xref: /linux/include/rdma/ib_verbs.h (revision 0d08df6c493898e679d9c517e77ea95c063d40ec)
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 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
54 #include <net/ipv6.h>
55 #include <net/ip.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 
59 #include <linux/if_link.h>
60 #include <linux/atomic.h>
61 #include <linux/mmu_notifier.h>
62 #include <asm/uaccess.h>
63 
64 extern struct workqueue_struct *ib_wq;
65 extern struct workqueue_struct *ib_comp_wq;
66 
67 union ib_gid {
68 	u8	raw[16];
69 	struct {
70 		__be64	subnet_prefix;
71 		__be64	interface_id;
72 	} global;
73 };
74 
75 extern union ib_gid zgid;
76 
77 enum ib_gid_type {
78 	/* If link layer is Ethernet, this is RoCE V1 */
79 	IB_GID_TYPE_IB        = 0,
80 	IB_GID_TYPE_ROCE      = 0,
81 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
82 	IB_GID_TYPE_SIZE
83 };
84 
85 #define ROCE_V2_UDP_DPORT      4791
86 struct ib_gid_attr {
87 	enum ib_gid_type	gid_type;
88 	struct net_device	*ndev;
89 };
90 
91 enum rdma_node_type {
92 	/* IB values map to NodeInfo:NodeType. */
93 	RDMA_NODE_IB_CA 	= 1,
94 	RDMA_NODE_IB_SWITCH,
95 	RDMA_NODE_IB_ROUTER,
96 	RDMA_NODE_RNIC,
97 	RDMA_NODE_USNIC,
98 	RDMA_NODE_USNIC_UDP,
99 };
100 
101 enum {
102 	/* set the local administered indication */
103 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
104 };
105 
106 enum rdma_transport_type {
107 	RDMA_TRANSPORT_IB,
108 	RDMA_TRANSPORT_IWARP,
109 	RDMA_TRANSPORT_USNIC,
110 	RDMA_TRANSPORT_USNIC_UDP
111 };
112 
113 enum rdma_protocol_type {
114 	RDMA_PROTOCOL_IB,
115 	RDMA_PROTOCOL_IBOE,
116 	RDMA_PROTOCOL_IWARP,
117 	RDMA_PROTOCOL_USNIC_UDP
118 };
119 
120 __attribute_const__ enum rdma_transport_type
121 rdma_node_get_transport(enum rdma_node_type node_type);
122 
123 enum rdma_network_type {
124 	RDMA_NETWORK_IB,
125 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
126 	RDMA_NETWORK_IPV4,
127 	RDMA_NETWORK_IPV6
128 };
129 
130 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
131 {
132 	if (network_type == RDMA_NETWORK_IPV4 ||
133 	    network_type == RDMA_NETWORK_IPV6)
134 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
135 
136 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
137 	return IB_GID_TYPE_IB;
138 }
139 
140 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
141 							    union ib_gid *gid)
142 {
143 	if (gid_type == IB_GID_TYPE_IB)
144 		return RDMA_NETWORK_IB;
145 
146 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
147 		return RDMA_NETWORK_IPV4;
148 	else
149 		return RDMA_NETWORK_IPV6;
150 }
151 
152 enum rdma_link_layer {
153 	IB_LINK_LAYER_UNSPECIFIED,
154 	IB_LINK_LAYER_INFINIBAND,
155 	IB_LINK_LAYER_ETHERNET,
156 };
157 
158 enum ib_device_cap_flags {
159 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
160 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
161 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
162 	IB_DEVICE_RAW_MULTI			= (1 << 3),
163 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
164 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
165 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
166 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
167 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
168 	IB_DEVICE_INIT_TYPE			= (1 << 9),
169 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
170 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
171 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
172 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
173 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
174 
175 	/*
176 	 * This device supports a per-device lkey or stag that can be
177 	 * used without performing a memory registration for the local
178 	 * memory.  Note that ULPs should never check this flag, but
179 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
180 	 * which will always contain a usable lkey.
181 	 */
182 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
183 	IB_DEVICE_RESERVED /* old SEND_W_INV */	= (1 << 16),
184 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
185 	/*
186 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
187 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
188 	 * messages and can verify the validity of checksum for
189 	 * incoming messages.  Setting this flag implies that the
190 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
191 	 */
192 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
193 	IB_DEVICE_UD_TSO			= (1 << 19),
194 	IB_DEVICE_XRC				= (1 << 20),
195 
196 	/*
197 	 * This device supports the IB "base memory management extension",
198 	 * which includes support for fast registrations (IB_WR_REG_MR,
199 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
200 	 * also be set by any iWarp device which must support FRs to comply
201 	 * to the iWarp verbs spec.  iWarp devices also support the
202 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
203 	 * stag.
204 	 */
205 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
206 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
207 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
208 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
209 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
210 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
211 	/*
212 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
213 	 * support execution of WQEs that involve synchronization
214 	 * of I/O operations with single completion queue managed
215 	 * by hardware.
216 	 */
217 	IB_DEVICE_CROSS_CHANNEL		= (1 << 27),
218 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
219 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
220 	IB_DEVICE_ON_DEMAND_PAGING		= (1 << 31),
221 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
222 	IB_DEVICE_VIRTUAL_FUNCTION		= ((u64)1 << 33),
223 	IB_DEVICE_RAW_SCATTER_FCS		= ((u64)1 << 34),
224 };
225 
226 enum ib_signature_prot_cap {
227 	IB_PROT_T10DIF_TYPE_1 = 1,
228 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
229 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
230 };
231 
232 enum ib_signature_guard_cap {
233 	IB_GUARD_T10DIF_CRC	= 1,
234 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
235 };
236 
237 enum ib_atomic_cap {
238 	IB_ATOMIC_NONE,
239 	IB_ATOMIC_HCA,
240 	IB_ATOMIC_GLOB
241 };
242 
243 enum ib_odp_general_cap_bits {
244 	IB_ODP_SUPPORT = 1 << 0,
245 };
246 
247 enum ib_odp_transport_cap_bits {
248 	IB_ODP_SUPPORT_SEND	= 1 << 0,
249 	IB_ODP_SUPPORT_RECV	= 1 << 1,
250 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
251 	IB_ODP_SUPPORT_READ	= 1 << 3,
252 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
253 };
254 
255 struct ib_odp_caps {
256 	uint64_t general_caps;
257 	struct {
258 		uint32_t  rc_odp_caps;
259 		uint32_t  uc_odp_caps;
260 		uint32_t  ud_odp_caps;
261 	} per_transport_caps;
262 };
263 
264 enum ib_cq_creation_flags {
265 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
266 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
267 };
268 
269 struct ib_cq_init_attr {
270 	unsigned int	cqe;
271 	int		comp_vector;
272 	u32		flags;
273 };
274 
275 struct ib_device_attr {
276 	u64			fw_ver;
277 	__be64			sys_image_guid;
278 	u64			max_mr_size;
279 	u64			page_size_cap;
280 	u32			vendor_id;
281 	u32			vendor_part_id;
282 	u32			hw_ver;
283 	int			max_qp;
284 	int			max_qp_wr;
285 	u64			device_cap_flags;
286 	int			max_sge;
287 	int			max_sge_rd;
288 	int			max_cq;
289 	int			max_cqe;
290 	int			max_mr;
291 	int			max_pd;
292 	int			max_qp_rd_atom;
293 	int			max_ee_rd_atom;
294 	int			max_res_rd_atom;
295 	int			max_qp_init_rd_atom;
296 	int			max_ee_init_rd_atom;
297 	enum ib_atomic_cap	atomic_cap;
298 	enum ib_atomic_cap	masked_atomic_cap;
299 	int			max_ee;
300 	int			max_rdd;
301 	int			max_mw;
302 	int			max_raw_ipv6_qp;
303 	int			max_raw_ethy_qp;
304 	int			max_mcast_grp;
305 	int			max_mcast_qp_attach;
306 	int			max_total_mcast_qp_attach;
307 	int			max_ah;
308 	int			max_fmr;
309 	int			max_map_per_fmr;
310 	int			max_srq;
311 	int			max_srq_wr;
312 	int			max_srq_sge;
313 	unsigned int		max_fast_reg_page_list_len;
314 	u16			max_pkeys;
315 	u8			local_ca_ack_delay;
316 	int			sig_prot_cap;
317 	int			sig_guard_cap;
318 	struct ib_odp_caps	odp_caps;
319 	uint64_t		timestamp_mask;
320 	uint64_t		hca_core_clock; /* in KHZ */
321 };
322 
323 enum ib_mtu {
324 	IB_MTU_256  = 1,
325 	IB_MTU_512  = 2,
326 	IB_MTU_1024 = 3,
327 	IB_MTU_2048 = 4,
328 	IB_MTU_4096 = 5
329 };
330 
331 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
332 {
333 	switch (mtu) {
334 	case IB_MTU_256:  return  256;
335 	case IB_MTU_512:  return  512;
336 	case IB_MTU_1024: return 1024;
337 	case IB_MTU_2048: return 2048;
338 	case IB_MTU_4096: return 4096;
339 	default: 	  return -1;
340 	}
341 }
342 
343 enum ib_port_state {
344 	IB_PORT_NOP		= 0,
345 	IB_PORT_DOWN		= 1,
346 	IB_PORT_INIT		= 2,
347 	IB_PORT_ARMED		= 3,
348 	IB_PORT_ACTIVE		= 4,
349 	IB_PORT_ACTIVE_DEFER	= 5
350 };
351 
352 enum ib_port_cap_flags {
353 	IB_PORT_SM				= 1 <<  1,
354 	IB_PORT_NOTICE_SUP			= 1 <<  2,
355 	IB_PORT_TRAP_SUP			= 1 <<  3,
356 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
357 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
358 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
359 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
360 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
361 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
362 	IB_PORT_SM_DISABLED			= 1 << 10,
363 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
364 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
365 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
366 	IB_PORT_CM_SUP				= 1 << 16,
367 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
368 	IB_PORT_REINIT_SUP			= 1 << 18,
369 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
370 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
371 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
372 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
373 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
374 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
375 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
376 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
377 };
378 
379 enum ib_port_width {
380 	IB_WIDTH_1X	= 1,
381 	IB_WIDTH_4X	= 2,
382 	IB_WIDTH_8X	= 4,
383 	IB_WIDTH_12X	= 8
384 };
385 
386 static inline int ib_width_enum_to_int(enum ib_port_width width)
387 {
388 	switch (width) {
389 	case IB_WIDTH_1X:  return  1;
390 	case IB_WIDTH_4X:  return  4;
391 	case IB_WIDTH_8X:  return  8;
392 	case IB_WIDTH_12X: return 12;
393 	default: 	  return -1;
394 	}
395 }
396 
397 enum ib_port_speed {
398 	IB_SPEED_SDR	= 1,
399 	IB_SPEED_DDR	= 2,
400 	IB_SPEED_QDR	= 4,
401 	IB_SPEED_FDR10	= 8,
402 	IB_SPEED_FDR	= 16,
403 	IB_SPEED_EDR	= 32
404 };
405 
406 struct ib_protocol_stats {
407 	/* TBD... */
408 };
409 
410 struct iw_protocol_stats {
411 	u64	ipInReceives;
412 	u64	ipInHdrErrors;
413 	u64	ipInTooBigErrors;
414 	u64	ipInNoRoutes;
415 	u64	ipInAddrErrors;
416 	u64	ipInUnknownProtos;
417 	u64	ipInTruncatedPkts;
418 	u64	ipInDiscards;
419 	u64	ipInDelivers;
420 	u64	ipOutForwDatagrams;
421 	u64	ipOutRequests;
422 	u64	ipOutDiscards;
423 	u64	ipOutNoRoutes;
424 	u64	ipReasmTimeout;
425 	u64	ipReasmReqds;
426 	u64	ipReasmOKs;
427 	u64	ipReasmFails;
428 	u64	ipFragOKs;
429 	u64	ipFragFails;
430 	u64	ipFragCreates;
431 	u64	ipInMcastPkts;
432 	u64	ipOutMcastPkts;
433 	u64	ipInBcastPkts;
434 	u64	ipOutBcastPkts;
435 
436 	u64	tcpRtoAlgorithm;
437 	u64	tcpRtoMin;
438 	u64	tcpRtoMax;
439 	u64	tcpMaxConn;
440 	u64	tcpActiveOpens;
441 	u64	tcpPassiveOpens;
442 	u64	tcpAttemptFails;
443 	u64	tcpEstabResets;
444 	u64	tcpCurrEstab;
445 	u64	tcpInSegs;
446 	u64	tcpOutSegs;
447 	u64	tcpRetransSegs;
448 	u64	tcpInErrs;
449 	u64	tcpOutRsts;
450 };
451 
452 union rdma_protocol_stats {
453 	struct ib_protocol_stats	ib;
454 	struct iw_protocol_stats	iw;
455 };
456 
457 /* Define bits for the various functionality this port needs to be supported by
458  * the core.
