xref: /openbmc/linux/include/rdma/ib_verbs.h (revision baa7eb025ab14f3cba2e35c0a8648f9c9f01d24f)
1 /*
2  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
3  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
4  * Copyright (c) 2004 Intel Corporation.  All rights reserved.
5  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
6  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
7  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
9  *
10  * This software is available to you under a choice of one of two
11  * licenses.  You may choose to be licensed under the terms of the GNU
12  * General Public License (GPL) Version 2, available from the file
13  * COPYING in the main directory of this source tree, or the
14  * OpenIB.org BSD license below:
15  *
16  *     Redistribution and use in source and binary forms, with or
17  *     without modification, are permitted provided that the following
18  *     conditions are met:
19  *
20  *      - Redistributions of source code must retain the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer.
23  *
24  *      - Redistributions in binary form must reproduce the above
25  *        copyright notice, this list of conditions and the following
26  *        disclaimer in the documentation and/or other materials
27  *        provided with the distribution.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36  * SOFTWARE.
37  */
38 
39 #if !defined(IB_VERBS_H)
40 #define IB_VERBS_H
41 
42 #include <linux/types.h>
43 #include <linux/device.h>
44 #include <linux/mm.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 
51 #include <asm/atomic.h>
52 #include <asm/uaccess.h>
53 
54 union ib_gid {
55 	u8	raw[16];
56 	struct {
57 		__be64	subnet_prefix;
58 		__be64	interface_id;
59 	} global;
60 };
61 
62 enum rdma_node_type {
63 	/* IB values map to NodeInfo:NodeType. */
64 	RDMA_NODE_IB_CA 	= 1,
65 	RDMA_NODE_IB_SWITCH,
66 	RDMA_NODE_IB_ROUTER,
67 	RDMA_NODE_RNIC
68 };
69 
70 enum rdma_transport_type {
71 	RDMA_TRANSPORT_IB,
72 	RDMA_TRANSPORT_IWARP
73 };
74 
75 enum rdma_transport_type
76 rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__;
77 
78 enum rdma_link_layer {
79 	IB_LINK_LAYER_UNSPECIFIED,
80 	IB_LINK_LAYER_INFINIBAND,
81 	IB_LINK_LAYER_ETHERNET,
82 };
83 
84 enum ib_device_cap_flags {
85 	IB_DEVICE_RESIZE_MAX_WR		= 1,
86 	IB_DEVICE_BAD_PKEY_CNTR		= (1<<1),
87 	IB_DEVICE_BAD_QKEY_CNTR		= (1<<2),
88 	IB_DEVICE_RAW_MULTI		= (1<<3),
89 	IB_DEVICE_AUTO_PATH_MIG		= (1<<4),
90 	IB_DEVICE_CHANGE_PHY_PORT	= (1<<5),
91 	IB_DEVICE_UD_AV_PORT_ENFORCE	= (1<<6),
92 	IB_DEVICE_CURR_QP_STATE_MOD	= (1<<7),
93 	IB_DEVICE_SHUTDOWN_PORT		= (1<<8),
94 	IB_DEVICE_INIT_TYPE		= (1<<9),
95 	IB_DEVICE_PORT_ACTIVE_EVENT	= (1<<10),
96 	IB_DEVICE_SYS_IMAGE_GUID	= (1<<11),
97 	IB_DEVICE_RC_RNR_NAK_GEN	= (1<<12),
98 	IB_DEVICE_SRQ_RESIZE		= (1<<13),
99 	IB_DEVICE_N_NOTIFY_CQ		= (1<<14),
100 	IB_DEVICE_LOCAL_DMA_LKEY	= (1<<15),
101 	IB_DEVICE_RESERVED		= (1<<16), /* old SEND_W_INV */
102 	IB_DEVICE_MEM_WINDOW		= (1<<17),
103 	/*
104 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
105 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
106 	 * messages and can verify the validity of checksum for
107 	 * incoming messages.  Setting this flag implies that the
108 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
109 	 */
110 	IB_DEVICE_UD_IP_CSUM		= (1<<18),
111 	IB_DEVICE_UD_TSO		= (1<<19),
112 	IB_DEVICE_MEM_MGT_EXTENSIONS	= (1<<21),
113 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
114 };
115 
116 enum ib_atomic_cap {
117 	IB_ATOMIC_NONE,
118 	IB_ATOMIC_HCA,
119 	IB_ATOMIC_GLOB
120 };
121 
122 struct ib_device_attr {
123 	u64			fw_ver;
124 	__be64			sys_image_guid;
125 	u64			max_mr_size;
126 	u64			page_size_cap;
127 	u32			vendor_id;
128 	u32			vendor_part_id;
129 	u32			hw_ver;
130 	int			max_qp;
131 	int			max_qp_wr;
132 	int			device_cap_flags;
133 	int			max_sge;
134 	int			max_sge_rd;
135 	int			max_cq;
136 	int			max_cqe;
137 	int			max_mr;
138 	int			max_pd;
139 	int			max_qp_rd_atom;
140 	int			max_ee_rd_atom;
141 	int			max_res_rd_atom;
142 	int			max_qp_init_rd_atom;
143 	int			max_ee_init_rd_atom;
144 	enum ib_atomic_cap	atomic_cap;
145 	enum ib_atomic_cap	masked_atomic_cap;
146 	int			max_ee;
147 	int			max_rdd;
148 	int			max_mw;
149 	int			max_raw_ipv6_qp;
150 	int			max_raw_ethy_qp;
151 	int			max_mcast_grp;
152 	int			max_mcast_qp_attach;
153 	int			max_total_mcast_qp_attach;
154 	int			max_ah;
155 	int			max_fmr;
156 	int			max_map_per_fmr;
157 	int			max_srq;
158 	int			max_srq_wr;
159 	int			max_srq_sge;
160 	unsigned int		max_fast_reg_page_list_len;
161 	u16			max_pkeys;
162 	u8			local_ca_ack_delay;
163 };
164 
165 enum ib_mtu {
166 	IB_MTU_256  = 1,
167 	IB_MTU_512  = 2,
168 	IB_MTU_1024 = 3,
169 	IB_MTU_2048 = 4,
170 	IB_MTU_4096 = 5
171 };
172 
173 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
174 {
175 	switch (mtu) {
176 	case IB_MTU_256:  return  256;
177 	case IB_MTU_512:  return  512;
178 	case IB_MTU_1024: return 1024;
179 	case IB_MTU_2048: return 2048;
180 	case IB_MTU_4096: return 4096;
181 	default: 	  return -1;
182 	}
183 }
184 
185 enum ib_port_state {
186 	IB_PORT_NOP		= 0,
187 	IB_PORT_DOWN		= 1,
188 	IB_PORT_INIT		= 2,
189 	IB_PORT_ARMED		= 3,
190 	IB_PORT_ACTIVE		= 4,
191 	IB_PORT_ACTIVE_DEFER	= 5
192 };
193 
194 enum ib_port_cap_flags {
195 	IB_PORT_SM				= 1 <<  1,
196 	IB_PORT_NOTICE_SUP			= 1 <<  2,
197 	IB_PORT_TRAP_SUP			= 1 <<  3,
198 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
199 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
200 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
201 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
202 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
203 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
204 	IB_PORT_SM_DISABLED			= 1 << 10,
205 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
206 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
207 	IB_PORT_CM_SUP				= 1 << 16,
208 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
209 	IB_PORT_REINIT_SUP			= 1 << 18,
210 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
211 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
212 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
213 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
214 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
215 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
216 	IB_PORT_CLIENT_REG_SUP			= 1 << 25
217 };
218 
219 enum ib_port_width {
220 	IB_WIDTH_1X	= 1,
221 	IB_WIDTH_4X	= 2,
222 	IB_WIDTH_8X	= 4,
223 	IB_WIDTH_12X	= 8
224 };
225 
226 static inline int ib_width_enum_to_int(enum ib_port_width width)
227 {
228 	switch (width) {
229 	case IB_WIDTH_1X:  return  1;
230 	case IB_WIDTH_4X:  return  4;
231 	case IB_WIDTH_8X:  return  8;
232 	case IB_WIDTH_12X: return 12;
233 	default: 	  return -1;
234 	}
235 }
236 
237 struct ib_protocol_stats {
238 	/* TBD... */
239 };
240 
241 struct iw_protocol_stats {
242 	u64	ipInReceives;
243 	u64	ipInHdrErrors;
244 	u64	ipInTooBigErrors;
245 	u64	ipInNoRoutes;
246 	u64	ipInAddrErrors;
247 	u64	ipInUnknownProtos;
248 	u64	ipInTruncatedPkts;
