struct ib_pkey_cache  **pkey_cache;
1125	struct ib_gid_cache   **gid_cache;
1126	u8                     *lmc_cache;
1127};
1128
1129struct ib_dma_mapping_ops {
1130	int		(*mapping_error)(struct ib_device *dev,
1131					 u64 dma_addr);
1132	u64		(*map_single)(struct ib_device *dev,
1133				      void *ptr, size_t size,
1134				      enum dma_data_direction direction);
1135	void		(*unmap_single)(struct ib_device *dev,
1136					u64 addr, size_t size,
1137					enum dma_data_direction direction);
1138	u64		(*map_page)(struct ib_device *dev,
1139				    struct page *page, unsigned long offset,
1140				    size_t size,
1141				    enum dma_data_direction direction);
1142	void		(*unmap_page)(struct ib_device *dev,
1143				      u64 addr, size_t size,
1144				      enum dma_data_direction direction);
1145	int		(*map_sg)(struct ib_device *dev,
1146				  struct scatterlist *sg, int nents,
1147				  enum dma_data_direction direction);
1148	void		(*unmap_sg)(struct ib_device *dev,
1149				    struct scatterlist *sg, int nents,
1150				    enum dma_data_direction direction);
1151	u64		(*dma_address)(struct ib_device *dev,
1152				       struct scatterlist *sg);
1153	unsigned int	(*dma_len)(struct ib_device *dev,
1154				   struct scatterlist *sg);
1155	void		(*sync_single_for_cpu)(struct ib_device *dev,
1156					       u64 dma_handle,
1157					       size_t size,
1158					       enum dma_data_direction dir);
1159	void		(*sync_single_for_device)(struct ib_device *dev,
1160						  u64 dma_handle,
1161						  size_t size,
1162						  enum dma_data_direction dir);
1163	void		*(*alloc_coherent)(struct ib_device *dev,
1164					   size_t size,
1165					   u64 *dma_handle,
1166					   gfp_t flag);
1167	void		(*free_coherent)(struct ib_device *dev,
1168					 size_t size, void *cpu_addr,
1169					 u64 dma_handle);
1170};
1171
1172struct iw_cm_verbs;
1173
1174struct ib_device {
1175	struct device                *dma_device;
1176
1177	char                          name[IB_DEVICE_NAME_MAX];
1178
1179	struct list_head              event_handler_list;
1180	spinlock_t                    event_handler_lock;
1181
1182	spinlock_t                    client_data_lock;
1183	struct list_head              core_list;
1184	struct list_head              client_data_list;
1185
1186	struct ib_cache               cache;
1187	int                          *pkey_tbl_len;
1188	int                          *gid_tbl_len;
1189
1190	int			      num_comp_vectors;
1191
1192	struct iw_cm_verbs	     *iwcm;
1193
1194	int		           (*get_protocol_stats)(struct ib_device *device,
1195							 union rdma_protocol_stats *stats);
1196	int		           (*query_device)(struct ib_device *device,
1197						   struct ib_device_attr *device_attr);
1198	int		           (*query_port)(struct ib_device *device,
1199						 u8 port_num,
1200						 struct ib_port_attr *port_attr);
1201	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1202						     u8 port_num);
1203	int		           (*query_gid)(struct ib_device *device,
1204						u8 port_num, int index,
1205						union ib_gid *gid);
1206	int		           (*query_pkey)(struct ib_device *device,
1207						 u8 port_num, u16 index, u16 *pkey);
1208	int		           (*modify_device)(struct ib_device *device,
1209						    int device_modify_mask,
1210						    struct ib_device_modify *device_modify);
1211	int		           (*modify_port)(struct ib_device *device,
1212						  u8 port_num, int port_modify_mask,
1213						  struct ib_port_modify *port_modify);
1214	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1215						     struct ib_udata *udata);
1216	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1217	int                        (*mmap)(struct ib_ucontext *context,
1218					   struct vm_area_struct *vma);
1219	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1220					       struct ib_ucontext *context,
1221					       struct ib_udata *udata);
1222	int                        (*dealloc_pd)(struct ib_pd *pd);
1223	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1224						struct ib_ah_attr *ah_attr);
1225	int                        (*modify_ah)(struct ib_ah *ah,
1226						struct ib_ah_attr *ah_attr);
1227	int                        (*query_ah)(struct ib_ah *ah,
1228					       struct ib_ah_attr *ah_attr);
1229	int                        (*destroy_ah)(struct ib_ah *ah);
1230	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1231						 struct ib_srq_init_attr *srq_init_attr,
1232						 struct ib_udata *udata);
1233	int                        (*modify_srq)(struct ib_srq *srq,
1234						 struct ib_srq_attr *srq_attr,
1235						 enum ib_srq_attr_mask srq_attr_mask,
1236						 struct ib_udata *udata);
1237	int                        (*query_srq)(struct ib_srq *srq,
1238						struct ib_srq_attr *srq_attr);
1239	int                        (*destroy_srq)(struct ib_srq *srq);
1240	int                        (*post_srq_recv)(struct ib_srq *srq,
1241						    struct ib_recv_wr *recv_wr,
1242						    struct ib_recv_wr **bad_recv_wr);
1243	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1244						struct ib_qp_init_attr *qp_init_attr,
1245						struct ib_udata *udata);
1246	int                        (*modify_qp)(struct ib_qp *qp,
1247						struct ib_qp_attr *qp_attr,
1248						int qp_attr_mask,
1249						struct ib_udata *udata);
1250	int                        (*query_qp)(struct ib_qp *qp,
1251					       struct ib_qp_attr *qp_attr,
1252					       int qp_attr_mask,
1253					       struct ib_qp_init_attr *qp_init_attr);
1254	int                        (*destroy_qp)(struct ib_qp *qp);
1255	int                        (*post_send)(struct ib_qp *qp,
1256						struct ib_send_wr *send_wr,
1257						struct ib_send_wr **bad_send_wr);
1258	int                        (*post_recv)(struct ib_qp *qp,
1259						struct ib_recv_wr *recv_wr,
1260						struct ib_recv_wr **bad_recv_wr);
1261	struct ib_cq *             (*create_cq)(struct ib_device *device, int cqe,
1262						int comp_vector,
1263						struct ib_ucontext *context,
1264						struct ib_udata *udata);
1265	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1266						u16 cq_period);
1267	int                        (*destroy_cq)(struct ib_cq *cq);
1268	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1269						struct ib_udata *udata);