459  */
460 /* Management                           0x00000FFF */
461 #define RDMA_CORE_CAP_IB_MAD            0x00000001
462 #define RDMA_CORE_CAP_IB_SMI            0x00000002
463 #define RDMA_CORE_CAP_IB_CM             0x00000004
464 #define RDMA_CORE_CAP_IW_CM             0x00000008
465 #define RDMA_CORE_CAP_IB_SA             0x00000010
466 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
467 
468 /* Address format                       0x000FF000 */
469 #define RDMA_CORE_CAP_AF_IB             0x00001000
470 #define RDMA_CORE_CAP_ETH_AH            0x00002000
471 
472 /* Protocol                             0xFFF00000 */
473 #define RDMA_CORE_CAP_PROT_IB           0x00100000
474 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
475 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
476 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
477 
478 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
479 					| RDMA_CORE_CAP_IB_MAD \
480 					| RDMA_CORE_CAP_IB_SMI \
481 					| RDMA_CORE_CAP_IB_CM  \
482 					| RDMA_CORE_CAP_IB_SA  \
483 					| RDMA_CORE_CAP_AF_IB)
484 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
485 					| RDMA_CORE_CAP_IB_MAD  \
486 					| RDMA_CORE_CAP_IB_CM   \
487 					| RDMA_CORE_CAP_AF_IB   \
488 					| RDMA_CORE_CAP_ETH_AH)
489 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
490 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
491 					| RDMA_CORE_CAP_IB_MAD  \
492 					| RDMA_CORE_CAP_IB_CM   \
493 					| RDMA_CORE_CAP_AF_IB   \
494 					| RDMA_CORE_CAP_ETH_AH)
495 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
496 					| RDMA_CORE_CAP_IW_CM)
497 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
498 					| RDMA_CORE_CAP_OPA_MAD)
499 
500 struct ib_port_attr {
501 	u64			subnet_prefix;
502 	enum ib_port_state	state;
503 	enum ib_mtu		max_mtu;
504 	enum ib_mtu		active_mtu;
505 	int			gid_tbl_len;
506 	u32			port_cap_flags;
507 	u32			max_msg_sz;
508 	u32			bad_pkey_cntr;
509 	u32			qkey_viol_cntr;
510 	u16			pkey_tbl_len;
511 	u16			lid;
512 	u16			sm_lid;
513 	u8			lmc;
514 	u8			max_vl_num;
515 	u8			sm_sl;
516 	u8			subnet_timeout;
517 	u8			init_type_reply;
518 	u8			active_width;
519 	u8			active_speed;
520 	u8                      phys_state;
521 	bool			grh_required;
522 };
523 
524 enum ib_device_modify_flags {
525 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
526 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
527 };
528 
529 struct ib_device_modify {
530 	u64	sys_image_guid;
531 	char	node_desc[64];
532 };
533 
534 enum ib_port_modify_flags {
535 	IB_PORT_SHUTDOWN		= 1,
536 	IB_PORT_INIT_TYPE		= (1<<2),
537 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
538 };
539 
540 struct ib_port_modify {
541 	u32	set_port_cap_mask;
542 	u32	clr_port_cap_mask;
543 	u8	init_type;
544 };
545 
546 enum ib_event_type {
547 	IB_EVENT_CQ_ERR,
548 	IB_EVENT_QP_FATAL,
549 	IB_EVENT_QP_REQ_ERR,
550 	IB_EVENT_QP_ACCESS_ERR,
551 	IB_EVENT_COMM_EST,
552 	IB_EVENT_SQ_DRAINED,
553 	IB_EVENT_PATH_MIG,
554 	IB_EVENT_PATH_MIG_ERR,
555 	IB_EVENT_DEVICE_FATAL,
556 	IB_EVENT_PORT_ACTIVE,
557 	IB_EVENT_PORT_ERR,
558 	IB_EVENT_LID_CHANGE,
559 	IB_EVENT_PKEY_CHANGE,
560 	IB_EVENT_SM_CHANGE,
561 	IB_EVENT_SRQ_ERR,
562 	IB_EVENT_SRQ_LIMIT_REACHED,
563 	IB_EVENT_QP_LAST_WQE_REACHED,
564 	IB_EVENT_CLIENT_REREGISTER,
565 	IB_EVENT_GID_CHANGE,
566 };
567 
568 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
569 
570 struct ib_event {
571 	struct ib_device	*device;
572 	union {
573 		struct ib_cq	*cq;
574 		struct ib_qp	*qp;
575 		struct ib_srq	*srq;
576 		u8		port_num;
577 	} element;
578 	enum ib_event_type	event;
579 };
580 
581 struct ib_event_handler {
582 	struct ib_device *device;
583 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
584 	struct list_head  list;
585 };
586 
587 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
588 	do {							\
589 		(_ptr)->device  = _device;			\
590 		(_ptr)->handler = _handler;			\
591 		INIT_LIST_HEAD(&(_ptr)->list);			\
592 	} while (0)
593 
594 struct ib_global_route {
595 	union ib_gid	dgid;
596 	u32		flow_label;
597 	u8		sgid_index;
598 	u8		hop_limit;
599 	u8		traffic_class;
600 };
601 
602 struct ib_grh {
603 	__be32		version_tclass_flow;
604 	__be16		paylen;
605 	u8		next_hdr;
606 	u8		hop_limit;
607 	union ib_gid	sgid;
608 	union ib_gid	dgid;
609 };
610 
611 union rdma_network_hdr {
612 	struct ib_grh ibgrh;
613 	struct {
614 		/* The IB spec states that if it's IPv4, the header
615 		 * is located in the last 20 bytes of the header.
616 		 */
617 		u8		reserved[20];
618 		struct iphdr	roce4grh;
619 	};
620 };
621 
622 enum {
623 	IB_MULTICAST_QPN = 0xffffff
624 };
625 
626 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
627 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
628 
629 enum ib_ah_flags {
630 	IB_AH_GRH	= 1
631 };
632 
633 enum ib_rate {
634 	IB_RATE_PORT_CURRENT = 0,
635 	IB_RATE_2_5_GBPS = 2,
636 	IB_RATE_5_GBPS   = 5,
637 	IB_RATE_10_GBPS  = 3,
638 	IB_RATE_20_GBPS  = 6,
639 	IB_RATE_30_GBPS  = 4,
640 	IB_RATE_40_GBPS  = 7,
641 	IB_RATE_60_GBPS  = 8,
642 	IB_RATE_80_GBPS  = 9,
643 	IB_RATE_120_GBPS = 10,
644 	IB_RATE_14_GBPS  = 11,
645 	IB_RATE_56_GBPS  = 12,
646 	IB_RATE_112_GBPS = 13,
647 	IB_RATE_168_GBPS = 14,
648 	IB_RATE_25_GBPS  = 15,
649 	IB_RATE_100_GBPS = 16,
650 	IB_RATE_200_GBPS = 17,
651 	IB_RATE_300_GBPS = 18
652 };
653 
654 /**
655  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
656  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
657  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
658  * @rate: rate to convert.
659  */
660 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
661 
662 /**
663  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
664  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
665  * @rate: rate to convert.
666  */
667 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
668 
669 
670 /**
671  * enum ib_mr_type - memory region type
672  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
673  *                            normal registration
674  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
675  *                            signature operations (data-integrity
676  *                            capable regions)
677  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
678  *                            register any arbitrary sg lists (without
679  *                            the normal mr constraints - see
680  *                            ib_map_mr_sg)
681  */
682 enum ib_mr_type {
683 	IB_MR_TYPE_MEM_REG,
684 	IB_MR_TYPE_SIGNATURE,
685 	IB_MR_TYPE_SG_GAPS,
686 };
687 
688 /**
689  * Signature types
690  * IB_SIG_TYPE_NONE: Unprotected.
691  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
692  */
693 enum ib_signature_type {
694 	IB_SIG_TYPE_NONE,
695 	IB_SIG_TYPE_T10_DIF,
696 };
697 
698 /**
699  * Signature T10-DIF block-guard types
700  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
701  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
702  */
703 enum ib_t10_dif_bg_type {
704 	IB_T10DIF_CRC,
705 	IB_T10DIF_CSUM
706 };
707 
708 /**
709  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
710  *     domain.
711  * @bg_type: T10-DIF block guard type (CRC|CSUM)
712  * @pi_interval: protection information interval.
713  * @bg: seed of guard computation.
714  * @app_tag: application tag of guard block
715  * @ref_tag: initial guard block reference tag.
716  * @ref_remap: Indicate wethear the reftag increments each block
717  * @app_escape: Indicate to skip block check if apptag=0xffff
718  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
719  * @apptag_check_mask: check bitmask of application tag.
720  */
721 struct ib_t10_dif_domain {
722 	enum ib_t10_dif_bg_type bg_type;
723 	u16			pi_interval;
724 	u16			bg;
725 	u16			app_tag;
726 	u32			ref_tag;
727 	bool			ref_remap;
728 	bool			app_escape;
729 	bool			ref_escape;
730 	u16			apptag_check_mask;
731 };
732 
733 /**
734  * struct ib_sig_domain - Parameters for signature domain
735  * @sig_type: specific signauture type
736  * @sig: union of all signature domain attributes that may
737  *     be used to set domain layout.
738  */
739 struct ib_sig_domain {
740 	enum ib_signature_type sig_type;
741 	union {
742 		struct ib_t10_dif_domain dif;
743 	} sig;
744 };
745 
746 /**
747  * struct ib_sig_attrs - Parameters for signature handover operation
748  * @check_mask: bitmask for signature byte check (8 bytes)
749  * @mem: memory domain layout desciptor.
750  * @wire: wire domain layout desciptor.
751  */
752 struct ib_sig_attrs {
753 	u8			check_mask;
754 	struct ib_sig_domain	mem;
755 	struct ib_sig_domain	wire;
756 };
757 
758 enum ib_sig_err_type {
759 	IB_SIG_BAD_GUARD,
760 	IB_SIG_BAD_REFTAG,
761 	IB_SIG_BAD_APPTAG,
762 };
763 
764 /**
765  * struct ib_sig_err - signature error descriptor
766  */
767 struct ib_sig_err {
768 	enum ib_sig_err_type	err_type;
769 	u32			expected;
770 	u32			actual;
771 	u64			sig_err_offset;
772 	u32			key;
773 };
774 
775 enum ib_mr_status_check {
776 	IB_MR_CHECK_SIG_STATUS = 1,
777 };
778 
779 /**
780  * struct ib_mr_status - Memory region status container
781  *
782  * @fail_status: Bitmask of MR checks status. For each
783  *     failed check a corresponding status bit is set.
784  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
785  *     failure.
786  */
787 struct ib_mr_status {
788 	u32		    fail_status;
789 	struct ib_sig_err   sig_err;
790 };
791 
792 /**
793  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
794  * enum.
795  * @mult: multiple to convert.
796  */
797 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
798 
799 struct ib_ah_attr {
800 	struct ib_global_route	grh;
801 	u16			dlid;
802 	u8			sl;
803 	u8			src_path_bits;
804 	u8			static_rate;
805 	u8			ah_flags;
806 	u8			port_num;
807 	u8			dmac[ETH_ALEN];
808 };
809 
810 enum ib_wc_status {
811 	IB_WC_SUCCESS,
812 	IB_WC_LOC_LEN_ERR,
813 	IB_WC_LOC_QP_OP_ERR,
814 	IB_WC_LOC_EEC_OP_ERR,
815 	IB_WC_LOC_PROT_ERR,
816 	IB_WC_WR_FLUSH_ERR,
817 	IB_WC_MW_BIND_ERR,
818 	IB_WC_BAD_RESP_ERR,
819 	IB_WC_LOC_ACCESS_ERR,
820 	IB_WC_REM_INV_REQ_ERR,
821 	IB_WC_REM_ACCESS_ERR,
822 	IB_WC_REM_OP_ERR,
823 	IB_WC_RETRY_EXC_ERR,
824 	IB_WC_RNR_RETRY_EXC_ERR,
825 	IB_WC_LOC_RDD_VIOL_ERR,
826 	IB_WC_REM_INV_RD_REQ_ERR,
827 	IB_WC_REM_ABORT_ERR,
828 	IB_WC_INV_EECN_ERR,
829 	IB_WC_INV_EEC_STATE_ERR,
830 	IB_WC_FATAL_ERR,
831 	IB_WC_RESP_TIMEOUT_ERR,
832 	IB_WC_GENERAL_ERR
833 };
834 
835 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
836 
837 enum ib_wc_opcode {
838 	IB_WC_SEND,
839 	IB_WC_RDMA_WRITE,
840 	IB_WC_RDMA_READ,
841 	IB_WC_COMP_SWAP,
842 	IB_WC_FETCH_ADD,
843 	IB_WC_LSO,
844 	IB_WC_LOCAL_INV,
845 	IB_WC_REG_MR,
846 	IB_WC_MASKED_COMP_SWAP,
847 	IB_WC_MASKED_FETCH_ADD,
848 /*
849  * Set value of IB_WC_RECV so consumers can test if a completion is a
850  * receive by testing (opcode & IB_WC_RECV).
