xref: /linux/include/rdma/ib_verbs.h (revision 60e13231561b3a4c5269bfa1ef6c0569ad6f28ec)
1 /*
2  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
3  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
4  * Copyright (c) 2004 Intel Corporation.  All rights reserved.
5  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
6  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
7  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
9  *
10  * This software is available to you under a choice of one of two
11  * licenses.  You may choose to be licensed under the terms of the GNU
12  * General Public License (GPL) Version 2, available from the file
13  * COPYING in the main directory of this source tree, or the
14  * OpenIB.org BSD license below:
15  *
16  *     Redistribution and use in source and binary forms, with or
17  *     without modification, are permitted provided that the following
18  *     conditions are met:
19  *
20  *      - Redistributions of source code must retain the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer.
23  *
24  *      - Redistributions in binary form must reproduce the above
25  *        copyright notice, this list of conditions and the following
26  *        disclaimer in the documentation and/or other materials
27  *        provided with the distribution.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36  * SOFTWARE.
37  */
38 
39 #if !defined(IB_VERBS_H)
40 #define IB_VERBS_H
41 
42 #include <linux/types.h>
43 #include <linux/device.h>
44 #include <linux/mm.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 
52 #include <linux/atomic.h>
53 #include <asm/uaccess.h>
54 
55 extern struct workqueue_struct *ib_wq;
56 
57 union ib_gid {
58 	u8	raw[16];
59 	struct {
60 		__be64	subnet_prefix;
61 		__be64	interface_id;
62 	} global;
63 };
64 
65 enum rdma_node_type {
66 	/* IB values map to NodeInfo:NodeType. */
67 	RDMA_NODE_IB_CA 	= 1,
68 	RDMA_NODE_IB_SWITCH,
69 	RDMA_NODE_IB_ROUTER,
70 	RDMA_NODE_RNIC
71 };
72 
73 enum rdma_transport_type {
74 	RDMA_TRANSPORT_IB,
75 	RDMA_TRANSPORT_IWARP
76 };
77 
78 enum rdma_transport_type
79 rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__;
80 
81 enum rdma_link_layer {
82 	IB_LINK_LAYER_UNSPECIFIED,
83 	IB_LINK_LAYER_INFINIBAND,
84 	IB_LINK_LAYER_ETHERNET,
85 };
86 
87 enum ib_device_cap_flags {
88 	IB_DEVICE_RESIZE_MAX_WR		= 1,
89 	IB_DEVICE_BAD_PKEY_CNTR		= (1<<1),
90 	IB_DEVICE_BAD_QKEY_CNTR		= (1<<2),
91 	IB_DEVICE_RAW_MULTI		= (1<<3),
92 	IB_DEVICE_AUTO_PATH_MIG		= (1<<4),
93 	IB_DEVICE_CHANGE_PHY_PORT	= (1<<5),
94 	IB_DEVICE_UD_AV_PORT_ENFORCE	= (1<<6),
95 	IB_DEVICE_CURR_QP_STATE_MOD	= (1<<7),
96 	IB_DEVICE_SHUTDOWN_PORT		= (1<<8),
97 	IB_DEVICE_INIT_TYPE		= (1<<9),
98 	IB_DEVICE_PORT_ACTIVE_EVENT	= (1<<10),
99 	IB_DEVICE_SYS_IMAGE_GUID	= (1<<11),
100 	IB_DEVICE_RC_RNR_NAK_GEN	= (1<<12),
101 	IB_DEVICE_SRQ_RESIZE		= (1<<13),
102 	IB_DEVICE_N_NOTIFY_CQ		= (1<<14),
103 	IB_DEVICE_LOCAL_DMA_LKEY	= (1<<15),
104 	IB_DEVICE_RESERVED		= (1<<16), /* old SEND_W_INV */
105 	IB_DEVICE_MEM_WINDOW		= (1<<17),
106 	/*
107 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
108 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
109 	 * messages and can verify the validity of checksum for
110 	 * incoming messages.  Setting this flag implies that the
111 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
112 	 */
113 	IB_DEVICE_UD_IP_CSUM		= (1<<18),
114 	IB_DEVICE_UD_TSO		= (1<<19),
115 	IB_DEVICE_MEM_MGT_EXTENSIONS	= (1<<21),
116 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
117 };
118 
119 enum ib_atomic_cap {
120 	IB_ATOMIC_NONE,
121 	IB_ATOMIC_HCA,
122 	IB_ATOMIC_GLOB
123 };
124 
125 struct ib_device_attr {
126 	u64			fw_ver;
127 	__be64			sys_image_guid;
128 	u64			max_mr_size;
129 	u64			page_size_cap;
130 	u32			vendor_id;
131 	u32			vendor_part_id;
132 	u32			hw_ver;
133 	int			max_qp;
134 	int			max_qp_wr;
135 	int			device_cap_flags;
136 	int			max_sge;
137 	int			max_sge_rd;
138 	int			max_cq;
139 	int			max_cqe;
140 	int			max_mr;
141 	int			max_pd;
142 	int			max_qp_rd_atom;
143 	int			max_ee_rd_atom;
144 	int			max_res_rd_atom;
145 	int			max_qp_init_rd_atom;
146 	int			max_ee_init_rd_atom;
147 	enum ib_atomic_cap	atomic_cap;
148 	enum ib_atomic_cap	masked_atomic_cap;
149 	int			max_ee;
150 	int			max_rdd;
151 	int			max_mw;
152 	int			max_raw_ipv6_qp;
153 	int			max_raw_ethy_qp;
154 	int			max_mcast_grp;
155 	int			max_mcast_qp_attach;
156 	int			max_total_mcast_qp_attach;
157 	int			max_ah;
158 	int			max_fmr;
159 	int			max_map_per_fmr;
160 	int			max_srq;
161 	int			max_srq_wr;
162 	int			max_srq_sge;
163 	unsigned int		max_fast_reg_page_list_len;
164 	u16			max_pkeys;
165 	u8			local_ca_ack_delay;
166 };
167 
168 enum ib_mtu {
169 	IB_MTU_256  = 1,
170 	IB_MTU_512  = 2,
171 	IB_MTU_1024 = 3,
172 	IB_MTU_2048 = 4,
173 	IB_MTU_4096 = 5
174 };
175 
176 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
177 {
178 	switch (mtu) {
179 	case IB_MTU_256:  return  256;
180 	case IB_MTU_512:  return  512;
181 	case IB_MTU_1024: return 1024;
182 	case IB_MTU_2048: return 2048;
183 	case IB_MTU_4096: return 4096;
184 	default: 	  return -1;
185 	}
186 }
187 
188 enum ib_port_state {
189 	IB_PORT_NOP		= 0,
190 	IB_PORT_DOWN		= 1,
191 	IB_PORT_INIT		= 2,
192 	IB_PORT_ARMED		= 3,
193 	IB_PORT_ACTIVE		= 4,
194 	IB_PORT_ACTIVE_DEFER	= 5
195 };
196 
197 enum ib_port_cap_flags {
198 	IB_PORT_SM				= 1 <<  1,
199 	IB_PORT_NOTICE_SUP			= 1 <<  2,
200 	IB_PORT_TRAP_SUP			= 1 <<  3,
201 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
202 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
203 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
204 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
205 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
206 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
207 	IB_PORT_SM_DISABLED			= 1 << 10,
208 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
209 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
210 	IB_PORT_CM_SUP				= 1 << 16,
211 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
212 	IB_PORT_REINIT_SUP			= 1 << 18,
213 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
214 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
215 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
216 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
217 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
218 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
219 	IB_PORT_CLIENT_REG_SUP			= 1 << 25
220 };
221 
222 enum ib_port_width {
223 	IB_WIDTH_1X	= 1,
224 	IB_WIDTH_4X	= 2,
225 	IB_WIDTH_8X	= 4,
226 	IB_WIDTH_12X	= 8
227 };
228 
229 static inline int ib_width_enum_to_int(enum ib_port_width width)
230 {
231 	switch (width) {
232 	case IB_WIDTH_1X:  return  1;
233 	case IB_WIDTH_4X:  return  4;
234 	case IB_WIDTH_8X:  return  8;
235 	case IB_WIDTH_12X: return 12;
236 	default: 	  return -1;
237 	}
238 }
239 
240 struct ib_protocol_stats {
241 	/* TBD... */
242 };
243 
244 struct iw_protocol_stats {
245 	u64	ipInReceives;
246 	u64	ipInHdrErrors;
247 	u64	ipInTooBigErrors;
248 	u64	ipInNoRoutes;
249 	u64	ipInAddrErrors;
250 	u64	ipInUnknownProtos;
251 	u64	ipInTruncatedPkts;
