xref: /linux/include/rdma/ib_verbs.h (revision 55f3538c4923e9dfca132e99ebec370e8094afda)
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 #include <linux/netdevice.h>
59 
60 #include <linux/if_link.h>
61 #include <linux/atomic.h>
62 #include <linux/mmu_notifier.h>
63 #include <linux/uaccess.h>
64 #include <linux/cgroup_rdma.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 #include <rdma/restrack.h>
67 
68 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
69 
70 extern struct workqueue_struct *ib_wq;
71 extern struct workqueue_struct *ib_comp_wq;
72 
73 union ib_gid {
74 	u8	raw[16];
75 	struct {
76 		__be64	subnet_prefix;
77 		__be64	interface_id;
78 	} global;
79 };
80 
81 extern union ib_gid zgid;
82 
83 enum ib_gid_type {
84 	/* If link layer is Ethernet, this is RoCE V1 */
85 	IB_GID_TYPE_IB        = 0,
86 	IB_GID_TYPE_ROCE      = 0,
87 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
88 	IB_GID_TYPE_SIZE
89 };
90 
91 #define ROCE_V2_UDP_DPORT      4791
92 struct ib_gid_attr {
93 	enum ib_gid_type	gid_type;
94 	struct net_device	*ndev;
95 };
96 
97 enum rdma_node_type {
98 	/* IB values map to NodeInfo:NodeType. */
99 	RDMA_NODE_IB_CA 	= 1,
100 	RDMA_NODE_IB_SWITCH,
101 	RDMA_NODE_IB_ROUTER,
102 	RDMA_NODE_RNIC,
103 	RDMA_NODE_USNIC,
104 	RDMA_NODE_USNIC_UDP,
105 };
106 
107 enum {
108 	/* set the local administered indication */
109 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
110 };
111 
112 enum rdma_transport_type {
113 	RDMA_TRANSPORT_IB,
114 	RDMA_TRANSPORT_IWARP,
115 	RDMA_TRANSPORT_USNIC,
116 	RDMA_TRANSPORT_USNIC_UDP
117 };
118 
119 enum rdma_protocol_type {
120 	RDMA_PROTOCOL_IB,
121 	RDMA_PROTOCOL_IBOE,
122 	RDMA_PROTOCOL_IWARP,
123 	RDMA_PROTOCOL_USNIC_UDP
124 };
125 
126 __attribute_const__ enum rdma_transport_type
127 rdma_node_get_transport(enum rdma_node_type node_type);
128 
129 enum rdma_network_type {
130 	RDMA_NETWORK_IB,
131 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
132 	RDMA_NETWORK_IPV4,
133 	RDMA_NETWORK_IPV6
134 };
135 
136 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
137 {
138 	if (network_type == RDMA_NETWORK_IPV4 ||
139 	    network_type == RDMA_NETWORK_IPV6)
140 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
141 
142 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
143 	return IB_GID_TYPE_IB;
144 }
145 
146 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
147 							    union ib_gid *gid)
148 {
149 	if (gid_type == IB_GID_TYPE_IB)
150 		return RDMA_NETWORK_IB;
151 
152 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
153 		return RDMA_NETWORK_IPV4;
154 	else
155 		return RDMA_NETWORK_IPV6;
156 }
157 
158 enum rdma_link_layer {
159 	IB_LINK_LAYER_UNSPECIFIED,
160 	IB_LINK_LAYER_INFINIBAND,
161 	IB_LINK_LAYER_ETHERNET,
162 };
163 
164 enum ib_device_cap_flags {
165 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
166 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
167 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
168 	IB_DEVICE_RAW_MULTI			= (1 << 3),
169 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
170 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
171 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
172 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
173 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
174 	/* Not in use, former INIT_TYPE		= (1 << 9),*/
175 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
176 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
177 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
178 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
179 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
180 
181 	/*
182 	 * This device supports a per-device lkey or stag that can be
183 	 * used without performing a memory registration for the local
184 	 * memory.  Note that ULPs should never check this flag, but
185 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
186 	 * which will always contain a usable lkey.
187 	 */
188 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
189 	/* Reserved, old SEND_W_INV		= (1 << 16),*/
190 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
191 	/*
192 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
193 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
194 	 * messages and can verify the validity of checksum for
195 	 * incoming messages.  Setting this flag implies that the
196 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
197 	 */
198 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
199 	IB_DEVICE_UD_TSO			= (1 << 19),
200 	IB_DEVICE_XRC				= (1 << 20),
201 
202 	/*
203 	 * This device supports the IB "base memory management extension",
204 	 * which includes support for fast registrations (IB_WR_REG_MR,
205 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
206 	 * also be set by any iWarp device which must support FRs to comply
207 	 * to the iWarp verbs spec.  iWarp devices also support the
208 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
209 	 * stag.
210 	 */
211 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
212 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
213 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
214 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
215 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
216 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
217 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
218 	/*
219 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
220 	 * support execution of WQEs that involve synchronization
221 	 * of I/O operations with single completion queue managed
222 	 * by hardware.
223 	 */
224 	IB_DEVICE_CROSS_CHANNEL			= (1 << 27),
225 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
226 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
227 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
228 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
229 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
230 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
231 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
232 	IB_DEVICE_RDMA_NETDEV_OPA_VNIC		= (1ULL << 35),
233 	/* The device supports padding incoming writes to cacheline. */
234 	IB_DEVICE_PCI_WRITE_END_PADDING		= (1ULL << 36),
235 };
236 
237 enum ib_signature_prot_cap {
238 	IB_PROT_T10DIF_TYPE_1 = 1,
239 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
240 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
241 };
242 
243 enum ib_signature_guard_cap {
244 	IB_GUARD_T10DIF_CRC	= 1,
245 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
246 };
247 
248 enum ib_atomic_cap {
249 	IB_ATOMIC_NONE,
250 	IB_ATOMIC_HCA,
251 	IB_ATOMIC_GLOB
252 };
253 
254 enum ib_odp_general_cap_bits {
255 	IB_ODP_SUPPORT		= 1 << 0,
256 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
257 };
258 
259 enum ib_odp_transport_cap_bits {
260 	IB_ODP_SUPPORT_SEND	= 1 << 0,
261 	IB_ODP_SUPPORT_RECV	= 1 << 1,
262 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
263 	IB_ODP_SUPPORT_READ	= 1 << 3,
264 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
265 };
266 
267 struct ib_odp_caps {
268 	uint64_t general_caps;
269 	struct {
270 		uint32_t  rc_odp_caps;
271 		uint32_t  uc_odp_caps;
272 		uint32_t  ud_odp_caps;
273 	} per_transport_caps;
274 };
275 
276 struct ib_rss_caps {
277 	/* Corresponding bit will be set if qp type from
278 	 * 'enum ib_qp_type' is supported, e.g.
279 	 * supported_qpts |= 1 << IB_QPT_UD
280 	 */
281 	u32 supported_qpts;
282 	u32 max_rwq_indirection_tables;
283 	u32 max_rwq_indirection_table_size;
284 };
285 
286 enum ib_tm_cap_flags {
287 	/*  Support tag matching on RC transport */
288 	IB_TM_CAP_RC		    = 1 << 0,
289 };
290 
291 struct ib_tm_caps {
292 	/* Max size of RNDV header */
293 	u32 max_rndv_hdr_size;
294 	/* Max number of entries in tag matching list */
295 	u32 max_num_tags;
296 	/* From enum ib_tm_cap_flags */
297 	u32 flags;
298 	/* Max number of outstanding list operations */
299 	u32 max_ops;
300 	/* Max number of SGE in tag matching entry */
301 	u32 max_sge;
302 };
303 
304 struct ib_cq_init_attr {
305 	unsigned int	cqe;
306 	int		comp_vector;
307 	u32		flags;
308 };
309 
310 enum ib_cq_attr_mask {
311 	IB_CQ_MODERATE = 1 << 0,
312 };
313 
314 struct ib_cq_caps {
315 	u16     max_cq_moderation_count;
316 	u16     max_cq_moderation_period;
317 };
318 
319 struct ib_device_attr {
320 	u64			fw_ver;
321 	__be64			sys_image_guid;
322 	u64			max_mr_size;
323 	u64			page_size_cap;
324 	u32			vendor_id;
325 	u32			vendor_part_id;
326 	u32			hw_ver;
327 	int			max_qp;
328 	int			max_qp_wr;
329 	u64			device_cap_flags;
330 	int			max_sge;
331 	int			max_sge_rd;
332 	int			max_cq;
333 	int			max_cqe;
334 	int			max_mr;
335 	int			max_pd;
336 	int			max_qp_rd_atom;
337 	int			max_ee_rd_atom;
338 	int			max_res_rd_atom;
339 	int			max_qp_init_rd_atom;
340 	int			max_ee_init_rd_atom;
341 	enum ib_atomic_cap	atomic_cap;
342 	enum ib_atomic_cap	masked_atomic_cap;
343 	int			max_ee;
344 	int			max_rdd;
345 	int			max_mw;
346 	int			max_raw_ipv6_qp;
347 	int			max_raw_ethy_qp;
348 	int			max_mcast_grp;
349 	int			max_mcast_qp_attach;
350 	int			max_total_mcast_qp_attach;
351 	int			max_ah;
352 	int			max_fmr;
353 	int			max_map_per_fmr;
354 	int			max_srq;
355 	int			max_srq_wr;
356 	int			max_srq_sge;
357 	unsigned int		max_fast_reg_page_list_len;
358 	u16			max_pkeys;
359 	u8			local_ca_ack_delay;
360 	int			sig_prot_cap;
361 	int			sig_guard_cap;
362 	struct ib_odp_caps	odp_caps;
363 	uint64_t		timestamp_mask;
364 	uint64_t		hca_core_clock; /* in KHZ */
365 	struct ib_rss_caps	rss_caps;
366 	u32			max_wq_type_rq;
367 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
368 	struct ib_tm_caps	tm_caps;
369 	struct ib_cq_caps       cq_caps;
370 };
371 
372 enum ib_mtu {
373 	IB_MTU_256  = 1,
374 	IB_MTU_512  = 2,
375 	IB_MTU_1024 = 3,
376 	IB_MTU_2048 = 4,
377 	IB_MTU_4096 = 5
378 };
379 
380 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
381 {
382 	switch (mtu) {
383 	case IB_MTU_256:  return  256;
384 	case IB_MTU_512:  return  512;
385 	case IB_MTU_1024: return 1024;
386 	case IB_MTU_2048: return 2048;
387 	case IB_MTU_4096: return 4096;
388 	default: 	  return -1;
389 	}
390 }
391 
392 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
393 {
394 	if (mtu >= 4096)
395 		return IB_MTU_4096;
396 	else if (mtu >= 2048)
397 		return IB_MTU_2048;
398 	else if (mtu >= 1024)
399 		return IB_MTU_1024;
400 	else if (mtu >= 512)
401 		return IB_MTU_512;
402 	else
403 		return IB_MTU_256;
404 }
405 
406 enum ib_port_state {
407 	IB_PORT_NOP		= 0,
408 	IB_PORT_DOWN		= 1,
409 	IB_PORT_INIT		= 2,
410 	IB_PORT_ARMED		= 3,
411 	IB_PORT_ACTIVE		= 4,
412 	IB_PORT_ACTIVE_DEFER	= 5
413 };
414 
415 enum ib_port_cap_flags {
416 	IB_PORT_SM				= 1 <<  1,
417 	IB_PORT_NOTICE_SUP			= 1 <<  2,
418 	IB_PORT_TRAP_SUP			= 1 <<  3,
419 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
420 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
421 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
422 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
423 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
424 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
425 	IB_PORT_SM_DISABLED			= 1 << 10,
426 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
427 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
428 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
429 	IB_PORT_CM_SUP				= 1 << 16,
430 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
431 	IB_PORT_REINIT_SUP			= 1 << 18,
432 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
433 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
434 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
435 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
436 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
437 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
438 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
439 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
440 };
441 
442 enum ib_port_width {
443 	IB_WIDTH_1X	= 1,
444 	IB_WIDTH_4X	= 2,
445 	IB_WIDTH_8X	= 4,
446 	IB_WIDTH_12X	= 8
447 };
448 
449 static inline int ib_width_enum_to_int(enum ib_port_width width)
450 {
451 	switch (width) {
452 	case IB_WIDTH_1X:  return  1;
453 	case IB_WIDTH_4X:  return  4;
454 	case IB_WIDTH_8X:  return  8;
455 	case IB_WIDTH_12X: return 12;
456 	default: 	  return -1;
457 	}
458 }
459 
460 enum ib_port_speed {
461 	IB_SPEED_SDR	= 1,
462 	IB_SPEED_DDR	= 2,
463 	IB_SPEED_QDR	= 4,
464 	IB_SPEED_FDR10	= 8,
465 	IB_SPEED_FDR	= 16,
466 	IB_SPEED_EDR	= 32,
467 	IB_SPEED_HDR	= 64
468 };
469 
470 /**
471  * struct rdma_hw_stats
472  * @timestamp - Used by the core code to track when the last update was
473  * @lifespan - Used by the core code to determine how old the counters
474  *   should be before being updated again.  Stored in jiffies, defaults
475  *   to 10 milliseconds, drivers can override the default be specifying
476  *   their own value during their allocation routine.
477  * @name - Array of pointers to static names used for the counters in
478  *   directory.
479  * @num_counters - How many hardware counters there are.  If name is
480  *   shorter than this number, a kernel oops will result.  Driver authors
481  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
482  *   in their code to prevent this.
483  * @value - Array of u64 counters that are accessed by the sysfs code and
484  *   filled in by the drivers get_stats routine
485  */
486 struct rdma_hw_stats {
487 	unsigned long	timestamp;
488 	unsigned long	lifespan;
489 	const char * const *names;
490 	int		num_counters;
491 	u64		value[];
492 };
493 
494 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
495 /**
496  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
497  *   for drivers.
498  * @names - Array of static const char *
499  * @num_counters - How many elements in array
500  * @lifespan - How many milliseconds between updates
501  */
502 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
503 		const char * const *names, int num_counters,
504 		unsigned long lifespan)
505 {
506 	struct rdma_hw_stats *stats;
507 
508 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
509 			GFP_KERNEL);
510 	if (!stats)
511 		return NULL;
512 	stats->names = names;
513 	stats->num_counters = num_counters;
514 	stats->lifespan = msecs_to_jiffies(lifespan);
515 
516 	return stats;
517 }
518 
519 
520 /* Define bits for the various functionality this port needs to be supported by
521  * the core.