1270	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1271					      struct ib_wc *wc);
1272	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1273	int                        (*req_notify_cq)(struct ib_cq *cq,
1274						    enum ib_cq_notify_flags flags);
1275	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1276						      int wc_cnt);
1277	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1278						 int mr_access_flags);
1279	struct ib_mr *             (*reg_phys_mr)(struct ib_pd *pd,
1280						  struct ib_phys_buf *phys_buf_array,
1281						  int num_phys_buf,
1282						  int mr_access_flags,
1283						  u64 *iova_start);
1284	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1285						  u64 start, u64 length,
1286						  u64 virt_addr,
1287						  int mr_access_flags,
1288						  struct ib_udata *udata,
1289							int mr_id);
1290	int                        (*query_mr)(struct ib_mr *mr,
1291					       struct ib_mr_attr *mr_attr);
1292	int                        (*dereg_mr)(struct ib_mr *mr);
1293	struct ib_mr *		   (*alloc_fast_reg_mr)(struct ib_pd *pd,
1294					       int max_page_list_len);
1295	struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1296								   int page_list_len);
1297	void			   (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1298	int                        (*rereg_phys_mr)(struct ib_mr *mr,
1299						    int mr_rereg_mask,
1300						    struct ib_pd *pd,
1301						    struct ib_phys_buf *phys_buf_array,
1302						    int num_phys_buf,
1303						    int mr_access_flags,
1304						    u64 *iova_start);
1305	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd);
1306	int                        (*bind_mw)(struct ib_qp *qp,
1307					      struct ib_mw *mw,
1308					      struct ib_mw_bind *mw_bind);
1309	int                        (*dealloc_mw)(struct ib_mw *mw);
1310	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1311						int mr_access_flags,
1312						struct ib_fmr_attr *fmr_attr);
1313	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1314						   u64 *page_list, int list_len,
1315						   u64 iova);
1316	int		           (*unmap_fmr)(struct list_head *fmr_list);
1317	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1318	int                        (*attach_mcast)(struct ib_qp *qp,
1319						   union ib_gid *gid,
1320						   u16 lid);
1321	int                        (*detach_mcast)(struct ib_qp *qp,
1322						   union ib_gid *gid,
1323						   u16 lid);
1324	int                        (*process_mad)(struct ib_device *device,
1325						  int process_mad_flags,
1326						  u8 port_num,
1327						  struct ib_wc *in_wc,
1328						  struct ib_grh *in_grh,
1329						  struct ib_mad *in_mad,
1330						  struct ib_mad *out_mad);
1331	struct ib_srq *		   (*create_xrc_srq)(struct ib_pd *pd,
1332						     struct ib_cq *xrc_cq,
1333						     struct ib_xrcd *xrcd,
1334						     struct ib_srq_init_attr *srq_init_attr,
1335						     struct ib_udata *udata);
1336	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
1337						 struct ib_ucontext *ucontext,
1338						 struct ib_udata *udata);
1339	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1340	int			   (*create_xrc_rcv_qp)(struct ib_qp_init_attr *init_attr,
1341							u32 *qp_num);
1342	int			   (*modify_xrc_rcv_qp)(struct ib_xrcd *xrcd,
1343							u32 qp_num,
1344							struct ib_qp_attr *attr,
1345							int attr_mask);
1346	int			   (*query_xrc_rcv_qp)(struct ib_xrcd *xrcd,
1347						       u32 qp_num,
1348						       struct ib_qp_attr *attr,
1349						       int attr_mask,
1350						       struct ib_qp_init_attr *init_attr);
1351	int 			   (*reg_xrc_rcv_qp)(struct ib_xrcd *xrcd,
1352						     void *context,
1353						     u32 qp_num);
1354	int 			   (*unreg_xrc_rcv_qp)(struct ib_xrcd *xrcd,
1355						       void *context,
1356						       u32 qp_num);
1357	int                        (*attach_flow)(struct ib_qp *qp,
1358						  struct ib_flow_spec *spec,
1359						  int priority);
1360	int                        (*detach_flow)(struct ib_qp *qp,
1361						  struct ib_flow_spec *spec,
1362						  int priority);
1363
1364	unsigned long		   (*get_unmapped_area)(struct file *file,
1365					unsigned long addr,
1366					unsigned long len, unsigned long pgoff,
1367					unsigned long flags);
1368	struct ib_dma_mapping_ops   *dma_ops;
1369
1370	struct module               *owner;
1371	struct device                dev;
1372	struct kobject               *ports_parent;
1373	struct list_head             port_list;
1374
1375	enum {
1376		IB_DEV_UNINITIALIZED,
1377		IB_DEV_REGISTERED,
1378		IB_DEV_UNREGISTERED
1379	}                            reg_state;
1380
1381	int			     uverbs_abi_ver;
1382	u64			     uverbs_cmd_mask;
1383
1384	char			     node_desc[64];
1385	__be64			     node_guid;
1386	u32			     local_dma_lkey;
1387	u8                           node_type;
1388	u8                           phys_port_cnt;
1389	struct rb_root		     ib_uverbs_xrcd_table;
1390	struct mutex		     xrcd_table_mutex;
1391};
1392
1393struct ib_client {
1394	char  *name;
1395	void (*add)   (struct ib_device *);
1396	void (*remove)(struct ib_device *);
1397
1398	struct list_head list;
1399};
1400
1401struct ib_device *ib_alloc_device(size_t size);
1402void ib_dealloc_device(struct ib_device *device);
1403
1404int ib_register_device(struct ib_device *device,
1405		       int (*port_callback)(struct ib_device *,
1406					    u8, struct kobject *));
1407void ib_unregister_device(struct ib_device *device);
1408
1409int ib_register_client   (struct ib_client *client);
1410void ib_unregister_client(struct ib_client *client);
1411
1412void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1413void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
1414			 void *data);
1415
1416static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1417{
1418	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1419}
1420
1421static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1422{
1423	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1424}
1425
1426/**
1427 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1428 * contains all required attributes and no attributes not allowed for
1429 * the given QP state transition.