252 	u64	ipInDiscards;
253 	u64	ipInDelivers;
254 	u64	ipOutForwDatagrams;
255 	u64	ipOutRequests;
256 	u64	ipOutDiscards;
257 	u64	ipOutNoRoutes;
258 	u64	ipReasmTimeout;
259 	u64	ipReasmReqds;
260 	u64	ipReasmOKs;
261 	u64	ipReasmFails;
262 	u64	ipFragOKs;
263 	u64	ipFragFails;
264 	u64	ipFragCreates;
265 	u64	ipInMcastPkts;
266 	u64	ipOutMcastPkts;
267 	u64	ipInBcastPkts;
268 	u64	ipOutBcastPkts;
269 
270 	u64	tcpRtoAlgorithm;
271 	u64	tcpRtoMin;
272 	u64	tcpRtoMax;
273 	u64	tcpMaxConn;
274 	u64	tcpActiveOpens;
275 	u64	tcpPassiveOpens;
276 	u64	tcpAttemptFails;
277 	u64	tcpEstabResets;
278 	u64	tcpCurrEstab;
279 	u64	tcpInSegs;
280 	u64	tcpOutSegs;
281 	u64	tcpRetransSegs;
282 	u64	tcpInErrs;
283 	u64	tcpOutRsts;
284 };
285 
286 union rdma_protocol_stats {
287 	struct ib_protocol_stats	ib;
288 	struct iw_protocol_stats	iw;
289 };
290 
291 struct ib_port_attr {
292 	enum ib_port_state	state;
293 	enum ib_mtu		max_mtu;
294 	enum ib_mtu		active_mtu;
295 	int			gid_tbl_len;
296 	u32			port_cap_flags;
297 	u32			max_msg_sz;
298 	u32			bad_pkey_cntr;
299 	u32			qkey_viol_cntr;
300 	u16			pkey_tbl_len;
301 	u16			lid;
302 	u16			sm_lid;
303 	u8			lmc;
304 	u8			max_vl_num;
305 	u8			sm_sl;
306 	u8			subnet_timeout;
307 	u8			init_type_reply;
308 	u8			active_width;
309 	u8			active_speed;
310 	u8                      phys_state;
311 };
312 
313 enum ib_device_modify_flags {
314 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
315 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
316 };
317 
318 struct ib_device_modify {
319 	u64	sys_image_guid;
320 	char	node_desc[64];
321 };
322 
323 enum ib_port_modify_flags {
324 	IB_PORT_SHUTDOWN		= 1,
325 	IB_PORT_INIT_TYPE		= (1<<2),
326 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
327 };
328 
329 struct ib_port_modify {
330 	u32	set_port_cap_mask;
331 	u32	clr_port_cap_mask;
332 	u8	init_type;
333 };
334 
335 enum ib_event_type {
336 	IB_EVENT_CQ_ERR,
337 	IB_EVENT_QP_FATAL,
338 	IB_EVENT_QP_REQ_ERR,
339 	IB_EVENT_QP_ACCESS_ERR,
340 	IB_EVENT_COMM_EST,
341 	IB_EVENT_SQ_DRAINED,
342 	IB_EVENT_PATH_MIG,
343 	IB_EVENT_PATH_MIG_ERR,
344 	IB_EVENT_DEVICE_FATAL,
345 	IB_EVENT_PORT_ACTIVE,
346 	IB_EVENT_PORT_ERR,
347 	IB_EVENT_LID_CHANGE,
348 	IB_EVENT_PKEY_CHANGE,
349 	IB_EVENT_SM_CHANGE,
350 	IB_EVENT_SRQ_ERR,
351 	IB_EVENT_SRQ_LIMIT_REACHED,
352 	IB_EVENT_QP_LAST_WQE_REACHED,
353 	IB_EVENT_CLIENT_REREGISTER,
354 	IB_EVENT_GID_CHANGE,
355 };
356 
357 struct ib_event {
358 	struct ib_device	*device;
359 	union {
360 		struct ib_cq	*cq;
361 		struct ib_qp	*qp;
362 		struct ib_srq	*srq;
363 		u8		port_num;
364 	} element;
365 	enum ib_event_type	event;
366 };
367 
368 struct ib_event_handler {
369 	struct ib_device *device;
370 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
371 	struct list_head  list;
372 };
373 
374 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
375 	do {							\
376 		(_ptr)->device  = _device;			\
377 		(_ptr)->handler = _handler;			\
378 		INIT_LIST_HEAD(&(_ptr)->list);			\
379 	} while (0)
380 
381 struct ib_global_route {
382 	union ib_gid	dgid;
383 	u32		flow_label;
384 	u8		sgid_index;
385 	u8		hop_limit;
386 	u8		traffic_class;
387 };
388 
389 struct ib_grh {
390 	__be32		version_tclass_flow;
391 	__be16		paylen;
392 	u8		next_hdr;
393 	u8		hop_limit;
394 	union ib_gid	sgid;
395 	union ib_gid	dgid;
396 };
397 
398 enum {
399 	IB_MULTICAST_QPN = 0xffffff
400 };
401 
402 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
403 
404 enum ib_ah_flags {
405 	IB_AH_GRH	= 1
406 };
407 
408 enum ib_rate {
409 	IB_RATE_PORT_CURRENT = 0,
410 	IB_RATE_2_5_GBPS = 2,
411 	IB_RATE_5_GBPS   = 5,
412 	IB_RATE_10_GBPS  = 3,
413 	IB_RATE_20_GBPS  = 6,
414 	IB_RATE_30_GBPS  = 4,
415 	IB_RATE_40_GBPS  = 7,
416 	IB_RATE_60_GBPS  = 8,
417 	IB_RATE_80_GBPS  = 9,
418 	IB_RATE_120_GBPS = 10
419 };
420 
421 /**
422  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
423  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
424  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
425  * @rate: rate to convert.
426  */
427 int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
428 
429 /**
430  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
431  * enum.
432  * @mult: multiple to convert.
433  */
434 enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
435 
436 struct ib_ah_attr {
437 	struct ib_global_route	grh;
438 	u16			dlid;
439 	u8			sl;
440 	u8			src_path_bits;
441 	u8			static_rate;
442 	u8			ah_flags;
443 	u8			port_num;
444 };
445 
446 enum ib_wc_status {
447 	IB_WC_SUCCESS,
448 	IB_WC_LOC_LEN_ERR,
449 	IB_WC_LOC_QP_OP_ERR,
450 	IB_WC_LOC_EEC_OP_ERR,
451 	IB_WC_LOC_PROT_ERR,
452 	IB_WC_WR_FLUSH_ERR,
453 	IB_WC_MW_BIND_ERR,
454 	IB_WC_BAD_RESP_ERR,
455 	IB_WC_LOC_ACCESS_ERR,
456 	IB_WC_REM_INV_REQ_ERR,
457 	IB_WC_REM_ACCESS_ERR,
458 	IB_WC_REM_OP_ERR,
459 	IB_WC_RETRY_EXC_ERR,
460 	IB_WC_RNR_RETRY_EXC_ERR,
461 	IB_WC_LOC_RDD_VIOL_ERR,
462 	IB_WC_REM_INV_RD_REQ_ERR,
463 	IB_WC_REM_ABORT_ERR,
464 	IB_WC_INV_EECN_ERR,
465 	IB_WC_INV_EEC_STATE_ERR,
466 	IB_WC_FATAL_ERR,
467 	IB_WC_RESP_TIMEOUT_ERR,
468 	IB_WC_GENERAL_ERR
469 };
470 
471 enum ib_wc_opcode {
472 	IB_WC_SEND,
473 	IB_WC_RDMA_WRITE,
474 	IB_WC_RDMA_READ,
475 	IB_WC_COMP_SWAP,
476 	IB_WC_FETCH_ADD,
477 	IB_WC_BIND_MW,
478 	IB_WC_LSO,
479 	IB_WC_LOCAL_INV,
480 	IB_WC_FAST_REG_MR,
481 	IB_WC_MASKED_COMP_SWAP,
482 	IB_WC_MASKED_FETCH_ADD,
483 /*
484  * Set value of IB_WC_RECV so consumers can test if a completion is a
485  * receive by testing (opcode & IB_WC_RECV).
486  */
487 	IB_WC_RECV			= 1 << 7,
488 	IB_WC_RECV_RDMA_WITH_IMM
489 };
490 
491 enum ib_wc_flags {
492 	IB_WC_GRH		= 1,
493 	IB_WC_WITH_IMM		= (1<<1),
494 	IB_WC_WITH_INVALIDATE	= (1<<2),
495 };
496 
497 struct ib_wc {
498 	u64			wr_id;
499 	enum ib_wc_status	status;
500 	enum ib_wc_opcode	opcode;
501 	u32			vendor_err;
502 	u32			byte_len;
503 	struct ib_qp	       *qp;
504 	union {
505 		__be32		imm_data;
506 		u32		invalidate_rkey;
507 	} ex;
508 	u32			src_qp;
509 	int			wc_flags;
510 	u16			pkey_index;
511 	u16			slid;
512 	u8			sl;
513 	u8			dlid_path_bits;
514 	u8			port_num;	/* valid only for DR SMPs on switches */
515 	int			csum_ok;
516 };
517 
518 enum ib_cq_notify_flags {
519 	IB_CQ_SOLICITED			= 1 << 0,
520 	IB_CQ_NEXT_COMP			= 1 << 1,
521 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
522 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
523 };
524 
525 enum ib_srq_attr_mask {
526 	IB_SRQ_MAX_WR	= 1 << 0,
527 	IB_SRQ_LIMIT	= 1 << 1,
528 };
529 
530 struct ib_srq_attr {
531 	u32	max_wr;
532 	u32	max_sge;
533 	u32	srq_limit;
534 };
535 
536 struct ib_srq_init_attr {
537 	void		      (*event_handler)(struct ib_event *, void *);
538 	void		       *srq_context;
539 	struct ib_srq_attr	attr;
540 };
541 
542 struct ib_qp_cap {
543 	u32	max_send_wr;
544 	u32	max_recv_wr;
545 	u32	max_send_sge;
546 	u32	max_recv_sge;
547 	u32	max_inline_data;
548 };
549 
550 enum ib_sig_type {
551 	IB_SIGNAL_ALL_WR,
552 	IB_SIGNAL_REQ_WR
553 };
554 
555 enum ib_qp_type {
556 	/*
557 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
558 	 * here (and in that order) since the MAD layer uses them as
559 	 * indices into a 2-entry table.