522  */
523 /* Management                           0x00000FFF */
524 #define RDMA_CORE_CAP_IB_MAD            0x00000001
525 #define RDMA_CORE_CAP_IB_SMI            0x00000002
526 #define RDMA_CORE_CAP_IB_CM             0x00000004
527 #define RDMA_CORE_CAP_IW_CM             0x00000008
528 #define RDMA_CORE_CAP_IB_SA             0x00000010
529 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
530 
531 /* Address format                       0x000FF000 */
532 #define RDMA_CORE_CAP_AF_IB             0x00001000
533 #define RDMA_CORE_CAP_ETH_AH            0x00002000
534 #define RDMA_CORE_CAP_OPA_AH            0x00004000
535 
536 /* Protocol                             0xFFF00000 */
537 #define RDMA_CORE_CAP_PROT_IB           0x00100000
538 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
539 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
540 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
541 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
542 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
543 
544 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
545 					| RDMA_CORE_CAP_IB_MAD \
546 					| RDMA_CORE_CAP_IB_SMI \
547 					| RDMA_CORE_CAP_IB_CM  \
548 					| RDMA_CORE_CAP_IB_SA  \
549 					| RDMA_CORE_CAP_AF_IB)
550 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
551 					| RDMA_CORE_CAP_IB_MAD  \
552 					| RDMA_CORE_CAP_IB_CM   \
553 					| RDMA_CORE_CAP_AF_IB   \
554 					| RDMA_CORE_CAP_ETH_AH)
555 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
556 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
557 					| RDMA_CORE_CAP_IB_MAD  \
558 					| RDMA_CORE_CAP_IB_CM   \
559 					| RDMA_CORE_CAP_AF_IB   \
560 					| RDMA_CORE_CAP_ETH_AH)
561 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
562 					| RDMA_CORE_CAP_IW_CM)
563 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
564 					| RDMA_CORE_CAP_OPA_MAD)
565 
566 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
567 
568 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
569 
570 struct ib_port_attr {
571 	u64			subnet_prefix;
572 	enum ib_port_state	state;
573 	enum ib_mtu		max_mtu;
574 	enum ib_mtu		active_mtu;
575 	int			gid_tbl_len;
576 	u32			port_cap_flags;
577 	u32			max_msg_sz;
578 	u32			bad_pkey_cntr;
579 	u32			qkey_viol_cntr;
580 	u16			pkey_tbl_len;
581 	u32			sm_lid;
582 	u32			lid;
583 	u8			lmc;
584 	u8			max_vl_num;
585 	u8			sm_sl;
586 	u8			subnet_timeout;
587 	u8			init_type_reply;
588 	u8			active_width;
589 	u8			active_speed;
590 	u8                      phys_state;
591 	bool			grh_required;
592 };
593 
594 enum ib_device_modify_flags {
595 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
596 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
597 };
598 
599 #define IB_DEVICE_NODE_DESC_MAX 64
600 
601 struct ib_device_modify {
602 	u64	sys_image_guid;
603 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
604 };
605 
606 enum ib_port_modify_flags {
607 	IB_PORT_SHUTDOWN		= 1,
608 	IB_PORT_INIT_TYPE		= (1<<2),
609 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
610 	IB_PORT_OPA_MASK_CHG		= (1<<4)
611 };
612 
613 struct ib_port_modify {
614 	u32	set_port_cap_mask;
615 	u32	clr_port_cap_mask;
616 	u8	init_type;
617 };
618 
619 enum ib_event_type {
620 	IB_EVENT_CQ_ERR,
621 	IB_EVENT_QP_FATAL,
622 	IB_EVENT_QP_REQ_ERR,
623 	IB_EVENT_QP_ACCESS_ERR,
624 	IB_EVENT_COMM_EST,
625 	IB_EVENT_SQ_DRAINED,
626 	IB_EVENT_PATH_MIG,
627 	IB_EVENT_PATH_MIG_ERR,
628 	IB_EVENT_DEVICE_FATAL,
629 	IB_EVENT_PORT_ACTIVE,
630 	IB_EVENT_PORT_ERR,
631 	IB_EVENT_LID_CHANGE,
632 	IB_EVENT_PKEY_CHANGE,
633 	IB_EVENT_SM_CHANGE,
634 	IB_EVENT_SRQ_ERR,
635 	IB_EVENT_SRQ_LIMIT_REACHED,
636 	IB_EVENT_QP_LAST_WQE_REACHED,
637 	IB_EVENT_CLIENT_REREGISTER,
638 	IB_EVENT_GID_CHANGE,
639 	IB_EVENT_WQ_FATAL,
640 };
641 
642 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
643 
644 struct ib_event {
645 	struct ib_device	*device;
646 	union {
647 		struct ib_cq	*cq;
648 		struct ib_qp	*qp;
649 		struct ib_srq	*srq;
650 		struct ib_wq	*wq;
651 		u8		port_num;
652 	} element;
653 	enum ib_event_type	event;
654 };
655 
656 struct ib_event_handler {
657 	struct ib_device *device;
658 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
659 	struct list_head  list;
660 };
661 
662 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
663 	do {							\
664 		(_ptr)->device  = _device;			\
665 		(_ptr)->handler = _handler;			\
666 		INIT_LIST_HEAD(&(_ptr)->list);			\
667 	} while (0)
668 
669 struct ib_global_route {
670 	union ib_gid	dgid;
671 	u32		flow_label;
672 	u8		sgid_index;
673 	u8		hop_limit;
674 	u8		traffic_class;
675 };
676 
677 struct ib_grh {
678 	__be32		version_tclass_flow;
679 	__be16		paylen;
680 	u8		next_hdr;
681 	u8		hop_limit;
682 	union ib_gid	sgid;
683 	union ib_gid	dgid;
684 };
685 
686 union rdma_network_hdr {
687 	struct ib_grh ibgrh;
688 	struct {
689 		/* The IB spec states that if it's IPv4, the header
690 		 * is located in the last 20 bytes of the header.
691 		 */
692 		u8		reserved[20];
693 		struct iphdr	roce4grh;
694 	};
695 };
696 
697 #define IB_QPN_MASK		0xFFFFFF
698 
699 enum {
700 	IB_MULTICAST_QPN = 0xffffff
701 };
702 
703 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
704 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
705 
706 enum ib_ah_flags {
707 	IB_AH_GRH	= 1
708 };
709 
710 enum ib_rate {
711 	IB_RATE_PORT_CURRENT = 0,
712 	IB_RATE_2_5_GBPS = 2,
713 	IB_RATE_5_GBPS   = 5,
714 	IB_RATE_10_GBPS  = 3,
715 	IB_RATE_20_GBPS  = 6,
716 	IB_RATE_30_GBPS  = 4,
717 	IB_RATE_40_GBPS  = 7,
718 	IB_RATE_60_GBPS  = 8,
719 	IB_RATE_80_GBPS  = 9,
720 	IB_RATE_120_GBPS = 10,
721 	IB_RATE_14_GBPS  = 11,
722 	IB_RATE_56_GBPS  = 12,
723 	IB_RATE_112_GBPS = 13,
724 	IB_RATE_168_GBPS = 14,
725 	IB_RATE_25_GBPS  = 15,
726 	IB_RATE_100_GBPS = 16,
727 	IB_RATE_200_GBPS = 17,
728 	IB_RATE_300_GBPS = 18
729 };
730 
731 /**
732  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
733  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
734  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
735  * @rate: rate to convert.
736  */
737 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
738 
739 /**
740  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
741  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
742  * @rate: rate to convert.
743  */
744 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
745 
746 
747 /**
748  * enum ib_mr_type - memory region type
749  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
750  *                            normal registration
751  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
752  *                            signature operations (data-integrity
753  *                            capable regions)
754  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
755  *                            register any arbitrary sg lists (without
756  *                            the normal mr constraints - see
757  *                            ib_map_mr_sg)
758  */
759 enum ib_mr_type {
760 	IB_MR_TYPE_MEM_REG,
761 	IB_MR_TYPE_SIGNATURE,
762 	IB_MR_TYPE_SG_GAPS,
763 };
764 
765 /**
766  * Signature types
767  * IB_SIG_TYPE_NONE: Unprotected.
768  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
769  */
770 enum ib_signature_type {
771 	IB_SIG_TYPE_NONE,
772 	IB_SIG_TYPE_T10_DIF,
773 };
774 
775 /**
776  * Signature T10-DIF block-guard types
777  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
778  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
779  */
780 enum ib_t10_dif_bg_type {
781 	IB_T10DIF_CRC,
782 	IB_T10DIF_CSUM
783 };
784 
785 /**
786  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
787  *     domain.
788  * @bg_type: T10-DIF block guard type (CRC|CSUM)
789  * @pi_interval: protection information interval.
790  * @bg: seed of guard computation.
791  * @app_tag: application tag of guard block
792  * @ref_tag: initial guard block reference tag.
793  * @ref_remap: Indicate wethear the reftag increments each block
794  * @app_escape: Indicate to skip block check if apptag=0xffff
795  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
796  * @apptag_check_mask: check bitmask of application tag.
797  */
798 struct ib_t10_dif_domain {
799 	enum ib_t10_dif_bg_type bg_type;
800 	u16			pi_interval;
801 	u16			bg;
802 	u16			app_tag;
803 	u32			ref_tag;
804 	bool			ref_remap;
805 	bool			app_escape;
806 	bool			ref_escape;
807 	u16			apptag_check_mask;
808 };
809 
810 /**
811  * struct ib_sig_domain - Parameters for signature domain
812  * @sig_type: specific signauture type
813  * @sig: union of all signature domain attributes that may
814  *     be used to set domain layout.
815  */
816 struct ib_sig_domain {
817 	enum ib_signature_type sig_type;
818 	union {
819 		struct ib_t10_dif_domain dif;
820 	} sig;
821 };
822 
823 /**
824  * struct ib_sig_attrs - Parameters for signature handover operation
825  * @check_mask: bitmask for signature byte check (8 bytes)
826  * @mem: memory domain layout desciptor.
827  * @wire: wire domain layout desciptor.
828  */
829 struct ib_sig_attrs {
830 	u8			check_mask;
831 	struct ib_sig_domain	mem;
832 	struct ib_sig_domain	wire;
833 };
834 
835 enum ib_sig_err_type {
836 	IB_SIG_BAD_GUARD,
837 	IB_SIG_BAD_REFTAG,
838 	IB_SIG_BAD_APPTAG,
839 };
840 
841 /**
842  * struct ib_sig_err - signature error descriptor
843  */
844 struct ib_sig_err {
845 	enum ib_sig_err_type	err_type;
846 	u32			expected;
847 	u32			actual;
848 	u64			sig_err_offset;
849 	u32			key;
850 };
851 
852 enum ib_mr_status_check {
853 	IB_MR_CHECK_SIG_STATUS = 1,
854 };
855 
856 /**
857  * struct ib_mr_status - Memory region status container
858  *
859  * @fail_status: Bitmask of MR checks status. For each
860  *     failed check a corresponding status bit is set.
861  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
862  *     failure.
863  */
864 struct ib_mr_status {
865 	u32		    fail_status;
866 	struct ib_sig_err   sig_err;
867 };
868 
869 /**
870  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
871  * enum.
872  * @mult: multiple to convert.
873  */
874 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
875 
876 enum rdma_ah_attr_type {
877 	RDMA_AH_ATTR_TYPE_IB,
878 	RDMA_AH_ATTR_TYPE_ROCE,
879 	RDMA_AH_ATTR_TYPE_OPA,
880 };
881 
882 struct ib_ah_attr {
883 	u16			dlid;
884 	u8			src_path_bits;
885 };
886 
887 struct roce_ah_attr {
888 	u8			dmac[ETH_ALEN];
889 };
890 
891 struct opa_ah_attr {
892 	u32			dlid;
893 	u8			src_path_bits;
894 	bool			make_grd;
895 };
896 
897 struct rdma_ah_attr {
898 	struct ib_global_route	grh;
899 	u8			sl;
900 	u8			static_rate;
901 	u8			port_num;
902 	u8			ah_flags;
903 	enum rdma_ah_attr_type type;
904 	union {
905 		struct ib_ah_attr ib;
906 		struct roce_ah_attr roce;
907 		struct opa_ah_attr opa;
908 	};
909 };
910 
911 enum ib_wc_status {
912 	IB_WC_SUCCESS,
913 	IB_WC_LOC_LEN_ERR,
914 	IB_WC_LOC_QP_OP_ERR,
915 	IB_WC_LOC_EEC_OP_ERR,
916 	IB_WC_LOC_PROT_ERR,
917 	IB_WC_WR_FLUSH_ERR,
918 	IB_WC_MW_BIND_ERR,
919 	IB_WC_BAD_RESP_ERR,
920 	IB_WC_LOC_ACCESS_ERR,
921 	IB_WC_REM_INV_REQ_ERR,
922 	IB_WC_REM_ACCESS_ERR,
923 	IB_WC_REM_OP_ERR,
924 	IB_WC_RETRY_EXC_ERR,
925 	IB_WC_RNR_RETRY_EXC_ERR,
926 	IB_WC_LOC_RDD_VIOL_ERR,
927 	IB_WC_REM_INV_RD_REQ_ERR,
928 	IB_WC_REM_ABORT_ERR,
929 	IB_WC_INV_EECN_ERR,
930 	IB_WC_INV_EEC_STATE_ERR,
931 	IB_WC_FATAL_ERR,
932 	IB_WC_RESP_TIMEOUT_ERR,
933 	IB_WC_GENERAL_ERR
934 };
935 
936 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
937 
938 enum ib_wc_opcode {
939 	IB_WC_SEND,
940 	IB_WC_RDMA_WRITE,
941 	IB_WC_RDMA_READ,
942 	IB_WC_COMP_SWAP,
943 	IB_WC_FETCH_ADD,
944 	IB_WC_LSO,
945 	IB_WC_LOCAL_INV,
946 	IB_WC_REG_MR,
947 	IB_WC_MASKED_COMP_SWAP,
948 	IB_WC_MASKED_FETCH_ADD,
949 /*
950  * Set value of IB_WC_RECV so consumers can test if a completion is a
951  * receive by testing (opcode & IB_WC_RECV).
952  */
953 	IB_WC_RECV			= 1 << 7,
954 	IB_WC_RECV_RDMA_WITH_IMM
955 };
956 
957 enum ib_wc_flags {
958 	IB_WC_GRH		= 1,
959 	IB_WC_WITH_IMM		= (1<<1),
960 	IB_WC_WITH_INVALIDATE	= (1<<2),
961 	IB_WC_IP_CSUM_OK	= (1<<3),
962 	IB_WC_WITH_SMAC		= (1<<4),
963 	IB_WC_WITH_VLAN		= (1<<5),
964 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
965 };
966 
967 struct ib_wc {
968 	union {
969 		u64		wr_id;
970 		struct ib_cqe	*wr_cqe;
971 	};
972 	enum ib_wc_status	status;
973 	enum ib_wc_opcode	opcode;
974 	u32			vendor_err;
975 	u32			byte_len;
976 	struct ib_qp	       *qp;
977 	union {
978 		__be32		imm_data;
979 		u32		invalidate_rkey;
980 	} ex;
981 	u32			src_qp;
982 	u32			slid;
983 	int			wc_flags;
984 	u16			pkey_index;
985 	u8			sl;
986 	u8			dlid_path_bits;
987 	u8			port_num;	/* valid only for DR SMPs on switches */
988 	u8			smac[ETH_ALEN];
989 	u16			vlan_id;
990 	u8			network_hdr_type;
991 };
992 
993 enum ib_cq_notify_flags {
994 	IB_CQ_SOLICITED			= 1 << 0,
995 	IB_CQ_NEXT_COMP			= 1 << 1,
996 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
997 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
998 };
999 
1000 enum ib_srq_type {
1001 	IB_SRQT_BASIC,
1002 	IB_SRQT_XRC,
1003 	IB_SRQT_TM,
1004 };
1005 
1006 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1007 {
1008 	return srq_type == IB_SRQT_XRC ||
1009 	       srq_type == IB_SRQT_TM;
1010 }
1011 
1012 enum ib_srq_attr_mask {
1013 	IB_SRQ_MAX_WR	= 1 << 0,
1014 	IB_SRQ_LIMIT	= 1 << 1,
1015 };
1016 
1017 struct ib_srq_attr {
1018 	u32	max_wr;
1019 	u32	max_sge;
1020 	u32	srq_limit;
1021 };
1022 
1023 struct ib_srq_init_attr {
1024 	void		      (*event_handler)(struct ib_event *, void *);
1025 	void		       *srq_context;
1026 	struct ib_srq_attr	attr;
1027 	enum ib_srq_type	srq_type;
1028 
1029 	struct {
1030 		struct ib_cq   *cq;
1031 		union {
1032 			struct {
1033 				struct ib_xrcd *xrcd;
1034 			} xrc;
1035 
1036 			struct {
1037 				u32		max_num_tags;
1038 			} tag_matching;
1039 		};
1040 	} ext;
1041 };
1042 
1043 struct ib_qp_cap {
1044 	u32	max_send_wr;
1045 	u32	max_recv_wr;
1046 	u32	max_send_sge;
1047 	u32	max_recv_sge;
1048 	u32	max_inline_data;
1049 
1050 	/*
1051 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1052 	 * ib_create_qp() will calculate the right amount of neededed WRs
1053 	 * and MRs based on this.