1430 * @cur_state: Current QP state
1431 * @next_state: Next QP state
1432 * @type: QP type
1433 * @mask: Mask of supplied QP attributes
1434 *
1435 * This function is a helper function that a low-level driver's
1436 * modify_qp method can use to validate the consumer's input.  It
1437 * checks that cur_state and next_state are valid QP states, that a
1438 * transition from cur_state to next_state is allowed by the IB spec,
1439 * and that the attribute mask supplied is allowed for the transition.
1440 */
1441int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1442		       enum ib_qp_type type, enum ib_qp_attr_mask mask);
1443
1444int ib_register_event_handler  (struct ib_event_handler *event_handler);
1445int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1446void ib_dispatch_event(struct ib_event *event);
1447
1448int ib_query_device(struct ib_device *device,
1449		    struct ib_device_attr *device_attr);
1450
1451int ib_query_port(struct ib_device *device,
1452		  u8 port_num, struct ib_port_attr *port_attr);
1453
1454enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1455					       u8 port_num);
1456
1457int ib_query_gid(struct ib_device *device,
1458		 u8 port_num, int index, union ib_gid *gid);
1459
1460int ib_query_pkey(struct ib_device *device,
1461		  u8 port_num, u16 index, u16 *pkey);
1462
1463int ib_modify_device(struct ib_device *device,
1464		     int device_modify_mask,
1465		     struct ib_device_modify *device_modify);
1466
1467int ib_modify_port(struct ib_device *device,
1468		   u8 port_num, int port_modify_mask,
1469		   struct ib_port_modify *port_modify);
1470
1471int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1472		u8 *port_num, u16 *index);
1473
1474int ib_find_pkey(struct ib_device *device,
1475		 u8 port_num, u16 pkey, u16 *index);
1476
1477/**
1478 * ib_alloc_pd - Allocates an unused protection domain.
1479 * @device: The device on which to allocate the protection domain.
1480 *
1481 * A protection domain object provides an association between QPs, shared
1482 * receive queues, address handles, memory regions, and memory windows.
1483 */
1484struct ib_pd *ib_alloc_pd(struct ib_device *device);
1485
1486/**
1487 * ib_dealloc_pd - Deallocates a protection domain.
1488 * @pd: The protection domain to deallocate.
1489 */
1490int ib_dealloc_pd(struct ib_pd *pd);
1491
1492/**
1493 * ib_create_ah - Creates an address handle for the given address vector.
1494 * @pd: The protection domain associated with the address handle.
1495 * @ah_attr: The attributes of the address vector.
1496 *
1497 * The address handle is used to reference a local or global destination
1498 * in all UD QP post sends.
1499 */
1500struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
1501
1502/**
1503 * ib_init_ah_from_wc - Initializes address handle attributes from a
1504 *   work completion.
1505 * @device: Device on which the received message arrived.
1506 * @port_num: Port on which the received message arrived.
1507 * @wc: Work completion associated with the received message.
1508 * @grh: References the received global route header.  This parameter is
1509 *   ignored unless the work completion indicates that the GRH is valid.
1510 * @ah_attr: Returned attributes that can be used when creating an address
1511 *   handle for replying to the message.
1512 */
1513int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
1514		       struct ib_grh *grh, struct ib_ah_attr *ah_attr);
1515
1516/**
1517 * ib_create_ah_from_wc - Creates an address handle associated with the
1518 *   sender of the specified work completion.
1519 * @pd: The protection domain associated with the address handle.
1520 * @wc: Work completion information associated with a received message.
1521 * @grh: References the received global route header.  This parameter is
1522 *   ignored unless the work completion indicates that the GRH is valid.
1523 * @port_num: The outbound port number to associate with the address.
1524 *
1525 * The address handle is used to reference a local or global destination
1526 * in all UD QP post sends.
1527 */
1528struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
1529				   struct ib_grh *grh, u8 port_num);
1530
1531/**
1532 * ib_modify_ah - Modifies the address vector associated with an address
1533 *   handle.
1534 * @ah: The address handle to modify.
1535 * @ah_attr: The new address vector attributes to associate with the
1536 *   address handle.
1537 */
1538int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1539
1540/**
1541 * ib_query_ah - Queries the address vector associated with an address
1542 *   handle.
1543 * @ah: The address handle to query.
1544 * @ah_attr: The address vector attributes associated with the address
1545 *   handle.