249 	u64	ipInDiscards;
250 	u64	ipInDelivers;
251 	u64	ipOutForwDatagrams;
252 	u64	ipOutRequests;
253 	u64	ipOutDiscards;
254 	u64	ipOutNoRoutes;
255 	u64	ipReasmTimeout;
256 	u64	ipReasmReqds;
257 	u64	ipReasmOKs;
258 	u64	ipReasmFails;
259 	u64	ipFragOKs;
260 	u64	ipFragFails;
261 	u64	ipFragCreates;
262 	u64	ipInMcastPkts;
263 	u64	ipOutMcastPkts;
264 	u64	ipInBcastPkts;
265 	u64	ipOutBcastPkts;
266 
267 	u64	tcpRtoAlgorithm;
268 	u64	tcpRtoMin;
269 	u64	tcpRtoMax;
270 	u64	tcpMaxConn;
271 	u64	tcpActiveOpens;
272 	u64	tcpPassiveOpens;
273 	u64	tcpAttemptFails;
274 	u64	tcpEstabResets;
275 	u64	tcpCurrEstab;
276 	u64	tcpInSegs;
277 	u64	tcpOutSegs;
278 	u64	tcpRetransSegs;
279 	u64	tcpInErrs;
280 	u64	tcpOutRsts;
281 };
282 
283 union rdma_protocol_stats {
284 	struct ib_protocol_stats	ib;
285 	struct iw_protocol_stats	iw;
286 };
287 
288 struct ib_port_attr {
289 	enum ib_port_state	state;
290 	enum ib_mtu		max_mtu;
291 	enum ib_mtu		active_mtu;
292 	int			gid_tbl_len;
293 	u32			port_cap_flags;
294 	u32			max_msg_sz;
295 	u32			bad_pkey_cntr;
296 	u32			qkey_viol_cntr;
297 	u16			pkey_tbl_len;
298 	u16			lid;
299 	u16			sm_lid;
300 	u8			lmc;
301 	u8			max_vl_num;
302 	u8			sm_sl;
303 	u8			subnet_timeout;
304 	u8			init_type_reply;
305 	u8			active_width;
306 	u8			active_speed;
307 	u8                      phys_state;
308 };
309 
310 enum ib_device_modify_flags {
311 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
312 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
313 };
314 
315 struct ib_device_modify {
316 	u64	sys_image_guid;
317 	char	node_desc[64];
318 };
319 
320 enum ib_port_modify_flags {
321 	IB_PORT_SHUTDOWN		= 1,
322 	IB_PORT_INIT_TYPE		= (1<<2),
323 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
324 };
325 
326 struct ib_port_modify {
327 	u32	set_port_cap_mask;
328 	u32	clr_port_cap_mask;
329 	u8	init_type;
330 };
331 
332 enum ib_event_type {
333 	IB_EVENT_CQ_ERR,
334 	IB_EVENT_QP_FATAL,
335 	IB_EVENT_QP_REQ_ERR,
336 	IB_EVENT_QP_ACCESS_ERR,
337 	IB_EVENT_COMM_EST,
338 	IB_EVENT_SQ_DRAINED,
339 	IB_EVENT_PATH_MIG,
340 	IB_EVENT_PATH_MIG_ERR,
341 	IB_EVENT_DEVICE_FATAL,
342 	IB_EVENT_PORT_ACTIVE,
343 	IB_EVENT_PORT_ERR,
344 	IB_EVENT_LID_CHANGE,
345 	IB_EVENT_PKEY_CHANGE,
346 	IB_EVENT_SM_CHANGE,
347 	IB_EVENT_SRQ_ERR,
348 	IB_EVENT_SRQ_LIMIT_REACHED,
349 	IB_EVENT_QP_LAST_WQE_REACHED,
350 	IB_EVENT_CLIENT_REREGISTER
351 };
352 
353 struct ib_event {
354 	struct ib_device	*device;
355 	union {
356 		struct ib_cq	*cq;
357 		struct ib_qp	*qp;
358 		struct ib_srq	*srq;
359 		u8		port_num;
360 	} element;
361 	enum ib_event_type	event;
362 };
363 
364 struct ib_event_handler {
365 	struct ib_device *device;
366 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
367 	struct list_head  list;
368 };
369 
370 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
371 	do {							\
372 		(_ptr)->device  = _device;			\
373 		(_ptr)->handler = _handler;			\
374 		INIT_LIST_HEAD(&(_ptr)->list);			\
375 	} while (0)
376 
377 struct ib_global_route {
378 	union ib_gid	dgid;
379 	u32		flow_label;
380 	u8		sgid_index;
381 	u8		hop_limit;
382 	u8		traffic_class;
383 };
384 
385 struct ib_grh {
386 	__be32		version_tclass_flow;
387 	__be16		paylen;
388 	u8		next_hdr;
389 	u8		hop_limit;
390 	union ib_gid	sgid;
391 	union ib_gid	dgid;
392 };
393 
394 enum {
395 	IB_MULTICAST_QPN = 0xffffff
396 };
397 
398 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
399 
400 enum ib_ah_flags {
401 	IB_AH_GRH	= 1
402 };
403 
404 enum ib_rate {
405 	IB_RATE_PORT_CURRENT = 0,
406 	IB_RATE_2_5_GBPS = 2,
407 	IB_RATE_5_GBPS   = 5,
408 	IB_RATE_10_GBPS  = 3,
409 	IB_RATE_20_GBPS  = 6,
410 	IB_RATE_30_GBPS  = 4,
411 	IB_RATE_40_GBPS  = 7,
412 	IB_RATE_60_GBPS  = 8,
413 	IB_RATE_80_GBPS  = 9,
414 	IB_RATE_120_GBPS = 10
415 };
416 
417 /**
418  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
419  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
420  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
421  * @rate: rate to convert.
422  */
423 int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
424 
425 /**
426  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
427  * enum.
428  * @mult: multiple to convert.
429  */
430 enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
431 
432 struct ib_ah_attr {
433 	struct ib_global_route	grh;
434 	u16			dlid;
435 	u8			sl;
436 	u8			src_path_bits;
437 	u8			static_rate;
438 	u8			ah_flags;
439 	u8			port_num;
440 };
441 
442 enum ib_wc_status {
443 	IB_WC_SUCCESS,
444 	IB_WC_LOC_LEN_ERR,
445 	IB_WC_LOC_QP_OP_ERR,
446 	IB_WC_LOC_EEC_OP_ERR,
447 	IB_WC_LOC_PROT_ERR,
448 	IB_WC_WR_FLUSH_ERR,
449 	IB_WC_MW_BIND_ERR,
450 	IB_WC_BAD_RESP_ERR,
451 	IB_WC_LOC_ACCESS_ERR,
452 	IB_WC_REM_INV_REQ_ERR,
453 	IB_WC_REM_ACCESS_ERR,
454 	IB_WC_REM_OP_ERR,
455 	IB_WC_RETRY_EXC_ERR,
456 	IB_WC_RNR_RETRY_EXC_ERR,
457 	IB_WC_LOC_RDD_VIOL_ERR,
458 	IB_WC_REM_INV_RD_REQ_ERR,
459 	IB_WC_REM_ABORT_ERR,
460 	IB_WC_INV_EECN_ERR,
461 	IB_WC_INV_EEC_STATE_ERR,
462 	IB_WC_FATAL_ERR,
463 	IB_WC_RESP_TIMEOUT_ERR,
464 	IB_WC_GENERAL_ERR
465 };
466 
467 enum ib_wc_opcode {
468 	IB_WC_SEND,
469 	IB_WC_RDMA_WRITE,
470 	IB_WC_RDMA_READ,
471 	IB_WC_COMP_SWAP,
472 	IB_WC_FETCH_ADD,
473 	IB_WC_BIND_MW,
474 	IB_WC_LSO,
475 	IB_WC_LOCAL_INV,
476 	IB_WC_FAST_REG_MR,
477 	IB_WC_MASKED_COMP_SWAP,
478 	IB_WC_MASKED_FETCH_ADD,
479 /*
480  * Set value of IB_WC_RECV so consumers can test if a completion is a
481  * receive by testing (opcode & IB_WC_RECV).
482  */
483 	IB_WC_RECV			= 1 << 7,
484 	IB_WC_RECV_RDMA_WITH_IMM
485 };
486 
487 enum ib_wc_flags {
488 	IB_WC_GRH		= 1,
489 	IB_WC_WITH_IMM		= (1<<1),
490 	IB_WC_WITH_INVALIDATE	= (1<<2),
491 };
492 
493 struct ib_wc {
494 	u64			wr_id;
495 	enum ib_wc_status	status;
496 	enum ib_wc_opcode	opcode;
497 	u32			vendor_err;
498 	u32			byte_len;
499 	struct ib_qp	       *qp;
500 	union {
501 		__be32		imm_data;
502 		u32		invalidate_rkey;
503 	} ex;
504 	u32			src_qp;
505 	int			wc_flags;
506 	u16			pkey_index;
507 	u16			slid;
508 	u8			sl;
509 	u8			dlid_path_bits;
510 	u8			port_num;	/* valid only for DR SMPs on switches */
511 	int			csum_ok;
512 };
513 
514 enum ib_cq_notify_flags {
515 	IB_CQ_SOLICITED			= 1 << 0,
516 	IB_CQ_NEXT_COMP			= 1 << 1,
517 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
518 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
519 };
520 
521 enum ib_srq_attr_mask {
522 	IB_SRQ_MAX_WR	= 1 << 0,
523 	IB_SRQ_LIMIT	= 1 << 1,
524 };
525 
526 struct ib_srq_attr {
527 	u32	max_wr;
528 	u32	max_sge;
529 	u32	srq_limit;
530 };
531 
532 struct ib_srq_init_attr {
533 	void		      (*event_handler)(struct ib_event *, void *);
534 	void		       *srq_context;
535 	struct ib_srq_attr	attr;
536 };
537 
538 struct ib_qp_cap {
539 	u32	max_send_wr;
540 	u32	max_recv_wr;
541 	u32	max_send_sge;
542 	u32	max_recv_sge;
543 	u32	max_inline_data;
544 };
545 
546 enum ib_sig_type {
547 	IB_SIGNAL_ALL_WR,
548 	IB_SIGNAL_REQ_WR
549 };
550 
551 enum ib_qp_type {
552 	/*
553 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
554 	 * here (and in that order) since the MAD layer uses them as
555 	 * indices into a 2-entry table.