1054 	 */
1055 	u32	max_rdma_ctxs;
1056 };
1057 
1058 enum ib_sig_type {
1059 	IB_SIGNAL_ALL_WR,
1060 	IB_SIGNAL_REQ_WR
1061 };
1062 
1063 enum ib_qp_type {
1064 	/*
1065 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1066 	 * here (and in that order) since the MAD layer uses them as
1067 	 * indices into a 2-entry table.
1068 	 */
1069 	IB_QPT_SMI,
1070 	IB_QPT_GSI,
1071 
1072 	IB_QPT_RC,
1073 	IB_QPT_UC,
1074 	IB_QPT_UD,
1075 	IB_QPT_RAW_IPV6,
1076 	IB_QPT_RAW_ETHERTYPE,
1077 	IB_QPT_RAW_PACKET = 8,
1078 	IB_QPT_XRC_INI = 9,
1079 	IB_QPT_XRC_TGT,
1080 	IB_QPT_MAX,
1081 	IB_QPT_DRIVER = 0xFF,
1082 	/* Reserve a range for qp types internal to the low level driver.
1083 	 * These qp types will not be visible at the IB core layer, so the
1084 	 * IB_QPT_MAX usages should not be affected in the core layer
1085 	 */
1086 	IB_QPT_RESERVED1 = 0x1000,
1087 	IB_QPT_RESERVED2,
1088 	IB_QPT_RESERVED3,
1089 	IB_QPT_RESERVED4,
1090 	IB_QPT_RESERVED5,
1091 	IB_QPT_RESERVED6,
1092 	IB_QPT_RESERVED7,
1093 	IB_QPT_RESERVED8,
1094 	IB_QPT_RESERVED9,
1095 	IB_QPT_RESERVED10,
1096 };
1097 
1098 enum ib_qp_create_flags {
1099 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1100 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1101 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1102 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1103 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1104 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1105 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1106 	/* FREE					= 1 << 7, */
1107 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1108 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1109 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1110 	IB_QP_CREATE_PCI_WRITE_END_PADDING	= 1 << 11,
1111 	/* reserve bits 26-31 for low level drivers' internal use */
1112 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1113 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1114 };
1115 
1116 /*
1117  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1118  * callback to destroy the passed in QP.
1119  */
1120 
1121 struct ib_qp_init_attr {
1122 	void                  (*event_handler)(struct ib_event *, void *);
1123 	void		       *qp_context;
1124 	struct ib_cq	       *send_cq;
1125 	struct ib_cq	       *recv_cq;
1126 	struct ib_srq	       *srq;
1127 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1128 	struct ib_qp_cap	cap;
1129 	enum ib_sig_type	sq_sig_type;
1130 	enum ib_qp_type		qp_type;
1131 	enum ib_qp_create_flags	create_flags;
1132 
1133 	/*
1134 	 * Only needed for special QP types, or when using the RW API.
1135 	 */
1136 	u8			port_num;
1137 	struct ib_rwq_ind_table *rwq_ind_tbl;
1138 	u32			source_qpn;
1139 };
1140 
1141 struct ib_qp_open_attr {
1142 	void                  (*event_handler)(struct ib_event *, void *);
1143 	void		       *qp_context;
1144 	u32			qp_num;
1145 	enum ib_qp_type		qp_type;
1146 };
1147 
1148 enum ib_rnr_timeout {
1149 	IB_RNR_TIMER_655_36 =  0,
1150 	IB_RNR_TIMER_000_01 =  1,
1151 	IB_RNR_TIMER_000_02 =  2,
1152 	IB_RNR_TIMER_000_03 =  3,
1153 	IB_RNR_TIMER_000_04 =  4,
1154 	IB_RNR_TIMER_000_06 =  5,
1155 	IB_RNR_TIMER_000_08 =  6,
1156 	IB_RNR_TIMER_000_12 =  7,
1157 	IB_RNR_TIMER_000_16 =  8,
1158 	IB_RNR_TIMER_000_24 =  9,
1159 	IB_RNR_TIMER_000_32 = 10,
1160 	IB_RNR_TIMER_000_48 = 11,
1161 	IB_RNR_TIMER_000_64 = 12,
1162 	IB_RNR_TIMER_000_96 = 13,
1163 	IB_RNR_TIMER_001_28 = 14,
1164 	IB_RNR_TIMER_001_92 = 15,
1165 	IB_RNR_TIMER_002_56 = 16,
1166 	IB_RNR_TIMER_003_84 = 17,
1167 	IB_RNR_TIMER_005_12 = 18,
1168 	IB_RNR_TIMER_007_68 = 19,
1169 	IB_RNR_TIMER_010_24 = 20,
1170 	IB_RNR_TIMER_015_36 = 21,
1171 	IB_RNR_TIMER_020_48 = 22,
1172 	IB_RNR_TIMER_030_72 = 23,
1173 	IB_RNR_TIMER_040_96 = 24,
1174 	IB_RNR_TIMER_061_44 = 25,
1175 	IB_RNR_TIMER_081_92 = 26,
1176 	IB_RNR_TIMER_122_88 = 27,
1177 	IB_RNR_TIMER_163_84 = 28,
1178 	IB_RNR_TIMER_245_76 = 29,
1179 	IB_RNR_TIMER_327_68 = 30,
1180 	IB_RNR_TIMER_491_52 = 31
1181 };
1182 
1183 enum ib_qp_attr_mask {
1184 	IB_QP_STATE			= 1,
1185 	IB_QP_CUR_STATE			= (1<<1),
1186 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1187 	IB_QP_ACCESS_FLAGS		= (1<<3),
1188 	IB_QP_PKEY_INDEX		= (1<<4),
1189 	IB_QP_PORT			= (1<<5),
1190 	IB_QP_QKEY			= (1<<6),
1191 	IB_QP_AV			= (1<<7),
1192 	IB_QP_PATH_MTU			= (1<<8),
1193 	IB_QP_TIMEOUT			= (1<<9),
1194 	IB_QP_RETRY_CNT			= (1<<10),
1195 	IB_QP_RNR_RETRY			= (1<<11),
1196 	IB_QP_RQ_PSN			= (1<<12),
1197 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1198 	IB_QP_ALT_PATH			= (1<<14),
1199 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1200 	IB_QP_SQ_PSN			= (1<<16),
1201 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1202 	IB_QP_PATH_MIG_STATE		= (1<<18),
1203 	IB_QP_CAP			= (1<<19),
1204 	IB_QP_DEST_QPN			= (1<<20),
1205 	IB_QP_RESERVED1			= (1<<21),
1206 	IB_QP_RESERVED2			= (1<<22),
1207 	IB_QP_RESERVED3			= (1<<23),
1208 	IB_QP_RESERVED4			= (1<<24),
1209 	IB_QP_RATE_LIMIT		= (1<<25),
1210 };
1211 
1212 enum ib_qp_state {
1213 	IB_QPS_RESET,
1214 	IB_QPS_INIT,
1215 	IB_QPS_RTR,
1216 	IB_QPS_RTS,
1217 	IB_QPS_SQD,
1218 	IB_QPS_SQE,
1219 	IB_QPS_ERR
1220 };
1221 
1222 enum ib_mig_state {
1223 	IB_MIG_MIGRATED,
1224 	IB_MIG_REARM,
1225 	IB_MIG_ARMED
1226 };
1227 
1228 enum ib_mw_type {
1229 	IB_MW_TYPE_1 = 1,
1230 	IB_MW_TYPE_2 = 2
1231 };
1232 
1233 struct ib_qp_attr {
1234 	enum ib_qp_state	qp_state;
1235 	enum ib_qp_state	cur_qp_state;
1236 	enum ib_mtu		path_mtu;
1237 	enum ib_mig_state	path_mig_state;
1238 	u32			qkey;
1239 	u32			rq_psn;
1240 	u32			sq_psn;
1241 	u32			dest_qp_num;
1242 	int			qp_access_flags;
1243 	struct ib_qp_cap	cap;
1244 	struct rdma_ah_attr	ah_attr;
1245 	struct rdma_ah_attr	alt_ah_attr;
1246 	u16			pkey_index;
1247 	u16			alt_pkey_index;
1248 	u8			en_sqd_async_notify;
1249 	u8			sq_draining;
1250 	u8			max_rd_atomic;
1251 	u8			max_dest_rd_atomic;
1252 	u8			min_rnr_timer;
1253 	u8			port_num;
1254 	u8			timeout;
1255 	u8			retry_cnt;
1256 	u8			rnr_retry;
1257 	u8			alt_port_num;
1258 	u8			alt_timeout;
1259 	u32			rate_limit;
1260 };
1261 
1262 enum ib_wr_opcode {
1263 	IB_WR_RDMA_WRITE,
1264 	IB_WR_RDMA_WRITE_WITH_IMM,
1265 	IB_WR_SEND,
1266 	IB_WR_SEND_WITH_IMM,
1267 	IB_WR_RDMA_READ,
1268 	IB_WR_ATOMIC_CMP_AND_SWP,
1269 	IB_WR_ATOMIC_FETCH_AND_ADD,
1270 	IB_WR_LSO,
1271 	IB_WR_SEND_WITH_INV,
1272 	IB_WR_RDMA_READ_WITH_INV,
1273 	IB_WR_LOCAL_INV,
1274 	IB_WR_REG_MR,
1275 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1276 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1277 	IB_WR_REG_SIG_MR,
1278 	/* reserve values for low level drivers' internal use.
1279 	 * These values will not be used at all in the ib core layer.
1280 	 */
1281 	IB_WR_RESERVED1 = 0xf0,
1282 	IB_WR_RESERVED2,
1283 	IB_WR_RESERVED3,
1284 	IB_WR_RESERVED4,
1285 	IB_WR_RESERVED5,
1286 	IB_WR_RESERVED6,
1287 	IB_WR_RESERVED7,
1288 	IB_WR_RESERVED8,
1289 	IB_WR_RESERVED9,
1290 	IB_WR_RESERVED10,
1291 };
1292 
1293 enum ib_send_flags {
1294 	IB_SEND_FENCE		= 1,
1295 	IB_SEND_SIGNALED	= (1<<1),
1296 	IB_SEND_SOLICITED	= (1<<2),
1297 	IB_SEND_INLINE		= (1<<3),
1298 	IB_SEND_IP_CSUM		= (1<<4),
1299 
1300 	/* reserve bits 26-31 for low level drivers' internal use */
1301 	IB_SEND_RESERVED_START	= (1 << 26),
1302 	IB_SEND_RESERVED_END	= (1 << 31),
1303 };
1304 
1305 struct ib_sge {
1306 	u64	addr;
1307 	u32	length;
1308 	u32	lkey;
1309 };
1310 
1311 struct ib_cqe {
1312 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1313 };
1314 
1315 struct ib_send_wr {
1316 	struct ib_send_wr      *next;
1317 	union {
1318 		u64		wr_id;
1319 		struct ib_cqe	*wr_cqe;
1320 	};
1321 	struct ib_sge	       *sg_list;
1322 	int			num_sge;
1323 	enum ib_wr_opcode	opcode;
1324 	int			send_flags;
1325 	union {
1326 		__be32		imm_data;
1327 		u32		invalidate_rkey;
1328 	} ex;
1329 };
1330 
1331 struct ib_rdma_wr {
1332 	struct ib_send_wr	wr;
1333 	u64			remote_addr;
1334 	u32			rkey;
1335 };
1336 
1337 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1338 {
1339 	return container_of(wr, struct ib_rdma_wr, wr);
1340 }
1341 
1342 struct ib_atomic_wr {
1343 	struct ib_send_wr	wr;
1344 	u64			remote_addr;
1345 	u64			compare_add;
1346 	u64			swap;
1347 	u64			compare_add_mask;
1348 	u64			swap_mask;
1349 	u32			rkey;
1350 };
1351 
1352 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1353 {
1354 	return container_of(wr, struct ib_atomic_wr, wr);
1355 }
1356 
1357 struct ib_ud_wr {
1358 	struct ib_send_wr	wr;
1359 	struct ib_ah		*ah;
1360 	void			*header;
1361 	int			hlen;
1362 	int			mss;
1363 	u32			remote_qpn;
1364 	u32			remote_qkey;
1365 	u16			pkey_index; /* valid for GSI only */
1366 	u8			port_num;   /* valid for DR SMPs on switch only */
1367 };
1368 
1369 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1370 {
1371 	return container_of(wr, struct ib_ud_wr, wr);
1372 }
1373 
1374 struct ib_reg_wr {
1375 	struct ib_send_wr	wr;
1376 	struct ib_mr		*mr;
1377 	u32			key;
1378 	int			access;
1379 };
1380 
1381 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1382 {
1383 	return container_of(wr, struct ib_reg_wr, wr);
1384 }
1385 
1386 struct ib_sig_handover_wr {
1387 	struct ib_send_wr	wr;
1388 	struct ib_sig_attrs    *sig_attrs;
1389 	struct ib_mr	       *sig_mr;
1390 	int			access_flags;
1391 	struct ib_sge	       *prot;
1392 };
1393 
1394 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1395 {
1396 	return container_of(wr, struct ib_sig_handover_wr, wr);
1397 }
1398 
1399 struct ib_recv_wr {
1400 	struct ib_recv_wr      *next;
1401 	union {
1402 		u64		wr_id;
1403 		struct ib_cqe	*wr_cqe;
1404 	};
1405 	struct ib_sge	       *sg_list;
1406 	int			num_sge;
1407 };
1408 
1409 enum ib_access_flags {
1410 	IB_ACCESS_LOCAL_WRITE	= 1,
1411 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1412 	IB_ACCESS_REMOTE_READ	= (1<<2),
1413 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1414 	IB_ACCESS_MW_BIND	= (1<<4),
1415 	IB_ZERO_BASED		= (1<<5),
1416 	IB_ACCESS_ON_DEMAND     = (1<<6),
1417 	IB_ACCESS_HUGETLB	= (1<<7),
1418 };
1419 
1420 /*
1421  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1422  * are hidden here instead of a uapi header!
1423  */
1424 enum ib_mr_rereg_flags {
1425 	IB_MR_REREG_TRANS	= 1,
1426 	IB_MR_REREG_PD		= (1<<1),
1427 	IB_MR_REREG_ACCESS	= (1<<2),
1428 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1429 };
1430 
1431 struct ib_fmr_attr {
1432 	int	max_pages;
1433 	int	max_maps;
1434 	u8	page_shift;
1435 };
1436 
1437 struct ib_umem;
1438 
1439 enum rdma_remove_reason {
1440 	/* Userspace requested uobject deletion. Call could fail */
1441 	RDMA_REMOVE_DESTROY,
1442 	/* Context deletion. This call should delete the actual object itself */
1443 	RDMA_REMOVE_CLOSE,
1444 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1445 	RDMA_REMOVE_DRIVER_REMOVE,
1446 	/* Context is being cleaned-up, but commit was just completed */
1447 	RDMA_REMOVE_DURING_CLEANUP,
1448 };
1449 
1450 struct ib_rdmacg_object {
1451 #ifdef CONFIG_CGROUP_RDMA
1452 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1453 #endif
1454 };
1455 
1456 struct ib_ucontext {
1457 	struct ib_device       *device;
1458 	struct ib_uverbs_file  *ufile;
1459 	int			closing;
1460 
1461 	/* locking the uobjects_list */
1462 	struct mutex		uobjects_lock;
1463 	struct list_head	uobjects;
1464 	/* protects cleanup process from other actions */
1465 	struct rw_semaphore	cleanup_rwsem;
1466 	enum rdma_remove_reason cleanup_reason;
1467 
1468 	struct pid             *tgid;
1469 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1470 	struct rb_root_cached   umem_tree;
1471 	/*
1472 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1473 	 * mmu notifiers registration.