560 	 */
561 	IB_QPT_SMI,
562 	IB_QPT_GSI,
563 
564 	IB_QPT_RC,
565 	IB_QPT_UC,
566 	IB_QPT_UD,
567 	IB_QPT_RAW_IPV6,
568 	IB_QPT_RAW_ETHERTYPE
569 };
570 
571 enum ib_qp_create_flags {
572 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
573 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
574 };
575 
576 struct ib_qp_init_attr {
577 	void                  (*event_handler)(struct ib_event *, void *);
578 	void		       *qp_context;
579 	struct ib_cq	       *send_cq;
580 	struct ib_cq	       *recv_cq;
581 	struct ib_srq	       *srq;
582 	struct ib_qp_cap	cap;
583 	enum ib_sig_type	sq_sig_type;
584 	enum ib_qp_type		qp_type;
585 	enum ib_qp_create_flags	create_flags;
586 	u8			port_num; /* special QP types only */
587 };
588 
589 enum ib_rnr_timeout {
590 	IB_RNR_TIMER_655_36 =  0,
591 	IB_RNR_TIMER_000_01 =  1,
592 	IB_RNR_TIMER_000_02 =  2,
593 	IB_RNR_TIMER_000_03 =  3,
594 	IB_RNR_TIMER_000_04 =  4,
595 	IB_RNR_TIMER_000_06 =  5,
596 	IB_RNR_TIMER_000_08 =  6,
597 	IB_RNR_TIMER_000_12 =  7,
598 	IB_RNR_TIMER_000_16 =  8,
599 	IB_RNR_TIMER_000_24 =  9,
600 	IB_RNR_TIMER_000_32 = 10,
601 	IB_RNR_TIMER_000_48 = 11,
602 	IB_RNR_TIMER_000_64 = 12,
603 	IB_RNR_TIMER_000_96 = 13,
604 	IB_RNR_TIMER_001_28 = 14,
605 	IB_RNR_TIMER_001_92 = 15,
606 	IB_RNR_TIMER_002_56 = 16,
607 	IB_RNR_TIMER_003_84 = 17,
608 	IB_RNR_TIMER_005_12 = 18,
609 	IB_RNR_TIMER_007_68 = 19,
610 	IB_RNR_TIMER_010_24 = 20,
611 	IB_RNR_TIMER_015_36 = 21,
612 	IB_RNR_TIMER_020_48 = 22,
613 	IB_RNR_TIMER_030_72 = 23,
614 	IB_RNR_TIMER_040_96 = 24,
615 	IB_RNR_TIMER_061_44 = 25,
616 	IB_RNR_TIMER_081_92 = 26,
617 	IB_RNR_TIMER_122_88 = 27,
618 	IB_RNR_TIMER_163_84 = 28,
619 	IB_RNR_TIMER_245_76 = 29,
620 	IB_RNR_TIMER_327_68 = 30,
621 	IB_RNR_TIMER_491_52 = 31
622 };
623 
624 enum ib_qp_attr_mask {
625 	IB_QP_STATE			= 1,
626 	IB_QP_CUR_STATE			= (1<<1),
627 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
628 	IB_QP_ACCESS_FLAGS		= (1<<3),
629 	IB_QP_PKEY_INDEX		= (1<<4),
630 	IB_QP_PORT			= (1<<5),
631 	IB_QP_QKEY			= (1<<6),
632 	IB_QP_AV			= (1<<7),
633 	IB_QP_PATH_MTU			= (1<<8),
634 	IB_QP_TIMEOUT			= (1<<9),
635 	IB_QP_RETRY_CNT			= (1<<10),
636 	IB_QP_RNR_RETRY			= (1<<11),
637 	IB_QP_RQ_PSN			= (1<<12),
638 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
639 	IB_QP_ALT_PATH			= (1<<14),
640 	IB_QP_MIN_RNR_TIMER		= (1<<15),
641 	IB_QP_SQ_PSN			= (1<<16),
642 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
643 	IB_QP_PATH_MIG_STATE		= (1<<18),
644 	IB_QP_CAP			= (1<<19),
645 	IB_QP_DEST_QPN			= (1<<20)
646 };
647 
648 enum ib_qp_state {
649 	IB_QPS_RESET,
650 	IB_QPS_INIT,
651 	IB_QPS_RTR,
652 	IB_QPS_RTS,
653 	IB_QPS_SQD,
654 	IB_QPS_SQE,
655 	IB_QPS_ERR
656 };
657 
658 enum ib_mig_state {
659 	IB_MIG_MIGRATED,
660 	IB_MIG_REARM,
661 	IB_MIG_ARMED
662 };
663 
664 struct ib_qp_attr {
665 	enum ib_qp_state	qp_state;
666 	enum ib_qp_state	cur_qp_state;
667 	enum ib_mtu		path_mtu;
668 	enum ib_mig_state	path_mig_state;
669 	u32			qkey;
670 	u32			rq_psn;
671 	u32			sq_psn;
672 	u32			dest_qp_num;
673 	int			qp_access_flags;
674 	struct ib_qp_cap	cap;
675 	struct ib_ah_attr	ah_attr;
676 	struct ib_ah_attr	alt_ah_attr;
677 	u16			pkey_index;
678 	u16			alt_pkey_index;
679 	u8			en_sqd_async_notify;
680 	u8			sq_draining;
681 	u8			max_rd_atomic;
682 	u8			max_dest_rd_atomic;
683 	u8			min_rnr_timer;
684 	u8			port_num;
685 	u8			timeout;
686 	u8			retry_cnt;
687 	u8			rnr_retry;
688 	u8			alt_port_num;
689 	u8			alt_timeout;
690 };
691 
692 enum ib_wr_opcode {
693 	IB_WR_RDMA_WRITE,
694 	IB_WR_RDMA_WRITE_WITH_IMM,
695 	IB_WR_SEND,
696 	IB_WR_SEND_WITH_IMM,
697 	IB_WR_RDMA_READ,
698 	IB_WR_ATOMIC_CMP_AND_SWP,
699 	IB_WR_ATOMIC_FETCH_AND_ADD,
700 	IB_WR_LSO,
701 	IB_WR_SEND_WITH_INV,
702 	IB_WR_RDMA_READ_WITH_INV,
703 	IB_WR_LOCAL_INV,
704 	IB_WR_FAST_REG_MR,
705 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
706 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
707 };
708 
709 enum ib_send_flags {
710 	IB_SEND_FENCE		= 1,
711 	IB_SEND_SIGNALED	= (1<<1),
712 	IB_SEND_SOLICITED	= (1<<2),
713 	IB_SEND_INLINE		= (1<<3),
714 	IB_SEND_IP_CSUM		= (1<<4)
715 };
716 
717 struct ib_sge {
718 	u64	addr;
719 	u32	length;
720 	u32	lkey;
721 };
722 
723 struct ib_fast_reg_page_list {
724 	struct ib_device       *device;
725 	u64		       *page_list;
726 	unsigned int		max_page_list_len;
727 };
728 
729 struct ib_send_wr {
730 	struct ib_send_wr      *next;
731 	u64			wr_id;
732 	struct ib_sge	       *sg_list;
733 	int			num_sge;
734 	enum ib_wr_opcode	opcode;
735 	int			send_flags;
736 	union {
737 		__be32		imm_data;
738 		u32		invalidate_rkey;
739 	} ex;
740 	union {
741 		struct {
742 			u64	remote_addr;
743 			u32	rkey;
744 		} rdma;
745 		struct {
746 			u64	remote_addr;
747 			u64	compare_add;
748 			u64	swap;
749 			u64	compare_add_mask;
750 			u64	swap_mask;
751 			u32	rkey;
752 		} atomic;
753 		struct {
754 			struct ib_ah *ah;
755 			void   *header;
756 			int     hlen;
757 			int     mss;
758 			u32	remote_qpn;
759 			u32	remote_qkey;
760 			u16	pkey_index; /* valid for GSI only */
761 			u8	port_num;   /* valid for DR SMPs on switch only */
762 		} ud;
763 		struct {
764 			u64				iova_start;
765 			struct ib_fast_reg_page_list   *page_list;
766 			unsigned int			page_shift;
767 			unsigned int			page_list_len;
768 			u32				length;
769 			int				access_flags;
770 			u32				rkey;
771 		} fast_reg;
772 	} wr;
773 };
774 
775 struct ib_recv_wr {
776 	struct ib_recv_wr      *next;
777 	u64			wr_id;
778 	struct ib_sge	       *sg_list;
779 	int			num_sge;
780 };
781 
782 enum ib_access_flags {
783 	IB_ACCESS_LOCAL_WRITE	= 1,
784 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
785 	IB_ACCESS_REMOTE_READ	= (1<<2),
786 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
787 	IB_ACCESS_MW_BIND	= (1<<4)
788 };
789 
790 struct ib_phys_buf {
791 	u64      addr;
792 	u64      size;
793 };
794 
795 struct ib_mr_attr {
796 	struct ib_pd	*pd;
797 	u64		device_virt_addr;
798 	u64		size;
799 	int		mr_access_flags;
800 	u32		lkey;
801 	u32		rkey;
802 };
803 
804 enum ib_mr_rereg_flags {
805 	IB_MR_REREG_TRANS	= 1,
806 	IB_MR_REREG_PD		= (1<<1),
807 	IB_MR_REREG_ACCESS	= (1<<2)
808 };
809 
810 struct ib_mw_bind {
811 	struct ib_mr   *mr;
812 	u64		wr_id;
813 	u64		addr;
814 	u32		length;
815 	int		send_flags;
816 	int		mw_access_flags;
817 };
818 
819 struct ib_fmr_attr {
820 	int	max_pages;
821 	int	max_maps;
822 	u8	page_shift;
823 };
824 
825 struct ib_ucontext {
826 	struct ib_device       *device;
827 	struct list_head	pd_list;
828 	struct list_head	mr_list;
829 	struct list_head	mw_list;
830 	struct list_head	cq_list;
831 	struct list_head	qp_list;
832 	struct list_head	srq_list;
833 	struct list_head	ah_list;
834 	int			closing;
835 };
836 
837 struct ib_uobject {
838 	u64			user_handle;	/* handle given to us by userspace */
839 	struct ib_ucontext     *context;	/* associated user context */
840 	void		       *object;		/* containing object */
841 	struct list_head	list;		/* link to context's list */
842 	int			id;		/* index into kernel idr */
843 	struct kref		ref;
844 	struct rw_semaphore	mutex;		/* protects .live */
845 	int			live;
846 };
847 
848 struct ib_udata {
849 	void __user *inbuf;
850 	void __user *outbuf;
851 	size_t       inlen;
852 	size_t       outlen;
853 };
854 
855 struct ib_pd {
856 	struct ib_device       *device;
857 	struct ib_uobject      *uobject;
858 	atomic_t          	usecnt; /* count all resources */
859 };
860 
861 struct ib_ah {
862 	struct ib_device	*device;
863 	struct ib_pd		*pd;
864 	struct ib_uobject	*uobject;
865 };
866 
867 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
868 
869 struct ib_cq {
870 	struct ib_device       *device;
871 	struct ib_uobject      *uobject;
872 	ib_comp_handler   	comp_handler;
873 	void                  (*event_handler)(struct ib_event *, void *);