556 	 */
557 	IB_QPT_SMI,
558 	IB_QPT_GSI,
559 
560 	IB_QPT_RC,
561 	IB_QPT_UC,
562 	IB_QPT_UD,
563 	IB_QPT_RAW_IPV6,
564 	IB_QPT_RAW_ETHERTYPE
565 };
566 
567 enum ib_qp_create_flags {
568 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
569 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
570 };
571 
572 struct ib_qp_init_attr {
573 	void                  (*event_handler)(struct ib_event *, void *);
574 	void		       *qp_context;
575 	struct ib_cq	       *send_cq;
576 	struct ib_cq	       *recv_cq;
577 	struct ib_srq	       *srq;
578 	struct ib_qp_cap	cap;
579 	enum ib_sig_type	sq_sig_type;
580 	enum ib_qp_type		qp_type;
581 	enum ib_qp_create_flags	create_flags;
582 	u8			port_num; /* special QP types only */
583 };
584 
585 enum ib_rnr_timeout {
586 	IB_RNR_TIMER_655_36 =  0,
587 	IB_RNR_TIMER_000_01 =  1,
588 	IB_RNR_TIMER_000_02 =  2,
589 	IB_RNR_TIMER_000_03 =  3,
590 	IB_RNR_TIMER_000_04 =  4,
591 	IB_RNR_TIMER_000_06 =  5,
592 	IB_RNR_TIMER_000_08 =  6,
593 	IB_RNR_TIMER_000_12 =  7,
594 	IB_RNR_TIMER_000_16 =  8,
595 	IB_RNR_TIMER_000_24 =  9,
596 	IB_RNR_TIMER_000_32 = 10,
597 	IB_RNR_TIMER_000_48 = 11,
598 	IB_RNR_TIMER_000_64 = 12,
599 	IB_RNR_TIMER_000_96 = 13,
600 	IB_RNR_TIMER_001_28 = 14,
601 	IB_RNR_TIMER_001_92 = 15,
602 	IB_RNR_TIMER_002_56 = 16,
603 	IB_RNR_TIMER_003_84 = 17,
604 	IB_RNR_TIMER_005_12 = 18,
605 	IB_RNR_TIMER_007_68 = 19,
606 	IB_RNR_TIMER_010_24 = 20,
607 	IB_RNR_TIMER_015_36 = 21,
608 	IB_RNR_TIMER_020_48 = 22,
609 	IB_RNR_TIMER_030_72 = 23,
610 	IB_RNR_TIMER_040_96 = 24,
611 	IB_RNR_TIMER_061_44 = 25,
612 	IB_RNR_TIMER_081_92 = 26,
613 	IB_RNR_TIMER_122_88 = 27,
614 	IB_RNR_TIMER_163_84 = 28,
615 	IB_RNR_TIMER_245_76 = 29,
616 	IB_RNR_TIMER_327_68 = 30,
617 	IB_RNR_TIMER_491_52 = 31
618 };
619 
620 enum ib_qp_attr_mask {
621 	IB_QP_STATE			= 1,
622 	IB_QP_CUR_STATE			= (1<<1),
623 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
624 	IB_QP_ACCESS_FLAGS		= (1<<3),
625 	IB_QP_PKEY_INDEX		= (1<<4),
626 	IB_QP_PORT			= (1<<5),
627 	IB_QP_QKEY			= (1<<6),
628 	IB_QP_AV			= (1<<7),
629 	IB_QP_PATH_MTU			= (1<<8),
630 	IB_QP_TIMEOUT			= (1<<9),
631 	IB_QP_RETRY_CNT			= (1<<10),
632 	IB_QP_RNR_RETRY			= (1<<11),
633 	IB_QP_RQ_PSN			= (1<<12),
634 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
635 	IB_QP_ALT_PATH			= (1<<14),
636 	IB_QP_MIN_RNR_TIMER		= (1<<15),
637 	IB_QP_SQ_PSN			= (1<<16),
638 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
639 	IB_QP_PATH_MIG_STATE		= (1<<18),
640 	IB_QP_CAP			= (1<<19),
641 	IB_QP_DEST_QPN			= (1<<20)
642 };
643 
644 enum ib_qp_state {
645 	IB_QPS_RESET,
646 	IB_QPS_INIT,
647 	IB_QPS_RTR,
648 	IB_QPS_RTS,
649 	IB_QPS_SQD,
650 	IB_QPS_SQE,
651 	IB_QPS_ERR
652 };
653 
654 enum ib_mig_state {
655 	IB_MIG_MIGRATED,
656 	IB_MIG_REARM,
657 	IB_MIG_ARMED
658 };
659 
660 struct ib_qp_attr {
661 	enum ib_qp_state	qp_state;
662 	enum ib_qp_state	cur_qp_state;
663 	enum ib_mtu		path_mtu;
664 	enum ib_mig_state	path_mig_state;
665 	u32			qkey;
666 	u32			rq_psn;
667 	u32			sq_psn;
668 	u32			dest_qp_num;
669 	int			qp_access_flags;
670 	struct ib_qp_cap	cap;
671 	struct ib_ah_attr	ah_attr;
672 	struct ib_ah_attr	alt_ah_attr;
673 	u16			pkey_index;
674 	u16			alt_pkey_index;
675 	u8			en_sqd_async_notify;
676 	u8			sq_draining;
677 	u8			max_rd_atomic;
678 	u8			max_dest_rd_atomic;
679 	u8			min_rnr_timer;
680 	u8			port_num;
681 	u8			timeout;
682 	u8			retry_cnt;
683 	u8			rnr_retry;
684 	u8			alt_port_num;
685 	u8			alt_timeout;
686 };
687 
688 enum ib_wr_opcode {
689 	IB_WR_RDMA_WRITE,
690 	IB_WR_RDMA_WRITE_WITH_IMM,
691 	IB_WR_SEND,
692 	IB_WR_SEND_WITH_IMM,
693 	IB_WR_RDMA_READ,
694 	IB_WR_ATOMIC_CMP_AND_SWP,
695 	IB_WR_ATOMIC_FETCH_AND_ADD,
696 	IB_WR_LSO,
697 	IB_WR_SEND_WITH_INV,
698 	IB_WR_RDMA_READ_WITH_INV,
699 	IB_WR_LOCAL_INV,
700 	IB_WR_FAST_REG_MR,
701 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
702 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
703 };
704 
705 enum ib_send_flags {
706 	IB_SEND_FENCE		= 1,
707 	IB_SEND_SIGNALED	= (1<<1),
708 	IB_SEND_SOLICITED	= (1<<2),
709 	IB_SEND_INLINE		= (1<<3),
710 	IB_SEND_IP_CSUM		= (1<<4)
711 };
712 
713 struct ib_sge {
714 	u64	addr;
715 	u32	length;
716 	u32	lkey;
717 };
718 
719 struct ib_fast_reg_page_list {
720 	struct ib_device       *device;
721 	u64		       *page_list;
722 	unsigned int		max_page_list_len;
723 };
724 
725 struct ib_send_wr {
726 	struct ib_send_wr      *next;
727 	u64			wr_id;
728 	struct ib_sge	       *sg_list;
729 	int			num_sge;
730 	enum ib_wr_opcode	opcode;
731 	int			send_flags;
732 	union {
733 		__be32		imm_data;
734 		u32		invalidate_rkey;
735 	} ex;
736 	union {
737 		struct {
738 			u64	remote_addr;
739 			u32	rkey;
740 		} rdma;
741 		struct {
742 			u64	remote_addr;
743 			u64	compare_add;
744 			u64	swap;
745 			u64	compare_add_mask;
746 			u64	swap_mask;
747 			u32	rkey;
748 		} atomic;
749 		struct {
750 			struct ib_ah *ah;
751 			void   *header;
752 			int     hlen;
753 			int     mss;
754 			u32	remote_qpn;
755 			u32	remote_qkey;
756 			u16	pkey_index; /* valid for GSI only */
757 			u8	port_num;   /* valid for DR SMPs on switch only */
758 		} ud;
759 		struct {
760 			u64				iova_start;
761 			struct ib_fast_reg_page_list   *page_list;
762 			unsigned int			page_shift;
763 			unsigned int			page_list_len;
764 			u32				length;
765 			int				access_flags;
766 			u32				rkey;
767 		} fast_reg;
768 	} wr;
769 };
770 
771 struct ib_recv_wr {
772 	struct ib_recv_wr      *next;
773 	u64			wr_id;
774 	struct ib_sge	       *sg_list;
775 	int			num_sge;
776 };
777 
778 enum ib_access_flags {
779 	IB_ACCESS_LOCAL_WRITE	= 1,
780 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
781 	IB_ACCESS_REMOTE_READ	= (1<<2),
782 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
783 	IB_ACCESS_MW_BIND	= (1<<4)
784 };
785 
786 struct ib_phys_buf {
787 	u64      addr;
788 	u64      size;
789 };
790 
791 struct ib_mr_attr {
792 	struct ib_pd	*pd;
793 	u64		device_virt_addr;
794 	u64		size;
795 	int		mr_access_flags;
796 	u32		lkey;
797 	u32		rkey;
798 };
799 
800 enum ib_mr_rereg_flags {
801 	IB_MR_REREG_TRANS	= 1,
802 	IB_MR_REREG_PD		= (1<<1),
803 	IB_MR_REREG_ACCESS	= (1<<2)
804 };
805 
806 struct ib_mw_bind {
807 	struct ib_mr   *mr;
808 	u64		wr_id;
809 	u64		addr;
810 	u32		length;
811 	int		send_flags;
812 	int		mw_access_flags;
813 };
814 
815 struct ib_fmr_attr {
816 	int	max_pages;
817 	int	max_maps;
818 	u8	page_shift;
819 };
820 
821 struct ib_ucontext {
822 	struct ib_device       *device;
823 	struct list_head	pd_list;
824 	struct list_head	mr_list;
825 	struct list_head	mw_list;
826 	struct list_head	cq_list;
827 	struct list_head	qp_list;
828 	struct list_head	srq_list;
829 	struct list_head	ah_list;
830 	int			closing;
831 };
832 
833 struct ib_uobject {
834 	u64			user_handle;	/* handle given to us by userspace */
835 	struct ib_ucontext     *context;	/* associated user context */
836 	void		       *object;		/* containing object */
837 	struct list_head	list;		/* link to context's list */
838 	int			id;		/* index into kernel idr */
839 	struct kref		ref;
840 	struct rw_semaphore	mutex;		/* protects .live */
841 	int			live;
842 };
843 
844 struct ib_udata {
845 	void __user *inbuf;
846 	void __user *outbuf;
847 	size_t       inlen;
848 	size_t       outlen;
849 };
850 
851 struct ib_pd {
852 	struct ib_device       *device;
853 	struct ib_uobject      *uobject;
854 	atomic_t          	usecnt; /* count all resources */
855 };
856 
857 struct ib_ah {
858 	struct ib_device	*device;
859 	struct ib_pd		*pd;
860 	struct ib_uobject	*uobject;
861 };
862 
863 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
864 
865 struct ib_cq {
866 	struct ib_device       *device;
867 	struct ib_uobject      *uobject;
868 	ib_comp_handler   	comp_handler;
869 	void                  (*event_handler)(struct ib_event *, void *);