851  */
852 	IB_WC_RECV			= 1 << 7,
853 	IB_WC_RECV_RDMA_WITH_IMM
854 };
855 
856 enum ib_wc_flags {
857 	IB_WC_GRH		= 1,
858 	IB_WC_WITH_IMM		= (1<<1),
859 	IB_WC_WITH_INVALIDATE	= (1<<2),
860 	IB_WC_IP_CSUM_OK	= (1<<3),
861 	IB_WC_WITH_SMAC		= (1<<4),
862 	IB_WC_WITH_VLAN		= (1<<5),
863 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
864 };
865 
866 struct ib_wc {
867 	union {
868 		u64		wr_id;
869 		struct ib_cqe	*wr_cqe;
870 	};
871 	enum ib_wc_status	status;
872 	enum ib_wc_opcode	opcode;
873 	u32			vendor_err;
874 	u32			byte_len;
875 	struct ib_qp	       *qp;
876 	union {
877 		__be32		imm_data;
878 		u32		invalidate_rkey;
879 	} ex;
880 	u32			src_qp;
881 	int			wc_flags;
882 	u16			pkey_index;
883 	u16			slid;
884 	u8			sl;
885 	u8			dlid_path_bits;
886 	u8			port_num;	/* valid only for DR SMPs on switches */
887 	u8			smac[ETH_ALEN];
888 	u16			vlan_id;
889 	u8			network_hdr_type;
890 };
891 
892 enum ib_cq_notify_flags {
893 	IB_CQ_SOLICITED			= 1 << 0,
894 	IB_CQ_NEXT_COMP			= 1 << 1,
895 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
896 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
897 };
898 
899 enum ib_srq_type {
900 	IB_SRQT_BASIC,
901 	IB_SRQT_XRC
902 };
903 
904 enum ib_srq_attr_mask {
905 	IB_SRQ_MAX_WR	= 1 << 0,
906 	IB_SRQ_LIMIT	= 1 << 1,
907 };
908 
909 struct ib_srq_attr {
910 	u32	max_wr;
911 	u32	max_sge;
912 	u32	srq_limit;
913 };
914 
915 struct ib_srq_init_attr {
916 	void		      (*event_handler)(struct ib_event *, void *);
917 	void		       *srq_context;
918 	struct ib_srq_attr	attr;
919 	enum ib_srq_type	srq_type;
920 
921 	union {
922 		struct {
923 			struct ib_xrcd *xrcd;
924 			struct ib_cq   *cq;
925 		} xrc;
926 	} ext;
927 };
928 
929 struct ib_qp_cap {
930 	u32	max_send_wr;
931 	u32	max_recv_wr;
932 	u32	max_send_sge;
933 	u32	max_recv_sge;
934 	u32	max_inline_data;
935 
936 	/*
937 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
938 	 * ib_create_qp() will calculate the right amount of neededed WRs
939 	 * and MRs based on this.
940 	 */
941 	u32	max_rdma_ctxs;
942 };
943 
944 enum ib_sig_type {
945 	IB_SIGNAL_ALL_WR,
946 	IB_SIGNAL_REQ_WR
947 };
948 
949 enum ib_qp_type {
950 	/*
951 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
952 	 * here (and in that order) since the MAD layer uses them as
953 	 * indices into a 2-entry table.
954 	 */
955 	IB_QPT_SMI,
956 	IB_QPT_GSI,
957 
958 	IB_QPT_RC,
959 	IB_QPT_UC,
960 	IB_QPT_UD,
961 	IB_QPT_RAW_IPV6,
962 	IB_QPT_RAW_ETHERTYPE,
963 	IB_QPT_RAW_PACKET = 8,
964 	IB_QPT_XRC_INI = 9,
965 	IB_QPT_XRC_TGT,
966 	IB_QPT_MAX,
967 	/* Reserve a range for qp types internal to the low level driver.
968 	 * These qp types will not be visible at the IB core layer, so the
969 	 * IB_QPT_MAX usages should not be affected in the core layer
970 	 */
971 	IB_QPT_RESERVED1 = 0x1000,
972 	IB_QPT_RESERVED2,
973 	IB_QPT_RESERVED3,
974 	IB_QPT_RESERVED4,
975 	IB_QPT_RESERVED5,
976 	IB_QPT_RESERVED6,
977 	IB_QPT_RESERVED7,
978 	IB_QPT_RESERVED8,
979 	IB_QPT_RESERVED9,
980 	IB_QPT_RESERVED10,
981 };
982 
983 enum ib_qp_create_flags {
984 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
985 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
986 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
987 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
988 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
989 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
990 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
991 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
992 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
993 	/* reserve bits 26-31 for low level drivers' internal use */
994 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
995 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
996 };
997 
998 /*
999  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1000  * callback to destroy the passed in QP.
1001  */
1002 
1003 struct ib_qp_init_attr {
1004 	void                  (*event_handler)(struct ib_event *, void *);
1005 	void		       *qp_context;
1006 	struct ib_cq	       *send_cq;
1007 	struct ib_cq	       *recv_cq;
1008 	struct ib_srq	       *srq;
1009 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1010 	struct ib_qp_cap	cap;
1011 	enum ib_sig_type	sq_sig_type;
1012 	enum ib_qp_type		qp_type;
1013 	enum ib_qp_create_flags	create_flags;
1014 
1015 	/*
1016 	 * Only needed for special QP types, or when using the RW API.
1017 	 */
1018 	u8			port_num;
1019 };
1020 
1021 struct ib_qp_open_attr {
1022 	void                  (*event_handler)(struct ib_event *, void *);
1023 	void		       *qp_context;
1024 	u32			qp_num;
1025 	enum ib_qp_type		qp_type;
1026 };
1027 
1028 enum ib_rnr_timeout {
1029 	IB_RNR_TIMER_655_36 =  0,
1030 	IB_RNR_TIMER_000_01 =  1,
1031 	IB_RNR_TIMER_000_02 =  2,
1032 	IB_RNR_TIMER_000_03 =  3,
1033 	IB_RNR_TIMER_000_04 =  4,
1034 	IB_RNR_TIMER_000_06 =  5,
1035 	IB_RNR_TIMER_000_08 =  6,
1036 	IB_RNR_TIMER_000_12 =  7,
1037 	IB_RNR_TIMER_000_16 =  8,
1038 	IB_RNR_TIMER_000_24 =  9,
1039 	IB_RNR_TIMER_000_32 = 10,
1040 	IB_RNR_TIMER_000_48 = 11,
1041 	IB_RNR_TIMER_000_64 = 12,
1042 	IB_RNR_TIMER_000_96 = 13,
1043 	IB_RNR_TIMER_001_28 = 14,
1044 	IB_RNR_TIMER_001_92 = 15,
1045 	IB_RNR_TIMER_002_56 = 16,
1046 	IB_RNR_TIMER_003_84 = 17,
1047 	IB_RNR_TIMER_005_12 = 18,
1048 	IB_RNR_TIMER_007_68 = 19,
1049 	IB_RNR_TIMER_010_24 = 20,
1050 	IB_RNR_TIMER_015_36 = 21,
1051 	IB_RNR_TIMER_020_48 = 22,
1052 	IB_RNR_TIMER_030_72 = 23,
1053 	IB_RNR_TIMER_040_96 = 24,
1054 	IB_RNR_TIMER_061_44 = 25,
1055 	IB_RNR_TIMER_081_92 = 26,
1056 	IB_RNR_TIMER_122_88 = 27,
1057 	IB_RNR_TIMER_163_84 = 28,
1058 	IB_RNR_TIMER_245_76 = 29,
1059 	IB_RNR_TIMER_327_68 = 30,
1060 	IB_RNR_TIMER_491_52 = 31
1061 };
1062 
1063 enum ib_qp_attr_mask {
1064 	IB_QP_STATE			= 1,
1065 	IB_QP_CUR_STATE			= (1<<1),
1066 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1067 	IB_QP_ACCESS_FLAGS		= (1<<3),
1068 	IB_QP_PKEY_INDEX		= (1<<4),
1069 	IB_QP_PORT			= (1<<5),
1070 	IB_QP_QKEY			= (1<<6),
1071 	IB_QP_AV			= (1<<7),
1072 	IB_QP_PATH_MTU			= (1<<8),
1073 	IB_QP_TIMEOUT			= (1<<9),
1074 	IB_QP_RETRY_CNT			= (1<<10),
1075 	IB_QP_RNR_RETRY			= (1<<11),
1076 	IB_QP_RQ_PSN			= (1<<12),
1077 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1078 	IB_QP_ALT_PATH			= (1<<14),
1079 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1080 	IB_QP_SQ_PSN			= (1<<16),
1081 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1082 	IB_QP_PATH_MIG_STATE		= (1<<18),
1083 	IB_QP_CAP			= (1<<19),
1084 	IB_QP_DEST_QPN			= (1<<20),
1085 	IB_QP_RESERVED1			= (1<<21),
1086 	IB_QP_RESERVED2			= (1<<22),
1087 	IB_QP_RESERVED3			= (1<<23),
1088 	IB_QP_RESERVED4			= (1<<24),
1089 };
1090 
1091 enum ib_qp_state {
1092 	IB_QPS_RESET,
1093 	IB_QPS_INIT,
1094 	IB_QPS_RTR,
1095 	IB_QPS_RTS,
1096 	IB_QPS_SQD,
1097 	IB_QPS_SQE,
1098 	IB_QPS_ERR
1099 };
1100 
1101 enum ib_mig_state {
1102 	IB_MIG_MIGRATED,
1103 	IB_MIG_REARM,
1104 	IB_MIG_ARMED
1105 };
1106 
1107 enum ib_mw_type {
1108 	IB_MW_TYPE_1 = 1,
1109 	IB_MW_TYPE_2 = 2
1110 };
1111 
1112 struct ib_qp_attr {
1113 	enum ib_qp_state	qp_state;
1114 	enum ib_qp_state	cur_qp_state;
1115 	enum ib_mtu		path_mtu;
1116 	enum ib_mig_state	path_mig_state;
1117 	u32			qkey;
1118 	u32			rq_psn;
1119 	u32			sq_psn;
1120 	u32			dest_qp_num;
1121 	int			qp_access_flags;
1122 	struct ib_qp_cap	cap;
1123 	struct ib_ah_attr	ah_attr;
1124 	struct ib_ah_attr	alt_ah_attr;
1125 	u16			pkey_index;
1126 	u16			alt_pkey_index;
1127 	u8			en_sqd_async_notify;
1128 	u8			sq_draining;
1129 	u8			max_rd_atomic;
1130 	u8			max_dest_rd_atomic;
1131 	u8			min_rnr_timer;
1132 	u8			port_num;
1133 	u8			timeout;
1134 	u8			retry_cnt;
1135 	u8			rnr_retry;
1136 	u8			alt_port_num;
1137 	u8			alt_timeout;
1138 };
1139 
1140 enum ib_wr_opcode {
1141 	IB_WR_RDMA_WRITE,
1142 	IB_WR_RDMA_WRITE_WITH_IMM,
1143 	IB_WR_SEND,
1144 	IB_WR_SEND_WITH_IMM,
1145 	IB_WR_RDMA_READ,
1146 	IB_WR_ATOMIC_CMP_AND_SWP,
1147 	IB_WR_ATOMIC_FETCH_AND_ADD,
1148 	IB_WR_LSO,
1149 	IB_WR_SEND_WITH_INV,
1150 	IB_WR_RDMA_READ_WITH_INV,
1151 	IB_WR_LOCAL_INV,
1152 	IB_WR_REG_MR,
1153 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1154 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1155 	IB_WR_REG_SIG_MR,
1156 	/* reserve values for low level drivers' internal use.
1157 	 * These values will not be used at all in the ib core layer.