1474 	 */
1475 	struct rw_semaphore	umem_rwsem;
1476 	void (*invalidate_range)(struct ib_umem *umem,
1477 				 unsigned long start, unsigned long end);
1478 
1479 	struct mmu_notifier	mn;
1480 	atomic_t		notifier_count;
1481 	/* A list of umems that don't have private mmu notifier counters yet. */
1482 	struct list_head	no_private_counters;
1483 	int                     odp_mrs_count;
1484 #endif
1485 
1486 	struct ib_rdmacg_object	cg_obj;
1487 };
1488 
1489 struct ib_uobject {
1490 	u64			user_handle;	/* handle given to us by userspace */
1491 	struct ib_ucontext     *context;	/* associated user context */
1492 	void		       *object;		/* containing object */
1493 	struct list_head	list;		/* link to context's list */
1494 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1495 	int			id;		/* index into kernel idr */
1496 	struct kref		ref;
1497 	atomic_t		usecnt;		/* protects exclusive access */
1498 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1499 
1500 	const struct uverbs_obj_type *type;
1501 };
1502 
1503 struct ib_uobject_file {
1504 	struct ib_uobject	uobj;
1505 	/* ufile contains the lock between context release and file close */
1506 	struct ib_uverbs_file	*ufile;
1507 };
1508 
1509 struct ib_udata {
1510 	const void __user *inbuf;
1511 	void __user *outbuf;
1512 	size_t       inlen;
1513 	size_t       outlen;
1514 };
1515 
1516 struct ib_pd {
1517 	u32			local_dma_lkey;
1518 	u32			flags;
1519 	struct ib_device       *device;
1520 	struct ib_uobject      *uobject;
1521 	atomic_t          	usecnt; /* count all resources */
1522 
1523 	u32			unsafe_global_rkey;
1524 
1525 	/*
1526 	 * Implementation details of the RDMA core, don't use in drivers:
1527 	 */
1528 	struct ib_mr	       *__internal_mr;
1529 	struct rdma_restrack_entry res;
1530 };
1531 
1532 struct ib_xrcd {
1533 	struct ib_device       *device;
1534 	atomic_t		usecnt; /* count all exposed resources */
1535 	struct inode	       *inode;
1536 
1537 	struct mutex		tgt_qp_mutex;
1538 	struct list_head	tgt_qp_list;
1539 	/*
1540 	 * Implementation details of the RDMA core, don't use in drivers:
1541 	 */
1542 	struct rdma_restrack_entry res;
1543 };
1544 
1545 struct ib_ah {
1546 	struct ib_device	*device;
1547 	struct ib_pd		*pd;
1548 	struct ib_uobject	*uobject;
1549 	enum rdma_ah_attr_type	type;
1550 };
1551 
1552 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1553 
1554 enum ib_poll_context {
1555 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1556 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1557 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1558 };
1559 
1560 struct ib_cq {
1561 	struct ib_device       *device;
1562 	struct ib_uobject      *uobject;
1563 	ib_comp_handler   	comp_handler;
1564 	void                  (*event_handler)(struct ib_event *, void *);
1565 	void                   *cq_context;
1566 	int               	cqe;
1567 	atomic_t          	usecnt; /* count number of work queues */
1568 	enum ib_poll_context	poll_ctx;
1569 	struct ib_wc		*wc;
1570 	union {
1571 		struct irq_poll		iop;
1572 		struct work_struct	work;
1573 	};
1574 	/*
1575 	 * Implementation details of the RDMA core, don't use in drivers:
1576 	 */
1577 	struct rdma_restrack_entry res;
1578 };
1579 
1580 struct ib_srq {
1581 	struct ib_device       *device;
1582 	struct ib_pd	       *pd;
1583 	struct ib_uobject      *uobject;
1584 	void		      (*event_handler)(struct ib_event *, void *);
1585 	void		       *srq_context;
1586 	enum ib_srq_type	srq_type;
1587 	atomic_t		usecnt;
1588 
1589 	struct {
1590 		struct ib_cq   *cq;
1591 		union {
1592 			struct {
1593 				struct ib_xrcd *xrcd;
1594 				u32		srq_num;
1595 			} xrc;
1596 		};
1597 	} ext;
1598 };
1599 
1600 enum ib_raw_packet_caps {
1601 	/* Strip cvlan from incoming packet and report it in the matching work
1602 	 * completion is supported.
1603 	 */
1604 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1605 	/* Scatter FCS field of an incoming packet to host memory is supported.
1606 	 */
1607 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1608 	/* Checksum offloads are supported (for both send and receive). */
1609 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1610 	/* When a packet is received for an RQ with no receive WQEs, the
1611 	 * packet processing is delayed.
1612 	 */
1613 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1614 };
1615 
1616 enum ib_wq_type {
1617 	IB_WQT_RQ
1618 };
1619 
1620 enum ib_wq_state {
1621 	IB_WQS_RESET,
1622 	IB_WQS_RDY,
1623 	IB_WQS_ERR
1624 };
1625 
1626 struct ib_wq {
1627 	struct ib_device       *device;
1628 	struct ib_uobject      *uobject;
1629 	void		    *wq_context;
1630 	void		    (*event_handler)(struct ib_event *, void *);
1631 	struct ib_pd	       *pd;
1632 	struct ib_cq	       *cq;
1633 	u32		wq_num;
1634 	enum ib_wq_state       state;
1635 	enum ib_wq_type	wq_type;
1636 	atomic_t		usecnt;
1637 };
1638 
1639 enum ib_wq_flags {
1640 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1641 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1642 	IB_WQ_FLAGS_DELAY_DROP		= 1 << 2,
1643 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1644 };
1645 
1646 struct ib_wq_init_attr {
1647 	void		       *wq_context;
1648 	enum ib_wq_type	wq_type;
1649 	u32		max_wr;
1650 	u32		max_sge;
1651 	struct	ib_cq	       *cq;
1652 	void		    (*event_handler)(struct ib_event *, void *);
1653 	u32		create_flags; /* Use enum ib_wq_flags */
1654 };
1655 
1656 enum ib_wq_attr_mask {
1657 	IB_WQ_STATE		= 1 << 0,
1658 	IB_WQ_CUR_STATE		= 1 << 1,
1659 	IB_WQ_FLAGS		= 1 << 2,
1660 };
1661 
1662 struct ib_wq_attr {
1663 	enum	ib_wq_state	wq_state;
1664 	enum	ib_wq_state	curr_wq_state;
1665 	u32			flags; /* Use enum ib_wq_flags */
1666 	u32			flags_mask; /* Use enum ib_wq_flags */
1667 };
1668 
1669 struct ib_rwq_ind_table {
1670 	struct ib_device	*device;
1671 	struct ib_uobject      *uobject;
1672 	atomic_t		usecnt;
1673 	u32		ind_tbl_num;
1674 	u32		log_ind_tbl_size;
1675 	struct ib_wq	**ind_tbl;
1676 };
1677 
1678 struct ib_rwq_ind_table_init_attr {
1679 	u32		log_ind_tbl_size;
1680 	/* Each entry is a pointer to Receive Work Queue */
1681 	struct ib_wq	**ind_tbl;
1682 };
1683 
1684 enum port_pkey_state {
1685 	IB_PORT_PKEY_NOT_VALID = 0,
1686 	IB_PORT_PKEY_VALID = 1,
1687 	IB_PORT_PKEY_LISTED = 2,
1688 };
1689 
1690 struct ib_qp_security;
1691 
1692 struct ib_port_pkey {
1693 	enum port_pkey_state	state;
1694 	u16			pkey_index;
1695 	u8			port_num;
1696 	struct list_head	qp_list;
1697 	struct list_head	to_error_list;
1698 	struct ib_qp_security  *sec;
1699 };
1700 
1701 struct ib_ports_pkeys {
1702 	struct ib_port_pkey	main;
1703 	struct ib_port_pkey	alt;
1704 };
1705 
1706 struct ib_qp_security {
1707 	struct ib_qp	       *qp;
1708 	struct ib_device       *dev;
1709 	/* Hold this mutex when changing port and pkey settings. */
1710 	struct mutex		mutex;
1711 	struct ib_ports_pkeys  *ports_pkeys;
1712 	/* A list of all open shared QP handles.  Required to enforce security
1713 	 * properly for all users of a shared QP.
1714 	 */
1715 	struct list_head        shared_qp_list;
1716 	void                   *security;
1717 	bool			destroying;
1718 	atomic_t		error_list_count;
1719 	struct completion	error_complete;
1720 	int			error_comps_pending;
1721 };
1722 
1723 /*
1724  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1725  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1726  */
1727 struct ib_qp {
1728 	struct ib_device       *device;
1729 	struct ib_pd	       *pd;
1730 	struct ib_cq	       *send_cq;
1731 	struct ib_cq	       *recv_cq;
1732 	spinlock_t		mr_lock;
1733 	int			mrs_used;
1734 	struct list_head	rdma_mrs;
1735 	struct list_head	sig_mrs;
1736 	struct ib_srq	       *srq;
1737 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1738 	struct list_head	xrcd_list;
1739 
1740 	/* count times opened, mcast attaches, flow attaches */
1741 	atomic_t		usecnt;
1742 	struct list_head	open_list;
1743 	struct ib_qp           *real_qp;
1744 	struct ib_uobject      *uobject;
1745 	void                  (*event_handler)(struct ib_event *, void *);
1746 	void		       *qp_context;
1747 	u32			qp_num;
1748 	u32			max_write_sge;
1749 	u32			max_read_sge;
1750 	enum ib_qp_type		qp_type;
1751 	struct ib_rwq_ind_table *rwq_ind_tbl;
1752 	struct ib_qp_security  *qp_sec;
1753 	u8			port;
1754 
1755 	/*
1756 	 * Implementation details of the RDMA core, don't use in drivers:
1757 	 */
1758 	struct rdma_restrack_entry     res;
1759 };
1760 
1761 struct ib_mr {
1762 	struct ib_device  *device;
1763 	struct ib_pd	  *pd;
1764 	u32		   lkey;
1765 	u32		   rkey;
1766 	u64		   iova;
1767 	u64		   length;
1768 	unsigned int	   page_size;
1769 	bool		   need_inval;
1770 	union {
1771 		struct ib_uobject	*uobject;	/* user */
1772 		struct list_head	qp_entry;	/* FR */
1773 	};
1774 };
1775 
1776 struct ib_mw {
1777 	struct ib_device	*device;
1778 	struct ib_pd		*pd;
1779 	struct ib_uobject	*uobject;
1780 	u32			rkey;
1781 	enum ib_mw_type         type;
1782 };
1783 
1784 struct ib_fmr {
1785 	struct ib_device	*device;
1786 	struct ib_pd		*pd;
1787 	struct list_head	list;
1788 	u32			lkey;
1789 	u32			rkey;
1790 };
1791 
1792 /* Supported steering options */
1793 enum ib_flow_attr_type {
1794 	/* steering according to rule specifications */
1795 	IB_FLOW_ATTR_NORMAL		= 0x0,
1796 	/* default unicast and multicast rule -
1797 	 * receive all Eth traffic which isn't steered to any QP
1798 	 */
1799 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1800 	/* default multicast rule -
1801 	 * receive all Eth multicast traffic which isn't steered to any QP
1802 	 */
1803 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1804 	/* sniffer rule - receive all port traffic */
1805 	IB_FLOW_ATTR_SNIFFER		= 0x3
1806 };
1807 
1808 /* Supported steering header types */
1809 enum ib_flow_spec_type {
1810 	/* L2 headers*/
1811 	IB_FLOW_SPEC_ETH		= 0x20,
1812 	IB_FLOW_SPEC_IB			= 0x22,
1813 	/* L3 header*/
1814 	IB_FLOW_SPEC_IPV4		= 0x30,
1815 	IB_FLOW_SPEC_IPV6		= 0x31,
1816 	/* L4 headers*/
1817 	IB_FLOW_SPEC_TCP		= 0x40,
1818 	IB_FLOW_SPEC_UDP		= 0x41,
1819 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1820 	IB_FLOW_SPEC_INNER		= 0x100,
1821 	/* Actions */
1822 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1823 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1824 };
1825 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1826 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1827 
1828 /* Flow steering rule priority is set according to it's domain.
1829  * Lower domain value means higher priority.