870 	void                   *cq_context;
871 	int               	cqe;
872 	atomic_t          	usecnt; /* count number of work queues */
873 };
874 
875 struct ib_srq {
876 	struct ib_device       *device;
877 	struct ib_pd	       *pd;
878 	struct ib_uobject      *uobject;
879 	void		      (*event_handler)(struct ib_event *, void *);
880 	void		       *srq_context;
881 	atomic_t		usecnt;
882 };
883 
884 struct ib_qp {
885 	struct ib_device       *device;
886 	struct ib_pd	       *pd;
887 	struct ib_cq	       *send_cq;
888 	struct ib_cq	       *recv_cq;
889 	struct ib_srq	       *srq;
890 	struct ib_uobject      *uobject;
891 	void                  (*event_handler)(struct ib_event *, void *);
892 	void		       *qp_context;
893 	u32			qp_num;
894 	enum ib_qp_type		qp_type;
895 };
896 
897 struct ib_mr {
898 	struct ib_device  *device;
899 	struct ib_pd	  *pd;
900 	struct ib_uobject *uobject;
901 	u32		   lkey;
902 	u32		   rkey;
903 	atomic_t	   usecnt; /* count number of MWs */
904 };
905 
906 struct ib_mw {
907 	struct ib_device	*device;
908 	struct ib_pd		*pd;
909 	struct ib_uobject	*uobject;
910 	u32			rkey;
911 };
912 
913 struct ib_fmr {
914 	struct ib_device	*device;
915 	struct ib_pd		*pd;
916 	struct list_head	list;
917 	u32			lkey;
918 	u32			rkey;
919 };
920 
921 struct ib_mad;
922 struct ib_grh;
923 
924 enum ib_process_mad_flags {
925 	IB_MAD_IGNORE_MKEY	= 1,
926 	IB_MAD_IGNORE_BKEY	= 2,
927 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
928 };
929 
930 enum ib_mad_result {
931 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
932 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
933 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
934 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
935 };
936 
937 #define IB_DEVICE_NAME_MAX 64
938 
939 struct ib_cache {
940 	rwlock_t                lock;
941 	struct ib_event_handler event_handler;
942 	struct ib_pkey_cache  **pkey_cache;
943 	struct ib_gid_cache   **gid_cache;
944 	u8                     *lmc_cache;
945 };
946 
947 struct ib_dma_mapping_ops {
948 	int		(*mapping_error)(struct ib_device *dev,
949 					 u64 dma_addr);
950 	u64		(*map_single)(struct ib_device *dev,
951 				      void *ptr, size_t size,
952 				      enum dma_data_direction direction);
953 	void		(*unmap_single)(struct ib_device *dev,
954 					u64 addr, size_t size,
955 					enum dma_data_direction direction);
956 	u64		(*map_page)(struct ib_device *dev,
957 				    struct page *page, unsigned long offset,
958 				    size_t size,
959 				    enum dma_data_direction direction);
960 	void		(*unmap_page)(struct ib_device *dev,
961 				      u64 addr, size_t size,
962 				      enum dma_data_direction direction);
963 	int		(*map_sg)(struct ib_device *dev,
964 				  struct scatterlist *sg, int nents,
965 				  enum dma_data_direction direction);
966 	void		(*unmap_sg)(struct ib_device *dev,
967 				    struct scatterlist *sg, int nents,
968 				    enum dma_data_direction direction);
969 	u64		(*dma_address)(struct ib_device *dev,
970 				       struct scatterlist *sg);
971 	unsigned int	(*dma_len)(struct ib_device *dev,
972 				   struct scatterlist *sg);
973 	void		(*sync_single_for_cpu)(struct ib_device *dev,
974 					       u64 dma_handle,
975 					       size_t size,
976 					       enum dma_data_direction dir);
977 	void		(*sync_single_for_device)(struct ib_device *dev,
978 						  u64 dma_handle,
979 						  size_t size,
980 						  enum dma_data_direction dir);
981 	void		*(*alloc_coherent)(struct ib_device *dev,
982 					   size_t size,
983 					   u64 *dma_handle,
984 					   gfp_t flag);
985 	void		(*free_coherent)(struct ib_device *dev,
986 					 size_t size, void *cpu_addr,
987 					 u64 dma_handle);
988 };
989 
990 struct iw_cm_verbs;
991 
992 struct ib_device {
993 	struct device                *dma_device;
994 
995 	char                          name[IB_DEVICE_NAME_MAX];
996 
997 	struct list_head              event_handler_list;
998 	spinlock_t                    event_handler_lock;
999 
1000 	spinlock_t                    client_data_lock;
1001 	struct list_head              core_list;
1002 	struct list_head              client_data_list;
1003 
1004 	struct ib_cache               cache;
1005 	int                          *pkey_tbl_len;
1006 	int                          *gid_tbl_len;
1007 
1008 	int			      num_comp_vectors;
1009 
1010 	struct iw_cm_verbs	     *iwcm;
1011 
1012 	int		           (*get_protocol_stats)(struct ib_device *device,
1013 							 union rdma_protocol_stats *stats);
1014 	int		           (*query_device)(struct ib_device *device,
1015 						   struct ib_device_attr *device_attr);
1016 	int		           (*query_port)(struct ib_device *device,
1017 						 u8 port_num,
1018 						 struct ib_port_attr *port_attr);
1019 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1020 						     u8 port_num);
1021 	int		           (*query_gid)(struct ib_device *device,
1022 						u8 port_num, int index,
1023 						union ib_gid *gid);
1024 	int		           (*query_pkey)(struct ib_device *device,
1025 						 u8 port_num, u16 index, u16 *pkey);
1026 	int		           (*modify_device)(struct ib_device *device,
1027 						    int device_modify_mask,
1028 						    struct ib_device_modify *device_modify);
1029 	int		           (*modify_port)(struct ib_device *device,
1030 						  u8 port_num, int port_modify_mask,
1031 						  struct ib_port_modify *port_modify);
1032 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1033 						     struct ib_udata *udata);
1034 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1035 	int                        (*mmap)(struct ib_ucontext *context,
1036 					   struct vm_area_struct *vma);
1037 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1038 					       struct ib_ucontext *context,
1039 					       struct ib_udata *udata);
1040 	int                        (*dealloc_pd)(struct ib_pd *pd);
1041 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1042 						struct ib_ah_attr *ah_attr);
1043 	int                        (*modify_ah)(struct ib_ah *ah,
1044 						struct ib_ah_attr *ah_attr);
1045 	int                        (*query_ah)(struct ib_ah *ah,
1046 					       struct ib_ah_attr *ah_attr);
1047 	int                        (*destroy_ah)(struct ib_ah *ah);
1048 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1049 						 struct ib_srq_init_attr *srq_init_attr,
1050 						 struct ib_udata *udata);
1051 	int                        (*modify_srq)(struct ib_srq *srq,
1052 						 struct ib_srq_attr *srq_attr,
1053 						 enum ib_srq_attr_mask srq_attr_mask,
1054 						 struct ib_udata *udata);
1055 	int                        (*query_srq)(struct ib_srq *srq,
1056 						struct ib_srq_attr *srq_attr);
1057 	int                        (*destroy_srq)(struct ib_srq *srq);
1058 	int                        (*post_srq_recv)(struct ib_srq *srq,
1059 						    struct ib_recv_wr *recv_wr,
1060 						    struct ib_recv_wr **bad_recv_wr);
1061 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1062 						struct ib_qp_init_attr *qp_init_attr,
1063 						struct ib_udata *udata);
1064 	int                        (*modify_qp)(struct ib_qp *qp,
1065 						struct ib_qp_attr *qp_attr,
1066 						int qp_attr_mask,
1067 						struct ib_udata *udata);
1068 	int                        (*query_qp)(struct ib_qp *qp,
1069 					       struct ib_qp_attr *qp_attr,
1070 					       int qp_attr_mask,
1071 					       struct ib_qp_init_attr *qp_init_attr);
1072 	int                        (*destroy_qp)(struct ib_qp *qp);
1073 	int                        (*post_send)(struct ib_qp *qp,
1074 						struct ib_send_wr *send_wr,
1075 						struct ib_send_wr **bad_send_wr);
1076 	int                        (*post_recv)(struct ib_qp *qp,
1077 						struct ib_recv_wr *recv_wr,
1078 						struct ib_recv_wr **bad_recv_wr);