1158 	 */
1159 	IB_WR_RESERVED1 = 0xf0,
1160 	IB_WR_RESERVED2,
1161 	IB_WR_RESERVED3,
1162 	IB_WR_RESERVED4,
1163 	IB_WR_RESERVED5,
1164 	IB_WR_RESERVED6,
1165 	IB_WR_RESERVED7,
1166 	IB_WR_RESERVED8,
1167 	IB_WR_RESERVED9,
1168 	IB_WR_RESERVED10,
1169 };
1170 
1171 enum ib_send_flags {
1172 	IB_SEND_FENCE		= 1,
1173 	IB_SEND_SIGNALED	= (1<<1),
1174 	IB_SEND_SOLICITED	= (1<<2),
1175 	IB_SEND_INLINE		= (1<<3),
1176 	IB_SEND_IP_CSUM		= (1<<4),
1177 
1178 	/* reserve bits 26-31 for low level drivers' internal use */
1179 	IB_SEND_RESERVED_START	= (1 << 26),
1180 	IB_SEND_RESERVED_END	= (1 << 31),
1181 };
1182 
1183 struct ib_sge {
1184 	u64	addr;
1185 	u32	length;
1186 	u32	lkey;
1187 };
1188 
1189 struct ib_cqe {
1190 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1191 };
1192 
1193 struct ib_send_wr {
1194 	struct ib_send_wr      *next;
1195 	union {
1196 		u64		wr_id;
1197 		struct ib_cqe	*wr_cqe;
1198 	};
1199 	struct ib_sge	       *sg_list;
1200 	int			num_sge;
1201 	enum ib_wr_opcode	opcode;
1202 	int			send_flags;
1203 	union {
1204 		__be32		imm_data;
1205 		u32		invalidate_rkey;
1206 	} ex;
1207 };
1208 
1209 struct ib_rdma_wr {
1210 	struct ib_send_wr	wr;
1211 	u64			remote_addr;
1212 	u32			rkey;
1213 };
1214 
1215 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1216 {
1217 	return container_of(wr, struct ib_rdma_wr, wr);
1218 }
1219 
1220 struct ib_atomic_wr {
1221 	struct ib_send_wr	wr;
1222 	u64			remote_addr;
1223 	u64			compare_add;
1224 	u64			swap;
1225 	u64			compare_add_mask;
1226 	u64			swap_mask;
1227 	u32			rkey;
1228 };
1229 
1230 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1231 {
1232 	return container_of(wr, struct ib_atomic_wr, wr);
1233 }
1234 
1235 struct ib_ud_wr {
1236 	struct ib_send_wr	wr;
1237 	struct ib_ah		*ah;
1238 	void			*header;
1239 	int			hlen;
1240 	int			mss;
1241 	u32			remote_qpn;
1242 	u32			remote_qkey;
1243 	u16			pkey_index; /* valid for GSI only */
1244 	u8			port_num;   /* valid for DR SMPs on switch only */
1245 };
1246 
1247 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1248 {
1249 	return container_of(wr, struct ib_ud_wr, wr);
1250 }
1251 
1252 struct ib_reg_wr {
1253 	struct ib_send_wr	wr;
1254 	struct ib_mr		*mr;
1255 	u32			key;
1256 	int			access;
1257 };
1258 
1259 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1260 {
1261 	return container_of(wr, struct ib_reg_wr, wr);
1262 }
1263 
1264 struct ib_sig_handover_wr {
1265 	struct ib_send_wr	wr;
1266 	struct ib_sig_attrs    *sig_attrs;
1267 	struct ib_mr	       *sig_mr;
1268 	int			access_flags;
1269 	struct ib_sge	       *prot;
1270 };
1271 
1272 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1273 {
1274 	return container_of(wr, struct ib_sig_handover_wr, wr);
1275 }
1276 
1277 struct ib_recv_wr {
1278 	struct ib_recv_wr      *next;
1279 	union {
1280 		u64		wr_id;
1281 		struct ib_cqe	*wr_cqe;
1282 	};
1283 	struct ib_sge	       *sg_list;
1284 	int			num_sge;
1285 };
1286 
1287 enum ib_access_flags {
1288 	IB_ACCESS_LOCAL_WRITE	= 1,
1289 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1290 	IB_ACCESS_REMOTE_READ	= (1<<2),
1291 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1292 	IB_ACCESS_MW_BIND	= (1<<4),
1293 	IB_ZERO_BASED		= (1<<5),
1294 	IB_ACCESS_ON_DEMAND     = (1<<6),
1295 };
1296 
1297 /*
1298  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1299  * are hidden here instead of a uapi header!
1300  */
1301 enum ib_mr_rereg_flags {
1302 	IB_MR_REREG_TRANS	= 1,
1303 	IB_MR_REREG_PD		= (1<<1),
1304 	IB_MR_REREG_ACCESS	= (1<<2),
1305 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1306 };
1307 
1308 struct ib_fmr_attr {
1309 	int	max_pages;
1310 	int	max_maps;
1311 	u8	page_shift;
1312 };
1313 
1314 struct ib_umem;
1315 
1316 struct ib_ucontext {
1317 	struct ib_device       *device;
1318 	struct list_head	pd_list;
1319 	struct list_head	mr_list;
1320 	struct list_head	mw_list;
1321 	struct list_head	cq_list;
1322 	struct list_head	qp_list;
1323 	struct list_head	srq_list;
1324 	struct list_head	ah_list;
1325 	struct list_head	xrcd_list;
1326 	struct list_head	rule_list;
1327 	int			closing;
1328 
1329 	struct pid             *tgid;
1330 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1331 	struct rb_root      umem_tree;
1332 	/*
1333 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1334 	 * mmu notifiers registration.
1335 	 */
1336 	struct rw_semaphore	umem_rwsem;
1337 	void (*invalidate_range)(struct ib_umem *umem,
1338 				 unsigned long start, unsigned long end);
1339 
1340 	struct mmu_notifier	mn;
1341 	atomic_t		notifier_count;
1342 	/* A list of umems that don't have private mmu notifier counters yet. */
1343 	struct list_head	no_private_counters;
1344 	int                     odp_mrs_count;
1345 #endif
1346 };
1347 
1348 struct ib_uobject {
1349 	u64			user_handle;	/* handle given to us by userspace */
1350 	struct ib_ucontext     *context;	/* associated user context */
1351 	void		       *object;		/* containing object */
1352 	struct list_head	list;		/* link to context's list */
1353 	int			id;		/* index into kernel idr */
1354 	struct kref		ref;
1355 	struct rw_semaphore	mutex;		/* protects .live */
1356 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1357 	int			live;
1358 };
1359 
1360 struct ib_udata {
1361 	const void __user *inbuf;
1362 	void __user *outbuf;
1363 	size_t       inlen;
1364 	size_t       outlen;
1365 };
1366 
1367 struct ib_pd {
1368 	u32			local_dma_lkey;
1369 	struct ib_device       *device;
1370 	struct ib_uobject      *uobject;
1371 	atomic_t          	usecnt; /* count all resources */
1372 	struct ib_mr	       *local_mr;
1373 };
1374 
1375 struct ib_xrcd {
1376 	struct ib_device       *device;
1377 	atomic_t		usecnt; /* count all exposed resources */
1378 	struct inode	       *inode;
1379 
1380 	struct mutex		tgt_qp_mutex;
1381 	struct list_head	tgt_qp_list;
1382 };
1383 
1384 struct ib_ah {
1385 	struct ib_device	*device;
1386 	struct ib_pd		*pd;
1387 	struct ib_uobject	*uobject;
1388 };
1389 
1390 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1391 
1392 enum ib_poll_context {
1393 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1394 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1395 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1396 };
1397 
1398 struct ib_cq {
1399 	struct ib_device       *device;
1400 	struct ib_uobject      *uobject;
1401 	ib_comp_handler   	comp_handler;
1402 	void                  (*event_handler)(struct ib_event *, void *);
1403 	void                   *cq_context;
1404 	int               	cqe;
1405 	atomic_t          	usecnt; /* count number of work queues */
1406 	enum ib_poll_context	poll_ctx;
1407 	struct ib_wc		*wc;
1408 	union {
1409 		struct irq_poll		iop;
1410 		struct work_struct	work;
1411 	};
1412 };
1413 
1414 struct ib_srq {
1415 	struct ib_device       *device;
1416 	struct ib_pd	       *pd;
1417 	struct ib_uobject      *uobject;
1418 	void		      (*event_handler)(struct ib_event *, void *);
1419 	void		       *srq_context;
1420 	enum ib_srq_type	srq_type;
1421 	atomic_t		usecnt;
1422 
1423 	union {
1424 		struct {
1425 			struct ib_xrcd *xrcd;
1426 			struct ib_cq   *cq;
1427 			u32		srq_num;
1428 		} xrc;
1429 	} ext;
1430 };
1431 
1432 struct ib_qp {
1433 	struct ib_device       *device;
1434 	struct ib_pd	       *pd;
1435 	struct ib_cq	       *send_cq;
1436 	struct ib_cq	       *recv_cq;
1437 	spinlock_t		mr_lock;
1438 	int			mrs_used;
1439 	struct list_head	rdma_mrs;
1440 	struct list_head	sig_mrs;
1441 	struct ib_srq	       *srq;
1442 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1443 	struct list_head	xrcd_list;
1444 
1445 	/* count times opened, mcast attaches, flow attaches */
1446 	atomic_t		usecnt;
1447 	struct list_head	open_list;
1448 	struct ib_qp           *real_qp;
1449 	struct ib_uobject      *uobject;
1450 	void                  (*event_handler)(struct ib_event *, void *);
1451 	void		       *qp_context;
1452 	u32			qp_num;
1453 	enum ib_qp_type		qp_type;
1454 };
1455 
1456 struct ib_mr {
1457 	struct ib_device  *device;
1458 	struct ib_pd	  *pd;
1459 	u32		   lkey;
1460 	u32		   rkey;
1461 	u64		   iova;
1462 	u32		   length;
1463 	unsigned int	   page_size;
1464 	bool		   need_inval;
1465 	union {
1466 		struct ib_uobject	*uobject;	/* user */
1467 		struct list_head	qp_entry;	/* FR */
1468 	};
1469 };
1470 
1471 struct ib_mw {
1472 	struct ib_device	*device;
1473 	struct ib_pd		*pd;
1474 	struct ib_uobject	*uobject;
1475 	u32			rkey;
1476 	enum ib_mw_type         type;
1477 };
1478 
1479 struct ib_fmr {
1480 	struct ib_device	*device;
1481 	struct ib_pd		*pd;
1482 	struct list_head	list;
1483 	u32			lkey;
1484 	u32			rkey;
1485 };
1486 
1487 /* Supported steering options */
1488 enum ib_flow_attr_type {
1489 	/* steering according to rule specifications */
1490 	IB_FLOW_ATTR_NORMAL		= 0x0,
1491 	/* default unicast and multicast rule -
1492 	 * receive all Eth traffic which isn't steered to any QP
1493 	 */
1494 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1495 	/* default multicast rule -
1496 	 * receive all Eth multicast traffic which isn't steered to any QP
1497 	 */
1498 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1499 	/* sniffer rule - receive all port traffic */
1500 	IB_FLOW_ATTR_SNIFFER		= 0x3
1501 };
1502 
1503 /* Supported steering header types */
1504 enum ib_flow_spec_type {
1505 	/* L2 headers*/
1506 	IB_FLOW_SPEC_ETH	= 0x20,
1507 	IB_FLOW_SPEC_IB		= 0x22,
1508 	/* L3 header*/
1509 	IB_FLOW_SPEC_IPV4	= 0x30,
1510 	/* L4 headers*/
1511 	IB_FLOW_SPEC_TCP	= 0x40,
1512 	IB_FLOW_SPEC_UDP	= 0x41
1513 };
1514 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1515 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1516 
1517 /* Flow steering rule priority is set according to it's domain.
1518  * Lower domain value means higher priority.