874 	void                   *cq_context;
875 	int               	cqe;
876 	atomic_t          	usecnt; /* count number of work queues */
877 };
878 
879 struct ib_srq {
880 	struct ib_device       *device;
881 	struct ib_pd	       *pd;
882 	struct ib_uobject      *uobject;
883 	void		      (*event_handler)(struct ib_event *, void *);
884 	void		       *srq_context;
885 	atomic_t		usecnt;
886 };
887 
888 struct ib_qp {
889 	struct ib_device       *device;
890 	struct ib_pd	       *pd;
891 	struct ib_cq	       *send_cq;
892 	struct ib_cq	       *recv_cq;
893 	struct ib_srq	       *srq;
894 	struct ib_uobject      *uobject;
895 	void                  (*event_handler)(struct ib_event *, void *);
896 	void		       *qp_context;
897 	u32			qp_num;
898 	enum ib_qp_type		qp_type;
899 };
900 
901 struct ib_mr {
902 	struct ib_device  *device;
903 	struct ib_pd	  *pd;
904 	struct ib_uobject *uobject;
905 	u32		   lkey;
906 	u32		   rkey;
907 	atomic_t	   usecnt; /* count number of MWs */
908 };
909 
910 struct ib_mw {
911 	struct ib_device	*device;
912 	struct ib_pd		*pd;
913 	struct ib_uobject	*uobject;
914 	u32			rkey;
915 };
916 
917 struct ib_fmr {
918 	struct ib_device	*device;
919 	struct ib_pd		*pd;
920 	struct list_head	list;
921 	u32			lkey;
922 	u32			rkey;
923 };
924 
925 struct ib_mad;
926 struct ib_grh;
927 
928 enum ib_process_mad_flags {
929 	IB_MAD_IGNORE_MKEY	= 1,
930 	IB_MAD_IGNORE_BKEY	= 2,
931 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
932 };
933 
934 enum ib_mad_result {
935 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
936 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
937 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
938 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
939 };
940 
941 #define IB_DEVICE_NAME_MAX 64
942 
943 struct ib_cache {
944 	rwlock_t                lock;
945 	struct ib_event_handler event_handler;
946 	struct ib_pkey_cache  **pkey_cache;
947 	struct ib_gid_cache   **gid_cache;
948 	u8                     *lmc_cache;
949 };
950 
951 struct ib_dma_mapping_ops {
952 	int		(*mapping_error)(struct ib_device *dev,
953 					 u64 dma_addr);
954 	u64		(*map_single)(struct ib_device *dev,
955 				      void *ptr, size_t size,
956 				      enum dma_data_direction direction);
957 	void		(*unmap_single)(struct ib_device *dev,
958 					u64 addr, size_t size,
959 					enum dma_data_direction direction);
960 	u64		(*map_page)(struct ib_device *dev,
961 				    struct page *page, unsigned long offset,
962 				    size_t size,
963 				    enum dma_data_direction direction);
964 	void		(*unmap_page)(struct ib_device *dev,
965 				      u64 addr, size_t size,
966 				      enum dma_data_direction direction);
967 	int		(*map_sg)(struct ib_device *dev,
968 				  struct scatterlist *sg, int nents,
969 				  enum dma_data_direction direction);
970 	void		(*unmap_sg)(struct ib_device *dev,
971 				    struct scatterlist *sg, int nents,
972 				    enum dma_data_direction direction);
973 	u64		(*dma_address)(struct ib_device *dev,
974 				       struct scatterlist *sg);
975 	unsigned int	(*dma_len)(struct ib_device *dev,
976 				   struct scatterlist *sg);
977 	void		(*sync_single_for_cpu)(struct ib_device *dev,
978 					       u64 dma_handle,
979 					       size_t size,
980 					       enum dma_data_direction dir);
981 	void		(*sync_single_for_device)(struct ib_device *dev,
982 						  u64 dma_handle,
983 						  size_t size,
984 						  enum dma_data_direction dir);
985 	void		*(*alloc_coherent)(struct ib_device *dev,
986 					   size_t size,
987 					   u64 *dma_handle,
988 					   gfp_t flag);
989 	void		(*free_coherent)(struct ib_device *dev,
990 					 size_t size, void *cpu_addr,
991 					 u64 dma_handle);
992 };
993 
994 struct iw_cm_verbs;
995 
996 struct ib_device {
997 	struct device                *dma_device;
998 
999 	char                          name[IB_DEVICE_NAME_MAX];
1000 
1001 	struct list_head              event_handler_list;
1002 	spinlock_t                    event_handler_lock;
1003 
1004 	spinlock_t                    client_data_lock;
1005 	struct list_head              core_list;
1006 	struct list_head              client_data_list;
1007 
1008 	struct ib_cache               cache;
1009 	int                          *pkey_tbl_len;
1010 	int                          *gid_tbl_len;
1011 
1012 	int			      num_comp_vectors;
1013 
1014 	struct iw_cm_verbs	     *iwcm;
1015 
1016 	int		           (*get_protocol_stats)(struct ib_device *device,
1017 							 union rdma_protocol_stats *stats);
1018 	int		           (*query_device)(struct ib_device *device,
1019 						   struct ib_device_attr *device_attr);
1020 	int		           (*query_port)(struct ib_device *device,
1021 						 u8 port_num,
1022 						 struct ib_port_attr *port_attr);
1023 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1024 						     u8 port_num);
1025 	int		           (*query_gid)(struct ib_device *device,
1026 						u8 port_num, int index,
1027 						union ib_gid *gid);
1028 	int		           (*query_pkey)(struct ib_device *device,
1029 						 u8 port_num, u16 index, u16 *pkey);
1030 	int		           (*modify_device)(struct ib_device *device,
1031 						    int device_modify_mask,
1032 						    struct ib_device_modify *device_modify);
1033 	int		           (*modify_port)(struct ib_device *device,
1034 						  u8 port_num, int port_modify_mask,
1035 						  struct ib_port_modify *port_modify);
1036 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1037 						     struct ib_udata *udata);
1038 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1039 	int                        (*mmap)(struct ib_ucontext *context,
1040 					   struct vm_area_struct *vma);
1041 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1042 					       struct ib_ucontext *context,
1043 					       struct ib_udata *udata);
1044 	int                        (*dealloc_pd)(struct ib_pd *pd);
1045 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1046 						struct ib_ah_attr *ah_attr);
1047 	int                        (*modify_ah)(struct ib_ah *ah,
1048 						struct ib_ah_attr *ah_attr);
1049 	int                        (*query_ah)(struct ib_ah *ah,
1050 					       struct ib_ah_attr *ah_attr);
1051 	int                        (*destroy_ah)(struct ib_ah *ah);
1052 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1053 						 struct ib_srq_init_attr *srq_init_attr,
1054 						 struct ib_udata *udata);
1055 	int                        (*modify_srq)(struct ib_srq *srq,
1056 						 struct ib_srq_attr *srq_attr,
1057 						 enum ib_srq_attr_mask srq_attr_mask,
1058 						 struct ib_udata *udata);
1059 	int                        (*query_srq)(struct ib_srq *srq,
1060 						struct ib_srq_attr *srq_attr);
1061 	int                        (*destroy_srq)(struct ib_srq *srq);
1062 	int                        (*post_srq_recv)(struct ib_srq *srq,
1063 						    struct ib_recv_wr *recv_wr,
1064 						    struct ib_recv_wr **bad_recv_wr);
1065 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1066 						struct ib_qp_init_attr *qp_init_attr,
1067 						struct ib_udata *udata);
1068 	int                        (*modify_qp)(struct ib_qp *qp,
1069 						struct ib_qp_attr *qp_attr,
1070 						int qp_attr_mask,
1071 						struct ib_udata *udata);
1072 	int                        (*query_qp)(struct ib_qp *qp,
1073 					       struct ib_qp_attr *qp_attr,
1074 					       int qp_attr_mask,
1075 					       struct ib_qp_init_attr *qp_init_attr);
1076 	int                        (*destroy_qp)(struct ib_qp *qp);
1077 	int                        (*post_send)(struct ib_qp *qp,
1078 						struct ib_send_wr *send_wr,
1079 						struct ib_send_wr **bad_send_wr);
1080 	int                        (*post_recv)(struct ib_qp *qp,
1081 						struct ib_recv_wr *recv_wr,
1082 						struct ib_recv_wr **bad_recv_wr);