1079 	struct ib_cq *             (*create_cq)(struct ib_device *device, int cqe,
1080 						int comp_vector,
1081 						struct ib_ucontext *context,
1082 						struct ib_udata *udata);
1083 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1084 						u16 cq_period);
1085 	int                        (*destroy_cq)(struct ib_cq *cq);
1086 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1087 						struct ib_udata *udata);
1088 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1089 					      struct ib_wc *wc);
1090 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1091 	int                        (*req_notify_cq)(struct ib_cq *cq,
1092 						    enum ib_cq_notify_flags flags);
1093 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1094 						      int wc_cnt);
1095 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1096 						 int mr_access_flags);
1097 	struct ib_mr *             (*reg_phys_mr)(struct ib_pd *pd,
1098 						  struct ib_phys_buf *phys_buf_array,
1099 						  int num_phys_buf,
1100 						  int mr_access_flags,
1101 						  u64 *iova_start);
1102 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1103 						  u64 start, u64 length,
1104 						  u64 virt_addr,
1105 						  int mr_access_flags,
1106 						  struct ib_udata *udata);
1107 	int                        (*query_mr)(struct ib_mr *mr,
1108 					       struct ib_mr_attr *mr_attr);
1109 	int                        (*dereg_mr)(struct ib_mr *mr);
1110 	struct ib_mr *		   (*alloc_fast_reg_mr)(struct ib_pd *pd,
1111 					       int max_page_list_len);
1112 	struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1113 								   int page_list_len);
1114 	void			   (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1115 	int                        (*rereg_phys_mr)(struct ib_mr *mr,
1116 						    int mr_rereg_mask,
1117 						    struct ib_pd *pd,
1118 						    struct ib_phys_buf *phys_buf_array,
1119 						    int num_phys_buf,
1120 						    int mr_access_flags,
1121 						    u64 *iova_start);
1122 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd);
1123 	int                        (*bind_mw)(struct ib_qp *qp,
1124 					      struct ib_mw *mw,
1125 					      struct ib_mw_bind *mw_bind);
1126 	int                        (*dealloc_mw)(struct ib_mw *mw);
1127 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1128 						int mr_access_flags,
1129 						struct ib_fmr_attr *fmr_attr);
1130 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1131 						   u64 *page_list, int list_len,
1132 						   u64 iova);
1133 	int		           (*unmap_fmr)(struct list_head *fmr_list);
1134 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1135 	int                        (*attach_mcast)(struct ib_qp *qp,
1136 						   union ib_gid *gid,
1137 						   u16 lid);
1138 	int                        (*detach_mcast)(struct ib_qp *qp,
1139 						   union ib_gid *gid,
1140 						   u16 lid);
1141 	int                        (*process_mad)(struct ib_device *device,
1142 						  int process_mad_flags,
1143 						  u8 port_num,
1144 						  struct ib_wc *in_wc,
1145 						  struct ib_grh *in_grh,
1146 						  struct ib_mad *in_mad,
1147 						  struct ib_mad *out_mad);
1148 
1149 	struct ib_dma_mapping_ops   *dma_ops;
1150 
1151 	struct module               *owner;
1152 	struct device                dev;
1153 	struct kobject               *ports_parent;
1154 	struct list_head             port_list;
1155 
1156 	enum {
1157 		IB_DEV_UNINITIALIZED,
1158 		IB_DEV_REGISTERED,
1159 		IB_DEV_UNREGISTERED
1160 	}                            reg_state;
1161 
1162 	int			     uverbs_abi_ver;
1163 	u64			     uverbs_cmd_mask;
1164 
1165 	char			     node_desc[64];
1166 	__be64			     node_guid;
1167 	u32			     local_dma_lkey;
1168 	u8                           node_type;
1169 	u8                           phys_port_cnt;
1170 };
1171 
1172 struct ib_client {
1173 	char  *name;
1174 	void (*add)   (struct ib_device *);
1175 	void (*remove)(struct ib_device *);
1176 
1177 	struct list_head list;
1178 };
1179 
1180 struct ib_device *ib_alloc_device(size_t size);
1181 void ib_dealloc_device(struct ib_device *device);
1182 
1183 int ib_register_device(struct ib_device *device,
1184 		       int (*port_callback)(struct ib_device *,
1185 					    u8, struct kobject *));
1186 void ib_unregister_device(struct ib_device *device);
1187 
1188 int ib_register_client   (struct ib_client *client);
1189 void ib_unregister_client(struct ib_client *client);
1190 
1191 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1192 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
1193 			 void *data);
1194 
1195 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1196 {
1197 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1198 }
1199 
1200 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1201 {
1202 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1203 }
1204 
1205 /**
1206  * ib_modify_qp_is_ok - Check that the supplied attribute mask
1207  * contains all required attributes and no attributes not allowed for
1208  * the given QP state transition.
1209  * @cur_state: Current QP state
1210  * @next_state: Next QP state
1211  * @type: QP type
1212  * @mask: Mask of supplied QP attributes
1213  *
1214  * This function is a helper function that a low-level driver's
1215  * modify_qp method can use to validate the consumer's input.  It
1216  * checks that cur_state and next_state are valid QP states, that a
1217  * transition from cur_state to next_state is allowed by the IB spec,
1218  * and that the attribute mask supplied is allowed for the transition.
1219  */
1220 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1221 		       enum ib_qp_type type, enum ib_qp_attr_mask mask);
1222 
1223 int ib_register_event_handler  (struct ib_event_handler *event_handler);
1224 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1225 void ib_dispatch_event(struct ib_event *event);
1226 
1227 int ib_query_device(struct ib_device *device,
1228 		    struct ib_device_attr *device_attr);
1229 
1230 int ib_query_port(struct ib_device *device,
1231 		  u8 port_num, struct ib_port_attr *port_attr);
1232 
1233 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1234 					       u8 port_num);
1235 
1236 int ib_query_gid(struct ib_device *device,
1237 		 u8 port_num, int index, union ib_gid *gid);
1238 
1239 int ib_query_pkey(struct ib_device *device,
1240 		  u8 port_num, u16 index, u16 *pkey);
1241 
1242 int ib_modify_device(struct ib_device *device,
1243 		     int device_modify_mask,
1244 		     struct ib_device_modify *device_modify);
1245 
1246 int ib_modify_port(struct ib_device *device,
1247 		   u8 port_num, int port_modify_mask,
1248 		   struct ib_port_modify *port_modify);
1249 
1250 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1251 		u8 *port_num, u16 *index);
1252 
1253 int ib_find_pkey(struct ib_device *device,
1254 		 u8 port_num, u16 pkey, u16 *index);
1255 
1256 /**
1257  * ib_alloc_pd - Allocates an unused protection domain.
1258  * @device: The device on which to allocate the protection domain.
1259  *
1260  * A protection domain object provides an association between QPs, shared
1261  * receive queues, address handles, memory regions, and memory windows.
1262  */
1263 struct ib_pd *ib_alloc_pd(struct ib_device *device);
1264 
1265 /**
1266  * ib_dealloc_pd - Deallocates a protection domain.
1267  * @pd: The protection domain to deallocate.
1268  */
1269 int ib_dealloc_pd(struct ib_pd *pd);
1270 
1271 /**
1272  * ib_create_ah - Creates an address handle for the given address vector.
1273  * @pd: The protection domain associated with the address handle.
1274  * @ah_attr: The attributes of the address vector.
1275  *
1276  * The address handle is used to reference a local or global destination
1277  * in all UD QP post sends.