1519  */
1520 enum ib_flow_domain {
1521 	IB_FLOW_DOMAIN_USER,
1522 	IB_FLOW_DOMAIN_ETHTOOL,
1523 	IB_FLOW_DOMAIN_RFS,
1524 	IB_FLOW_DOMAIN_NIC,
1525 	IB_FLOW_DOMAIN_NUM /* Must be last */
1526 };
1527 
1528 enum ib_flow_flags {
1529 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1530 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1531 };
1532 
1533 struct ib_flow_eth_filter {
1534 	u8	dst_mac[6];
1535 	u8	src_mac[6];
1536 	__be16	ether_type;
1537 	__be16	vlan_tag;
1538 };
1539 
1540 struct ib_flow_spec_eth {
1541 	enum ib_flow_spec_type	  type;
1542 	u16			  size;
1543 	struct ib_flow_eth_filter val;
1544 	struct ib_flow_eth_filter mask;
1545 };
1546 
1547 struct ib_flow_ib_filter {
1548 	__be16 dlid;
1549 	__u8   sl;
1550 };
1551 
1552 struct ib_flow_spec_ib {
1553 	enum ib_flow_spec_type	 type;
1554 	u16			 size;
1555 	struct ib_flow_ib_filter val;
1556 	struct ib_flow_ib_filter mask;
1557 };
1558 
1559 struct ib_flow_ipv4_filter {
1560 	__be32	src_ip;
1561 	__be32	dst_ip;
1562 };
1563 
1564 struct ib_flow_spec_ipv4 {
1565 	enum ib_flow_spec_type	   type;
1566 	u16			   size;
1567 	struct ib_flow_ipv4_filter val;
1568 	struct ib_flow_ipv4_filter mask;
1569 };
1570 
1571 struct ib_flow_tcp_udp_filter {
1572 	__be16	dst_port;
1573 	__be16	src_port;
1574 };
1575 
1576 struct ib_flow_spec_tcp_udp {
1577 	enum ib_flow_spec_type	      type;
1578 	u16			      size;
1579 	struct ib_flow_tcp_udp_filter val;
1580 	struct ib_flow_tcp_udp_filter mask;
1581 };
1582 
1583 union ib_flow_spec {
1584 	struct {
1585 		enum ib_flow_spec_type	type;
1586 		u16			size;
1587 	};
1588 	struct ib_flow_spec_eth		eth;
1589 	struct ib_flow_spec_ib		ib;
1590 	struct ib_flow_spec_ipv4        ipv4;
1591 	struct ib_flow_spec_tcp_udp	tcp_udp;
1592 };
1593 
1594 struct ib_flow_attr {
1595 	enum ib_flow_attr_type type;
1596 	u16	     size;
1597 	u16	     priority;
1598 	u32	     flags;
1599 	u8	     num_of_specs;
1600 	u8	     port;
1601 	/* Following are the optional layers according to user request
1602 	 * struct ib_flow_spec_xxx
1603 	 * struct ib_flow_spec_yyy
1604 	 */
1605 };
1606 
1607 struct ib_flow {
1608 	struct ib_qp		*qp;
1609 	struct ib_uobject	*uobject;
1610 };
1611 
1612 struct ib_mad_hdr;
1613 struct ib_grh;
1614 
1615 enum ib_process_mad_flags {
1616 	IB_MAD_IGNORE_MKEY	= 1,
1617 	IB_MAD_IGNORE_BKEY	= 2,
1618 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1619 };
1620 
1621 enum ib_mad_result {
1622 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1623 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1624 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1625 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1626 };
1627 
1628 #define IB_DEVICE_NAME_MAX 64
1629 
1630 struct ib_cache {
1631 	rwlock_t                lock;
1632 	struct ib_event_handler event_handler;
1633 	struct ib_pkey_cache  **pkey_cache;
1634 	struct ib_gid_table   **gid_cache;
1635 	u8                     *lmc_cache;
1636 };
1637 
1638 struct ib_dma_mapping_ops {
1639 	int		(*mapping_error)(struct ib_device *dev,
1640 					 u64 dma_addr);
1641 	u64		(*map_single)(struct ib_device *dev,
1642 				      void *ptr, size_t size,
1643 				      enum dma_data_direction direction);
1644 	void		(*unmap_single)(struct ib_device *dev,
1645 					u64 addr, size_t size,
1646 					enum dma_data_direction direction);
1647 	u64		(*map_page)(struct ib_device *dev,
1648 				    struct page *page, unsigned long offset,
1649 				    size_t size,
1650 				    enum dma_data_direction direction);
1651 	void		(*unmap_page)(struct ib_device *dev,
1652 				      u64 addr, size_t size,
1653 				      enum dma_data_direction direction);
1654 	int		(*map_sg)(struct ib_device *dev,
1655 				  struct scatterlist *sg, int nents,
1656 				  enum dma_data_direction direction);
1657 	void		(*unmap_sg)(struct ib_device *dev,
1658 				    struct scatterlist *sg, int nents,
1659 				    enum dma_data_direction direction);
1660 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1661 					       u64 dma_handle,
1662 					       size_t size,
1663 					       enum dma_data_direction dir);
1664 	void		(*sync_single_for_device)(struct ib_device *dev,
1665 						  u64 dma_handle,
1666 						  size_t size,
1667 						  enum dma_data_direction dir);
1668 	void		*(*alloc_coherent)(struct ib_device *dev,
1669 					   size_t size,
1670 					   u64 *dma_handle,
1671 					   gfp_t flag);
1672 	void		(*free_coherent)(struct ib_device *dev,
1673 					 size_t size, void *cpu_addr,
1674 					 u64 dma_handle);
1675 };
1676 
1677 struct iw_cm_verbs;
1678 
1679 struct ib_port_immutable {
1680 	int                           pkey_tbl_len;
1681 	int                           gid_tbl_len;
1682 	u32                           core_cap_flags;
1683 	u32                           max_mad_size;
1684 };
1685 
1686 struct ib_device {
1687 	struct device                *dma_device;
1688 
1689 	char                          name[IB_DEVICE_NAME_MAX];
1690 
1691 	struct list_head              event_handler_list;
1692 	spinlock_t                    event_handler_lock;
1693 
1694 	spinlock_t                    client_data_lock;
1695 	struct list_head              core_list;
1696 	/* Access to the client_data_list is protected by the client_data_lock
1697 	 * spinlock and the lists_rwsem read-write semaphore */
1698 	struct list_head              client_data_list;
1699 
1700 	struct ib_cache               cache;
1701 	/**
1702 	 * port_immutable is indexed by port number
1703 	 */
1704 	struct ib_port_immutable     *port_immutable;
1705 
1706 	int			      num_comp_vectors;
1707 
1708 	struct iw_cm_verbs	     *iwcm;
1709 
1710 	int		           (*get_protocol_stats)(struct ib_device *device,
1711 							 union rdma_protocol_stats *stats);
1712 	int		           (*query_device)(struct ib_device *device,
1713 						   struct ib_device_attr *device_attr,
1714 						   struct ib_udata *udata);
1715 	int		           (*query_port)(struct ib_device *device,
1716 						 u8 port_num,
1717 						 struct ib_port_attr *port_attr);
1718 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1719 						     u8 port_num);
1720 	/* When calling get_netdev, the HW vendor's driver should return the
1721 	 * net device of device @device at port @port_num or NULL if such
1722 	 * a net device doesn't exist. The vendor driver should call dev_hold
1723 	 * on this net device. The HW vendor's device driver must guarantee
1724 	 * that this function returns NULL before the net device reaches
1725 	 * NETDEV_UNREGISTER_FINAL state.
1726 	 */
1727 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1728 						 u8 port_num);
1729 	int		           (*query_gid)(struct ib_device *device,
1730 						u8 port_num, int index,
1731 						union ib_gid *gid);
1732 	/* When calling add_gid, the HW vendor's driver should
1733 	 * add the gid of device @device at gid index @index of
1734 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1735 	 * the network device related to this gid is available
1736 	 * at @attr. @context allows the HW vendor driver to store extra
1737 	 * information together with a GID entry. The HW vendor may allocate
1738 	 * memory to contain this information and store it in @context when a
1739 	 * new GID entry is written to. Params are consistent until the next
1740 	 * call of add_gid or delete_gid. The function should return 0 on
1741 	 * success or error otherwise. The function could be called
1742 	 * concurrently for different ports. This function is only called
1743 	 * when roce_gid_table is used.
1744 	 */
1745 	int		           (*add_gid)(struct ib_device *device,
1746 					      u8 port_num,
1747 					      unsigned int index,
1748 					      const union ib_gid *gid,
1749 					      const struct ib_gid_attr *attr,
1750 					      void **context);
1751 	/* When calling del_gid, the HW vendor's driver should delete the
1752 	 * gid of device @device at gid index @index of port @port_num.
1753 	 * Upon the deletion of a GID entry, the HW vendor must free any
1754 	 * allocated memory. The caller will clear @context afterwards.
1755 	 * This function is only called when roce_gid_table is used.
1756 	 */
1757 	int		           (*del_gid)(struct ib_device *device,
1758 					      u8 port_num,
1759 					      unsigned int index,
1760 					      void **context);
1761 	int		           (*query_pkey)(struct ib_device *device,
1762 						 u8 port_num, u16 index, u16 *pkey);
1763 	int		           (*modify_device)(struct ib_device *device,
1764 						    int device_modify_mask,
1765 						    struct ib_device_modify *device_modify);
1766 	int		           (*modify_port)(struct ib_device *device,
1767 						  u8 port_num, int port_modify_mask,
1768 						  struct ib_port_modify *port_modify);
1769 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1770 						     struct ib_udata *udata);
1771 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1772 	int                        (*mmap)(struct ib_ucontext *context,
1773 					   struct vm_area_struct *vma);
1774 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1775 					       struct ib_ucontext *context,
1776 					       struct ib_udata *udata);
1777 	int                        (*dealloc_pd)(struct ib_pd *pd);
1778 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1779 						struct ib_ah_attr *ah_attr);
1780 	int                        (*modify_ah)(struct ib_ah *ah,
1781 						struct ib_ah_attr *ah_attr);
1782 	int                        (*query_ah)(struct ib_ah *ah,
1783 					       struct ib_ah_attr *ah_attr);
1784 	int                        (*destroy_ah)(struct ib_ah *ah);
1785 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1786 						 struct ib_srq_init_attr *srq_init_attr,
1787 						 struct ib_udata *udata);
1788 	int                        (*modify_srq)(struct ib_srq *srq,
1789 						 struct ib_srq_attr *srq_attr,
1790 						 enum ib_srq_attr_mask srq_attr_mask,
1791 						 struct ib_udata *udata);
1792 	int                        (*query_srq)(struct ib_srq *srq,
1793 						struct ib_srq_attr *srq_attr);
1794 	int                        (*destroy_srq)(struct ib_srq *srq);
1795 	int                        (*post_srq_recv)(struct ib_srq *srq,
1796 						    struct ib_recv_wr *recv_wr,
1797 						    struct ib_recv_wr **bad_recv_wr);
1798 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1799 						struct ib_qp_init_attr *qp_init_attr,
1800 						struct ib_udata *udata);
1801 	int                        (*modify_qp)(struct ib_qp *qp,
1802 						struct ib_qp_attr *qp_attr,
1803 						int qp_attr_mask,
1804 						struct ib_udata *udata);
1805 	int                        (*query_qp)(struct ib_qp *qp,
1806 					       struct ib_qp_attr *qp_attr,
1807 					       int qp_attr_mask,
1808 					       struct ib_qp_init_attr *qp_init_attr);
1809 	int                        (*destroy_qp)(struct ib_qp *qp);
1810 	int                        (*post_send)(struct ib_qp *qp,
1811 						struct ib_send_wr *send_wr,
1812 						struct ib_send_wr **bad_send_wr);
1813 	int                        (*post_recv)(struct ib_qp *qp,
1814 						struct ib_recv_wr *recv_wr,
1815 						struct ib_recv_wr **bad_recv_wr);
1816 	struct ib_cq *             (*create_cq)(struct ib_device *device,
1817 						const struct ib_cq_init_attr *attr,
1818 						struct ib_ucontext *context,
1819 						struct ib_udata *udata);
1820 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1821 						u16 cq_period);
1822 	int                        (*destroy_cq)(struct ib_cq *cq);
1823 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1824 						struct ib_udata *udata);
1825 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1826 					      struct ib_wc *wc);
1827 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1828 	int                        (*req_notify_cq)(struct ib_cq *cq,
1829 						    enum ib_cq_notify_flags flags);
1830 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1831 						      int wc_cnt);
1832 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1833 						 int mr_access_flags);
1834 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1835 						  u64 start, u64 length,
1836 						  u64 virt_addr,
1837 						  int mr_access_flags,
1838 						  struct ib_udata *udata);
1839 	int			   (*rereg_user_mr)(struct ib_mr *mr,
1840 						    int flags,
1841 						    u64 start, u64 length,
1842 						    u64 virt_addr,
1843 						    int mr_access_flags,
1844 						    struct ib_pd *pd,
1845 						    struct ib_udata *udata);
1846 	int                        (*dereg_mr)(struct ib_mr *mr);
1847 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
1848 					       enum ib_mr_type mr_type,
1849 					       u32 max_num_sg);
1850 	int                        (*map_mr_sg)(struct ib_mr *mr,
1851 						struct scatterlist *sg,
1852 						int sg_nents,
1853 						unsigned int *sg_offset);
1854 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
1855 					       enum ib_mw_type type,
1856 					       struct ib_udata *udata);
1857 	int                        (*dealloc_mw)(struct ib_mw *mw);
1858 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1859 						int mr_access_flags,
1860 						struct ib_fmr_attr *fmr_attr);
1861 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1862 						   u64 *page_list, int list_len,
1863 						   u64 iova);
1864 	int		           (*unmap_fmr)(struct list_head *fmr_list);
1865 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1866 	int                        (*attach_mcast)(struct ib_qp *qp,
1867 						   union ib_gid *gid,
1868 						   u16 lid);
1869 	int                        (*detach_mcast)(struct ib_qp *qp,
1870 						   union ib_gid *gid,
1871 						   u16 lid);
1872 	int                        (*process_mad)(struct ib_device *device,
1873 						  int process_mad_flags,
1874 						  u8 port_num,
1875 						  const struct ib_wc *in_wc,
1876 						  const struct ib_grh *in_grh,
1877 						  const struct ib_mad_hdr *in_mad,
1878 						  size_t in_mad_size,
1879 						  struct ib_mad_hdr *out_mad,
1880 						  size_t *out_mad_size,
1881 						  u16 *out_mad_pkey_index);
1882 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
1883 						 struct ib_ucontext *ucontext,
1884 						 struct ib_udata *udata);
1885 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1886 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
1887 						  struct ib_flow_attr
1888 						  *flow_attr,
1889 						  int domain);
1890 	int			   (*destroy_flow)(struct ib_flow *flow_id);
1891 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1892 						      struct ib_mr_status *mr_status);
1893 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
1894 	void			   (*drain_rq)(struct ib_qp *qp);
1895 	void			   (*drain_sq)(struct ib_qp *qp);
1896 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
1897 							int state);
1898 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
1899 						   struct ifla_vf_info *ivf);
1900 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
1901 						   struct ifla_vf_stats *stats);
1902 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
1903 						  int type);
1904 
1905 	struct ib_dma_mapping_ops   *dma_ops;
1906 
1907 	struct module               *owner;
1908 	struct device                dev;
1909 	struct kobject               *ports_parent;
1910 	struct list_head             port_list;
1911 
1912 	enum {
1913 		IB_DEV_UNINITIALIZED,
1914 		IB_DEV_REGISTERED,
1915 		IB_DEV_UNREGISTERED
1916 	}                            reg_state;
1917 
1918 	int			     uverbs_abi_ver;
1919 	u64			     uverbs_cmd_mask;
1920 	u64			     uverbs_ex_cmd_mask;
1921 
1922 	char			     node_desc[64];
1923 	__be64			     node_guid;
1924 	u32			     local_dma_lkey;
1925 	u16                          is_switch:1;
1926 	u8                           node_type;
1927 	u8                           phys_port_cnt;
1928 	struct ib_device_attr        attrs;
1929 
1930 	/**
1931 	 * The following mandatory functions are used only at device
1932 	 * registration.  Keep functions such as these at the end of this
1933 	 * structure to avoid cache line misses when accessing struct ib_device
1934 	 * in fast paths.