1546 */
1547int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1548
1549/**
1550 * ib_destroy_ah - Destroys an address handle.
1551 * @ah: The address handle to destroy.
1552 */
1553int ib_destroy_ah(struct ib_ah *ah);
1554
1555/**
1556 * ib_create_xrc_srq - Creates an XRC SRQ associated with the specified
1557 *   protection domain, cq, and xrc domain.
1558 * @pd: The protection domain associated with the SRQ.
1559 * @xrc_cq: The cq to be associated with the XRC SRQ.
1560 * @xrcd: The XRC domain to be associated with the XRC SRQ.
1561 * @srq_init_attr: A list of initial attributes required to create the
1562 *   XRC SRQ.  If XRC SRQ creation succeeds, then the attributes are updated
1563 *   to the actual capabilities of the created XRC SRQ.
1564 *
1565 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1566 * requested size of the XRC SRQ, and set to the actual values allocated
1567 * on return.  If ib_create_xrc_srq() succeeds, then max_wr and max_sge
1568 * will always be at least as large as the requested values.
1569 */
1570struct ib_srq *ib_create_xrc_srq(struct ib_pd *pd,
1571				 struct ib_cq *xrc_cq,
1572				 struct ib_xrcd *xrcd,
1573				 struct ib_srq_init_attr *srq_init_attr);
1574
1575/**
1576 * ib_create_srq - Creates a SRQ associated with the specified protection
1577 *   domain.
1578 * @pd: The protection domain associated with the SRQ.
1579 * @srq_init_attr: A list of initial attributes required to create the
1580 *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
1581 *   the actual capabilities of the created SRQ.
1582 *
1583 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1584 * requested size of the SRQ, and set to the actual values allocated
1585 * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
1586 * will always be at least as large as the requested values.
1587 */
1588struct ib_srq *ib_create_srq(struct ib_pd *pd,
1589			     struct ib_srq_init_attr *srq_init_attr);
1590
1591/**
1592 * ib_modify_srq - Modifies the attributes for the specified SRQ.
1593 * @srq: The SRQ to modify.
1594 * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
1595 *   the current values of selected SRQ attributes are returned.
1596 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1597 *   are being modified.
1598 *
1599 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1600 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1601 * the number of receives queued drops below the limit.
1602 */
1603int ib_modify_srq(struct ib_srq *srq,
1604		  struct ib_srq_attr *srq_attr,
1605		  enum ib_srq_attr_mask srq_attr_mask);
1606
1607/**
1608 * ib_query_srq - Returns the attribute list and current values for the
1609 *   specified SRQ.
1610 * @srq: The SRQ to query.
1611 * @srq_attr: The attributes of the specified SRQ.
1612 */
1613int ib_query_srq(struct ib_srq *srq,
1614		 struct ib_srq_attr *srq_attr);
1615
1616/**
1617 * ib_destroy_srq - Destroys the specified SRQ.
1618 * @srq: The SRQ to destroy.
1619 */
1620int ib_destroy_srq(struct ib_srq *srq);
1621
1622/**
1623 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1624 * @srq: The SRQ to post the work request on.
1625 * @recv_wr: A list of work requests to post on the receive queue.
1626 * @bad_recv_wr: On an immediate failure, this parameter will reference
1627 *   the work request that failed to be posted on the QP.
1628 */
1629static inline int ib_post_srq_recv(struct ib_srq *srq,
1630				   struct ib_recv_wr *recv_wr,
1631				   struct ib_recv_wr **bad_recv_wr)
1632{
1633	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1634}
1635
1636/**
1637 * ib_create_qp - Creates a QP associated with the specified protection
1638 *   domain.
1639 * @pd: The protection domain associated with the QP.
1640 * @qp_init_attr: A list of initial attributes required to create the
1641 *   QP.  If QP creation succeeds, then the attributes are updated to
1642 *   the actual capabilities of the created QP.
1643 */
1644struct ib_qp *ib_create_qp(struct ib_pd *pd,
1645			   struct ib_qp_init_attr *qp_init_attr);
1646
1647/**
1648 * ib_modify_qp - Modifies the attributes for the specified QP and then
1649 *   transitions the QP to the given state.
1650 * @qp: The QP to modify.
1651 * @qp_attr: On input, specifies the QP attributes to modify.  On output,
1652 *   the current values of selected QP attributes are returned.
1653 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1654 *   are being modified.
1655 */
1656int ib_modify_qp(struct ib_qp *qp,
1657		 struct ib_qp_attr *qp_attr,
1658		 int qp_attr_mask);
1659
1660/**
1661 * ib_query_qp - Returns the attribute list and current values for the
1662 *   specified QP.
1663 * @qp: The QP to query.
1664 * @qp_attr: The attributes of the specified QP.
1665 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1666 * @qp_init_attr: Additional attributes of the selected QP.
1667 *
1668 * The qp_attr_mask may be used to limit the query to gathering only the
1669 * selected attributes.
1670 */
1671int ib_query_qp(struct ib_qp *qp,
1672		struct ib_qp_attr *qp_attr,
1673		int qp_attr_mask,
1674		struct ib_qp_init_attr *qp_init_attr);
1675
1676/**
1677 * ib_destroy_qp - Destroys the specified QP.
1678 * @qp: The QP to destroy.
1679 */
1680int ib_destroy_qp(struct ib_qp *qp);
1681
1682/**
1683 * ib_open_qp - Obtain a reference to an existing sharable QP.
1684 * @xrcd - XRC domain
1685 * @qp_open_attr: Attributes identifying the QP to open.
1686 *
1687 * Returns a reference to a sharable QP.