1830  */
1831 enum ib_flow_domain {
1832 	IB_FLOW_DOMAIN_USER,
1833 	IB_FLOW_DOMAIN_ETHTOOL,
1834 	IB_FLOW_DOMAIN_RFS,
1835 	IB_FLOW_DOMAIN_NIC,
1836 	IB_FLOW_DOMAIN_NUM /* Must be last */
1837 };
1838 
1839 enum ib_flow_flags {
1840 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1841 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1842 };
1843 
1844 struct ib_flow_eth_filter {
1845 	u8	dst_mac[6];
1846 	u8	src_mac[6];
1847 	__be16	ether_type;
1848 	__be16	vlan_tag;
1849 	/* Must be last */
1850 	u8	real_sz[0];
1851 };
1852 
1853 struct ib_flow_spec_eth {
1854 	u32			  type;
1855 	u16			  size;
1856 	struct ib_flow_eth_filter val;
1857 	struct ib_flow_eth_filter mask;
1858 };
1859 
1860 struct ib_flow_ib_filter {
1861 	__be16 dlid;
1862 	__u8   sl;
1863 	/* Must be last */
1864 	u8	real_sz[0];
1865 };
1866 
1867 struct ib_flow_spec_ib {
1868 	u32			 type;
1869 	u16			 size;
1870 	struct ib_flow_ib_filter val;
1871 	struct ib_flow_ib_filter mask;
1872 };
1873 
1874 /* IPv4 header flags */
1875 enum ib_ipv4_flags {
1876 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1877 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1878 				    last have this flag set */
1879 };
1880 
1881 struct ib_flow_ipv4_filter {
1882 	__be32	src_ip;
1883 	__be32	dst_ip;
1884 	u8	proto;
1885 	u8	tos;
1886 	u8	ttl;
1887 	u8	flags;
1888 	/* Must be last */
1889 	u8	real_sz[0];
1890 };
1891 
1892 struct ib_flow_spec_ipv4 {
1893 	u32			   type;
1894 	u16			   size;
1895 	struct ib_flow_ipv4_filter val;
1896 	struct ib_flow_ipv4_filter mask;
1897 };
1898 
1899 struct ib_flow_ipv6_filter {
1900 	u8	src_ip[16];
1901 	u8	dst_ip[16];
1902 	__be32	flow_label;
1903 	u8	next_hdr;
1904 	u8	traffic_class;
1905 	u8	hop_limit;
1906 	/* Must be last */
1907 	u8	real_sz[0];
1908 };
1909 
1910 struct ib_flow_spec_ipv6 {
1911 	u32			   type;
1912 	u16			   size;
1913 	struct ib_flow_ipv6_filter val;
1914 	struct ib_flow_ipv6_filter mask;
1915 };
1916 
1917 struct ib_flow_tcp_udp_filter {
1918 	__be16	dst_port;
1919 	__be16	src_port;
1920 	/* Must be last */
1921 	u8	real_sz[0];
1922 };
1923 
1924 struct ib_flow_spec_tcp_udp {
1925 	u32			      type;
1926 	u16			      size;
1927 	struct ib_flow_tcp_udp_filter val;
1928 	struct ib_flow_tcp_udp_filter mask;
1929 };
1930 
1931 struct ib_flow_tunnel_filter {
1932 	__be32	tunnel_id;
1933 	u8	real_sz[0];
1934 };
1935 
1936 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1937  * the tunnel_id from val has the vni value
1938  */
1939 struct ib_flow_spec_tunnel {
1940 	u32			      type;
1941 	u16			      size;
1942 	struct ib_flow_tunnel_filter  val;
1943 	struct ib_flow_tunnel_filter  mask;
1944 };
1945 
1946 struct ib_flow_spec_action_tag {
1947 	enum ib_flow_spec_type	      type;
1948 	u16			      size;
1949 	u32                           tag_id;
1950 };
1951 
1952 struct ib_flow_spec_action_drop {
1953 	enum ib_flow_spec_type	      type;
1954 	u16			      size;
1955 };
1956 
1957 union ib_flow_spec {
1958 	struct {
1959 		u32			type;
1960 		u16			size;
1961 	};
1962 	struct ib_flow_spec_eth		eth;
1963 	struct ib_flow_spec_ib		ib;
1964 	struct ib_flow_spec_ipv4        ipv4;
1965 	struct ib_flow_spec_tcp_udp	tcp_udp;
1966 	struct ib_flow_spec_ipv6        ipv6;
1967 	struct ib_flow_spec_tunnel      tunnel;
1968 	struct ib_flow_spec_action_tag  flow_tag;
1969 	struct ib_flow_spec_action_drop drop;
1970 };
1971 
1972 struct ib_flow_attr {
1973 	enum ib_flow_attr_type type;
1974 	u16	     size;
1975 	u16	     priority;
1976 	u32	     flags;
1977 	u8	     num_of_specs;
1978 	u8	     port;
1979 	/* Following are the optional layers according to user request
1980 	 * struct ib_flow_spec_xxx
1981 	 * struct ib_flow_spec_yyy
1982 	 */
1983 };
1984 
1985 struct ib_flow {
1986 	struct ib_qp		*qp;
1987 	struct ib_uobject	*uobject;
1988 };
1989 
1990 struct ib_mad_hdr;
1991 struct ib_grh;
1992 
1993 enum ib_process_mad_flags {
1994 	IB_MAD_IGNORE_MKEY	= 1,
1995 	IB_MAD_IGNORE_BKEY	= 2,
1996 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1997 };
1998 
1999 enum ib_mad_result {
2000 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2001 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2002 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2003 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2004 };
2005 
2006 struct ib_port_cache {
2007 	u64		      subnet_prefix;
2008 	struct ib_pkey_cache  *pkey;
2009 	struct ib_gid_table   *gid;
2010 	u8                     lmc;
2011 	enum ib_port_state     port_state;
2012 };
2013 
2014 struct ib_cache {
2015 	rwlock_t                lock;
2016 	struct ib_event_handler event_handler;
2017 	struct ib_port_cache   *ports;
2018 };
2019 
2020 struct iw_cm_verbs;
2021 
2022 struct ib_port_immutable {
2023 	int                           pkey_tbl_len;
2024 	int                           gid_tbl_len;
2025 	u32                           core_cap_flags;
2026 	u32                           max_mad_size;
2027 };
2028 
2029 /* rdma netdev type - specifies protocol type */
2030 enum rdma_netdev_t {
2031 	RDMA_NETDEV_OPA_VNIC,
2032 	RDMA_NETDEV_IPOIB,
2033 };
2034 
2035 /**
2036  * struct rdma_netdev - rdma netdev
2037  * For cases where netstack interfacing is required.
2038  */
2039 struct rdma_netdev {
2040 	void              *clnt_priv;
2041 	struct ib_device  *hca;
2042 	u8                 port_num;
2043 
2044 	/* cleanup function must be specified */
2045 	void (*free_rdma_netdev)(struct net_device *netdev);
2046 
2047 	/* control functions */
2048 	void (*set_id)(struct net_device *netdev, int id);
2049 	/* send packet */
2050 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2051 		    struct ib_ah *address, u32 dqpn);
2052 	/* multicast */
2053 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2054 			    union ib_gid *gid, u16 mlid,
2055 			    int set_qkey, u32 qkey);
2056 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2057 			    union ib_gid *gid, u16 mlid);
2058 };
2059 
2060 struct ib_port_pkey_list {
2061 	/* Lock to hold while modifying the list. */
2062 	spinlock_t                    list_lock;
2063 	struct list_head              pkey_list;
2064 };
2065 
2066 struct ib_device {
2067 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2068 	struct device                *dma_device;
2069 
2070 	char                          name[IB_DEVICE_NAME_MAX];
2071 
2072 	struct list_head              event_handler_list;
2073 	spinlock_t                    event_handler_lock;
2074 
2075 	spinlock_t                    client_data_lock;
2076 	struct list_head              core_list;
2077 	/* Access to the client_data_list is protected by the client_data_lock
2078 	 * spinlock and the lists_rwsem read-write semaphore */
2079 	struct list_head              client_data_list;
2080 
2081 	struct ib_cache               cache;
2082 	/**
2083 	 * port_immutable is indexed by port number
2084 	 */
2085 	struct ib_port_immutable     *port_immutable;
2086 
2087 	int			      num_comp_vectors;
2088 
2089 	struct ib_port_pkey_list     *port_pkey_list;
2090 
2091 	struct iw_cm_verbs	     *iwcm;
2092 
2093 	/**
2094 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2095 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2096 	 *   core when the device is removed.  A lifespan of -1 in the return
2097 	 *   struct tells the core to set a default lifespan.
2098 	 */
2099 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
2100 						     u8 port_num);
2101 	/**
2102 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2103 	 * @index - The index in the value array we wish to have updated, or
2104 	 *   num_counters if we want all stats updated
2105 	 * Return codes -
2106 	 *   < 0 - Error, no counters updated
2107 	 *   index - Updated the single counter pointed to by index
2108 	 *   num_counters - Updated all counters (will reset the timestamp
2109 	 *     and prevent further calls for lifespan milliseconds)
2110 	 * Drivers are allowed to update all counters in leiu of just the
2111 	 *   one given in index at their option
2112 	 */
2113 	int		           (*get_hw_stats)(struct ib_device *device,
2114 						   struct rdma_hw_stats *stats,
2115 						   u8 port, int index);
2116 	int		           (*query_device)(struct ib_device *device,
2117 						   struct ib_device_attr *device_attr,
2118 						   struct ib_udata *udata);
2119 	int		           (*query_port)(struct ib_device *device,
2120 						 u8 port_num,
2121 						 struct ib_port_attr *port_attr);
2122 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
2123 						     u8 port_num);
2124 	/* When calling get_netdev, the HW vendor's driver should return the
2125 	 * net device of device @device at port @port_num or NULL if such
2126 	 * a net device doesn't exist. The vendor driver should call dev_hold
2127 	 * on this net device. The HW vendor's device driver must guarantee
2128 	 * that this function returns NULL before the net device reaches
2129 	 * NETDEV_UNREGISTER_FINAL state.
2130 	 */
2131 	struct net_device	  *(*get_netdev)(struct ib_device *device,
2132 						 u8 port_num);
2133 	int		           (*query_gid)(struct ib_device *device,
2134 						u8 port_num, int index,
2135 						union ib_gid *gid);
2136 	/* When calling add_gid, the HW vendor's driver should
2137 	 * add the gid of device @device at gid index @index of
2138 	 * port @port_num to be @gid. Meta-info of that gid (for example,
2139 	 * the network device related to this gid is available
2140 	 * at @attr. @context allows the HW vendor driver to store extra
2141 	 * information together with a GID entry. The HW vendor may allocate
2142 	 * memory to contain this information and store it in @context when a
2143 	 * new GID entry is written to. Params are consistent until the next
2144 	 * call of add_gid or delete_gid. The function should return 0 on
2145 	 * success or error otherwise. The function could be called
2146 	 * concurrently for different ports. This function is only called
2147 	 * when roce_gid_table is used.
2148 	 */
2149 	int		           (*add_gid)(struct ib_device *device,
2150 					      u8 port_num,
2151 					      unsigned int index,
2152 					      const union ib_gid *gid,
2153 					      const struct ib_gid_attr *attr,
2154 					      void **context);
2155 	/* When calling del_gid, the HW vendor's driver should delete the
2156 	 * gid of device @device at gid index @index of port @port_num.
2157 	 * Upon the deletion of a GID entry, the HW vendor must free any
2158 	 * allocated memory. The caller will clear @context afterwards.
2159 	 * This function is only called when roce_gid_table is used.
2160 	 */
2161 	int		           (*del_gid)(struct ib_device *device,
2162 					      u8 port_num,
2163 					      unsigned int index,
2164 					      void **context);
2165 	int		           (*query_pkey)(struct ib_device *device,
2166 						 u8 port_num, u16 index, u16 *pkey);
2167 	int		           (*modify_device)(struct ib_device *device,
2168 						    int device_modify_mask,
2169 						    struct ib_device_modify *device_modify);
2170 	int		           (*modify_port)(struct ib_device *device,
2171 						  u8 port_num, int port_modify_mask,
2172 						  struct ib_port_modify *port_modify);
2173 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
2174 						     struct ib_udata *udata);
2175 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
2176 	int                        (*mmap)(struct ib_ucontext *context,
2177 					   struct vm_area_struct *vma);
2178 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
2179 					       struct ib_ucontext *context,
2180 					       struct ib_udata *udata);
2181 	int                        (*dealloc_pd)(struct ib_pd *pd);
2182 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
2183 						struct rdma_ah_attr *ah_attr,
2184 						struct ib_udata *udata);
2185 	int                        (*modify_ah)(struct ib_ah *ah,
2186 						struct rdma_ah_attr *ah_attr);
2187 	int                        (*query_ah)(struct ib_ah *ah,
2188 					       struct rdma_ah_attr *ah_attr);
2189 	int                        (*destroy_ah)(struct ib_ah *ah);
2190 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
2191 						 struct ib_srq_init_attr *srq_init_attr,
2192 						 struct ib_udata *udata);
2193 	int                        (*modify_srq)(struct ib_srq *srq,
2194 						 struct ib_srq_attr *srq_attr,
2195 						 enum ib_srq_attr_mask srq_attr_mask,
2196 						 struct ib_udata *udata);
2197 	int                        (*query_srq)(struct ib_srq *srq,
2198 						struct ib_srq_attr *srq_attr);
2199 	int                        (*destroy_srq)(struct ib_srq *srq);
2200 	int                        (*post_srq_recv)(struct ib_srq *srq,
2201 						    struct ib_recv_wr *recv_wr,
2202 						    struct ib_recv_wr **bad_recv_wr);
2203 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
2204 						struct ib_qp_init_attr *qp_init_attr,
2205 						struct ib_udata *udata);
2206 	int                        (*modify_qp)(struct ib_qp *qp,
2207 						struct ib_qp_attr *qp_attr,
2208 						int qp_attr_mask,
2209 						struct ib_udata *udata);
2210 	int                        (*query_qp)(struct ib_qp *qp,
2211 					       struct ib_qp_attr *qp_attr,
2212 					       int qp_attr_mask,
2213 					       struct ib_qp_init_attr *qp_init_attr);
2214 	int                        (*destroy_qp)(struct ib_qp *qp);
2215 	int                        (*post_send)(struct ib_qp *qp,
2216 						struct ib_send_wr *send_wr,
2217 						struct ib_send_wr **bad_send_wr);
2218 	int                        (*post_recv)(struct ib_qp *qp,
2219 						struct ib_recv_wr *recv_wr,
2220 						struct ib_recv_wr **bad_recv_wr);
2221 	struct ib_cq *             (*create_cq)(struct ib_device *device,
2222 						const struct ib_cq_init_attr *attr,
2223 						struct ib_ucontext *context,
2224 						struct ib_udata *udata);
2225 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2226 						u16 cq_period);
2227 	int                        (*destroy_cq)(struct ib_cq *cq);
2228 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
2229 						struct ib_udata *udata);
2230 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
2231 					      struct ib_wc *wc);
2232 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2233 	int                        (*req_notify_cq)(struct ib_cq *cq,
2234 						    enum ib_cq_notify_flags flags);
2235 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
2236 						      int wc_cnt);
2237 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
2238 						 int mr_access_flags);
2239 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
2240 						  u64 start, u64 length,
2241 						  u64 virt_addr,
2242 						  int mr_access_flags,
2243 						  struct ib_udata *udata);
2244 	int			   (*rereg_user_mr)(struct ib_mr *mr,
2245 						    int flags,
2246 						    u64 start, u64 length,
2247 						    u64 virt_addr,
2248 						    int mr_access_flags,
2249 						    struct ib_pd *pd,
2250 						    struct ib_udata *udata);
2251 	int                        (*dereg_mr)(struct ib_mr *mr);
2252 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2253 					       enum ib_mr_type mr_type,
2254 					       u32 max_num_sg);
2255 	int                        (*map_mr_sg)(struct ib_mr *mr,
2256 						struct scatterlist *sg,
2257 						int sg_nents,
2258 						unsigned int *sg_offset);
2259 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2260 					       enum ib_mw_type type,
2261 					       struct ib_udata *udata);
2262 	int                        (*dealloc_mw)(struct ib_mw *mw);
2263 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2264 						int mr_access_flags,
2265 						struct ib_fmr_attr *fmr_attr);
2266 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2267 						   u64 *page_list, int list_len,
2268 						   u64 iova);
2269 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2270 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2271 	int                        (*attach_mcast)(struct ib_qp *qp,
2272 						   union ib_gid *gid,
2273 						   u16 lid);
2274 	int                        (*detach_mcast)(struct ib_qp *qp,
2275 						   union ib_gid *gid,
2276 						   u16 lid);
2277 	int                        (*process_mad)(struct ib_device *device,
2278 						  int process_mad_flags,
2279 						  u8 port_num,
2280 						  const struct ib_wc *in_wc,
2281 						  const struct ib_grh *in_grh,
2282 						  const struct ib_mad_hdr *in_mad,
2283 						  size_t in_mad_size,
2284 						  struct ib_mad_hdr *out_mad,
2285 						  size_t *out_mad_size,
2286 						  u16 *out_mad_pkey_index);
2287 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2288 						 struct ib_ucontext *ucontext,
2289 						 struct ib_udata *udata);
2290 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2291 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2292 						  struct ib_flow_attr
2293 						  *flow_attr,
2294 						  int domain);
2295 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2296 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2297 						      struct ib_mr_status *mr_status);
2298 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2299 	void			   (*drain_rq)(struct ib_qp *qp);
2300 	void			   (*drain_sq)(struct ib_qp *qp);
2301 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2302 							int state);
2303 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2304 						   struct ifla_vf_info *ivf);
2305 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2306 						   struct ifla_vf_stats *stats);
2307 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2308 						  int type);
2309 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2310 						struct ib_wq_init_attr *init_attr,
2311 						struct ib_udata *udata);
2312 	int			   (*destroy_wq)(struct ib_wq *wq);
2313 	int			   (*modify_wq)(struct ib_wq *wq,
2314 						struct ib_wq_attr *attr,
2315 						u32 wq_attr_mask,
2316 						struct ib_udata *udata);
2317 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2318 							   struct ib_rwq_ind_table_init_attr *init_attr,
2319 							   struct ib_udata *udata);
2320 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2321 	/**
2322 	 * rdma netdev operation
2323 	 *
2324 	 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2325 	 * doesn't support the specified rdma netdev type.