1935 	 */
1936 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
1937 };
1938 
1939 struct ib_client {
1940 	char  *name;
1941 	void (*add)   (struct ib_device *);
1942 	void (*remove)(struct ib_device *, void *client_data);
1943 
1944 	/* Returns the net_dev belonging to this ib_client and matching the
1945 	 * given parameters.
1946 	 * @dev:	 An RDMA device that the net_dev use for communication.
1947 	 * @port:	 A physical port number on the RDMA device.
1948 	 * @pkey:	 P_Key that the net_dev uses if applicable.
1949 	 * @gid:	 A GID that the net_dev uses to communicate.
1950 	 * @addr:	 An IP address the net_dev is configured with.
1951 	 * @client_data: The device's client data set by ib_set_client_data().
1952 	 *
1953 	 * An ib_client that implements a net_dev on top of RDMA devices
1954 	 * (such as IP over IB) should implement this callback, allowing the
1955 	 * rdma_cm module to find the right net_dev for a given request.
1956 	 *
1957 	 * The caller is responsible for calling dev_put on the returned
1958 	 * netdev. */
1959 	struct net_device *(*get_net_dev_by_params)(
1960 			struct ib_device *dev,
1961 			u8 port,
1962 			u16 pkey,
1963 			const union ib_gid *gid,
1964 			const struct sockaddr *addr,
1965 			void *client_data);
1966 	struct list_head list;
1967 };
1968 
1969 struct ib_device *ib_alloc_device(size_t size);
1970 void ib_dealloc_device(struct ib_device *device);
1971 
1972 int ib_register_device(struct ib_device *device,
1973 		       int (*port_callback)(struct ib_device *,
1974 					    u8, struct kobject *));
1975 void ib_unregister_device(struct ib_device *device);
1976 
1977 int ib_register_client   (struct ib_client *client);
1978 void ib_unregister_client(struct ib_client *client);
1979 
1980 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1981 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
1982 			 void *data);
1983 
1984 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1985 {
1986 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1987 }
1988 
1989 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1990 {
1991 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1992 }
1993 
1994 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
1995 				       size_t offset,
1996 				       size_t len)
1997 {
1998 	const void __user *p = udata->inbuf + offset;
1999 	bool ret = false;
2000 	u8 *buf;
2001 
2002 	if (len > USHRT_MAX)
2003 		return false;
2004 
2005 	buf = kmalloc(len, GFP_KERNEL);
2006 	if (!buf)
2007 		return false;
2008 
2009 	if (copy_from_user(buf, p, len))
2010 		goto free;
2011 
2012 	ret = !memchr_inv(buf, 0, len);
2013 
2014 free:
2015 	kfree(buf);
2016 	return ret;
2017 }
2018 
2019 /**
2020  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2021  * contains all required attributes and no attributes not allowed for
2022  * the given QP state transition.
2023  * @cur_state: Current QP state
2024  * @next_state: Next QP state
2025  * @type: QP type
2026  * @mask: Mask of supplied QP attributes
2027  * @ll : link layer of port
2028  *
2029  * This function is a helper function that a low-level driver's
2030  * modify_qp method can use to validate the consumer's input.  It
2031  * checks that cur_state and next_state are valid QP states, that a
2032  * transition from cur_state to next_state is allowed by the IB spec,
2033  * and that the attribute mask supplied is allowed for the transition.
2034  */
2035 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2036 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2037 		       enum rdma_link_layer ll);
2038 
2039 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2040 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2041 void ib_dispatch_event(struct ib_event *event);
2042 
2043 int ib_query_port(struct ib_device *device,
2044 		  u8 port_num, struct ib_port_attr *port_attr);
2045 
2046 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2047 					       u8 port_num);
2048 
2049 /**
2050  * rdma_cap_ib_switch - Check if the device is IB switch
2051  * @device: Device to check
2052  *
2053  * Device driver is responsible for setting is_switch bit on
2054  * in ib_device structure at init time.
2055  *
2056  * Return: true if the device is IB switch.
2057  */
2058 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2059 {
2060 	return device->is_switch;
2061 }
2062 
2063 /**
2064  * rdma_start_port - Return the first valid port number for the device
2065  * specified
2066  *
2067  * @device: Device to be checked
2068  *
2069  * Return start port number
2070  */
2071 static inline u8 rdma_start_port(const struct ib_device *device)
2072 {
2073 	return rdma_cap_ib_switch(device) ? 0 : 1;
2074 }
2075 
2076 /**
2077  * rdma_end_port - Return the last valid port number for the device
2078  * specified
2079  *
2080  * @device: Device to be checked
2081  *
2082  * Return last port number
2083  */
2084 static inline u8 rdma_end_port(const struct ib_device *device)
2085 {
2086 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2087 }
2088 
2089 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2090 {
2091 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2092 }
2093 
2094 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2095 {
2096 	return device->port_immutable[port_num].core_cap_flags &
2097 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2098 }
2099 
2100 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2101 {
2102 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2103 }
2104 
2105 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2106 {
2107 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2108 }
2109 
2110 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2111 {
2112 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2113 }
2114 
2115 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2116 {
2117 	return rdma_protocol_ib(device, port_num) ||
2118 		rdma_protocol_roce(device, port_num);
2119 }
2120 
2121 /**
2122  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2123  * Management Datagrams.
2124  * @device: Device to check
2125  * @port_num: Port number to check
2126  *
2127  * Management Datagrams (MAD) are a required part of the InfiniBand
2128  * specification and are supported on all InfiniBand devices.  A slightly
2129  * extended version are also supported on OPA interfaces.
2130  *
2131  * Return: true if the port supports sending/receiving of MAD packets.
2132  */
2133 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2134 {
2135 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2136 }
2137 
2138 /**
2139  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2140  * Management Datagrams.
2141  * @device: Device to check
2142  * @port_num: Port number to check
2143  *
2144  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2145  * datagrams with their own versions.  These OPA MADs share many but not all of
2146  * the characteristics of InfiniBand MADs.
2147  *
2148  * OPA MADs differ in the following ways:
2149  *
2150  *    1) MADs are variable size up to 2K
2151  *       IBTA defined MADs remain fixed at 256 bytes
2152  *    2) OPA SMPs must carry valid PKeys
2153  *    3) OPA SMP packets are a different format
2154  *
2155  * Return: true if the port supports OPA MAD packet formats.
2156  */
2157 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2158 {
2159 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2160 		== RDMA_CORE_CAP_OPA_MAD;
2161 }
2162 
2163 /**
2164  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2165  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2166  * @device: Device to check
2167  * @port_num: Port number to check
2168  *
2169  * Each InfiniBand node is required to provide a Subnet Management Agent
2170  * that the subnet manager can access.  Prior to the fabric being fully
2171  * configured by the subnet manager, the SMA is accessed via a well known
2172  * interface called the Subnet Management Interface (SMI).  This interface
2173  * uses directed route packets to communicate with the SM to get around the
2174  * chicken and egg problem of the SM needing to know what's on the fabric
2175  * in order to configure the fabric, and needing to configure the fabric in
2176  * order to send packets to the devices on the fabric.  These directed
2177  * route packets do not need the fabric fully configured in order to reach
2178  * their destination.  The SMI is the only method allowed to send
2179  * directed route packets on an InfiniBand fabric.
2180  *
2181  * Return: true if the port provides an SMI.
2182  */
2183 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2184 {
2185 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2186 }
2187 
2188 /**
2189  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2190  * Communication Manager.
2191  * @device: Device to check
2192  * @port_num: Port number to check
2193  *
2194  * The InfiniBand Communication Manager is one of many pre-defined General
2195  * Service Agents (GSA) that are accessed via the General Service
2196  * Interface (GSI).  It's role is to facilitate establishment of connections
2197  * between nodes as well as other management related tasks for established
2198  * connections.
2199  *
2200  * Return: true if the port supports an IB CM (this does not guarantee that
2201  * a CM is actually running however).
2202  */
2203 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2204 {
2205 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2206 }
2207 
2208 /**
2209  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2210  * Communication Manager.
2211  * @device: Device to check
2212  * @port_num: Port number to check
2213  *
2214  * Similar to above, but specific to iWARP connections which have a different
2215  * managment protocol than InfiniBand.
2216  *
2217  * Return: true if the port supports an iWARP CM (this does not guarantee that
2218  * a CM is actually running however).
2219  */
2220 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2221 {
2222 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2223 }
2224 
2225 /**
2226  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2227  * Subnet Administration.
2228  * @device: Device to check
2229  * @port_num: Port number to check
2230  *
2231  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2232  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2233  * fabrics, devices should resolve routes to other hosts by contacting the
2234  * SA to query the proper route.
2235  *
2236  * Return: true if the port should act as a client to the fabric Subnet
2237  * Administration interface.  This does not imply that the SA service is
2238  * running locally.
2239  */
2240 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2241 {
2242 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2243 }
2244 
2245 /**
2246  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2247  * Multicast.
2248  * @device: Device to check
2249  * @port_num: Port number to check
2250  *
2251  * InfiniBand multicast registration is more complex than normal IPv4 or
2252  * IPv6 multicast registration.  Each Host Channel Adapter must register
2253  * with the Subnet Manager when it wishes to join a multicast group.  It
2254  * should do so only once regardless of how many queue pairs it subscribes
2255  * to this group.  And it should leave the group only after all queue pairs
2256  * attached to the group have been detached.
2257  *
2258  * Return: true if the port must undertake the additional adminstrative
2259  * overhead of registering/unregistering with the SM and tracking of the
2260  * total number of queue pairs attached to the multicast group.
2261  */
2262 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2263 {
2264 	return rdma_cap_ib_sa(device, port_num);
2265 }
2266 
2267 /**
2268  * rdma_cap_af_ib - Check if the port of device has the capability
2269  * Native Infiniband Address.
2270  * @device: Device to check
2271  * @port_num: Port number to check
2272  *
2273  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2274  * GID.  RoCE uses a different mechanism, but still generates a GID via
2275  * a prescribed mechanism and port specific data.
2276  *
2277  * Return: true if the port uses a GID address to identify devices on the
2278  * network.
2279  */
2280 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2281 {
2282 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2283 }
2284 
2285 /**
2286  * rdma_cap_eth_ah - Check if the port of device has the capability
2287  * Ethernet Address Handle.
2288  * @device: Device to check
2289  * @port_num: Port number to check
2290  *
2291  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2292  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2293  * port.  Normally, packet headers are generated by the sending host
2294  * adapter, but when sending connectionless datagrams, we must manually
2295  * inject the proper headers for the fabric we are communicating over.
2296  *
2297  * Return: true if we are running as a RoCE port and must force the
2298  * addition of a Global Route Header built from our Ethernet Address
2299  * Handle into our header list for connectionless packets.
2300  */
2301 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2302 {
2303 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2304 }
2305 
2306 /**
2307  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2308  *
2309  * @device: Device
2310  * @port_num: Port number
2311  *
2312  * This MAD size includes the MAD headers and MAD payload.  No other headers
2313  * are included.
2314  *
2315  * Return the max MAD size required by the Port.  Will return 0 if the port
2316  * does not support MADs
2317  */
2318 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2319 {
2320 	return device->port_immutable[port_num].max_mad_size;
2321 }
2322 
2323 /**
2324  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2325  * @device: Device to check
2326  * @port_num: Port number to check
2327  *
2328  * RoCE GID table mechanism manages the various GIDs for a device.
2329  *
2330  * NOTE: if allocating the port's GID table has failed, this call will still
2331  * return true, but any RoCE GID table API will fail.
2332  *
2333  * Return: true if the port uses RoCE GID table mechanism in order to manage
2334  * its GIDs.
2335  */
2336 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2337 					   u8 port_num)
2338 {
2339 	return rdma_protocol_roce(device, port_num) &&
2340 		device->add_gid && device->del_gid;
2341 }
2342 
2343 /*
2344  * Check if the device supports READ W/ INVALIDATE.
2345  */
2346 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2347 {
2348 	/*
2349 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2350 	 * has support for it yet.
2351 	 */
2352 	return rdma_protocol_iwarp(dev, port_num);
2353 }
2354 
2355 int ib_query_gid(struct ib_device *device,
2356 		 u8 port_num, int index, union ib_gid *gid,
2357 		 struct ib_gid_attr *attr);
2358 
2359 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2360 			 int state);
2361 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2362 		     struct ifla_vf_info *info);
2363 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2364 		    struct ifla_vf_stats *stats);
2365 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2366 		   int type);
2367 
2368 int ib_query_pkey(struct ib_device *device,
2369 		  u8 port_num, u16 index, u16 *pkey);
2370 
2371 int ib_modify_device(struct ib_device *device,
2372 		     int device_modify_mask,
2373 		     struct ib_device_modify *device_modify);
2374 
2375 int ib_modify_port(struct ib_device *device,
2376 		   u8 port_num, int port_modify_mask,
2377 		   struct ib_port_modify *port_modify);
2378 
2379 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2380 		enum ib_gid_type gid_type, struct net_device *ndev,
2381 		u8 *port_num, u16 *index);
2382 
2383 int ib_find_pkey(struct ib_device *device,
2384 		 u8 port_num, u16 pkey, u16 *index);
2385 
2386 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2387 
2388 void ib_dealloc_pd(struct ib_pd *pd);
2389 
2390 /**
2391  * ib_create_ah - Creates an address handle for the given address vector.