1688 */
1689struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
1690			 struct ib_qp_open_attr *qp_open_attr);
1691
1692/**
1693 * ib_close_qp - Release an external reference to a QP.
1694 * @qp: The QP handle to release
1695 *
1696 * The opened QP handle is released by the caller.  The underlying
1697 * shared QP is not destroyed until all internal references are released.
1698 */
1699int ib_close_qp(struct ib_qp *qp);
1700
1701/**
1702 * ib_post_send - Posts a list of work requests to the send queue of
1703 *   the specified QP.
1704 * @qp: The QP to post the work request on.
1705 * @send_wr: A list of work requests to post on the send queue.
1706 * @bad_send_wr: On an immediate failure, this parameter will reference
1707 *   the work request that failed to be posted on the QP.
1708 *
1709 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
1710 * error is returned, the QP state shall not be affected,
1711 * ib_post_send() will return an immediate error after queueing any
1712 * earlier work requests in the list.
1713 */
1714static inline int ib_post_send(struct ib_qp *qp,
1715			       struct ib_send_wr *send_wr,
1716			       struct ib_send_wr **bad_send_wr)
1717{
1718	return qp->device->post_send(qp, send_wr, bad_send_wr);
1719}
1720
1721/**
1722 * ib_post_recv - Posts a list of work requests to the receive queue of
1723 *   the specified QP.
1724 * @qp: The QP to post the work request on.
1725 * @recv_wr: A list of work requests to post on the receive queue.
1726 * @bad_recv_wr: On an immediate failure, this parameter will reference
1727 *   the work request that failed to be posted on the QP.
1728 */
1729static inline int ib_post_recv(struct ib_qp *qp,
1730			       struct ib_recv_wr *recv_wr,
1731			       struct ib_recv_wr **bad_recv_wr)
1732{
1733	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1734}
1735
1736/*
1737 * IB_CQ_VECTOR_LEAST_ATTACHED: The constant specifies that
1738 *	the CQ will be attached to the completion vector that has
1739 *	the least number of CQs already attached to it.
1740 */
1741#define IB_CQ_VECTOR_LEAST_ATTACHED	0xffffffff
1742
1743/**
1744 * ib_create_cq - Creates a CQ on the specified device.
1745 * @device: The device on which to create the CQ.
1746 * @comp_handler: A user-specified callback that is invoked when a
1747 *   completion event occurs on the CQ.
1748 * @event_handler: A user-specified callback that is invoked when an
1749 *   asynchronous event not associated with a completion occurs on the CQ.
1750 * @cq_context: Context associated with the CQ returned to the user via
1751 *   the associated completion and event handlers.
1752 * @cqe: The minimum size of the CQ.
1753 * @comp_vector - Completion vector used to signal completion events.
1754 *     Must be >= 0 and < context->num_comp_vectors.
1755 *
1756 * Users can examine the cq structure to determine the actual CQ size.
1757 */
1758struct ib_cq *ib_create_cq(struct ib_device *device,
1759			   ib_comp_handler comp_handler,
1760			   void (*event_handler)(struct ib_event *, void *),
1761			   void *cq_context, int cqe, int comp_vector);
1762
1763/**
1764 * ib_resize_cq - Modifies the capacity of the CQ.
1765 * @cq: The CQ to resize.
1766 * @cqe: The minimum size of the CQ.
1767 *
1768 * Users can examine the cq structure to determine the actual CQ size.
1769 */
1770int ib_resize_cq(struct ib_cq *cq, int cqe);
1771
1772/**
1773 * ib_modify_cq - Modifies moderation params of the CQ
1774 * @cq: The CQ to modify.
1775 * @cq_count: number of CQEs that will trigger an event
1776 * @cq_period: max period of time in usec before triggering an event
1777 *
1778 */
1779int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1780
1781/**
1782 * ib_destroy_cq - Destroys the specified CQ.
1783 * @cq: The CQ to destroy.
1784 */
1785int ib_destroy_cq(struct ib_cq *cq);
1786
1787/**
1788 * ib_poll_cq - poll a CQ for completion(s)
1789 * @cq:the CQ being polled
1790 * @num_entries:maximum number of completions to return
1791 * @wc:array of at least @num_entries &struct ib_wc where completions
1792 *   will be returned
1793 *
1794 * Poll a CQ for (possibly multiple) completions.  If the return value
1795 * is < 0, an error occurred.  If the return value is >= 0, it is the
1796 * number of completions returned.  If the return value is
1797 * non-negative and < num_entries, then the CQ was emptied.
1798 */
1799static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1800			     struct ib_wc *wc)
1801{
1802	return cq->device->poll_cq(cq, num_entries, wc);
1803}
1804
1805/**
1806 * ib_peek_cq - Returns the number of unreaped completions currently
1807 *   on the specified CQ.
1808 * @cq: The CQ to peek.
1809 * @wc_cnt: A minimum number of unreaped completions to check for.
1810 *
1811 * If the number of unreaped completions is greater than or equal to wc_cnt,
1812 * this function returns wc_cnt, otherwise, it returns the actual number of
1813 * unreaped completions.
1814 */
1815int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1816
1817/**
1818 * ib_req_notify_cq - Request completion notification on a CQ.
1819 * @cq: The CQ to generate an event for.
1820 * @flags:
1821 *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1822 *   to request an event on the next solicited event or next work
1823 *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1824 *   may also be |ed in to request a hint about missed events, as
1825 *   described below.
1826 *
1827 * Return Value:
1828 *    < 0 means an error occurred while requesting notification
1829 *   == 0 means notification was requested successfully, and if
1830 *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1831 *        were missed and it is safe to wait for another event.  In
1832 *        this case is it guaranteed that any work completions added
1833 *        to the CQ since the last CQ poll will trigger a completion
1834 *        notification event.