2326 	 */
2327 	struct net_device *(*alloc_rdma_netdev)(
2328 					struct ib_device *device,
2329 					u8 port_num,
2330 					enum rdma_netdev_t type,
2331 					const char *name,
2332 					unsigned char name_assign_type,
2333 					void (*setup)(struct net_device *));
2334 
2335 	struct module               *owner;
2336 	struct device                dev;
2337 	struct kobject               *ports_parent;
2338 	struct list_head             port_list;
2339 
2340 	enum {
2341 		IB_DEV_UNINITIALIZED,
2342 		IB_DEV_REGISTERED,
2343 		IB_DEV_UNREGISTERED
2344 	}                            reg_state;
2345 
2346 	int			     uverbs_abi_ver;
2347 	u64			     uverbs_cmd_mask;
2348 	u64			     uverbs_ex_cmd_mask;
2349 
2350 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2351 	__be64			     node_guid;
2352 	u32			     local_dma_lkey;
2353 	u16                          is_switch:1;
2354 	u8                           node_type;
2355 	u8                           phys_port_cnt;
2356 	struct ib_device_attr        attrs;
2357 	struct attribute_group	     *hw_stats_ag;
2358 	struct rdma_hw_stats         *hw_stats;
2359 
2360 #ifdef CONFIG_CGROUP_RDMA
2361 	struct rdmacg_device         cg_device;
2362 #endif
2363 
2364 	u32                          index;
2365 	/*
2366 	 * Implementation details of the RDMA core, don't use in drivers
2367 	 */
2368 	struct rdma_restrack_root     res;
2369 
2370 	/**
2371 	 * The following mandatory functions are used only at device
2372 	 * registration.  Keep functions such as these at the end of this
2373 	 * structure to avoid cache line misses when accessing struct ib_device
2374 	 * in fast paths.
2375 	 */
2376 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2377 	void (*get_dev_fw_str)(struct ib_device *, char *str);
2378 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2379 						     int comp_vector);
2380 
2381 	struct uverbs_root_spec		*specs_root;
2382 };
2383 
2384 struct ib_client {
2385 	char  *name;
2386 	void (*add)   (struct ib_device *);
2387 	void (*remove)(struct ib_device *, void *client_data);
2388 
2389 	/* Returns the net_dev belonging to this ib_client and matching the
2390 	 * given parameters.
2391 	 * @dev:	 An RDMA device that the net_dev use for communication.
2392 	 * @port:	 A physical port number on the RDMA device.
2393 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2394 	 * @gid:	 A GID that the net_dev uses to communicate.
2395 	 * @addr:	 An IP address the net_dev is configured with.
2396 	 * @client_data: The device's client data set by ib_set_client_data().
2397 	 *
2398 	 * An ib_client that implements a net_dev on top of RDMA devices
2399 	 * (such as IP over IB) should implement this callback, allowing the
2400 	 * rdma_cm module to find the right net_dev for a given request.
2401 	 *
2402 	 * The caller is responsible for calling dev_put on the returned
2403 	 * netdev. */
2404 	struct net_device *(*get_net_dev_by_params)(
2405 			struct ib_device *dev,
2406 			u8 port,
2407 			u16 pkey,
2408 			const union ib_gid *gid,
2409 			const struct sockaddr *addr,
2410 			void *client_data);
2411 	struct list_head list;
2412 };
2413 
2414 struct ib_device *ib_alloc_device(size_t size);
2415 void ib_dealloc_device(struct ib_device *device);
2416 
2417 void ib_get_device_fw_str(struct ib_device *device, char *str);
2418 
2419 int ib_register_device(struct ib_device *device,
2420 		       int (*port_callback)(struct ib_device *,
2421 					    u8, struct kobject *));
2422 void ib_unregister_device(struct ib_device *device);
2423 
2424 int ib_register_client   (struct ib_client *client);
2425 void ib_unregister_client(struct ib_client *client);
2426 
2427 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2428 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2429 			 void *data);
2430 
2431 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2432 {
2433 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2434 }
2435 
2436 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2437 {
2438 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2439 }
2440 
2441 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2442 				       size_t offset,
2443 				       size_t len)
2444 {
2445 	const void __user *p = udata->inbuf + offset;
2446 	bool ret;
2447 	u8 *buf;
2448 
2449 	if (len > USHRT_MAX)
2450 		return false;
2451 
2452 	buf = memdup_user(p, len);
2453 	if (IS_ERR(buf))
2454 		return false;
2455 
2456 	ret = !memchr_inv(buf, 0, len);
2457 	kfree(buf);
2458 	return ret;
2459 }
2460 
2461 /**
2462  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2463  * contains all required attributes and no attributes not allowed for
2464  * the given QP state transition.
2465  * @cur_state: Current QP state
2466  * @next_state: Next QP state
2467  * @type: QP type
2468  * @mask: Mask of supplied QP attributes
2469  * @ll : link layer of port
2470  *
2471  * This function is a helper function that a low-level driver's
2472  * modify_qp method can use to validate the consumer's input.  It
2473  * checks that cur_state and next_state are valid QP states, that a
2474  * transition from cur_state to next_state is allowed by the IB spec,
2475  * and that the attribute mask supplied is allowed for the transition.
2476  */
2477 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2478 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2479 		       enum rdma_link_layer ll);
2480 
2481 void ib_register_event_handler(struct ib_event_handler *event_handler);
2482 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2483 void ib_dispatch_event(struct ib_event *event);
2484 
2485 int ib_query_port(struct ib_device *device,
2486 		  u8 port_num, struct ib_port_attr *port_attr);
2487 
2488 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2489 					       u8 port_num);
2490 
2491 /**
2492  * rdma_cap_ib_switch - Check if the device is IB switch
2493  * @device: Device to check
2494  *
2495  * Device driver is responsible for setting is_switch bit on
2496  * in ib_device structure at init time.
2497  *
2498  * Return: true if the device is IB switch.
2499  */
2500 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2501 {
2502 	return device->is_switch;
2503 }
2504 
2505 /**
2506  * rdma_start_port - Return the first valid port number for the device
2507  * specified
2508  *
2509  * @device: Device to be checked
2510  *
2511  * Return start port number
2512  */
2513 static inline u8 rdma_start_port(const struct ib_device *device)
2514 {
2515 	return rdma_cap_ib_switch(device) ? 0 : 1;
2516 }
2517 
2518 /**
2519  * rdma_end_port - Return the last valid port number for the device
2520  * specified
2521  *
2522  * @device: Device to be checked
2523  *
2524  * Return last port number
2525  */
2526 static inline u8 rdma_end_port(const struct ib_device *device)
2527 {
2528 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2529 }
2530 
2531 static inline int rdma_is_port_valid(const struct ib_device *device,
2532 				     unsigned int port)
2533 {
2534 	return (port >= rdma_start_port(device) &&
2535 		port <= rdma_end_port(device));
2536 }
2537 
2538 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2539 {
2540 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2541 }
2542 
2543 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2544 {
2545 	return device->port_immutable[port_num].core_cap_flags &
2546 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2547 }
2548 
2549 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2550 {
2551 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2552 }
2553 
2554 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2555 {
2556 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2557 }
2558 
2559 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2560 {
2561 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2562 }
2563 
2564 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2565 {
2566 	return rdma_protocol_ib(device, port_num) ||
2567 		rdma_protocol_roce(device, port_num);
2568 }
2569 
2570 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2571 {
2572 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2573 }
2574 
2575 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2576 {
2577 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2578 }
2579 
2580 /**
2581  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2582  * Management Datagrams.
2583  * @device: Device to check
2584  * @port_num: Port number to check
2585  *
2586  * Management Datagrams (MAD) are a required part of the InfiniBand
2587  * specification and are supported on all InfiniBand devices.  A slightly
2588  * extended version are also supported on OPA interfaces.
2589  *
2590  * Return: true if the port supports sending/receiving of MAD packets.
2591  */
2592 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2593 {
2594 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2595 }
2596 
2597 /**
2598  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2599  * Management Datagrams.
2600  * @device: Device to check
2601  * @port_num: Port number to check
2602  *
2603  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2604  * datagrams with their own versions.  These OPA MADs share many but not all of
2605  * the characteristics of InfiniBand MADs.
2606  *
2607  * OPA MADs differ in the following ways:
2608  *
2609  *    1) MADs are variable size up to 2K
2610  *       IBTA defined MADs remain fixed at 256 bytes
2611  *    2) OPA SMPs must carry valid PKeys
2612  *    3) OPA SMP packets are a different format
2613  *
2614  * Return: true if the port supports OPA MAD packet formats.
2615  */
2616 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2617 {
2618 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2619 		== RDMA_CORE_CAP_OPA_MAD;
2620 }
2621 
2622 /**
2623  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2624  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2625  * @device: Device to check
2626  * @port_num: Port number to check
2627  *
2628  * Each InfiniBand node is required to provide a Subnet Management Agent
2629  * that the subnet manager can access.  Prior to the fabric being fully
2630  * configured by the subnet manager, the SMA is accessed via a well known
2631  * interface called the Subnet Management Interface (SMI).  This interface
2632  * uses directed route packets to communicate with the SM to get around the
2633  * chicken and egg problem of the SM needing to know what's on the fabric
2634  * in order to configure the fabric, and needing to configure the fabric in
2635  * order to send packets to the devices on the fabric.  These directed
2636  * route packets do not need the fabric fully configured in order to reach
2637  * their destination.  The SMI is the only method allowed to send
2638  * directed route packets on an InfiniBand fabric.
2639  *
2640  * Return: true if the port provides an SMI.
2641  */
2642 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2643 {
2644 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2645 }
2646 
2647 /**
2648  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2649  * Communication Manager.
2650  * @device: Device to check
2651  * @port_num: Port number to check
2652  *
2653  * The InfiniBand Communication Manager is one of many pre-defined General
2654  * Service Agents (GSA) that are accessed via the General Service
2655  * Interface (GSI).  It's role is to facilitate establishment of connections
2656  * between nodes as well as other management related tasks for established
2657  * connections.
2658  *
2659  * Return: true if the port supports an IB CM (this does not guarantee that
2660  * a CM is actually running however).
2661  */
2662 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2663 {
2664 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2665 }
2666 
2667 /**
2668  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2669  * Communication Manager.
2670  * @device: Device to check
2671  * @port_num: Port number to check
2672  *
2673  * Similar to above, but specific to iWARP connections which have a different
2674  * managment protocol than InfiniBand.
2675  *
2676  * Return: true if the port supports an iWARP CM (this does not guarantee that
2677  * a CM is actually running however).
2678  */
2679 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2680 {
2681 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2682 }
2683 
2684 /**
2685  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2686  * Subnet Administration.
2687  * @device: Device to check
2688  * @port_num: Port number to check
2689  *
2690  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2691  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2692  * fabrics, devices should resolve routes to other hosts by contacting the
2693  * SA to query the proper route.
2694  *
2695  * Return: true if the port should act as a client to the fabric Subnet
2696  * Administration interface.  This does not imply that the SA service is
2697  * running locally.
2698  */
2699 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2700 {
2701 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2702 }
2703 
2704 /**
2705  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2706  * Multicast.
2707  * @device: Device to check
2708  * @port_num: Port number to check
2709  *
2710  * InfiniBand multicast registration is more complex than normal IPv4 or
2711  * IPv6 multicast registration.  Each Host Channel Adapter must register
2712  * with the Subnet Manager when it wishes to join a multicast group.  It
2713  * should do so only once regardless of how many queue pairs it subscribes
2714  * to this group.  And it should leave the group only after all queue pairs
2715  * attached to the group have been detached.
2716  *
2717  * Return: true if the port must undertake the additional adminstrative
2718  * overhead of registering/unregistering with the SM and tracking of the
2719  * total number of queue pairs attached to the multicast group.
2720  */
2721 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2722 {
2723 	return rdma_cap_ib_sa(device, port_num);
2724 }
2725 
2726 /**
2727  * rdma_cap_af_ib - Check if the port of device has the capability
2728  * Native Infiniband Address.
2729  * @device: Device to check
2730  * @port_num: Port number to check
2731  *
2732  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2733  * GID.  RoCE uses a different mechanism, but still generates a GID via
2734  * a prescribed mechanism and port specific data.
2735  *
2736  * Return: true if the port uses a GID address to identify devices on the
2737  * network.
2738  */
2739 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2740 {
2741 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2742 }
2743 
2744 /**
2745  * rdma_cap_eth_ah - Check if the port of device has the capability
2746  * Ethernet Address Handle.
2747  * @device: Device to check
2748  * @port_num: Port number to check
2749  *
2750  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2751  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2752  * port.  Normally, packet headers are generated by the sending host
2753  * adapter, but when sending connectionless datagrams, we must manually
2754  * inject the proper headers for the fabric we are communicating over.
2755  *
2756  * Return: true if we are running as a RoCE port and must force the
2757  * addition of a Global Route Header built from our Ethernet Address
2758  * Handle into our header list for connectionless packets.
2759  */
2760 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2761 {
2762 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2763 }
2764 
2765 /**
2766  * rdma_cap_opa_ah - Check if the port of device supports
2767  * OPA Address handles
2768  * @device: Device to check
2769  * @port_num: Port number to check
2770  *
2771  * Return: true if we are running on an OPA device which supports
2772  * the extended OPA addressing.
2773  */
2774 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
2775 {
2776 	return (device->port_immutable[port_num].core_cap_flags &
2777 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
2778 }
2779 
2780 /**
2781  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2782  *
2783  * @device: Device
2784  * @port_num: Port number
2785  *
2786  * This MAD size includes the MAD headers and MAD payload.  No other headers
2787  * are included.
2788  *
2789  * Return the max MAD size required by the Port.  Will return 0 if the port
2790  * does not support MADs
2791  */
2792 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2793 {
2794 	return device->port_immutable[port_num].max_mad_size;
2795 }
2796 
2797 /**
2798  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2799  * @device: Device to check
2800  * @port_num: Port number to check
2801  *
2802  * RoCE GID table mechanism manages the various GIDs for a device.
2803  *
2804  * NOTE: if allocating the port's GID table has failed, this call will still
2805  * return true, but any RoCE GID table API will fail.
2806  *
2807  * Return: true if the port uses RoCE GID table mechanism in order to manage
2808  * its GIDs.
2809  */
2810 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2811 					   u8 port_num)
2812 {
2813 	return rdma_protocol_roce(device, port_num) &&
2814 		device->add_gid && device->del_gid;
2815 }
2816 
2817 /*
2818  * Check if the device supports READ W/ INVALIDATE.
2819  */
2820 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2821 {
2822 	/*
2823 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2824 	 * has support for it yet.
2825 	 */
2826 	return rdma_protocol_iwarp(dev, port_num);
2827 }
2828 
2829 int ib_query_gid(struct ib_device *device,
2830 		 u8 port_num, int index, union ib_gid *gid,
2831 		 struct ib_gid_attr *attr);
2832 
2833 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2834 			 int state);
2835 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2836 		     struct ifla_vf_info *info);
2837 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2838 		    struct ifla_vf_stats *stats);
2839 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2840 		   int type);
2841 
2842 int ib_query_pkey(struct ib_device *device,
2843 		  u8 port_num, u16 index, u16 *pkey);
2844 
2845 int ib_modify_device(struct ib_device *device,
2846 		     int device_modify_mask,
2847 		     struct ib_device_modify *device_modify);
2848 
2849 int ib_modify_port(struct ib_device *device,
2850 		   u8 port_num, int port_modify_mask,
2851 		   struct ib_port_modify *port_modify);
2852 
2853 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2854 		struct net_device *ndev, u8 *port_num, u16 *index);
2855 
2856 int ib_find_pkey(struct ib_device *device,
2857 		 u8 port_num, u16 pkey, u16 *index);
2858 
2859 enum ib_pd_flags {
2860 	/*
2861 	 * Create a memory registration for all memory in the system and place
2862 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2863 	 * ULPs to avoid the overhead of dynamic MRs.
2864 	 *
2865 	 * This flag is generally considered unsafe and must only be used in
2866 	 * extremly trusted environments.  Every use of it will log a warning
2867 	 * in the kernel log.
2868 	 */
2869 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2870 };
2871 
2872 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2873 		const char *caller);
2874 #define ib_alloc_pd(device, flags) \
2875 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
2876 void ib_dealloc_pd(struct ib_pd *pd);
2877 
2878 /**
2879  * rdma_create_ah - Creates an address handle for the given address vector.