2392  * @pd: The protection domain associated with the address handle.
2393  * @ah_attr: The attributes of the address vector.
2394  *
2395  * The address handle is used to reference a local or global destination
2396  * in all UD QP post sends.
2397  */
2398 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2399 
2400 /**
2401  * ib_init_ah_from_wc - Initializes address handle attributes from a
2402  *   work completion.
2403  * @device: Device on which the received message arrived.
2404  * @port_num: Port on which the received message arrived.
2405  * @wc: Work completion associated with the received message.
2406  * @grh: References the received global route header.  This parameter is
2407  *   ignored unless the work completion indicates that the GRH is valid.
2408  * @ah_attr: Returned attributes that can be used when creating an address
2409  *   handle for replying to the message.
2410  */
2411 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2412 		       const struct ib_wc *wc, const struct ib_grh *grh,
2413 		       struct ib_ah_attr *ah_attr);
2414 
2415 /**
2416  * ib_create_ah_from_wc - Creates an address handle associated with the
2417  *   sender of the specified work completion.
2418  * @pd: The protection domain associated with the address handle.
2419  * @wc: Work completion information associated with a received message.
2420  * @grh: References the received global route header.  This parameter is
2421  *   ignored unless the work completion indicates that the GRH is valid.
2422  * @port_num: The outbound port number to associate with the address.
2423  *
2424  * The address handle is used to reference a local or global destination
2425  * in all UD QP post sends.
2426  */
2427 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2428 				   const struct ib_grh *grh, u8 port_num);
2429 
2430 /**
2431  * ib_modify_ah - Modifies the address vector associated with an address
2432  *   handle.
2433  * @ah: The address handle to modify.
2434  * @ah_attr: The new address vector attributes to associate with the
2435  *   address handle.
2436  */
2437 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2438 
2439 /**
2440  * ib_query_ah - Queries the address vector associated with an address
2441  *   handle.
2442  * @ah: The address handle to query.
2443  * @ah_attr: The address vector attributes associated with the address
2444  *   handle.
2445  */
2446 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2447 
2448 /**
2449  * ib_destroy_ah - Destroys an address handle.
2450  * @ah: The address handle to destroy.
2451  */
2452 int ib_destroy_ah(struct ib_ah *ah);
2453 
2454 /**
2455  * ib_create_srq - Creates a SRQ associated with the specified protection
2456  *   domain.
2457  * @pd: The protection domain associated with the SRQ.
2458  * @srq_init_attr: A list of initial attributes required to create the
2459  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2460  *   the actual capabilities of the created SRQ.
2461  *
2462  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2463  * requested size of the SRQ, and set to the actual values allocated
2464  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2465  * will always be at least as large as the requested values.
2466  */
2467 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2468 			     struct ib_srq_init_attr *srq_init_attr);
2469 
2470 /**
2471  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2472  * @srq: The SRQ to modify.
2473  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2474  *   the current values of selected SRQ attributes are returned.
2475  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2476  *   are being modified.
2477  *
2478  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2479  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2480  * the number of receives queued drops below the limit.
2481  */
2482 int ib_modify_srq(struct ib_srq *srq,
2483 		  struct ib_srq_attr *srq_attr,
2484 		  enum ib_srq_attr_mask srq_attr_mask);
2485 
2486 /**
2487  * ib_query_srq - Returns the attribute list and current values for the
2488  *   specified SRQ.
2489  * @srq: The SRQ to query.
2490  * @srq_attr: The attributes of the specified SRQ.
2491  */
2492 int ib_query_srq(struct ib_srq *srq,
2493 		 struct ib_srq_attr *srq_attr);
2494 
2495 /**
2496  * ib_destroy_srq - Destroys the specified SRQ.
2497  * @srq: The SRQ to destroy.
2498  */
2499 int ib_destroy_srq(struct ib_srq *srq);
2500 
2501 /**
2502  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2503  * @srq: The SRQ to post the work request on.
2504  * @recv_wr: A list of work requests to post on the receive queue.
2505  * @bad_recv_wr: On an immediate failure, this parameter will reference
2506  *   the work request that failed to be posted on the QP.
2507  */
2508 static inline int ib_post_srq_recv(struct ib_srq *srq,
2509 				   struct ib_recv_wr *recv_wr,
2510 				   struct ib_recv_wr **bad_recv_wr)
2511 {
2512 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2513 }
2514 
2515 /**
2516  * ib_create_qp - Creates a QP associated with the specified protection
2517  *   domain.
2518  * @pd: The protection domain associated with the QP.
2519  * @qp_init_attr: A list of initial attributes required to create the
2520  *   QP.  If QP creation succeeds, then the attributes are updated to
2521  *   the actual capabilities of the created QP.
2522  */
2523 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2524 			   struct ib_qp_init_attr *qp_init_attr);
2525 
2526 /**
2527  * ib_modify_qp - Modifies the attributes for the specified QP and then
2528  *   transitions the QP to the given state.
2529  * @qp: The QP to modify.
2530  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2531  *   the current values of selected QP attributes are returned.
2532  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2533  *   are being modified.
2534  */
2535 int ib_modify_qp(struct ib_qp *qp,
2536 		 struct ib_qp_attr *qp_attr,
2537 		 int qp_attr_mask);
2538 
2539 /**
2540  * ib_query_qp - Returns the attribute list and current values for the
2541  *   specified QP.
2542  * @qp: The QP to query.
2543  * @qp_attr: The attributes of the specified QP.
2544  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2545  * @qp_init_attr: Additional attributes of the selected QP.
2546  *
2547  * The qp_attr_mask may be used to limit the query to gathering only the
2548  * selected attributes.
2549  */
2550 int ib_query_qp(struct ib_qp *qp,
2551 		struct ib_qp_attr *qp_attr,
2552 		int qp_attr_mask,
2553 		struct ib_qp_init_attr *qp_init_attr);
2554 
2555 /**
2556  * ib_destroy_qp - Destroys the specified QP.
2557  * @qp: The QP to destroy.
2558  */
2559 int ib_destroy_qp(struct ib_qp *qp);
2560 
2561 /**
2562  * ib_open_qp - Obtain a reference to an existing sharable QP.
2563  * @xrcd - XRC domain
2564  * @qp_open_attr: Attributes identifying the QP to open.
2565  *
2566  * Returns a reference to a sharable QP.
2567  */
2568 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2569 			 struct ib_qp_open_attr *qp_open_attr);
2570 
2571 /**
2572  * ib_close_qp - Release an external reference to a QP.
2573  * @qp: The QP handle to release
2574  *
2575  * The opened QP handle is released by the caller.  The underlying
2576  * shared QP is not destroyed until all internal references are released.
2577  */
2578 int ib_close_qp(struct ib_qp *qp);
2579 
2580 /**
2581  * ib_post_send - Posts a list of work requests to the send queue of
2582  *   the specified QP.
2583  * @qp: The QP to post the work request on.
2584  * @send_wr: A list of work requests to post on the send queue.
2585  * @bad_send_wr: On an immediate failure, this parameter will reference
2586  *   the work request that failed to be posted on the QP.
2587  *
2588  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2589  * error is returned, the QP state shall not be affected,
2590  * ib_post_send() will return an immediate error after queueing any
2591  * earlier work requests in the list.
2592  */
2593 static inline int ib_post_send(struct ib_qp *qp,
2594 			       struct ib_send_wr *send_wr,
2595 			       struct ib_send_wr **bad_send_wr)
2596 {
2597 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2598 }
2599 
2600 /**
2601  * ib_post_recv - Posts a list of work requests to the receive queue of
2602  *   the specified QP.
2603  * @qp: The QP to post the work request on.
2604  * @recv_wr: A list of work requests to post on the receive queue.
2605  * @bad_recv_wr: On an immediate failure, this parameter will reference
2606  *   the work request that failed to be posted on the QP.
2607  */
2608 static inline int ib_post_recv(struct ib_qp *qp,
2609 			       struct ib_recv_wr *recv_wr,
2610 			       struct ib_recv_wr **bad_recv_wr)
2611 {
2612 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2613 }
2614 
2615 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2616 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2617 void ib_free_cq(struct ib_cq *cq);
2618 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2619 
2620 /**
2621  * ib_create_cq - Creates a CQ on the specified device.
2622  * @device: The device on which to create the CQ.
2623  * @comp_handler: A user-specified callback that is invoked when a
2624  *   completion event occurs on the CQ.
2625  * @event_handler: A user-specified callback that is invoked when an
2626  *   asynchronous event not associated with a completion occurs on the CQ.
2627  * @cq_context: Context associated with the CQ returned to the user via
2628  *   the associated completion and event handlers.
2629  * @cq_attr: The attributes the CQ should be created upon.
2630  *
2631  * Users can examine the cq structure to determine the actual CQ size.
2632  */
2633 struct ib_cq *ib_create_cq(struct ib_device *device,
2634 			   ib_comp_handler comp_handler,
2635 			   void (*event_handler)(struct ib_event *, void *),
2636 			   void *cq_context,
2637 			   const struct ib_cq_init_attr *cq_attr);
2638 
2639 /**
2640  * ib_resize_cq - Modifies the capacity of the CQ.
2641  * @cq: The CQ to resize.
2642  * @cqe: The minimum size of the CQ.
2643  *
2644  * Users can examine the cq structure to determine the actual CQ size.
2645  */
2646 int ib_resize_cq(struct ib_cq *cq, int cqe);
2647 
2648 /**
2649  * ib_modify_cq - Modifies moderation params of the CQ
2650  * @cq: The CQ to modify.
2651  * @cq_count: number of CQEs that will trigger an event
2652  * @cq_period: max period of time in usec before triggering an event
2653  *
2654  */
2655 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2656 
2657 /**
2658  * ib_destroy_cq - Destroys the specified CQ.
2659  * @cq: The CQ to destroy.
2660  */
2661 int ib_destroy_cq(struct ib_cq *cq);
2662 
2663 /**
2664  * ib_poll_cq - poll a CQ for completion(s)
2665  * @cq:the CQ being polled
2666  * @num_entries:maximum number of completions to return
2667  * @wc:array of at least @num_entries &struct ib_wc where completions
2668  *   will be returned
2669  *
2670  * Poll a CQ for (possibly multiple) completions.  If the return value
2671  * is < 0, an error occurred.  If the return value is >= 0, it is the
2672  * number of completions returned.  If the return value is
2673  * non-negative and < num_entries, then the CQ was emptied.
2674  */
2675 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2676 			     struct ib_wc *wc)
2677 {
2678 	return cq->device->poll_cq(cq, num_entries, wc);
2679 }
2680 
2681 /**
2682  * ib_peek_cq - Returns the number of unreaped completions currently
2683  *   on the specified CQ.
2684  * @cq: The CQ to peek.
2685  * @wc_cnt: A minimum number of unreaped completions to check for.
2686  *
2687  * If the number of unreaped completions is greater than or equal to wc_cnt,
2688  * this function returns wc_cnt, otherwise, it returns the actual number of
2689  * unreaped completions.
2690  */
2691 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2692 
2693 /**
2694  * ib_req_notify_cq - Request completion notification on a CQ.
2695  * @cq: The CQ to generate an event for.
2696  * @flags:
2697  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2698  *   to request an event on the next solicited event or next work
2699  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2700  *   may also be |ed in to request a hint about missed events, as
2701  *   described below.
2702  *
2703  * Return Value:
2704  *    < 0 means an error occurred while requesting notification
2705  *   == 0 means notification was requested successfully, and if
2706  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2707  *        were missed and it is safe to wait for another event.  In
2708  *        this case is it guaranteed that any work completions added
2709  *        to the CQ since the last CQ poll will trigger a completion
2710  *        notification event.
2711  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2712  *        in.  It means that the consumer must poll the CQ again to
2713  *        make sure it is empty to avoid missing an event because of a
2714  *        race between requesting notification and an entry being
2715  *        added to the CQ.  This return value means it is possible
2716  *        (but not guaranteed) that a work completion has been added
2717  *        to the CQ since the last poll without triggering a
2718  *        completion notification event.
2719  */
2720 static inline int ib_req_notify_cq(struct ib_cq *cq,
2721 				   enum ib_cq_notify_flags flags)
2722 {
2723 	return cq->device->req_notify_cq(cq, flags);
2724 }
2725 
2726 /**
2727  * ib_req_ncomp_notif - Request completion notification when there are
2728  *   at least the specified number of unreaped completions on the CQ.
2729  * @cq: The CQ to generate an event for.