1835 *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1836 *        in.  It means that the consumer must poll the CQ again to
1837 *        make sure it is empty to avoid missing an event because of a
1838 *        race between requesting notification and an entry being
1839 *        added to the CQ.  This return value means it is possible
1840 *        (but not guaranteed) that a work completion has been added
1841 *        to the CQ since the last poll without triggering a
1842 *        completion notification event.
1843 */
1844static inline int ib_req_notify_cq(struct ib_cq *cq,
1845				   enum ib_cq_notify_flags flags)
1846{
1847	return cq->device->req_notify_cq(cq, flags);
1848}
1849
1850/**
1851 * ib_req_ncomp_notif - Request completion notification when there are
1852 *   at least the specified number of unreaped completions on the CQ.
1853 * @cq: The CQ to generate an event for.
1854 * @wc_cnt: The number of unreaped completions that should be on the
1855 *   CQ before an event is generated.
1856 */
1857static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1858{
1859	return cq->device->req_ncomp_notif ?
1860		cq->device->req_ncomp_notif(cq, wc_cnt) :
1861		-ENOSYS;
1862}
1863
1864/**
1865 * ib_get_dma_mr - Returns a memory region for system memory that is
1866 *   usable for DMA.
1867 * @pd: The protection domain associated with the memory region.
1868 * @mr_access_flags: Specifies the memory access rights.
1869 *
1870 * Note that the ib_dma_*() functions defined below must be used
1871 * to create/destroy addresses used with the Lkey or Rkey returned
1872 * by ib_get_dma_mr().
1873 */
1874struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1875
1876/**
1877 * ib_dma_mapping_error - check a DMA addr for error
1878 * @dev: The device for which the dma_addr was created
1879 * @dma_addr: The DMA address to check
1880 */
1881static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1882{
1883	if (dev->dma_ops)
1884		return dev->dma_ops->mapping_error(dev, dma_addr);
1885	return dma_mapping_error(dev->dma_device, dma_addr);
1886}
1887
1888/**
1889 * ib_dma_map_single - Map a kernel virtual address to DMA address
1890 * @dev: The device for which the dma_addr is to be created
1891 * @cpu_addr: The kernel virtual address
1892 * @size: The size of the region in bytes
1893 * @direction: The direction of the DMA
1894 */
1895static inline u64 ib_dma_map_single(struct ib_device *dev,
1896				    void *cpu_addr, size_t size,
1897				    enum dma_data_direction direction)
1898{
1899	if (dev->dma_ops)
1900		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1901	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1902}
1903
1904/**
1905 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1906 * @dev: The device for which the DMA address was created
1907 * @addr: The DMA address
1908 * @size: The size of the region in bytes
1909 * @direction: The direction of the DMA
1910 */
1911static inline void ib_dma_unmap_single(struct ib_device *dev,
1912				       u64 addr, size_t size,
1913				       enum dma_data_direction direction)
1914{
1915	if (dev->dma_ops)
1916		dev->dma_ops->unmap_single(dev, addr, size, direction);
1917	else
1918		dma_unmap_single(dev->dma_device, addr, size, direction);
1919}
1920
1921static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
1922					  void *cpu_addr, size_t size,
1923					  enum dma_data_direction direction,
1924					  struct dma_attrs *attrs)
1925{
1926	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
1927				    direction, attrs);
1928}
1929
1930static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
1931					     u64 addr, size_t size,
1932					     enum dma_data_direction direction,
1933					     struct dma_attrs *attrs)
1934{
1935	return dma_unmap_single_attrs(dev->dma_device, addr, size,
1936				      direction, attrs);
1937}
1938
1939/**
1940 * ib_dma_map_page - Map a physical page to DMA address
1941 * @dev: The device for which the dma_addr is to be created
1942 * @page: The page to be mapped
1943 * @offset: The offset within the page
1944 * @size: The size of the region in bytes
1945 * @direction: The direction of the DMA
1946 */
1947static inline u64 ib_dma_map_page(struct ib_device *dev,
1948				  struct page *page,
1949				  unsigned long offset,
1950				  size_t size,
1951					 enum dma_data_direction direction)
1952{
1953	if (dev->dma_ops)
1954		return dev->dma_ops->map_page(dev, page, offset, size, direction);
1955	return dma_map_page(dev->dma_device, page, offset, size, direction);
1956}
1957
1958/**
1959 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1960 * @dev: The device for which the DMA address was created
1961 * @addr: The DMA address
1962 * @size: The size of the region in bytes
1963 * @direction: The direction of the DMA
1964 */
1965static inline void ib_dma_unmap_page(struct ib_device *dev,
1966				     u64 addr, size_t size,
1967				     enum dma_data_direction direction)
1968{
1969	if (dev->dma_ops)
1970		dev->dma_ops->unmap_page(dev, addr, size, direction);
1971	else
1972		dma_unmap_page(dev->dma_device, addr, size, direction);
1973}
1974
1975/**
1976 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
1977 * @dev: The device for which the DMA addresses are to be created
1978 * @sg: The array of scatter/gather entries
1979 * @nents: The number of scatter/gather entries
1980 * @direction: The direction of the DMA
1981 */
1982static inline int ib_dma_map_sg(struct ib_device *dev,
1983				struct scatterlist *sg, int nents,
1984				enum dma_data_direction direction)
1985{
1986	if (dev->dma_ops)
1987		return dev->dma_ops->map_sg(dev, sg, nents, direction);
1988	return dma_map_sg(dev->dma_device, sg, nents, direction);
1989}
1990
1991/**
1992 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
1993 * @dev: The device for which the DMA addresses were created
1994 * @sg: The array of scatter/gather entries
1995 * @nents: The number of scatter/gather entries
1996 * @direction: The direction of the DMA
1997 */
1998static inline void ib_dma_unmap_sg(struct ib_device *dev,
1999				   struct scatterlist *sg, int nents,
2000				   enum dma_data_direction direction)
2001{
2002	if (dev->dma_ops)
2003		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2004	else