1083 	struct ib_cq *             (*create_cq)(struct ib_device *device, int cqe,
1084 						int comp_vector,
1085 						struct ib_ucontext *context,
1086 						struct ib_udata *udata);
1087 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1088 						u16 cq_period);
1089 	int                        (*destroy_cq)(struct ib_cq *cq);
1090 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1091 						struct ib_udata *udata);
1092 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1093 					      struct ib_wc *wc);
1094 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1095 	int                        (*req_notify_cq)(struct ib_cq *cq,
1096 						    enum ib_cq_notify_flags flags);
1097 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1098 						      int wc_cnt);
1099 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1100 						 int mr_access_flags);
1101 	struct ib_mr *             (*reg_phys_mr)(struct ib_pd *pd,
1102 						  struct ib_phys_buf *phys_buf_array,
1103 						  int num_phys_buf,
1104 						  int mr_access_flags,
1105 						  u64 *iova_start);
1106 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1107 						  u64 start, u64 length,
1108 						  u64 virt_addr,
1109 						  int mr_access_flags,
1110 						  struct ib_udata *udata);
1111 	int                        (*query_mr)(struct ib_mr *mr,
1112 					       struct ib_mr_attr *mr_attr);
1113 	int                        (*dereg_mr)(struct ib_mr *mr);
1114 	struct ib_mr *		   (*alloc_fast_reg_mr)(struct ib_pd *pd,
1115 					       int max_page_list_len);
1116 	struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1117 								   int page_list_len);
1118 	void			   (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1119 	int                        (*rereg_phys_mr)(struct ib_mr *mr,
1120 						    int mr_rereg_mask,
1121 						    struct ib_pd *pd,
1122 						    struct ib_phys_buf *phys_buf_array,
1123 						    int num_phys_buf,
1124 						    int mr_access_flags,
1125 						    u64 *iova_start);
1126 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd);
1127 	int                        (*bind_mw)(struct ib_qp *qp,
1128 					      struct ib_mw *mw,
1129 					      struct ib_mw_bind *mw_bind);
1130 	int                        (*dealloc_mw)(struct ib_mw *mw);
1131 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1132 						int mr_access_flags,
1133 						struct ib_fmr_attr *fmr_attr);
1134 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1135 						   u64 *page_list, int list_len,
1136 						   u64 iova);
1137 	int		           (*unmap_fmr)(struct list_head *fmr_list);
1138 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1139 	int                        (*attach_mcast)(struct ib_qp *qp,
1140 						   union ib_gid *gid,
1141 						   u16 lid);
1142 	int                        (*detach_mcast)(struct ib_qp *qp,
1143 						   union ib_gid *gid,
1144 						   u16 lid);
1145 	int                        (*process_mad)(struct ib_device *device,
1146 						  int process_mad_flags,
1147 						  u8 port_num,
1148 						  struct ib_wc *in_wc,
1149 						  struct ib_grh *in_grh,
1150 						  struct ib_mad *in_mad,
1151 						  struct ib_mad *out_mad);
1152 
1153 	struct ib_dma_mapping_ops   *dma_ops;
1154 
1155 	struct module               *owner;
1156 	struct device                dev;
1157 	struct kobject               *ports_parent;
1158 	struct list_head             port_list;
1159 
1160 	enum {
1161 		IB_DEV_UNINITIALIZED,
1162 		IB_DEV_REGISTERED,
1163 		IB_DEV_UNREGISTERED
1164 	}                            reg_state;
1165 
1166 	int			     uverbs_abi_ver;
1167 	u64			     uverbs_cmd_mask;
1168 
1169 	char			     node_desc[64];
1170 	__be64			     node_guid;
1171 	u32			     local_dma_lkey;
1172 	u8                           node_type;
1173 	u8                           phys_port_cnt;
1174 };
1175 
1176 struct ib_client {
1177 	char  *name;
1178 	void (*add)   (struct ib_device *);
1179 	void (*remove)(struct ib_device *);
1180 
1181 	struct list_head list;
1182 };
1183 
1184 struct ib_device *ib_alloc_device(size_t size);
1185 void ib_dealloc_device(struct ib_device *device);
1186 
1187 int ib_register_device(struct ib_device *device,
1188 		       int (*port_callback)(struct ib_device *,
1189 					    u8, struct kobject *));
1190 void ib_unregister_device(struct ib_device *device);
1191 
1192 int ib_register_client   (struct ib_client *client);
1193 void ib_unregister_client(struct ib_client *client);
1194 
1195 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1196 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
1197 			 void *data);
1198 
1199 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1200 {
1201 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1202 }
1203 
1204 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1205 {
1206 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1207 }
1208 
1209 /**
1210  * ib_modify_qp_is_ok - Check that the supplied attribute mask
1211  * contains all required attributes and no attributes not allowed for
1212  * the given QP state transition.
1213  * @cur_state: Current QP state
1214  * @next_state: Next QP state
1215  * @type: QP type
1216  * @mask: Mask of supplied QP attributes
1217  *
1218  * This function is a helper function that a low-level driver's
1219  * modify_qp method can use to validate the consumer's input.  It
1220  * checks that cur_state and next_state are valid QP states, that a
1221  * transition from cur_state to next_state is allowed by the IB spec,
1222  * and that the attribute mask supplied is allowed for the transition.
1223  */
1224 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1225 		       enum ib_qp_type type, enum ib_qp_attr_mask mask);
1226 
1227 int ib_register_event_handler  (struct ib_event_handler *event_handler);
1228 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1229 void ib_dispatch_event(struct ib_event *event);
1230 
1231 int ib_query_device(struct ib_device *device,
1232 		    struct ib_device_attr *device_attr);
1233 
1234 int ib_query_port(struct ib_device *device,
1235 		  u8 port_num, struct ib_port_attr *port_attr);
1236 
1237 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1238 					       u8 port_num);
1239 
1240 int ib_query_gid(struct ib_device *device,
1241 		 u8 port_num, int index, union ib_gid *gid);
1242 
1243 int ib_query_pkey(struct ib_device *device,
1244 		  u8 port_num, u16 index, u16 *pkey);
1245 
1246 int ib_modify_device(struct ib_device *device,
1247 		     int device_modify_mask,
1248 		     struct ib_device_modify *device_modify);
1249 
1250 int ib_modify_port(struct ib_device *device,
1251 		   u8 port_num, int port_modify_mask,
1252 		   struct ib_port_modify *port_modify);
1253 
1254 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1255 		u8 *port_num, u16 *index);
1256 
1257 int ib_find_pkey(struct ib_device *device,
1258 		 u8 port_num, u16 pkey, u16 *index);
1259 
1260 /**
1261  * ib_alloc_pd - Allocates an unused protection domain.
1262  * @device: The device on which to allocate the protection domain.
1263  *
1264  * A protection domain object provides an association between QPs, shared
1265  * receive queues, address handles, memory regions, and memory windows.
1266  */
1267 struct ib_pd *ib_alloc_pd(struct ib_device *device);
1268 
1269 /**
1270  * ib_dealloc_pd - Deallocates a protection domain.
1271  * @pd: The protection domain to deallocate.
1272  */
1273 int ib_dealloc_pd(struct ib_pd *pd);
1274 
1275 /**
1276  * ib_create_ah - Creates an address handle for the given address vector.
1277  * @pd: The protection domain associated with the address handle.
1278  * @ah_attr: The attributes of the address vector.
1279  *
1280  * The address handle is used to reference a local or global destination
1281  * in all UD QP post sends.