1278  */
1279 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
1280 
1281 /**
1282  * ib_init_ah_from_wc - Initializes address handle attributes from a
1283  *   work completion.
1284  * @device: Device on which the received message arrived.
1285  * @port_num: Port on which the received message arrived.
1286  * @wc: Work completion associated with the received message.
1287  * @grh: References the received global route header.  This parameter is
1288  *   ignored unless the work completion indicates that the GRH is valid.
1289  * @ah_attr: Returned attributes that can be used when creating an address
1290  *   handle for replying to the message.
1291  */
1292 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
1293 		       struct ib_grh *grh, struct ib_ah_attr *ah_attr);
1294 
1295 /**
1296  * ib_create_ah_from_wc - Creates an address handle associated with the
1297  *   sender of the specified work completion.
1298  * @pd: The protection domain associated with the address handle.
1299  * @wc: Work completion information associated with a received message.
1300  * @grh: References the received global route header.  This parameter is
1301  *   ignored unless the work completion indicates that the GRH is valid.
1302  * @port_num: The outbound port number to associate with the address.
1303  *
1304  * The address handle is used to reference a local or global destination
1305  * in all UD QP post sends.
1306  */
1307 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
1308 				   struct ib_grh *grh, u8 port_num);
1309 
1310 /**
1311  * ib_modify_ah - Modifies the address vector associated with an address
1312  *   handle.
1313  * @ah: The address handle to modify.
1314  * @ah_attr: The new address vector attributes to associate with the
1315  *   address handle.
1316  */
1317 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1318 
1319 /**
1320  * ib_query_ah - Queries the address vector associated with an address
1321  *   handle.
1322  * @ah: The address handle to query.
1323  * @ah_attr: The address vector attributes associated with the address
1324  *   handle.
1325  */
1326 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1327 
1328 /**
1329  * ib_destroy_ah - Destroys an address handle.
1330  * @ah: The address handle to destroy.
1331  */
1332 int ib_destroy_ah(struct ib_ah *ah);
1333 
1334 /**
1335  * ib_create_srq - Creates a SRQ associated with the specified protection
1336  *   domain.
1337  * @pd: The protection domain associated with the SRQ.
1338  * @srq_init_attr: A list of initial attributes required to create the
1339  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
1340  *   the actual capabilities of the created SRQ.
1341  *
1342  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1343  * requested size of the SRQ, and set to the actual values allocated
1344  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
1345  * will always be at least as large as the requested values.
1346  */
1347 struct ib_srq *ib_create_srq(struct ib_pd *pd,
1348 			     struct ib_srq_init_attr *srq_init_attr);
1349 
1350 /**
1351  * ib_modify_srq - Modifies the attributes for the specified SRQ.
1352  * @srq: The SRQ to modify.
1353  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
1354  *   the current values of selected SRQ attributes are returned.
1355  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1356  *   are being modified.
1357  *
1358  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1359  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1360  * the number of receives queued drops below the limit.
1361  */
1362 int ib_modify_srq(struct ib_srq *srq,
1363 		  struct ib_srq_attr *srq_attr,
1364 		  enum ib_srq_attr_mask srq_attr_mask);
1365 
1366 /**
1367  * ib_query_srq - Returns the attribute list and current values for the
1368  *   specified SRQ.
1369  * @srq: The SRQ to query.
1370  * @srq_attr: The attributes of the specified SRQ.
1371  */
1372 int ib_query_srq(struct ib_srq *srq,
1373 		 struct ib_srq_attr *srq_attr);
1374 
1375 /**
1376  * ib_destroy_srq - Destroys the specified SRQ.
1377  * @srq: The SRQ to destroy.
1378  */
1379 int ib_destroy_srq(struct ib_srq *srq);
1380 
1381 /**
1382  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1383  * @srq: The SRQ to post the work request on.
1384  * @recv_wr: A list of work requests to post on the receive queue.
1385  * @bad_recv_wr: On an immediate failure, this parameter will reference
1386  *   the work request that failed to be posted on the QP.
1387  */
1388 static inline int ib_post_srq_recv(struct ib_srq *srq,
1389 				   struct ib_recv_wr *recv_wr,
1390 				   struct ib_recv_wr **bad_recv_wr)
1391 {
1392 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1393 }
1394 
1395 /**
1396  * ib_create_qp - Creates a QP associated with the specified protection
1397  *   domain.
1398  * @pd: The protection domain associated with the QP.
1399  * @qp_init_attr: A list of initial attributes required to create the
1400  *   QP.  If QP creation succeeds, then the attributes are updated to
1401  *   the actual capabilities of the created QP.
1402  */
1403 struct ib_qp *ib_create_qp(struct ib_pd *pd,
1404 			   struct ib_qp_init_attr *qp_init_attr);
1405 
1406 /**
1407  * ib_modify_qp - Modifies the attributes for the specified QP and then
1408  *   transitions the QP to the given state.
1409  * @qp: The QP to modify.
1410  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
1411  *   the current values of selected QP attributes are returned.
1412  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1413  *   are being modified.
1414  */
1415 int ib_modify_qp(struct ib_qp *qp,
1416 		 struct ib_qp_attr *qp_attr,
1417 		 int qp_attr_mask);
1418 
1419 /**
1420  * ib_query_qp - Returns the attribute list and current values for the
1421  *   specified QP.
1422  * @qp: The QP to query.
1423  * @qp_attr: The attributes of the specified QP.
1424  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1425  * @qp_init_attr: Additional attributes of the selected QP.
1426  *
1427  * The qp_attr_mask may be used to limit the query to gathering only the
1428  * selected attributes.
1429  */
1430 int ib_query_qp(struct ib_qp *qp,
1431 		struct ib_qp_attr *qp_attr,
1432 		int qp_attr_mask,
1433 		struct ib_qp_init_attr *qp_init_attr);
1434 
1435 /**
1436  * ib_destroy_qp - Destroys the specified QP.
1437  * @qp: The QP to destroy.
1438  */
1439 int ib_destroy_qp(struct ib_qp *qp);
1440 
1441 /**
1442  * ib_post_send - Posts a list of work requests to the send queue of
1443  *   the specified QP.
1444  * @qp: The QP to post the work request on.
1445  * @send_wr: A list of work requests to post on the send queue.
1446  * @bad_send_wr: On an immediate failure, this parameter will reference
1447  *   the work request that failed to be posted on the QP.
1448  *
1449  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
1450  * error is returned, the QP state shall not be affected,
1451  * ib_post_send() will return an immediate error after queueing any
1452  * earlier work requests in the list.
1453  */
1454 static inline int ib_post_send(struct ib_qp *qp,
1455 			       struct ib_send_wr *send_wr,
1456 			       struct ib_send_wr **bad_send_wr)
1457 {
1458 	return qp->device->post_send(qp, send_wr, bad_send_wr);
1459 }
1460 
1461 /**
1462  * ib_post_recv - Posts a list of work requests to the receive queue of
1463  *   the specified QP.
1464  * @qp: The QP to post the work request on.
1465  * @recv_wr: A list of work requests to post on the receive queue.
1466  * @bad_recv_wr: On an immediate failure, this parameter will reference
1467  *   the work request that failed to be posted on the QP.
1468  */
1469 static inline int ib_post_recv(struct ib_qp *qp,
1470 			       struct ib_recv_wr *recv_wr,
1471 			       struct ib_recv_wr **bad_recv_wr)
1472 {
1473 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1474 }
1475 
1476 /**
1477  * ib_create_cq - Creates a CQ on the specified device.
1478  * @device: The device on which to create the CQ.
1479  * @comp_handler: A user-specified callback that is invoked when a
1480  *   completion event occurs on the CQ.
1481  * @event_handler: A user-specified callback that is invoked when an
1482  *   asynchronous event not associated with a completion occurs on the CQ.
1483  * @cq_context: Context associated with the CQ returned to the user via
1484  *   the associated completion and event handlers.
1485  * @cqe: The minimum size of the CQ.
1486  * @comp_vector - Completion vector used to signal completion events.
1487  *     Must be >= 0 and < context->num_comp_vectors.
1488  *
1489  * Users can examine the cq structure to determine the actual CQ size.
1490  */
1491 struct ib_cq *ib_create_cq(struct ib_device *device,
1492 			   ib_comp_handler comp_handler,
1493 			   void (*event_handler)(struct ib_event *, void *),
1494 			   void *cq_context, int cqe, int comp_vector);
1495 
1496 /**
1497  * ib_resize_cq - Modifies the capacity of the CQ.
1498  * @cq: The CQ to resize.
1499  * @cqe: The minimum size of the CQ.
1500  *
1501  * Users can examine the cq structure to determine the actual CQ size.
1502  */
1503 int ib_resize_cq(struct ib_cq *cq, int cqe);
1504 
1505 /**
1506  * ib_modify_cq - Modifies moderation params of the CQ
1507  * @cq: The CQ to modify.
1508  * @cq_count: number of CQEs that will trigger an event
1509  * @cq_period: max period of time in usec before triggering an event
1510  *
1511  */
1512 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1513 
1514 /**
1515  * ib_destroy_cq - Destroys the specified CQ.
1516  * @cq: The CQ to destroy.