2005		dma_unmap_sg(dev->dma_device, sg, nents, direction);
2006}
2007
2008static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2009				      struct scatterlist *sg, int nents,
2010				      enum dma_data_direction direction,
2011				      struct dma_attrs *attrs)
2012{
2013	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2014}
2015
2016static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2017					 struct scatterlist *sg, int nents,
2018					 enum dma_data_direction direction,
2019					 struct dma_attrs *attrs)
2020{
2021	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2022}
2023/**
2024 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2025 * @dev: The device for which the DMA addresses were created
2026 * @sg: The scatter/gather entry
2027 */
2028static inline u64 ib_sg_dma_address(struct ib_device *dev,
2029				    struct scatterlist *sg)
2030{
2031	if (dev->dma_ops)
2032		return dev->dma_ops->dma_address(dev, sg);
2033	return sg_dma_address(sg);
2034}
2035
2036/**
2037 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2038 * @dev: The device for which the DMA addresses were created
2039 * @sg: The scatter/gather entry
2040 */
2041static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2042					 struct scatterlist *sg)
2043{
2044	if (dev->dma_ops)
2045		return dev->dma_ops->dma_len(dev, sg);
2046	return sg_dma_len(sg);
2047}
2048
2049/**
2050 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2051 * @dev: The device for which the DMA address was created
2052 * @addr: The DMA address
2053 * @size: The size of the region in bytes
2054 * @dir: The direction of the DMA
2055 */
2056static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2057					      u64 addr,
2058					      size_t size,
2059					      enum dma_data_direction dir)
2060{
2061	if (dev->dma_ops)
2062		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2063	else
2064		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2065}
2066
2067/**
2068 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2069 * @dev: The device for which the DMA address was created
2070 * @addr: The DMA address
2071 * @size: The size of the region in bytes
2072 * @dir: The direction of the DMA
2073 */
2074static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2075						 u64 addr,
2076						 size_t size,
2077						 enum dma_data_direction dir)
2078{
2079	if (dev->dma_ops)
2080		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2081	else
2082		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2083}
2084
2085/**
2086 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2087 * @dev: The device for which the DMA address is requested
2088 * @size: The size of the region to allocate in bytes
2089 * @dma_handle: A pointer for returning the DMA address of the region
2090 * @flag: memory allocator flags
2091 */
2092static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2093					   size_t size,
2094					   u64 *dma_handle,
2095					   gfp_t flag)
2096{
2097	if (dev->dma_ops)
2098		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2099	else {
2100		dma_addr_t handle;
2101		void *ret;
2102
2103		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2104		*dma_handle = handle;
2105		return ret;
2106	}
2107}
2108
2109/**
2110 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2111 * @dev: The device for which the DMA addresses were allocated
2112 * @size: The size of the region
2113 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2114 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2115 */
2116static inline void ib_dma_free_coherent(struct ib_device *dev,
2117					size_t size, void *cpu_addr,
2118					u64 dma_handle)
2119{
2120	if (dev->dma_ops)
2121		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2122	else
2123		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2124}
2125
2126/**
2127 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
2128 *   by an HCA.
2129 * @pd: The protection domain associated assigned to the registered region.
2130 * @phys_buf_array: Specifies a list of physical buffers to use in the
2131 *   memory region.
2132 * @num_phys_buf: Specifies the size of the phys_buf_array.
2133 * @mr_access_flags: Specifies the memory access rights.
2134 * @iova_start: The offset of the region's starting I/O virtual address.
2135 */
2136struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
2137			     struct ib_phys_buf *phys_buf_array,
2138			     int num_phys_buf,
2139			     int mr_access_flags,
2140			     u64 *iova_start);
2141
2142/**
2143 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
2144 *   Conceptually, this call performs the functions deregister memory region
2145 *   followed by register physical memory region.  Where possible,
2146 *   resources are reused instead of deallocated and reallocated.
2147 * @mr: The memory region to modify.
2148 * @mr_rereg_mask: A bit-mask used to indicate which of the following
2149 *   properties of the memory region are being modified.
2150 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
2151 *   the new protection domain to associated with the memory region,
2152 *   otherwise, this parameter is ignored.
2153 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2154 *   field specifies a list of physical buffers to use in the new
2155 *   translation, otherwise, this parameter is ignored.
2156 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2157 *   field specifies the size of the phys_buf_array, otherwise, this
2158 *   parameter is ignored.
2159 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
2160 *   field specifies the new memory access rights, otherwise, this
2161 *   parameter is ignored.
2162 * @iova_start: The offset of the region's starting I/O virtual address.
2163 */
2164int ib_rereg_phys_mr(struct ib_mr *mr,
2165		     int mr_rereg_mask,
2166		     struct ib_pd *pd,
2167		     struct ib_phys_buf *phys_buf_array,
2168		     int num_phys_buf,
2169		     int mr_access_flags,
2170		     u64 *iova_start);
2171
2172/**
2173 * ib_query_mr - Retrieves information about a specific memory region.
2174 * @mr: The memory region to retrieve information about.
2175 * @mr_attr: The attributes of the specified memory region.