2730  * @wc_cnt: The number of unreaped completions that should be on the
2731  *   CQ before an event is generated.
2732  */
2733 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2734 {
2735 	return cq->device->req_ncomp_notif ?
2736 		cq->device->req_ncomp_notif(cq, wc_cnt) :
2737 		-ENOSYS;
2738 }
2739 
2740 /**
2741  * ib_get_dma_mr - Returns a memory region for system memory that is
2742  *   usable for DMA.
2743  * @pd: The protection domain associated with the memory region.
2744  * @mr_access_flags: Specifies the memory access rights.
2745  *
2746  * Note that the ib_dma_*() functions defined below must be used
2747  * to create/destroy addresses used with the Lkey or Rkey returned
2748  * by ib_get_dma_mr().
2749  */
2750 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2751 
2752 /**
2753  * ib_dma_mapping_error - check a DMA addr for error
2754  * @dev: The device for which the dma_addr was created
2755  * @dma_addr: The DMA address to check
2756  */
2757 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2758 {
2759 	if (dev->dma_ops)
2760 		return dev->dma_ops->mapping_error(dev, dma_addr);
2761 	return dma_mapping_error(dev->dma_device, dma_addr);
2762 }
2763 
2764 /**
2765  * ib_dma_map_single - Map a kernel virtual address to DMA address
2766  * @dev: The device for which the dma_addr is to be created
2767  * @cpu_addr: The kernel virtual address
2768  * @size: The size of the region in bytes
2769  * @direction: The direction of the DMA
2770  */
2771 static inline u64 ib_dma_map_single(struct ib_device *dev,
2772 				    void *cpu_addr, size_t size,
2773 				    enum dma_data_direction direction)
2774 {
2775 	if (dev->dma_ops)
2776 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2777 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2778 }
2779 
2780 /**
2781  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2782  * @dev: The device for which the DMA address was created
2783  * @addr: The DMA address
2784  * @size: The size of the region in bytes
2785  * @direction: The direction of the DMA
2786  */
2787 static inline void ib_dma_unmap_single(struct ib_device *dev,
2788 				       u64 addr, size_t size,
2789 				       enum dma_data_direction direction)
2790 {
2791 	if (dev->dma_ops)
2792 		dev->dma_ops->unmap_single(dev, addr, size, direction);
2793 	else
2794 		dma_unmap_single(dev->dma_device, addr, size, direction);
2795 }
2796 
2797 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2798 					  void *cpu_addr, size_t size,
2799 					  enum dma_data_direction direction,
2800 					  struct dma_attrs *attrs)
2801 {
2802 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2803 				    direction, attrs);
2804 }
2805 
2806 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2807 					     u64 addr, size_t size,
2808 					     enum dma_data_direction direction,
2809 					     struct dma_attrs *attrs)
2810 {
2811 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
2812 				      direction, attrs);
2813 }
2814 
2815 /**
2816  * ib_dma_map_page - Map a physical page to DMA address
2817  * @dev: The device for which the dma_addr is to be created
2818  * @page: The page to be mapped
2819  * @offset: The offset within the page
2820  * @size: The size of the region in bytes
2821  * @direction: The direction of the DMA
2822  */
2823 static inline u64 ib_dma_map_page(struct ib_device *dev,
2824 				  struct page *page,
2825 				  unsigned long offset,
2826 				  size_t size,
2827 					 enum dma_data_direction direction)
2828 {
2829 	if (dev->dma_ops)
2830 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
2831 	return dma_map_page(dev->dma_device, page, offset, size, direction);
2832 }
2833 
2834 /**
2835  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2836  * @dev: The device for which the DMA address was created
2837  * @addr: The DMA address
2838  * @size: The size of the region in bytes
2839  * @direction: The direction of the DMA
2840  */
2841 static inline void ib_dma_unmap_page(struct ib_device *dev,
2842 				     u64 addr, size_t size,
2843 				     enum dma_data_direction direction)
2844 {
2845 	if (dev->dma_ops)
2846 		dev->dma_ops->unmap_page(dev, addr, size, direction);
2847 	else
2848 		dma_unmap_page(dev->dma_device, addr, size, direction);
2849 }
2850 
2851 /**
2852  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2853  * @dev: The device for which the DMA addresses are to be created
2854  * @sg: The array of scatter/gather entries
2855  * @nents: The number of scatter/gather entries
2856  * @direction: The direction of the DMA
2857  */
2858 static inline int ib_dma_map_sg(struct ib_device *dev,
2859 				struct scatterlist *sg, int nents,
2860 				enum dma_data_direction direction)
2861 {
2862 	if (dev->dma_ops)
2863 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
2864 	return dma_map_sg(dev->dma_device, sg, nents, direction);
2865 }
2866 
2867 /**
2868  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2869  * @dev: The device for which the DMA addresses were created
2870  * @sg: The array of scatter/gather entries
2871  * @nents: The number of scatter/gather entries
2872  * @direction: The direction of the DMA
2873  */
2874 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2875 				   struct scatterlist *sg, int nents,
2876 				   enum dma_data_direction direction)
2877 {
2878 	if (dev->dma_ops)
2879 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2880 	else
2881 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
2882 }
2883 
2884 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2885 				      struct scatterlist *sg, int nents,
2886 				      enum dma_data_direction direction,
2887 				      struct dma_attrs *attrs)
2888 {
2889 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2890 }
2891 
2892 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2893 					 struct scatterlist *sg, int nents,
2894 					 enum dma_data_direction direction,
2895 					 struct dma_attrs *attrs)
2896 {
2897 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2898 }
2899 /**
2900  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2901  * @dev: The device for which the DMA addresses were created
2902  * @sg: The scatter/gather entry
2903  *
2904  * Note: this function is obsolete. To do: change all occurrences of
2905  * ib_sg_dma_address() into sg_dma_address().
2906  */
2907 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2908 				    struct scatterlist *sg)
2909 {
2910 	return sg_dma_address(sg);
2911 }
2912 
2913 /**
2914  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2915  * @dev: The device for which the DMA addresses were created
2916  * @sg: The scatter/gather entry
2917  *
2918  * Note: this function is obsolete. To do: change all occurrences of
2919  * ib_sg_dma_len() into sg_dma_len().
2920  */
2921 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2922 					 struct scatterlist *sg)
2923 {
2924 	return sg_dma_len(sg);
2925 }
2926 
2927 /**
2928  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2929  * @dev: The device for which the DMA address was created
2930  * @addr: The DMA address
2931  * @size: The size of the region in bytes
2932  * @dir: The direction of the DMA
2933  */
2934 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2935 					      u64 addr,
2936 					      size_t size,
2937 					      enum dma_data_direction dir)
2938 {
2939 	if (dev->dma_ops)
2940 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2941 	else
2942 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2943 }
2944 
2945 /**
2946  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2947  * @dev: The device for which the DMA address was created
2948  * @addr: The DMA address
2949  * @size: The size of the region in bytes
2950  * @dir: The direction of the DMA
2951  */
2952 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2953 						 u64 addr,
2954 						 size_t size,
2955 						 enum dma_data_direction dir)
2956 {
2957 	if (dev->dma_ops)
2958 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2959 	else
2960 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2961 }
2962 
2963 /**
2964  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2965  * @dev: The device for which the DMA address is requested
2966  * @size: The size of the region to allocate in bytes
2967  * @dma_handle: A pointer for returning the DMA address of the region
2968  * @flag: memory allocator flags
2969  */
2970 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2971 					   size_t size,
2972 					   u64 *dma_handle,
2973 					   gfp_t flag)
2974 {
2975 	if (dev->dma_ops)
2976 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2977 	else {
2978 		dma_addr_t handle;
2979 		void *ret;
2980 
2981 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2982 		*dma_handle = handle;
2983 		return ret;
2984 	}
2985 }
2986 
2987 /**
2988  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2989  * @dev: The device for which the DMA addresses were allocated
2990  * @size: The size of the region
2991  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2992  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2993  */
2994 static inline void ib_dma_free_coherent(struct ib_device *dev,
2995 					size_t size, void *cpu_addr,
2996 					u64 dma_handle)
2997 {
2998 	if (dev->dma_ops)
2999 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3000 	else
3001 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3002 }
3003 
3004 /**
3005  * ib_dereg_mr - Deregisters a memory region and removes it from the
3006  *   HCA translation table.
3007  * @mr: The memory region to deregister.
3008  *
3009  * This function can fail, if the memory region has memory windows bound to it.
3010  */
3011 int ib_dereg_mr(struct ib_mr *mr);
3012 
3013 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3014 			  enum ib_mr_type mr_type,
3015 			  u32 max_num_sg);
3016 
3017 /**
3018  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3019  *   R_Key and L_Key.
3020  * @mr - struct ib_mr pointer to be updated.
3021  * @newkey - new key to be used.
3022  */
3023 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3024 {
3025 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3026 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3027 }
3028 
3029 /**
3030  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3031  * for calculating a new rkey for type 2 memory windows.
3032  * @rkey - the rkey to increment.
3033  */
3034 static inline u32 ib_inc_rkey(u32 rkey)
3035 {
3036 	const u32 mask = 0x000000ff;
3037 	return ((rkey + 1) & mask) | (rkey & ~mask);
3038 }
3039 
3040 /**
3041  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3042  * @pd: The protection domain associated with the unmapped region.
3043  * @mr_access_flags: Specifies the memory access rights.
3044  * @fmr_attr: Attributes of the unmapped region.
3045  *
3046  * A fast memory region must be mapped before it can be used as part of
3047  * a work request.
3048  */
3049 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3050 			    int mr_access_flags,
3051 			    struct ib_fmr_attr *fmr_attr);
3052 
3053 /**
3054  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3055  * @fmr: The fast memory region to associate with the pages.
3056  * @page_list: An array of physical pages to map to the fast memory region.
3057  * @list_len: The number of pages in page_list.
3058  * @iova: The I/O virtual address to use with the mapped region.
3059  */
3060 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3061 				  u64 *page_list, int list_len,
3062 				  u64 iova)
3063 {
3064 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3065 }
3066 
3067 /**
3068  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3069  * @fmr_list: A linked list of fast memory regions to unmap.
3070  */
3071 int ib_unmap_fmr(struct list_head *fmr_list);
3072 
3073 /**
3074  * ib_dealloc_fmr - Deallocates a fast memory region.
3075  * @fmr: The fast memory region to deallocate.
3076  */
3077 int ib_dealloc_fmr(struct ib_fmr *fmr);
3078 
3079 /**
3080  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3081  * @qp: QP to attach to the multicast group.  The QP must be type
3082  *   IB_QPT_UD.
3083  * @gid: Multicast group GID.
3084  * @lid: Multicast group LID in host byte order.
3085  *
3086  * In order to send and receive multicast packets, subnet
3087  * administration must have created the multicast group and configured
3088  * the fabric appropriately.  The port associated with the specified
3089  * QP must also be a member of the multicast group.
3090  */
3091 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3092 
3093 /**
3094  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3095  * @qp: QP to detach from the multicast group.
3096  * @gid: Multicast group GID.
3097  * @lid: Multicast group LID in host byte order.
3098  */
3099 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3100 
3101 /**
3102  * ib_alloc_xrcd - Allocates an XRC domain.
3103  * @device: The device on which to allocate the XRC domain.
3104  */
3105 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3106 
3107 /**
3108  * ib_dealloc_xrcd - Deallocates an XRC domain.
3109  * @xrcd: The XRC domain to deallocate.
3110  */
3111 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3112 
3113 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3114 			       struct ib_flow_attr *flow_attr, int domain);
3115 int ib_destroy_flow(struct ib_flow *flow_id);
3116 
3117 static inline int ib_check_mr_access(int flags)
3118 {
3119 	/*
3120 	 * Local write permission is required if remote write or
3121 	 * remote atomic permission is also requested.
3122 	 */
3123 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3124 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3125 		return -EINVAL;
3126 
3127 	return 0;
3128 }
3129 
3130 /**
3131  * ib_check_mr_status: lightweight check of MR status.
3132  *     This routine may provide status checks on a selected
3133  *     ib_mr. first use is for signature status check.
3134  *
3135  * @mr: A memory region.
3136  * @check_mask: Bitmask of which checks to perform from
3137  *     ib_mr_status_check enumeration.
3138  * @mr_status: The container of relevant status checks.
3139  *     failed checks will be indicated in the status bitmask
3140  *     and the relevant info shall be in the error item.
3141  */
3142 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3143 		       struct ib_mr_status *mr_status);
3144 
3145 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3146 					    u16 pkey, const union ib_gid *gid,
3147 					    const struct sockaddr *addr);
3148 
3149 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3150 		 unsigned int *sg_offset, unsigned int page_size);
3151 
3152 static inline int
3153 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3154 		  unsigned int *sg_offset, unsigned int page_size)
3155 {
3156 	int n;
3157 
3158 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3159 	mr->iova = 0;
3160 
3161 	return n;
3162 }
3163 
3164 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3165 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3166 
3167 void ib_drain_rq(struct ib_qp *qp);
3168 void ib_drain_sq(struct ib_qp *qp);
3169 void ib_drain_qp(struct ib_qp *qp);
3170 #endif /* IB_VERBS_H */
3171