2880  * @pd: The protection domain associated with the address handle.
2881  * @ah_attr: The attributes of the address vector.
2882  *
2883  * The address handle is used to reference a local or global destination
2884  * in all UD QP post sends.
2885  */
2886 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
2887 
2888 /**
2889  * rdma_create_user_ah - Creates an address handle for the given address vector.
2890  * It resolves destination mac address for ah attribute of RoCE type.
2891  * @pd: The protection domain associated with the address handle.
2892  * @ah_attr: The attributes of the address vector.
2893  * @udata: pointer to user's input output buffer information need by
2894  *         provider driver.
2895  *
2896  * It returns 0 on success and returns appropriate error code on error.
2897  * The address handle is used to reference a local or global destination
2898  * in all UD QP post sends.
2899  */
2900 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
2901 				  struct rdma_ah_attr *ah_attr,
2902 				  struct ib_udata *udata);
2903 /**
2904  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2905  *   work completion.
2906  * @hdr: the L3 header to parse
2907  * @net_type: type of header to parse
2908  * @sgid: place to store source gid
2909  * @dgid: place to store destination gid
2910  */
2911 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2912 			      enum rdma_network_type net_type,
2913 			      union ib_gid *sgid, union ib_gid *dgid);
2914 
2915 /**
2916  * ib_get_rdma_header_version - Get the header version
2917  * @hdr: the L3 header to parse
2918  */
2919 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2920 
2921 /**
2922  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
2923  *   work completion.
2924  * @device: Device on which the received message arrived.
2925  * @port_num: Port on which the received message arrived.
2926  * @wc: Work completion associated with the received message.
2927  * @grh: References the received global route header.  This parameter is
2928  *   ignored unless the work completion indicates that the GRH is valid.
2929  * @ah_attr: Returned attributes that can be used when creating an address
2930  *   handle for replying to the message.
2931  */
2932 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
2933 			    const struct ib_wc *wc, const struct ib_grh *grh,
2934 			    struct rdma_ah_attr *ah_attr);
2935 
2936 /**
2937  * ib_create_ah_from_wc - Creates an address handle associated with the
2938  *   sender of the specified work completion.
2939  * @pd: The protection domain associated with the address handle.
2940  * @wc: Work completion information associated with a received message.
2941  * @grh: References the received global route header.  This parameter is
2942  *   ignored unless the work completion indicates that the GRH is valid.
2943  * @port_num: The outbound port number to associate with the address.
2944  *
2945  * The address handle is used to reference a local or global destination
2946  * in all UD QP post sends.
2947  */
2948 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2949 				   const struct ib_grh *grh, u8 port_num);
2950 
2951 /**
2952  * rdma_modify_ah - Modifies the address vector associated with an address
2953  *   handle.
2954  * @ah: The address handle to modify.
2955  * @ah_attr: The new address vector attributes to associate with the
2956  *   address handle.
2957  */
2958 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2959 
2960 /**
2961  * rdma_query_ah - Queries the address vector associated with an address
2962  *   handle.
2963  * @ah: The address handle to query.
2964  * @ah_attr: The address vector attributes associated with the address
2965  *   handle.
2966  */
2967 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2968 
2969 /**
2970  * rdma_destroy_ah - Destroys an address handle.
2971  * @ah: The address handle to destroy.
2972  */
2973 int rdma_destroy_ah(struct ib_ah *ah);
2974 
2975 /**
2976  * ib_create_srq - Creates a SRQ associated with the specified protection
2977  *   domain.
2978  * @pd: The protection domain associated with the SRQ.
2979  * @srq_init_attr: A list of initial attributes required to create the
2980  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2981  *   the actual capabilities of the created SRQ.
2982  *
2983  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2984  * requested size of the SRQ, and set to the actual values allocated
2985  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2986  * will always be at least as large as the requested values.
2987  */
2988 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2989 			     struct ib_srq_init_attr *srq_init_attr);
2990 
2991 /**
2992  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2993  * @srq: The SRQ to modify.
2994  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2995  *   the current values of selected SRQ attributes are returned.
2996  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2997  *   are being modified.
2998  *
2999  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3000  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3001  * the number of receives queued drops below the limit.
3002  */
3003 int ib_modify_srq(struct ib_srq *srq,
3004 		  struct ib_srq_attr *srq_attr,
3005 		  enum ib_srq_attr_mask srq_attr_mask);
3006 
3007 /**
3008  * ib_query_srq - Returns the attribute list and current values for the
3009  *   specified SRQ.
3010  * @srq: The SRQ to query.
3011  * @srq_attr: The attributes of the specified SRQ.
3012  */
3013 int ib_query_srq(struct ib_srq *srq,
3014 		 struct ib_srq_attr *srq_attr);
3015 
3016 /**
3017  * ib_destroy_srq - Destroys the specified SRQ.
3018  * @srq: The SRQ to destroy.
3019  */
3020 int ib_destroy_srq(struct ib_srq *srq);
3021 
3022 /**
3023  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3024  * @srq: The SRQ to post the work request on.
3025  * @recv_wr: A list of work requests to post on the receive queue.
3026  * @bad_recv_wr: On an immediate failure, this parameter will reference
3027  *   the work request that failed to be posted on the QP.
3028  */
3029 static inline int ib_post_srq_recv(struct ib_srq *srq,
3030 				   struct ib_recv_wr *recv_wr,
3031 				   struct ib_recv_wr **bad_recv_wr)
3032 {
3033 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
3034 }
3035 
3036 /**
3037  * ib_create_qp - Creates a QP associated with the specified protection
3038  *   domain.
3039  * @pd: The protection domain associated with the QP.
3040  * @qp_init_attr: A list of initial attributes required to create the
3041  *   QP.  If QP creation succeeds, then the attributes are updated to
3042  *   the actual capabilities of the created QP.
3043  */
3044 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3045 			   struct ib_qp_init_attr *qp_init_attr);
3046 
3047 /**
3048  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3049  * @qp: The QP to modify.
3050  * @attr: On input, specifies the QP attributes to modify.  On output,
3051  *   the current values of selected QP attributes are returned.
3052  * @attr_mask: A bit-mask used to specify which attributes of the QP
3053  *   are being modified.
3054  * @udata: pointer to user's input output buffer information
3055  *   are being modified.
3056  * It returns 0 on success and returns appropriate error code on error.
3057  */
3058 int ib_modify_qp_with_udata(struct ib_qp *qp,
3059 			    struct ib_qp_attr *attr,
3060 			    int attr_mask,
3061 			    struct ib_udata *udata);
3062 
3063 /**
3064  * ib_modify_qp - Modifies the attributes for the specified QP and then
3065  *   transitions the QP to the given state.
3066  * @qp: The QP to modify.
3067  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3068  *   the current values of selected QP attributes are returned.
3069  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3070  *   are being modified.
3071  */
3072 int ib_modify_qp(struct ib_qp *qp,
3073 		 struct ib_qp_attr *qp_attr,
3074 		 int qp_attr_mask);
3075 
3076 /**
3077  * ib_query_qp - Returns the attribute list and current values for the
3078  *   specified QP.
3079  * @qp: The QP to query.
3080  * @qp_attr: The attributes of the specified QP.
3081  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3082  * @qp_init_attr: Additional attributes of the selected QP.
3083  *
3084  * The qp_attr_mask may be used to limit the query to gathering only the
3085  * selected attributes.
3086  */
3087 int ib_query_qp(struct ib_qp *qp,
3088 		struct ib_qp_attr *qp_attr,
3089 		int qp_attr_mask,
3090 		struct ib_qp_init_attr *qp_init_attr);
3091 
3092 /**
3093  * ib_destroy_qp - Destroys the specified QP.
3094  * @qp: The QP to destroy.
3095  */
3096 int ib_destroy_qp(struct ib_qp *qp);
3097 
3098 /**
3099  * ib_open_qp - Obtain a reference to an existing sharable QP.
3100  * @xrcd - XRC domain
3101  * @qp_open_attr: Attributes identifying the QP to open.
3102  *
3103  * Returns a reference to a sharable QP.
3104  */
3105 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3106 			 struct ib_qp_open_attr *qp_open_attr);
3107 
3108 /**
3109  * ib_close_qp - Release an external reference to a QP.
3110  * @qp: The QP handle to release
3111  *
3112  * The opened QP handle is released by the caller.  The underlying
3113  * shared QP is not destroyed until all internal references are released.
3114  */
3115 int ib_close_qp(struct ib_qp *qp);
3116 
3117 /**
3118  * ib_post_send - Posts a list of work requests to the send queue of
3119  *   the specified QP.
3120  * @qp: The QP to post the work request on.
3121  * @send_wr: A list of work requests to post on the send queue.
3122  * @bad_send_wr: On an immediate failure, this parameter will reference
3123  *   the work request that failed to be posted on the QP.
3124  *
3125  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3126  * error is returned, the QP state shall not be affected,
3127  * ib_post_send() will return an immediate error after queueing any
3128  * earlier work requests in the list.
3129  */
3130 static inline int ib_post_send(struct ib_qp *qp,
3131 			       struct ib_send_wr *send_wr,
3132 			       struct ib_send_wr **bad_send_wr)
3133 {
3134 	return qp->device->post_send(qp, send_wr, bad_send_wr);
3135 }
3136 
3137 /**
3138  * ib_post_recv - Posts a list of work requests to the receive queue of
3139  *   the specified QP.
3140  * @qp: The QP to post the work request on.
3141  * @recv_wr: A list of work requests to post on the receive queue.
3142  * @bad_recv_wr: On an immediate failure, this parameter will reference
3143  *   the work request that failed to be posted on the QP.
3144  */
3145 static inline int ib_post_recv(struct ib_qp *qp,
3146 			       struct ib_recv_wr *recv_wr,
3147 			       struct ib_recv_wr **bad_recv_wr)
3148 {
3149 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
3150 }
3151 
3152 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3153 			    int nr_cqe, int comp_vector,
3154 			    enum ib_poll_context poll_ctx, const char *caller);
3155 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3156 	__ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3157 
3158 void ib_free_cq(struct ib_cq *cq);
3159 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3160 
3161 /**
3162  * ib_create_cq - Creates a CQ on the specified device.
3163  * @device: The device on which to create the CQ.
3164  * @comp_handler: A user-specified callback that is invoked when a
3165  *   completion event occurs on the CQ.
3166  * @event_handler: A user-specified callback that is invoked when an
3167  *   asynchronous event not associated with a completion occurs on the CQ.
3168  * @cq_context: Context associated with the CQ returned to the user via
3169  *   the associated completion and event handlers.
3170  * @cq_attr: The attributes the CQ should be created upon.
3171  *
3172  * Users can examine the cq structure to determine the actual CQ size.
3173  */
3174 struct ib_cq *ib_create_cq(struct ib_device *device,
3175 			   ib_comp_handler comp_handler,
3176 			   void (*event_handler)(struct ib_event *, void *),
3177 			   void *cq_context,
3178 			   const struct ib_cq_init_attr *cq_attr);
3179 
3180 /**
3181  * ib_resize_cq - Modifies the capacity of the CQ.
3182  * @cq: The CQ to resize.
3183  * @cqe: The minimum size of the CQ.
3184  *
3185  * Users can examine the cq structure to determine the actual CQ size.
3186  */
3187 int ib_resize_cq(struct ib_cq *cq, int cqe);
3188 
3189 /**
3190  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3191  * @cq: The CQ to modify.
3192  * @cq_count: number of CQEs that will trigger an event
3193  * @cq_period: max period of time in usec before triggering an event
3194  *
3195  */
3196 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3197 
3198 /**
3199  * ib_destroy_cq - Destroys the specified CQ.
3200  * @cq: The CQ to destroy.
3201  */
3202 int ib_destroy_cq(struct ib_cq *cq);
3203 
3204 /**
3205  * ib_poll_cq - poll a CQ for completion(s)
3206  * @cq:the CQ being polled
3207  * @num_entries:maximum number of completions to return
3208  * @wc:array of at least @num_entries &struct ib_wc where completions
3209  *   will be returned
3210  *
3211  * Poll a CQ for (possibly multiple) completions.  If the return value
3212  * is < 0, an error occurred.  If the return value is >= 0, it is the
3213  * number of completions returned.  If the return value is
3214  * non-negative and < num_entries, then the CQ was emptied.
3215  */
3216 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3217 			     struct ib_wc *wc)
3218 {
3219 	return cq->device->poll_cq(cq, num_entries, wc);
3220 }
3221 
3222 /**
3223  * ib_peek_cq - Returns the number of unreaped completions currently
3224  *   on the specified CQ.
3225  * @cq: The CQ to peek.
3226  * @wc_cnt: A minimum number of unreaped completions to check for.
3227  *
3228  * If the number of unreaped completions is greater than or equal to wc_cnt,
3229  * this function returns wc_cnt, otherwise, it returns the actual number of
3230  * unreaped completions.
3231  */
3232 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
3233 
3234 /**
3235  * ib_req_notify_cq - Request completion notification on a CQ.
3236  * @cq: The CQ to generate an event for.
3237  * @flags:
3238  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3239  *   to request an event on the next solicited event or next work
3240  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3241  *   may also be |ed in to request a hint about missed events, as
3242  *   described below.
3243  *
3244  * Return Value:
3245  *    < 0 means an error occurred while requesting notification
3246  *   == 0 means notification was requested successfully, and if
3247  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3248  *        were missed and it is safe to wait for another event.  In
3249  *        this case is it guaranteed that any work completions added
3250  *        to the CQ since the last CQ poll will trigger a completion
3251  *        notification event.
3252  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3253  *        in.  It means that the consumer must poll the CQ again to
3254  *        make sure it is empty to avoid missing an event because of a
3255  *        race between requesting notification and an entry being
3256  *        added to the CQ.  This return value means it is possible
3257  *        (but not guaranteed) that a work completion has been added
3258  *        to the CQ since the last poll without triggering a
3259  *        completion notification event.
3260  */
3261 static inline int ib_req_notify_cq(struct ib_cq *cq,
3262 				   enum ib_cq_notify_flags flags)
3263 {
3264 	return cq->device->req_notify_cq(cq, flags);
3265 }
3266 
3267 /**
3268  * ib_req_ncomp_notif - Request completion notification when there are
3269  *   at least the specified number of unreaped completions on the CQ.
3270  * @cq: The CQ to generate an event for.
3271  * @wc_cnt: The number of unreaped completions that should be on the
3272  *   CQ before an event is generated.
3273  */
3274 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3275 {
3276 	return cq->device->req_ncomp_notif ?