1282  */
1283 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
1284 
1285 /**
1286  * ib_init_ah_from_wc - Initializes address handle attributes from a
1287  *   work completion.
1288  * @device: Device on which the received message arrived.
1289  * @port_num: Port on which the received message arrived.
1290  * @wc: Work completion associated with the received message.
1291  * @grh: References the received global route header.  This parameter is
1292  *   ignored unless the work completion indicates that the GRH is valid.
1293  * @ah_attr: Returned attributes that can be used when creating an address
1294  *   handle for replying to the message.
1295  */
1296 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
1297 		       struct ib_grh *grh, struct ib_ah_attr *ah_attr);
1298 
1299 /**
1300  * ib_create_ah_from_wc - Creates an address handle associated with the
1301  *   sender of the specified work completion.
1302  * @pd: The protection domain associated with the address handle.
1303  * @wc: Work completion information associated with a received message.
1304  * @grh: References the received global route header.  This parameter is
1305  *   ignored unless the work completion indicates that the GRH is valid.
1306  * @port_num: The outbound port number to associate with the address.
1307  *
1308  * The address handle is used to reference a local or global destination
1309  * in all UD QP post sends.
1310  */
1311 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
1312 				   struct ib_grh *grh, u8 port_num);
1313 
1314 /**
1315  * ib_modify_ah - Modifies the address vector associated with an address
1316  *   handle.
1317  * @ah: The address handle to modify.
1318  * @ah_attr: The new address vector attributes to associate with the
1319  *   address handle.
1320  */
1321 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1322 
1323 /**
1324  * ib_query_ah - Queries the address vector associated with an address
1325  *   handle.
1326  * @ah: The address handle to query.
1327  * @ah_attr: The address vector attributes associated with the address
1328  *   handle.
1329  */
1330 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1331 
1332 /**
1333  * ib_destroy_ah - Destroys an address handle.
1334  * @ah: The address handle to destroy.
1335  */
1336 int ib_destroy_ah(struct ib_ah *ah);
1337 
1338 /**
1339  * ib_create_srq - Creates a SRQ associated with the specified protection
1340  *   domain.
1341  * @pd: The protection domain associated with the SRQ.
1342  * @srq_init_attr: A list of initial attributes required to create the
1343  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
1344  *   the actual capabilities of the created SRQ.
1345  *
1346  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1347  * requested size of the SRQ, and set to the actual values allocated
1348  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
1349  * will always be at least as large as the requested values.
1350  */
1351 struct ib_srq *ib_create_srq(struct ib_pd *pd,
1352 			     struct ib_srq_init_attr *srq_init_attr);
1353 
1354 /**
1355  * ib_modify_srq - Modifies the attributes for the specified SRQ.
1356  * @srq: The SRQ to modify.
1357  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
1358  *   the current values of selected SRQ attributes are returned.
1359  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1360  *   are being modified.
1361  *
1362  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1363  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1364  * the number of receives queued drops below the limit.
1365  */
1366 int ib_modify_srq(struct ib_srq *srq,
1367 		  struct ib_srq_attr *srq_attr,
1368 		  enum ib_srq_attr_mask srq_attr_mask);
1369 
1370 /**
1371  * ib_query_srq - Returns the attribute list and current values for the
1372  *   specified SRQ.
1373  * @srq: The SRQ to query.
1374  * @srq_attr: The attributes of the specified SRQ.
1375  */
1376 int ib_query_srq(struct ib_srq *srq,
1377 		 struct ib_srq_attr *srq_attr);
1378 
1379 /**
1380  * ib_destroy_srq - Destroys the specified SRQ.
1381  * @srq: The SRQ to destroy.
1382  */
1383 int ib_destroy_srq(struct ib_srq *srq);
1384 
1385 /**
1386  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1387  * @srq: The SRQ to post the work request on.
1388  * @recv_wr: A list of work requests to post on the receive queue.
1389  * @bad_recv_wr: On an immediate failure, this parameter will reference
1390  *   the work request that failed to be posted on the QP.
1391  */
1392 static inline int ib_post_srq_recv(struct ib_srq *srq,
1393 				   struct ib_recv_wr *recv_wr,
1394 				   struct ib_recv_wr **bad_recv_wr)
1395 {
1396 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1397 }
1398 
1399 /**
1400  * ib_create_qp - Creates a QP associated with the specified protection
1401  *   domain.
1402  * @pd: The protection domain associated with the QP.
1403  * @qp_init_attr: A list of initial attributes required to create the
1404  *   QP.  If QP creation succeeds, then the attributes are updated to
1405  *   the actual capabilities of the created QP.
1406  */
1407 struct ib_qp *ib_create_qp(struct ib_pd *pd,
1408 			   struct ib_qp_init_attr *qp_init_attr);
1409 
1410 /**
1411  * ib_modify_qp - Modifies the attributes for the specified QP and then
1412  *   transitions the QP to the given state.
1413  * @qp: The QP to modify.
1414  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
1415  *   the current values of selected QP attributes are returned.
1416  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1417  *   are being modified.
1418  */
1419 int ib_modify_qp(struct ib_qp *qp,
1420 		 struct ib_qp_attr *qp_attr,
1421 		 int qp_attr_mask);
1422 
1423 /**
1424  * ib_query_qp - Returns the attribute list and current values for the
1425  *   specified QP.
1426  * @qp: The QP to query.
1427  * @qp_attr: The attributes of the specified QP.
1428  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1429  * @qp_init_attr: Additional attributes of the selected QP.
1430  *
1431  * The qp_attr_mask may be used to limit the query to gathering only the
1432  * selected attributes.
1433  */
1434 int ib_query_qp(struct ib_qp *qp,
1435 		struct ib_qp_attr *qp_attr,
1436 		int qp_attr_mask,
1437 		struct ib_qp_init_attr *qp_init_attr);
1438 
1439 /**
1440  * ib_destroy_qp - Destroys the specified QP.
1441  * @qp: The QP to destroy.
1442  */
1443 int ib_destroy_qp(struct ib_qp *qp);
1444 
1445 /**
1446  * ib_post_send - Posts a list of work requests to the send queue of
1447  *   the specified QP.
1448  * @qp: The QP to post the work request on.
1449  * @send_wr: A list of work requests to post on the send queue.
1450  * @bad_send_wr: On an immediate failure, this parameter will reference
1451  *   the work request that failed to be posted on the QP.
1452  *
1453  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
1454  * error is returned, the QP state shall not be affected,
1455  * ib_post_send() will return an immediate error after queueing any
1456  * earlier work requests in the list.
1457  */
1458 static inline int ib_post_send(struct ib_qp *qp,
1459 			       struct ib_send_wr *send_wr,
1460 			       struct ib_send_wr **bad_send_wr)
1461 {
1462 	return qp->device->post_send(qp, send_wr, bad_send_wr);
1463 }
1464 
1465 /**
1466  * ib_post_recv - Posts a list of work requests to the receive queue of
1467  *   the specified QP.
1468  * @qp: The QP to post the work request on.
1469  * @recv_wr: A list of work requests to post on the receive queue.
1470  * @bad_recv_wr: On an immediate failure, this parameter will reference
1471  *   the work request that failed to be posted on the QP.
1472  */
1473 static inline int ib_post_recv(struct ib_qp *qp,
1474 			       struct ib_recv_wr *recv_wr,
1475 			       struct ib_recv_wr **bad_recv_wr)
1476 {
1477 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1478 }
1479 
1480 /**
1481  * ib_create_cq - Creates a CQ on the specified device.
1482  * @device: The device on which to create the CQ.
1483  * @comp_handler: A user-specified callback that is invoked when a
1484  *   completion event occurs on the CQ.
1485  * @event_handler: A user-specified callback that is invoked when an
1486  *   asynchronous event not associated with a completion occurs on the CQ.
1487  * @cq_context: Context associated with the CQ returned to the user via
1488  *   the associated completion and event handlers.
1489  * @cqe: The minimum size of the CQ.
1490  * @comp_vector - Completion vector used to signal completion events.
1491  *     Must be >= 0 and < context->num_comp_vectors.
1492  *
1493  * Users can examine the cq structure to determine the actual CQ size.
1494  */
1495 struct ib_cq *ib_create_cq(struct ib_device *device,
1496 			   ib_comp_handler comp_handler,
1497 			   void (*event_handler)(struct ib_event *, void *),
1498 			   void *cq_context, int cqe, int comp_vector);
1499 
1500 /**
1501  * ib_resize_cq - Modifies the capacity of the CQ.
1502  * @cq: The CQ to resize.
1503  * @cqe: The minimum size of the CQ.
1504  *
1505  * Users can examine the cq structure to determine the actual CQ size.
1506  */
1507 int ib_resize_cq(struct ib_cq *cq, int cqe);
1508 
1509 /**
1510  * ib_modify_cq - Modifies moderation params of the CQ
1511  * @cq: The CQ to modify.
1512  * @cq_count: number of CQEs that will trigger an event
1513  * @cq_period: max period of time in usec before triggering an event
1514  *
1515  */
1516 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1517 
1518 /**
1519  * ib_destroy_cq - Destroys the specified CQ.
1520  * @cq: The CQ to destroy.