2176 */
2177int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2178
2179/**
2180 * ib_dereg_mr - Deregisters a memory region and removes it from the
2181 *   HCA translation table.
2182 * @mr: The memory region to deregister.
2183 */
2184int ib_dereg_mr(struct ib_mr *mr);
2185
2186/**
2187 * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2188 *   IB_WR_FAST_REG_MR send work request.
2189 * @pd: The protection domain associated with the region.
2190 * @max_page_list_len: requested max physical buffer list length to be
2191 *   used with fast register work requests for this MR.
2192 */
2193struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
2194
2195/**
2196 * ib_alloc_fast_reg_page_list - Allocates a page list array
2197 * @device - ib device pointer.
2198 * @page_list_len - size of the page list array to be allocated.
2199 *
2200 * This allocates and returns a struct ib_fast_reg_page_list * and a
2201 * page_list array that is at least page_list_len in size.  The actual
2202 * size is returned in max_page_list_len.  The caller is responsible
2203 * for initializing the contents of the page_list array before posting
2204 * a send work request with the IB_WC_FAST_REG_MR opcode.
2205 *
2206 * The page_list array entries must be translated using one of the
2207 * ib_dma_*() functions just like the addresses passed to
2208 * ib_map_phys_fmr().  Once the ib_post_send() is issued, the struct
2209 * ib_fast_reg_page_list must not be modified by the caller until the
2210 * IB_WC_FAST_REG_MR work request completes.
2211 */
2212struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2213				struct ib_device *device, int page_list_len);
2214
2215/**
2216 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2217 *   page list array.
2218 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2219 */
2220void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2221
2222/**
2223 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2224 *   R_Key and L_Key.
2225 * @mr - struct ib_mr pointer to be updated.
2226 * @newkey - new key to be used.
2227 */
2228static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2229{
2230	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2231	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2232}
2233
2234/**
2235 * ib_alloc_mw - Allocates a memory window.
2236 * @pd: The protection domain associated with the memory window.
2237 */
2238struct ib_mw *ib_alloc_mw(struct ib_pd *pd);
2239
2240/**
2241 * ib_bind_mw - Posts a work request to the send queue of the specified
2242 *   QP, which binds the memory window to the given address range and
2243 *   remote access attributes.
2244 * @qp: QP to post the bind work request on.
2245 * @mw: The memory window to bind.
2246 * @mw_bind: Specifies information about the memory window, including
2247 *   its address range, remote access rights, and associated memory region.
2248 */
2249static inline int ib_bind_mw(struct ib_qp *qp,
2250			     struct ib_mw *mw,
2251			     struct ib_mw_bind *mw_bind)
2252{
2253	/* XXX reference counting in corresponding MR? */
2254	return mw->device->bind_mw ?
2255		mw->device->bind_mw(qp, mw, mw_bind) :
2256		-ENOSYS;
2257}
2258
2259/**
2260 * ib_dealloc_mw - Deallocates a memory window.
2261 * @mw: The memory window to deallocate.
2262 */
2263int ib_dealloc_mw(struct ib_mw *mw);
2264
2265/**
2266 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2267 * @pd: The protection domain associated with the unmapped region.
2268 * @mr_access_flags: Specifies the memory access rights.
2269 * @fmr_attr: Attributes of the unmapped region.
2270 *
2271 * A fast memory region must be mapped before it can be used as part of
2272 * a work request.
2273 */
2274struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2275			    int mr_access_flags,
2276			    struct ib_fmr_attr *fmr_attr);
2277
2278/**
2279 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2280 * @fmr: The fast memory region to associate with the pages.
2281 * @page_list: An array of physical pages to map to the fast memory region.
2282 * @list_len: The number of pages in page_list.
2283 * @iova: The I/O virtual address to use with the mapped region.
2284 */
2285static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2286				  u64 *page_list, int list_len,
2287				  u64 iova)
2288{
2289	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2290}
2291
2292/**
2293 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2294 * @fmr_list: A linked list of fast memory regions to unmap.
2295 */
2296int ib_unmap_fmr(struct list_head *fmr_list);
2297
2298/**
2299 * ib_dealloc_fmr - Deallocates a fast memory region.
2300 * @fmr: The fast memory region to deallocate.
2301 */
2302int ib_dealloc_fmr(struct ib_fmr *fmr);
2303
2304/**
2305 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2306 * @qp: QP to attach to the multicast group.  The QP must be type
2307 *   IB_QPT_UD.
2308 * @gid: Multicast group GID.
2309 * @lid: Multicast group LID in host byte order.
2310 *
2311 * In order to send and receive multicast packets, subnet
2312 * administration must have created the multicast group and configured
2313 * the fabric appropriately.  The port associated with the specified
2314 * QP must also be a member of the multicast group.
2315 */
2316int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2317
2318/**
2319 * ib_detach_mcast - Detaches the specified QP from a multicast group.
2320 * @qp: QP to detach from the multicast group.
2321 * @gid: Multicast group GID.
2322 * @lid: Multicast group LID in host byte order.
2323 */
2324int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2325
2326/**
2327 * ib_alloc_xrcd - Allocates an XRC domain.
2328 * @device: The device on which to allocate the XRC domain.
2329 */
2330struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2331
2332/**
2333 * ib_dealloc_xrcd - Deallocates an XRC domain.
2334 * @xrcd: The XRC domain to deallocate.
2335 */
2336int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2337
2338int ib_attach_flow(struct ib_qp *qp, struct ib_flow_spec *spec, int priority);
2339int ib_detach_flow(struct ib_qp *qp, struct ib_flow_spec *spec, int priority);
2340
2341#endif /* IB_VERBS_H */
2342