3277 		cq->device->req_ncomp_notif(cq, wc_cnt) :
3278 		-ENOSYS;
3279 }
3280 
3281 /**
3282  * ib_dma_mapping_error - check a DMA addr for error
3283  * @dev: The device for which the dma_addr was created
3284  * @dma_addr: The DMA address to check
3285  */
3286 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3287 {
3288 	return dma_mapping_error(dev->dma_device, dma_addr);
3289 }
3290 
3291 /**
3292  * ib_dma_map_single - Map a kernel virtual address to DMA address
3293  * @dev: The device for which the dma_addr is to be created
3294  * @cpu_addr: The kernel virtual address
3295  * @size: The size of the region in bytes
3296  * @direction: The direction of the DMA
3297  */
3298 static inline u64 ib_dma_map_single(struct ib_device *dev,
3299 				    void *cpu_addr, size_t size,
3300 				    enum dma_data_direction direction)
3301 {
3302 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3303 }
3304 
3305 /**
3306  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3307  * @dev: The device for which the DMA address was created
3308  * @addr: The DMA address
3309  * @size: The size of the region in bytes
3310  * @direction: The direction of the DMA
3311  */
3312 static inline void ib_dma_unmap_single(struct ib_device *dev,
3313 				       u64 addr, size_t size,
3314 				       enum dma_data_direction direction)
3315 {
3316 	dma_unmap_single(dev->dma_device, addr, size, direction);
3317 }
3318 
3319 /**
3320  * ib_dma_map_page - Map a physical page to DMA address
3321  * @dev: The device for which the dma_addr is to be created
3322  * @page: The page to be mapped
3323  * @offset: The offset within the page
3324  * @size: The size of the region in bytes
3325  * @direction: The direction of the DMA
3326  */
3327 static inline u64 ib_dma_map_page(struct ib_device *dev,
3328 				  struct page *page,
3329 				  unsigned long offset,
3330 				  size_t size,
3331 					 enum dma_data_direction direction)
3332 {
3333 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3334 }
3335 
3336 /**
3337  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3338  * @dev: The device for which the DMA address was created
3339  * @addr: The DMA address
3340  * @size: The size of the region in bytes
3341  * @direction: The direction of the DMA
3342  */
3343 static inline void ib_dma_unmap_page(struct ib_device *dev,
3344 				     u64 addr, size_t size,
3345 				     enum dma_data_direction direction)
3346 {
3347 	dma_unmap_page(dev->dma_device, addr, size, direction);
3348 }
3349 
3350 /**
3351  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3352  * @dev: The device for which the DMA addresses are to be created
3353  * @sg: The array of scatter/gather entries
3354  * @nents: The number of scatter/gather entries
3355  * @direction: The direction of the DMA
3356  */
3357 static inline int ib_dma_map_sg(struct ib_device *dev,
3358 				struct scatterlist *sg, int nents,
3359 				enum dma_data_direction direction)
3360 {
3361 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3362 }
3363 
3364 /**
3365  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3366  * @dev: The device for which the DMA addresses were created
3367  * @sg: The array of scatter/gather entries
3368  * @nents: The number of scatter/gather entries
3369  * @direction: The direction of the DMA
3370  */
3371 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3372 				   struct scatterlist *sg, int nents,
3373 				   enum dma_data_direction direction)
3374 {
3375 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
3376 }
3377 
3378 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3379 				      struct scatterlist *sg, int nents,
3380 				      enum dma_data_direction direction,
3381 				      unsigned long dma_attrs)
3382 {
3383 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3384 				dma_attrs);
3385 }
3386 
3387 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3388 					 struct scatterlist *sg, int nents,
3389 					 enum dma_data_direction direction,
3390 					 unsigned long dma_attrs)
3391 {
3392 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3393 }
3394 /**
3395  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3396  * @dev: The device for which the DMA addresses were created
3397  * @sg: The scatter/gather entry
3398  *
3399  * Note: this function is obsolete. To do: change all occurrences of
3400  * ib_sg_dma_address() into sg_dma_address().
3401  */
3402 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3403 				    struct scatterlist *sg)
3404 {
3405 	return sg_dma_address(sg);
3406 }
3407 
3408 /**
3409  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3410  * @dev: The device for which the DMA addresses were created
3411  * @sg: The scatter/gather entry
3412  *
3413  * Note: this function is obsolete. To do: change all occurrences of
3414  * ib_sg_dma_len() into sg_dma_len().
3415  */
3416 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3417 					 struct scatterlist *sg)
3418 {
3419 	return sg_dma_len(sg);
3420 }
3421 
3422 /**
3423  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3424  * @dev: The device for which the DMA address was created
3425  * @addr: The DMA address
3426  * @size: The size of the region in bytes
3427  * @dir: The direction of the DMA
3428  */
3429 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3430 					      u64 addr,
3431 					      size_t size,
3432 					      enum dma_data_direction dir)
3433 {
3434 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3435 }
3436 
3437 /**
3438  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3439  * @dev: The device for which the DMA address was created
3440  * @addr: The DMA address
3441  * @size: The size of the region in bytes
3442  * @dir: The direction of the DMA
3443  */
3444 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3445 						 u64 addr,
3446 						 size_t size,
3447 						 enum dma_data_direction dir)
3448 {
3449 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3450 }
3451 
3452 /**
3453  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3454  * @dev: The device for which the DMA address is requested
3455  * @size: The size of the region to allocate in bytes
3456  * @dma_handle: A pointer for returning the DMA address of the region
3457  * @flag: memory allocator flags
3458  */
3459 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3460 					   size_t size,
3461 					   dma_addr_t *dma_handle,
3462 					   gfp_t flag)
3463 {
3464 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3465 }
3466 
3467 /**
3468  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3469  * @dev: The device for which the DMA addresses were allocated
3470  * @size: The size of the region
3471  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3472  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3473  */
3474 static inline void ib_dma_free_coherent(struct ib_device *dev,
3475 					size_t size, void *cpu_addr,
3476 					dma_addr_t dma_handle)
3477 {
3478 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3479 }
3480 
3481 /**
3482  * ib_dereg_mr - Deregisters a memory region and removes it from the
3483  *   HCA translation table.
3484  * @mr: The memory region to deregister.
3485  *
3486  * This function can fail, if the memory region has memory windows bound to it.
3487  */
3488 int ib_dereg_mr(struct ib_mr *mr);
3489 
3490 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3491 			  enum ib_mr_type mr_type,
3492 			  u32 max_num_sg);
3493 
3494 /**
3495  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3496  *   R_Key and L_Key.
3497  * @mr - struct ib_mr pointer to be updated.
3498  * @newkey - new key to be used.
3499  */
3500 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3501 {
3502 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3503 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3504 }
3505 
3506 /**
3507  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3508  * for calculating a new rkey for type 2 memory windows.
3509  * @rkey - the rkey to increment.
3510  */
3511 static inline u32 ib_inc_rkey(u32 rkey)
3512 {
3513 	const u32 mask = 0x000000ff;
3514 	return ((rkey + 1) & mask) | (rkey & ~mask);
3515 }
3516 
3517 /**
3518  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3519  * @pd: The protection domain associated with the unmapped region.
3520  * @mr_access_flags: Specifies the memory access rights.
3521  * @fmr_attr: Attributes of the unmapped region.
3522  *
3523  * A fast memory region must be mapped before it can be used as part of
3524  * a work request.
3525  */
3526 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3527 			    int mr_access_flags,
3528 			    struct ib_fmr_attr *fmr_attr);
3529 
3530 /**
3531  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3532  * @fmr: The fast memory region to associate with the pages.
3533  * @page_list: An array of physical pages to map to the fast memory region.
3534  * @list_len: The number of pages in page_list.
3535  * @iova: The I/O virtual address to use with the mapped region.
3536  */
3537 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3538 				  u64 *page_list, int list_len,
3539 				  u64 iova)
3540 {
3541 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3542 }
3543 
3544 /**
3545  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3546  * @fmr_list: A linked list of fast memory regions to unmap.
3547  */
3548 int ib_unmap_fmr(struct list_head *fmr_list);
3549 
3550 /**
3551  * ib_dealloc_fmr - Deallocates a fast memory region.
3552  * @fmr: The fast memory region to deallocate.
3553  */
3554 int ib_dealloc_fmr(struct ib_fmr *fmr);
3555 
3556 /**
3557  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3558  * @qp: QP to attach to the multicast group.  The QP must be type
3559  *   IB_QPT_UD.
3560  * @gid: Multicast group GID.
3561  * @lid: Multicast group LID in host byte order.
3562  *
3563  * In order to send and receive multicast packets, subnet
3564  * administration must have created the multicast group and configured
3565  * the fabric appropriately.  The port associated with the specified
3566  * QP must also be a member of the multicast group.
3567  */
3568 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3569 
3570 /**
3571  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3572  * @qp: QP to detach from the multicast group.
3573  * @gid: Multicast group GID.
3574  * @lid: Multicast group LID in host byte order.
3575  */
3576 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3577 
3578 /**
3579  * ib_alloc_xrcd - Allocates an XRC domain.
3580  * @device: The device on which to allocate the XRC domain.
3581  * @caller: Module name for kernel consumers
3582  */
3583 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3584 #define ib_alloc_xrcd(device) \
3585 	__ib_alloc_xrcd((device), KBUILD_MODNAME)
3586 
3587 /**
3588  * ib_dealloc_xrcd - Deallocates an XRC domain.
3589  * @xrcd: The XRC domain to deallocate.
3590  */
3591 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3592 
3593 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3594 			       struct ib_flow_attr *flow_attr, int domain);
3595 int ib_destroy_flow(struct ib_flow *flow_id);
3596 
3597 static inline int ib_check_mr_access(int flags)
3598 {
3599 	/*
3600 	 * Local write permission is required if remote write or
3601 	 * remote atomic permission is also requested.
3602 	 */
3603 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3604 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3605 		return -EINVAL;
3606 
3607 	return 0;
3608 }
3609 
3610 /**
3611  * ib_check_mr_status: lightweight check of MR status.
3612  *     This routine may provide status checks on a selected
3613  *     ib_mr. first use is for signature status check.
3614  *
3615  * @mr: A memory region.
3616  * @check_mask: Bitmask of which checks to perform from
3617  *     ib_mr_status_check enumeration.
3618  * @mr_status: The container of relevant status checks.
3619  *     failed checks will be indicated in the status bitmask
3620  *     and the relevant info shall be in the error item.
3621  */
3622 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3623 		       struct ib_mr_status *mr_status);
3624 
3625 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3626 					    u16 pkey, const union ib_gid *gid,
3627 					    const struct sockaddr *addr);
3628 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3629 			   struct ib_wq_init_attr *init_attr);
3630 int ib_destroy_wq(struct ib_wq *wq);
3631 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3632 		 u32 wq_attr_mask);
3633 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3634 						 struct ib_rwq_ind_table_init_attr*
3635 						 wq_ind_table_init_attr);
3636 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3637 
3638 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3639 		 unsigned int *sg_offset, unsigned int page_size);
3640 
3641 static inline int
3642 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3643 		  unsigned int *sg_offset, unsigned int page_size)
3644 {
3645 	int n;
3646 
3647 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3648 	mr->iova = 0;
3649 
3650 	return n;
3651 }
3652 
3653 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3654 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3655 
3656 void ib_drain_rq(struct ib_qp *qp);
3657 void ib_drain_sq(struct ib_qp *qp);
3658 void ib_drain_qp(struct ib_qp *qp);
3659 
3660 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3661 
3662 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3663 {
3664 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3665 		return attr->roce.dmac;
3666 	return NULL;
3667 }
3668 
3669 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3670 {
3671 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3672 		attr->ib.dlid = (u16)dlid;
3673 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3674 		attr->opa.dlid = dlid;
3675 }
3676 
3677 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3678 {
3679 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3680 		return attr->ib.dlid;
3681 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3682 		return attr->opa.dlid;
3683 	return 0;
3684 }
3685 
3686 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3687 {
3688 	attr->sl = sl;
3689 }
3690 
3691 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3692 {
3693 	return attr->sl;
3694 }
3695 
3696 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3697 					 u8 src_path_bits)
3698 {
3699 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3700 		attr->ib.src_path_bits = src_path_bits;
3701 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3702 		attr->opa.src_path_bits = src_path_bits;
3703 }
3704 
3705 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3706 {
3707 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3708 		return attr->ib.src_path_bits;
3709 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3710 		return attr->opa.src_path_bits;
3711 	return 0;
3712 }
3713 
3714 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
3715 					bool make_grd)
3716 {
3717 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3718 		attr->opa.make_grd = make_grd;
3719 }
3720 
3721 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
3722 {
3723 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3724 		return attr->opa.make_grd;
3725 	return false;
3726 }
3727 
3728 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3729 {
3730 	attr->port_num = port_num;
3731 }
3732 
3733 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3734 {
3735 	return attr->port_num;
3736 }
3737 
3738 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3739 					   u8 static_rate)
3740 {
3741 	attr->static_rate = static_rate;
3742 }
3743 
3744 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3745 {
3746 	return attr->static_rate;
3747 }
3748 
3749 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3750 					enum ib_ah_flags flag)
3751 {
3752 	attr->ah_flags = flag;
3753 }
3754 
3755 static inline enum ib_ah_flags
3756 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3757 {
3758 	return attr->ah_flags;
3759 }
3760 
3761 static inline const struct ib_global_route
3762 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
3763 {
3764 	return &attr->grh;
3765 }
3766 
3767 /*To retrieve and modify the grh */
3768 static inline struct ib_global_route
3769 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
3770 {
3771 	return &attr->grh;
3772 }
3773 
3774 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
3775 {
3776 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3777 
3778 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
3779 }
3780 
3781 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
3782 					     __be64 prefix)
3783 {
3784 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3785 
3786 	grh->dgid.global.subnet_prefix = prefix;
3787 }
3788 
3789 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
3790 					    __be64 if_id)
3791 {
3792 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3793 
3794 	grh->dgid.global.interface_id = if_id;
3795 }
3796 
3797 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
3798 				   union ib_gid *dgid, u32 flow_label,
3799 				   u8 sgid_index, u8 hop_limit,
3800 				   u8 traffic_class)
3801 {
3802 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3803 
3804 	attr->ah_flags = IB_AH_GRH;
3805 	if (dgid)
3806 		grh->dgid = *dgid;
3807 	grh->flow_label = flow_label;
3808 	grh->sgid_index = sgid_index;
3809 	grh->hop_limit = hop_limit;
3810 	grh->traffic_class = traffic_class;
3811 }
3812 
3813 /*Get AH type */
3814 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
3815 						       u32 port_num)
3816 {
3817 	if (rdma_protocol_roce(dev, port_num))
3818 		return RDMA_AH_ATTR_TYPE_ROCE;
3819 	else if ((rdma_protocol_ib(dev, port_num)) &&
3820 		 (rdma_cap_opa_ah(dev, port_num)))
3821 		return RDMA_AH_ATTR_TYPE_OPA;
3822 	else
3823 		return RDMA_AH_ATTR_TYPE_IB;
3824 }
3825 
3826 /**
3827  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
3828  *     In the current implementation the only way to get
3829  *     get the 32bit lid is from other sources for OPA.
3830  *     For IB, lids will always be 16bits so cast the
3831  *     value accordingly.
3832  *
3833  * @lid: A 32bit LID
3834  */
3835 static inline u16 ib_lid_cpu16(u32 lid)
3836 {
3837 	WARN_ON_ONCE(lid & 0xFFFF0000);
3838 	return (u16)lid;
3839 }
3840 
3841 /**
3842  * ib_lid_be16 - Return lid in 16bit BE encoding.
3843  *
3844  * @lid: A 32bit LID
3845  */
3846 static inline __be16 ib_lid_be16(u32 lid)
3847 {
3848 	WARN_ON_ONCE(lid & 0xFFFF0000);
3849 	return cpu_to_be16((u16)lid);
3850 }
3851 
3852 /**
3853  * ib_get_vector_affinity - Get the affinity mappings of a given completion
3854  *   vector
3855  * @device:         the rdma device
3856  * @comp_vector:    index of completion vector
3857  *
3858  * Returns NULL on failure, otherwise a corresponding cpu map of the
3859  * completion vector (returns all-cpus map if the device driver doesn't
3860  * implement get_vector_affinity).
3861  */
3862 static inline const struct cpumask *
3863 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
3864 {
3865 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
3866 	    !device->get_vector_affinity)
3867 		return NULL;
3868 
3869 	return device->get_vector_affinity(device, comp_vector);
3870 
3871 }
3872 
3873 /**
3874  * rdma_roce_rescan_device - Rescan all of the network devices in the system
3875  * and add their gids, as needed, to the relevant RoCE devices.
3876  *
3877  * @device:         the rdma device
3878  */
3879 void rdma_roce_rescan_device(struct ib_device *ibdev);
3880 
3881 #endif /* IB_VERBS_H */
3882