1521  */
1522 int ib_destroy_cq(struct ib_cq *cq);
1523 
1524 /**
1525  * ib_poll_cq - poll a CQ for completion(s)
1526  * @cq:the CQ being polled
1527  * @num_entries:maximum number of completions to return
1528  * @wc:array of at least @num_entries &struct ib_wc where completions
1529  *   will be returned
1530  *
1531  * Poll a CQ for (possibly multiple) completions.  If the return value
1532  * is < 0, an error occurred.  If the return value is >= 0, it is the
1533  * number of completions returned.  If the return value is
1534  * non-negative and < num_entries, then the CQ was emptied.
1535  */
1536 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1537 			     struct ib_wc *wc)
1538 {
1539 	return cq->device->poll_cq(cq, num_entries, wc);
1540 }
1541 
1542 /**
1543  * ib_peek_cq - Returns the number of unreaped completions currently
1544  *   on the specified CQ.
1545  * @cq: The CQ to peek.
1546  * @wc_cnt: A minimum number of unreaped completions to check for.
1547  *
1548  * If the number of unreaped completions is greater than or equal to wc_cnt,
1549  * this function returns wc_cnt, otherwise, it returns the actual number of
1550  * unreaped completions.
1551  */
1552 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1553 
1554 /**
1555  * ib_req_notify_cq - Request completion notification on a CQ.
1556  * @cq: The CQ to generate an event for.
1557  * @flags:
1558  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1559  *   to request an event on the next solicited event or next work
1560  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1561  *   may also be |ed in to request a hint about missed events, as
1562  *   described below.
1563  *
1564  * Return Value:
1565  *    < 0 means an error occurred while requesting notification
1566  *   == 0 means notification was requested successfully, and if
1567  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1568  *        were missed and it is safe to wait for another event.  In
1569  *        this case is it guaranteed that any work completions added
1570  *        to the CQ since the last CQ poll will trigger a completion
1571  *        notification event.
1572  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1573  *        in.  It means that the consumer must poll the CQ again to
1574  *        make sure it is empty to avoid missing an event because of a
1575  *        race between requesting notification and an entry being
1576  *        added to the CQ.  This return value means it is possible
1577  *        (but not guaranteed) that a work completion has been added
1578  *        to the CQ since the last poll without triggering a
1579  *        completion notification event.
1580  */
1581 static inline int ib_req_notify_cq(struct ib_cq *cq,
1582 				   enum ib_cq_notify_flags flags)
1583 {
1584 	return cq->device->req_notify_cq(cq, flags);
1585 }
1586 
1587 /**
1588  * ib_req_ncomp_notif - Request completion notification when there are
1589  *   at least the specified number of unreaped completions on the CQ.
1590  * @cq: The CQ to generate an event for.
1591  * @wc_cnt: The number of unreaped completions that should be on the
1592  *   CQ before an event is generated.
1593  */
1594 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1595 {
1596 	return cq->device->req_ncomp_notif ?
1597 		cq->device->req_ncomp_notif(cq, wc_cnt) :
1598 		-ENOSYS;
1599 }
1600 
1601 /**
1602  * ib_get_dma_mr - Returns a memory region for system memory that is
1603  *   usable for DMA.
1604  * @pd: The protection domain associated with the memory region.
1605  * @mr_access_flags: Specifies the memory access rights.
1606  *
1607  * Note that the ib_dma_*() functions defined below must be used
1608  * to create/destroy addresses used with the Lkey or Rkey returned
1609  * by ib_get_dma_mr().
1610  */
1611 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1612 
1613 /**
1614  * ib_dma_mapping_error - check a DMA addr for error
1615  * @dev: The device for which the dma_addr was created
1616  * @dma_addr: The DMA address to check
1617  */
1618 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1619 {
1620 	if (dev->dma_ops)
1621 		return dev->dma_ops->mapping_error(dev, dma_addr);
1622 	return dma_mapping_error(dev->dma_device, dma_addr);
1623 }
1624 
1625 /**
1626  * ib_dma_map_single - Map a kernel virtual address to DMA address
1627  * @dev: The device for which the dma_addr is to be created
1628  * @cpu_addr: The kernel virtual address
1629  * @size: The size of the region in bytes
1630  * @direction: The direction of the DMA
1631  */
1632 static inline u64 ib_dma_map_single(struct ib_device *dev,
1633 				    void *cpu_addr, size_t size,
1634 				    enum dma_data_direction direction)
1635 {
1636 	if (dev->dma_ops)
1637 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1638 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1639 }
1640 
1641 /**
1642  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1643  * @dev: The device for which the DMA address was created
1644  * @addr: The DMA address
1645  * @size: The size of the region in bytes
1646  * @direction: The direction of the DMA
1647  */
1648 static inline void ib_dma_unmap_single(struct ib_device *dev,
1649 				       u64 addr, size_t size,
1650 				       enum dma_data_direction direction)
1651 {
1652 	if (dev->dma_ops)
1653 		dev->dma_ops->unmap_single(dev, addr, size, direction);
1654 	else
1655 		dma_unmap_single(dev->dma_device, addr, size, direction);
1656 }
1657 
1658 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
1659 					  void *cpu_addr, size_t size,
1660 					  enum dma_data_direction direction,
1661 					  struct dma_attrs *attrs)
1662 {
1663 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
1664 				    direction, attrs);
1665 }
1666 
1667 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
1668 					     u64 addr, size_t size,
1669 					     enum dma_data_direction direction,
1670 					     struct dma_attrs *attrs)
1671 {
1672 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
1673 				      direction, attrs);
1674 }
1675 
1676 /**
1677  * ib_dma_map_page - Map a physical page to DMA address
1678  * @dev: The device for which the dma_addr is to be created
1679  * @page: The page to be mapped
1680  * @offset: The offset within the page
1681  * @size: The size of the region in bytes
1682  * @direction: The direction of the DMA
1683  */
1684 static inline u64 ib_dma_map_page(struct ib_device *dev,
1685 				  struct page *page,
1686 				  unsigned long offset,
1687 				  size_t size,
1688 					 enum dma_data_direction direction)
1689 {
1690 	if (dev->dma_ops)
1691 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
1692 	return dma_map_page(dev->dma_device, page, offset, size, direction);
1693 }
1694 
1695 /**
1696  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1697  * @dev: The device for which the DMA address was created
1698  * @addr: The DMA address
1699  * @size: The size of the region in bytes
1700  * @direction: The direction of the DMA
1701  */
1702 static inline void ib_dma_unmap_page(struct ib_device *dev,
1703 				     u64 addr, size_t size,
1704 				     enum dma_data_direction direction)
1705 {
1706 	if (dev->dma_ops)
1707 		dev->dma_ops->unmap_page(dev, addr, size, direction);
1708 	else
1709 		dma_unmap_page(dev->dma_device, addr, size, direction);
1710 }
1711 
1712 /**
1713  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
1714  * @dev: The device for which the DMA addresses are to be created
1715  * @sg: The array of scatter/gather entries
1716  * @nents: The number of scatter/gather entries
1717  * @direction: The direction of the DMA
1718  */
1719 static inline int ib_dma_map_sg(struct ib_device *dev,
1720 				struct scatterlist *sg, int nents,
1721 				enum dma_data_direction direction)
1722 {
1723 	if (dev->dma_ops)
1724 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
1725 	return dma_map_sg(dev->dma_device, sg, nents, direction);
1726 }
1727 
1728 /**
1729  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
1730  * @dev: The device for which the DMA addresses were created
1731  * @sg: The array of scatter/gather entries
1732  * @nents: The number of scatter/gather entries
1733  * @direction: The direction of the DMA
1734  */
1735 static inline void ib_dma_unmap_sg(struct ib_device *dev,
1736 				   struct scatterlist *sg, int nents,
1737 				   enum dma_data_direction direction)
1738 {
1739 	if (dev->dma_ops)
1740 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
1741 	else
1742 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
1743 }
1744 
1745 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
1746 				      struct scatterlist *sg, int nents,
1747 				      enum dma_data_direction direction,
1748 				      struct dma_attrs *attrs)
1749 {
1750 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1751 }
1752 
1753 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
1754 					 struct scatterlist *sg, int nents,
1755 					 enum dma_data_direction direction,
1756 					 struct dma_attrs *attrs)
1757 {
1758 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1759 }
1760 /**
1761  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
1762  * @dev: The device for which the DMA addresses were created
1763  * @sg: The scatter/gather entry
1764  */
1765 static inline u64 ib_sg_dma_address(struct ib_device *dev,
1766 				    struct scatterlist *sg)
1767 {
1768 	if (dev->dma_ops)
1769 		return dev->dma_ops->dma_address(dev, sg);
1770 	return sg_dma_address(sg);
1771 }
1772 
1773 /**
1774  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
1775  * @dev: The device for which the DMA addresses were created
1776  * @sg: The scatter/gather entry
1777  */
1778 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
1779 					 struct scatterlist *sg)
1780 {
1781 	if (dev->dma_ops)
1782 		return dev->dma_ops->dma_len(dev, sg);
1783 	return sg_dma_len(sg);
1784 }
1785 
1786 /**
1787  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
1788  * @dev: The device for which the DMA address was created
1789  * @addr: The DMA address
1790  * @size: The size of the region in bytes
1791  * @dir: The direction of the DMA
1792  */
1793 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
1794 					      u64 addr,
1795 					      size_t size,
1796 					      enum dma_data_direction dir)
1797 {
1798 	if (dev->dma_ops)
1799 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
1800 	else
1801 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
1802 }
1803 
1804 /**
1805  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
1806  * @dev: The device for which the DMA address was created
1807  * @addr: The DMA address
1808  * @size: The size of the region in bytes
1809  * @dir: The direction of the DMA
1810  */
1811 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
1812 						 u64 addr,
1813 						 size_t size,
1814 						 enum dma_data_direction dir)
1815 {
1816 	if (dev->dma_ops)
1817 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
1818 	else
1819 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
1820 }
1821 
1822 /**
1823  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
1824  * @dev: The device for which the DMA address is requested
1825  * @size: The size of the region to allocate in bytes
1826  * @dma_handle: A pointer for returning the DMA address of the region
1827  * @flag: memory allocator flags
1828  */
1829 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
1830 					   size_t size,
1831 					   u64 *dma_handle,
1832 					   gfp_t flag)
1833 {
1834 	if (dev->dma_ops)
1835 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
1836 	else {
1837 		dma_addr_t handle;
1838 		void *ret;
1839 
1840 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
1841 		*dma_handle = handle;
1842 		return ret;
1843 	}
1844 }
1845 
1846 /**
1847  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
1848  * @dev: The device for which the DMA addresses were allocated
1849  * @size: The size of the region
1850  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
1851  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
1852  */
1853 static inline void ib_dma_free_coherent(struct ib_device *dev,
1854 					size_t size, void *cpu_addr,
1855 					u64 dma_handle)
1856 {
1857 	if (dev->dma_ops)
1858 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
1859 	else
1860 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
1861 }
1862 
1863 /**
1864  * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
1865  *   by an HCA.