1517  */
1518 int ib_destroy_cq(struct ib_cq *cq);
1519 
1520 /**
1521  * ib_poll_cq - poll a CQ for completion(s)
1522  * @cq:the CQ being polled
1523  * @num_entries:maximum number of completions to return
1524  * @wc:array of at least @num_entries &struct ib_wc where completions
1525  *   will be returned
1526  *
1527  * Poll a CQ for (possibly multiple) completions.  If the return value
1528  * is < 0, an error occurred.  If the return value is >= 0, it is the
1529  * number of completions returned.  If the return value is
1530  * non-negative and < num_entries, then the CQ was emptied.
1531  */
1532 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1533 			     struct ib_wc *wc)
1534 {
1535 	return cq->device->poll_cq(cq, num_entries, wc);
1536 }
1537 
1538 /**
1539  * ib_peek_cq - Returns the number of unreaped completions currently
1540  *   on the specified CQ.
1541  * @cq: The CQ to peek.
1542  * @wc_cnt: A minimum number of unreaped completions to check for.
1543  *
1544  * If the number of unreaped completions is greater than or equal to wc_cnt,
1545  * this function returns wc_cnt, otherwise, it returns the actual number of
1546  * unreaped completions.
1547  */
1548 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1549 
1550 /**
1551  * ib_req_notify_cq - Request completion notification on a CQ.
1552  * @cq: The CQ to generate an event for.
1553  * @flags:
1554  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1555  *   to request an event on the next solicited event or next work
1556  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1557  *   may also be |ed in to request a hint about missed events, as
1558  *   described below.
1559  *
1560  * Return Value:
1561  *    < 0 means an error occurred while requesting notification
1562  *   == 0 means notification was requested successfully, and if
1563  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1564  *        were missed and it is safe to wait for another event.  In
1565  *        this case is it guaranteed that any work completions added
1566  *        to the CQ since the last CQ poll will trigger a completion
1567  *        notification event.
1568  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1569  *        in.  It means that the consumer must poll the CQ again to
1570  *        make sure it is empty to avoid missing an event because of a
1571  *        race between requesting notification and an entry being
1572  *        added to the CQ.  This return value means it is possible
1573  *        (but not guaranteed) that a work completion has been added
1574  *        to the CQ since the last poll without triggering a
1575  *        completion notification event.
1576  */
1577 static inline int ib_req_notify_cq(struct ib_cq *cq,
1578 				   enum ib_cq_notify_flags flags)
1579 {
1580 	return cq->device->req_notify_cq(cq, flags);
1581 }
1582 
1583 /**
1584  * ib_req_ncomp_notif - Request completion notification when there are
1585  *   at least the specified number of unreaped completions on the CQ.
1586  * @cq: The CQ to generate an event for.
1587  * @wc_cnt: The number of unreaped completions that should be on the
1588  *   CQ before an event is generated.
1589  */
1590 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1591 {
1592 	return cq->device->req_ncomp_notif ?
1593 		cq->device->req_ncomp_notif(cq, wc_cnt) :
1594 		-ENOSYS;
1595 }
1596 
1597 /**
1598  * ib_get_dma_mr - Returns a memory region for system memory that is
1599  *   usable for DMA.
1600  * @pd: The protection domain associated with the memory region.
1601  * @mr_access_flags: Specifies the memory access rights.
1602  *
1603  * Note that the ib_dma_*() functions defined below must be used
1604  * to create/destroy addresses used with the Lkey or Rkey returned
1605  * by ib_get_dma_mr().
1606  */
1607 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1608 
1609 /**
1610  * ib_dma_mapping_error - check a DMA addr for error
1611  * @dev: The device for which the dma_addr was created
1612  * @dma_addr: The DMA address to check
1613  */
1614 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1615 {
1616 	if (dev->dma_ops)
1617 		return dev->dma_ops->mapping_error(dev, dma_addr);
1618 	return dma_mapping_error(dev->dma_device, dma_addr);
1619 }
1620 
1621 /**
1622  * ib_dma_map_single - Map a kernel virtual address to DMA address
1623  * @dev: The device for which the dma_addr is to be created
1624  * @cpu_addr: The kernel virtual address
1625  * @size: The size of the region in bytes
1626  * @direction: The direction of the DMA
1627  */
1628 static inline u64 ib_dma_map_single(struct ib_device *dev,
1629 				    void *cpu_addr, size_t size,
1630 				    enum dma_data_direction direction)
1631 {
1632 	if (dev->dma_ops)
1633 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1634 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1635 }
1636 
1637 /**
1638  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1639  * @dev: The device for which the DMA address was created
1640  * @addr: The DMA address
1641  * @size: The size of the region in bytes
1642  * @direction: The direction of the DMA
1643  */
1644 static inline void ib_dma_unmap_single(struct ib_device *dev,
1645 				       u64 addr, size_t size,
1646 				       enum dma_data_direction direction)
1647 {
1648 	if (dev->dma_ops)
1649 		dev->dma_ops->unmap_single(dev, addr, size, direction);
1650 	else
1651 		dma_unmap_single(dev->dma_device, addr, size, direction);
1652 }
1653 
1654 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
1655 					  void *cpu_addr, size_t size,
1656 					  enum dma_data_direction direction,
1657 					  struct dma_attrs *attrs)
1658 {
1659 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
1660 				    direction, attrs);
1661 }
1662 
1663 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
1664 					     u64 addr, size_t size,
1665 					     enum dma_data_direction direction,
1666 					     struct dma_attrs *attrs)
1667 {
1668 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
1669 				      direction, attrs);
1670 }
1671 
1672 /**
1673  * ib_dma_map_page - Map a physical page to DMA address
1674  * @dev: The device for which the dma_addr is to be created
1675  * @page: The page to be mapped
1676  * @offset: The offset within the page
1677  * @size: The size of the region in bytes
1678  * @direction: The direction of the DMA
1679  */
1680 static inline u64 ib_dma_map_page(struct ib_device *dev,
1681 				  struct page *page,
1682 				  unsigned long offset,
1683 				  size_t size,
1684 					 enum dma_data_direction direction)
1685 {
1686 	if (dev->dma_ops)
1687 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
1688 	return dma_map_page(dev->dma_device, page, offset, size, direction);
1689 }
1690 
1691 /**
1692  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1693  * @dev: The device for which the DMA address was created
1694  * @addr: The DMA address
1695  * @size: The size of the region in bytes
1696  * @direction: The direction of the DMA
1697  */
1698 static inline void ib_dma_unmap_page(struct ib_device *dev,
1699 				     u64 addr, size_t size,
1700 				     enum dma_data_direction direction)
1701 {
1702 	if (dev->dma_ops)
1703 		dev->dma_ops->unmap_page(dev, addr, size, direction);
1704 	else
1705 		dma_unmap_page(dev->dma_device, addr, size, direction);
1706 }
1707 
1708 /**
1709  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
1710  * @dev: The device for which the DMA addresses are to be created
1711  * @sg: The array of scatter/gather entries
1712  * @nents: The number of scatter/gather entries
1713  * @direction: The direction of the DMA
1714  */
1715 static inline int ib_dma_map_sg(struct ib_device *dev,
1716 				struct scatterlist *sg, int nents,
1717 				enum dma_data_direction direction)
1718 {
1719 	if (dev->dma_ops)
1720 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
1721 	return dma_map_sg(dev->dma_device, sg, nents, direction);
1722 }
1723 
1724 /**
1725  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
1726  * @dev: The device for which the DMA addresses were created
1727  * @sg: The array of scatter/gather entries
1728  * @nents: The number of scatter/gather entries
1729  * @direction: The direction of the DMA
1730  */
1731 static inline void ib_dma_unmap_sg(struct ib_device *dev,
1732 				   struct scatterlist *sg, int nents,
1733 				   enum dma_data_direction direction)
1734 {
1735 	if (dev->dma_ops)
1736 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
1737 	else
1738 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
1739 }
1740 
1741 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
1742 				      struct scatterlist *sg, int nents,
1743 				      enum dma_data_direction direction,
1744 				      struct dma_attrs *attrs)
1745 {
1746 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1747 }
1748 
1749 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
1750 					 struct scatterlist *sg, int nents,
1751 					 enum dma_data_direction direction,
1752 					 struct dma_attrs *attrs)
1753 {
1754 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1755 }
1756 /**
1757  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
1758  * @dev: The device for which the DMA addresses were created
1759  * @sg: The scatter/gather entry
1760  */
1761 static inline u64 ib_sg_dma_address(struct ib_device *dev,
1762 				    struct scatterlist *sg)
1763 {
1764 	if (dev->dma_ops)
1765 		return dev->dma_ops->dma_address(dev, sg);
1766 	return sg_dma_address(sg);
1767 }
1768 
1769 /**
1770  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
1771  * @dev: The device for which the DMA addresses were created
1772  * @sg: The scatter/gather entry
1773  */
1774 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
1775 					 struct scatterlist *sg)
1776 {
1777 	if (dev->dma_ops)
1778 		return dev->dma_ops->dma_len(dev, sg);
1779 	return sg_dma_len(sg);
1780 }
1781 
1782 /**
1783  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
1784  * @dev: The device for which the DMA address was created
1785  * @addr: The DMA address
1786  * @size: The size of the region in bytes
1787  * @dir: The direction of the DMA
1788  */
1789 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
1790 					      u64 addr,
1791 					      size_t size,
1792 					      enum dma_data_direction dir)
1793 {
1794 	if (dev->dma_ops)
1795 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
1796 	else
1797 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
1798 }
1799 
1800 /**
1801  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
1802  * @dev: The device for which the DMA address was created
1803  * @addr: The DMA address
1804  * @size: The size of the region in bytes
1805  * @dir: The direction of the DMA
1806  */
1807 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
1808 						 u64 addr,
1809 						 size_t size,
1810 						 enum dma_data_direction dir)
1811 {
1812 	if (dev->dma_ops)
1813 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
1814 	else
1815 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
1816 }
1817 
1818 /**
1819  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
1820  * @dev: The device for which the DMA address is requested
1821  * @size: The size of the region to allocate in bytes
1822  * @dma_handle: A pointer for returning the DMA address of the region
1823  * @flag: memory allocator flags
1824  */
1825 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
1826 					   size_t size,
1827 					   u64 *dma_handle,
1828 					   gfp_t flag)
1829 {
1830 	if (dev->dma_ops)
1831 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
1832 	else {
1833 		dma_addr_t handle;
1834 		void *ret;
1835 
1836 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
1837 		*dma_handle = handle;
1838 		return ret;
1839 	}
1840 }
1841 
1842 /**
1843  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
1844  * @dev: The device for which the DMA addresses were allocated
1845  * @size: The size of the region
1846  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
1847  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
1848  */
1849 static inline void ib_dma_free_coherent(struct ib_device *dev,
1850 					size_t size, void *cpu_addr,
1851 					u64 dma_handle)
1852 {
1853 	if (dev->dma_ops)
1854 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
1855 	else
1856 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
1857 }
1858 
1859 /**
1860  * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
1861  *   by an HCA.