1866  * @pd: The protection domain associated assigned to the registered region.
1867  * @phys_buf_array: Specifies a list of physical buffers to use in the
1868  *   memory region.
1869  * @num_phys_buf: Specifies the size of the phys_buf_array.
1870  * @mr_access_flags: Specifies the memory access rights.
1871  * @iova_start: The offset of the region's starting I/O virtual address.
1872  */
1873 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
1874 			     struct ib_phys_buf *phys_buf_array,
1875 			     int num_phys_buf,
1876 			     int mr_access_flags,
1877 			     u64 *iova_start);
1878 
1879 /**
1880  * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
1881  *   Conceptually, this call performs the functions deregister memory region
1882  *   followed by register physical memory region.  Where possible,
1883  *   resources are reused instead of deallocated and reallocated.
1884  * @mr: The memory region to modify.
1885  * @mr_rereg_mask: A bit-mask used to indicate which of the following
1886  *   properties of the memory region are being modified.
1887  * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
1888  *   the new protection domain to associated with the memory region,
1889  *   otherwise, this parameter is ignored.
1890  * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1891  *   field specifies a list of physical buffers to use in the new
1892  *   translation, otherwise, this parameter is ignored.
1893  * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1894  *   field specifies the size of the phys_buf_array, otherwise, this
1895  *   parameter is ignored.
1896  * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
1897  *   field specifies the new memory access rights, otherwise, this
1898  *   parameter is ignored.
1899  * @iova_start: The offset of the region's starting I/O virtual address.
1900  */
1901 int ib_rereg_phys_mr(struct ib_mr *mr,
1902 		     int mr_rereg_mask,
1903 		     struct ib_pd *pd,
1904 		     struct ib_phys_buf *phys_buf_array,
1905 		     int num_phys_buf,
1906 		     int mr_access_flags,
1907 		     u64 *iova_start);
1908 
1909 /**
1910  * ib_query_mr - Retrieves information about a specific memory region.
1911  * @mr: The memory region to retrieve information about.
1912  * @mr_attr: The attributes of the specified memory region.
1913  */
1914 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
1915 
1916 /**
1917  * ib_dereg_mr - Deregisters a memory region and removes it from the
1918  *   HCA translation table.
1919  * @mr: The memory region to deregister.
1920  */
1921 int ib_dereg_mr(struct ib_mr *mr);
1922 
1923 /**
1924  * ib_alloc_fast_reg_mr - Allocates memory region usable with the
1925  *   IB_WR_FAST_REG_MR send work request.
1926  * @pd: The protection domain associated with the region.
1927  * @max_page_list_len: requested max physical buffer list length to be
1928  *   used with fast register work requests for this MR.
1929  */
1930 struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
1931 
1932 /**
1933  * ib_alloc_fast_reg_page_list - Allocates a page list array
1934  * @device - ib device pointer.
1935  * @page_list_len - size of the page list array to be allocated.
1936  *
1937  * This allocates and returns a struct ib_fast_reg_page_list * and a
1938  * page_list array that is at least page_list_len in size.  The actual
1939  * size is returned in max_page_list_len.  The caller is responsible
1940  * for initializing the contents of the page_list array before posting
1941  * a send work request with the IB_WC_FAST_REG_MR opcode.
1942  *
1943  * The page_list array entries must be translated using one of the
1944  * ib_dma_*() functions just like the addresses passed to
1945  * ib_map_phys_fmr().  Once the ib_post_send() is issued, the struct
1946  * ib_fast_reg_page_list must not be modified by the caller until the
1947  * IB_WC_FAST_REG_MR work request completes.
1948  */
1949 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
1950 				struct ib_device *device, int page_list_len);
1951 
1952 /**
1953  * ib_free_fast_reg_page_list - Deallocates a previously allocated
1954  *   page list array.
1955  * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
1956  */
1957 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
1958 
1959 /**
1960  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
1961  *   R_Key and L_Key.
1962  * @mr - struct ib_mr pointer to be updated.
1963  * @newkey - new key to be used.
1964  */
1965 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
1966 {
1967 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
1968 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
1969 }
1970 
1971 /**
1972  * ib_alloc_mw - Allocates a memory window.
1973  * @pd: The protection domain associated with the memory window.
1974  */
1975 struct ib_mw *ib_alloc_mw(struct ib_pd *pd);
1976 
1977 /**
1978  * ib_bind_mw - Posts a work request to the send queue of the specified
1979  *   QP, which binds the memory window to the given address range and
1980  *   remote access attributes.
1981  * @qp: QP to post the bind work request on.
1982  * @mw: The memory window to bind.
1983  * @mw_bind: Specifies information about the memory window, including
1984  *   its address range, remote access rights, and associated memory region.
1985  */
1986 static inline int ib_bind_mw(struct ib_qp *qp,
1987 			     struct ib_mw *mw,
1988 			     struct ib_mw_bind *mw_bind)
1989 {
1990 	/* XXX reference counting in corresponding MR? */
1991 	return mw->device->bind_mw ?
1992 		mw->device->bind_mw(qp, mw, mw_bind) :
1993 		-ENOSYS;
1994 }
1995 
1996 /**
1997  * ib_dealloc_mw - Deallocates a memory window.
1998  * @mw: The memory window to deallocate.
1999  */
2000 int ib_dealloc_mw(struct ib_mw *mw);
2001 
2002 /**
2003  * ib_alloc_fmr - Allocates a unmapped fast memory region.
2004  * @pd: The protection domain associated with the unmapped region.
2005  * @mr_access_flags: Specifies the memory access rights.
2006  * @fmr_attr: Attributes of the unmapped region.
2007  *
2008  * A fast memory region must be mapped before it can be used as part of
2009  * a work request.
2010  */
2011 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2012 			    int mr_access_flags,
2013 			    struct ib_fmr_attr *fmr_attr);
2014 
2015 /**
2016  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2017  * @fmr: The fast memory region to associate with the pages.
2018  * @page_list: An array of physical pages to map to the fast memory region.
2019  * @list_len: The number of pages in page_list.
2020  * @iova: The I/O virtual address to use with the mapped region.
2021  */
2022 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2023 				  u64 *page_list, int list_len,
2024 				  u64 iova)
2025 {
2026 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2027 }
2028 
2029 /**
2030  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2031  * @fmr_list: A linked list of fast memory regions to unmap.
2032  */
2033 int ib_unmap_fmr(struct list_head *fmr_list);
2034 
2035 /**
2036  * ib_dealloc_fmr - Deallocates a fast memory region.
2037  * @fmr: The fast memory region to deallocate.
2038  */
2039 int ib_dealloc_fmr(struct ib_fmr *fmr);
2040 
2041 /**
2042  * ib_attach_mcast - Attaches the specified QP to a multicast group.
2043  * @qp: QP to attach to the multicast group.  The QP must be type
2044  *   IB_QPT_UD.
2045  * @gid: Multicast group GID.
2046  * @lid: Multicast group LID in host byte order.
2047  *
2048  * In order to send and receive multicast packets, subnet
2049  * administration must have created the multicast group and configured
2050  * the fabric appropriately.  The port associated with the specified
2051  * QP must also be a member of the multicast group.
2052  */
2053 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2054 
2055 /**
2056  * ib_detach_mcast - Detaches the specified QP from a multicast group.
2057  * @qp: QP to detach from the multicast group.
2058  * @gid: Multicast group GID.
2059  * @lid: Multicast group LID in host byte order.
2060  */
2061 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2062 
2063 #endif /* IB_VERBS_H */
2064