1862  * @pd: The protection domain associated assigned to the registered region.
1863  * @phys_buf_array: Specifies a list of physical buffers to use in the
1864  *   memory region.
1865  * @num_phys_buf: Specifies the size of the phys_buf_array.
1866  * @mr_access_flags: Specifies the memory access rights.
1867  * @iova_start: The offset of the region's starting I/O virtual address.
1868  */
1869 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
1870 			     struct ib_phys_buf *phys_buf_array,
1871 			     int num_phys_buf,
1872 			     int mr_access_flags,
1873 			     u64 *iova_start);
1874 
1875 /**
1876  * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
1877  *   Conceptually, this call performs the functions deregister memory region
1878  *   followed by register physical memory region.  Where possible,
1879  *   resources are reused instead of deallocated and reallocated.
1880  * @mr: The memory region to modify.
1881  * @mr_rereg_mask: A bit-mask used to indicate which of the following
1882  *   properties of the memory region are being modified.
1883  * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
1884  *   the new protection domain to associated with the memory region,
1885  *   otherwise, this parameter is ignored.
1886  * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1887  *   field specifies a list of physical buffers to use in the new
1888  *   translation, otherwise, this parameter is ignored.
1889  * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1890  *   field specifies the size of the phys_buf_array, otherwise, this
1891  *   parameter is ignored.
1892  * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
1893  *   field specifies the new memory access rights, otherwise, this
1894  *   parameter is ignored.
1895  * @iova_start: The offset of the region's starting I/O virtual address.
1896  */
1897 int ib_rereg_phys_mr(struct ib_mr *mr,
1898 		     int mr_rereg_mask,
1899 		     struct ib_pd *pd,
1900 		     struct ib_phys_buf *phys_buf_array,
1901 		     int num_phys_buf,
1902 		     int mr_access_flags,
1903 		     u64 *iova_start);
1904 
1905 /**
1906  * ib_query_mr - Retrieves information about a specific memory region.
1907  * @mr: The memory region to retrieve information about.
1908  * @mr_attr: The attributes of the specified memory region.
1909  */
1910 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
1911 
1912 /**
1913  * ib_dereg_mr - Deregisters a memory region and removes it from the
1914  *   HCA translation table.
1915  * @mr: The memory region to deregister.
1916  */
1917 int ib_dereg_mr(struct ib_mr *mr);
1918 
1919 /**
1920  * ib_alloc_fast_reg_mr - Allocates memory region usable with the
1921  *   IB_WR_FAST_REG_MR send work request.
1922  * @pd: The protection domain associated with the region.
1923  * @max_page_list_len: requested max physical buffer list length to be
1924  *   used with fast register work requests for this MR.
1925  */
1926 struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
1927 
1928 /**
1929  * ib_alloc_fast_reg_page_list - Allocates a page list array
1930  * @device - ib device pointer.
1931  * @page_list_len - size of the page list array to be allocated.
1932  *
1933  * This allocates and returns a struct ib_fast_reg_page_list * and a
1934  * page_list array that is at least page_list_len in size.  The actual
1935  * size is returned in max_page_list_len.  The caller is responsible
1936  * for initializing the contents of the page_list array before posting
1937  * a send work request with the IB_WC_FAST_REG_MR opcode.
1938  *
1939  * The page_list array entries must be translated using one of the
1940  * ib_dma_*() functions just like the addresses passed to
1941  * ib_map_phys_fmr().  Once the ib_post_send() is issued, the struct
1942  * ib_fast_reg_page_list must not be modified by the caller until the
1943  * IB_WC_FAST_REG_MR work request completes.
1944  */
1945 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
1946 				struct ib_device *device, int page_list_len);
1947 
1948 /**
1949  * ib_free_fast_reg_page_list - Deallocates a previously allocated
1950  *   page list array.
1951  * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
1952  */
1953 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
1954 
1955 /**
1956  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
1957  *   R_Key and L_Key.
1958  * @mr - struct ib_mr pointer to be updated.
1959  * @newkey - new key to be used.
1960  */
1961 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
1962 {
1963 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
1964 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
1965 }
1966 
1967 /**
1968  * ib_alloc_mw - Allocates a memory window.
1969  * @pd: The protection domain associated with the memory window.
1970  */
1971 struct ib_mw *ib_alloc_mw(struct ib_pd *pd);
1972 
1973 /**
1974  * ib_bind_mw - Posts a work request to the send queue of the specified
1975  *   QP, which binds the memory window to the given address range and
1976  *   remote access attributes.
1977  * @qp: QP to post the bind work request on.
1978  * @mw: The memory window to bind.
1979  * @mw_bind: Specifies information about the memory window, including
1980  *   its address range, remote access rights, and associated memory region.
1981  */
1982 static inline int ib_bind_mw(struct ib_qp *qp,
1983 			     struct ib_mw *mw,
1984 			     struct ib_mw_bind *mw_bind)
1985 {
1986 	/* XXX reference counting in corresponding MR? */
1987 	return mw->device->bind_mw ?
1988 		mw->device->bind_mw(qp, mw, mw_bind) :
1989 		-ENOSYS;
1990 }
1991 
1992 /**
1993  * ib_dealloc_mw - Deallocates a memory window.
1994  * @mw: The memory window to deallocate.
1995  */
1996 int ib_dealloc_mw(struct ib_mw *mw);
1997 
1998 /**
1999  * ib_alloc_fmr - Allocates a unmapped fast memory region.
2000  * @pd: The protection domain associated with the unmapped region.
2001  * @mr_access_flags: Specifies the memory access rights.
2002  * @fmr_attr: Attributes of the unmapped region.
2003  *
2004  * A fast memory region must be mapped before it can be used as part of
2005  * a work request.
2006  */
2007 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2008 			    int mr_access_flags,
2009 			    struct ib_fmr_attr *fmr_attr);
2010 
2011 /**
2012  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2013  * @fmr: The fast memory region to associate with the pages.
2014  * @page_list: An array of physical pages to map to the fast memory region.
2015  * @list_len: The number of pages in page_list.
2016  * @iova: The I/O virtual address to use with the mapped region.
2017  */
2018 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2019 				  u64 *page_list, int list_len,
2020 				  u64 iova)
2021 {
2022 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2023 }
2024 
2025 /**
2026  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2027  * @fmr_list: A linked list of fast memory regions to unmap.
2028  */
2029 int ib_unmap_fmr(struct list_head *fmr_list);
2030 
2031 /**
2032  * ib_dealloc_fmr - Deallocates a fast memory region.
2033  * @fmr: The fast memory region to deallocate.
2034  */
2035 int ib_dealloc_fmr(struct ib_fmr *fmr);
2036 
2037 /**
2038  * ib_attach_mcast - Attaches the specified QP to a multicast group.
2039  * @qp: QP to attach to the multicast group.  The QP must be type
2040  *   IB_QPT_UD.
2041  * @gid: Multicast group GID.
2042  * @lid: Multicast group LID in host byte order.
2043  *
2044  * In order to send and receive multicast packets, subnet
2045  * administration must have created the multicast group and configured
2046  * the fabric appropriately.  The port associated with the specified
2047  * QP must also be a member of the multicast group.
2048  */
2049 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2050 
2051 /**
2052  * ib_detach_mcast - Detaches the specified QP from a multicast group.
2053  * @qp: QP to detach from the multicast group.
2054  * @gid: Multicast group GID.
2055  * @lid: Multicast group LID in host byte order.
2056  */
2057 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2058 
2059 #endif /* IB_VERBS_H */
2060