xref: /openbmc/linux/include/rdma/ib_verbs.h (revision b240b419db5d624ce7a5a397d6f62a1a686009ec)
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_UNDEFINED,
878 	RDMA_AH_ATTR_TYPE_IB,
879 	RDMA_AH_ATTR_TYPE_ROCE,
880 	RDMA_AH_ATTR_TYPE_OPA,
881 };
882 
883 struct ib_ah_attr {
884 	u16			dlid;
885 	u8			src_path_bits;
886 };
887 
888 struct roce_ah_attr {
889 	u8			dmac[ETH_ALEN];
890 };
891 
892 struct opa_ah_attr {
893 	u32			dlid;
894 	u8			src_path_bits;
895 	bool			make_grd;
896 };
897 
898 struct rdma_ah_attr {
899 	struct ib_global_route	grh;
900 	u8			sl;
901 	u8			static_rate;
902 	u8			port_num;
903 	u8			ah_flags;
904 	enum rdma_ah_attr_type type;
905 	union {
906 		struct ib_ah_attr ib;
907 		struct roce_ah_attr roce;
908 		struct opa_ah_attr opa;
909 	};
910 };
911 
912 enum ib_wc_status {
913 	IB_WC_SUCCESS,
914 	IB_WC_LOC_LEN_ERR,
915 	IB_WC_LOC_QP_OP_ERR,
916 	IB_WC_LOC_EEC_OP_ERR,
917 	IB_WC_LOC_PROT_ERR,
918 	IB_WC_WR_FLUSH_ERR,
919 	IB_WC_MW_BIND_ERR,
920 	IB_WC_BAD_RESP_ERR,
921 	IB_WC_LOC_ACCESS_ERR,
922 	IB_WC_REM_INV_REQ_ERR,
923 	IB_WC_REM_ACCESS_ERR,
924 	IB_WC_REM_OP_ERR,
925 	IB_WC_RETRY_EXC_ERR,
926 	IB_WC_RNR_RETRY_EXC_ERR,
927 	IB_WC_LOC_RDD_VIOL_ERR,
928 	IB_WC_REM_INV_RD_REQ_ERR,
929 	IB_WC_REM_ABORT_ERR,
930 	IB_WC_INV_EECN_ERR,
931 	IB_WC_INV_EEC_STATE_ERR,
932 	IB_WC_FATAL_ERR,
933 	IB_WC_RESP_TIMEOUT_ERR,
934 	IB_WC_GENERAL_ERR
935 };
936 
937 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
938 
939 enum ib_wc_opcode {
940 	IB_WC_SEND,
941 	IB_WC_RDMA_WRITE,
942 	IB_WC_RDMA_READ,
943 	IB_WC_COMP_SWAP,
944 	IB_WC_FETCH_ADD,
945 	IB_WC_LSO,
946 	IB_WC_LOCAL_INV,
947 	IB_WC_REG_MR,
948 	IB_WC_MASKED_COMP_SWAP,
949 	IB_WC_MASKED_FETCH_ADD,
950 /*
951  * Set value of IB_WC_RECV so consumers can test if a completion is a
952  * receive by testing (opcode & IB_WC_RECV).
953  */
954 	IB_WC_RECV			= 1 << 7,
955 	IB_WC_RECV_RDMA_WITH_IMM
956 };
957 
958 enum ib_wc_flags {
959 	IB_WC_GRH		= 1,
960 	IB_WC_WITH_IMM		= (1<<1),
961 	IB_WC_WITH_INVALIDATE	= (1<<2),
962 	IB_WC_IP_CSUM_OK	= (1<<3),
963 	IB_WC_WITH_SMAC		= (1<<4),
964 	IB_WC_WITH_VLAN		= (1<<5),
965 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
966 };
967 
968 struct ib_wc {
969 	union {
970 		u64		wr_id;
971 		struct ib_cqe	*wr_cqe;
972 	};
973 	enum ib_wc_status	status;
974 	enum ib_wc_opcode	opcode;
975 	u32			vendor_err;
976 	u32			byte_len;
977 	struct ib_qp	       *qp;
978 	union {
979 		__be32		imm_data;
980 		u32		invalidate_rkey;
981 	} ex;
982 	u32			src_qp;
983 	u32			slid;
984 	int			wc_flags;
985 	u16			pkey_index;
986 	u8			sl;
987 	u8			dlid_path_bits;
988 	u8			port_num;	/* valid only for DR SMPs on switches */
989 	u8			smac[ETH_ALEN];
990 	u16			vlan_id;
991 	u8			network_hdr_type;
992 };
993 
994 enum ib_cq_notify_flags {
995 	IB_CQ_SOLICITED			= 1 << 0,
996 	IB_CQ_NEXT_COMP			= 1 << 1,
997 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
998 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
999 };
1000 
1001 enum ib_srq_type {
1002 	IB_SRQT_BASIC,
1003 	IB_SRQT_XRC,
1004 	IB_SRQT_TM,
1005 };
1006 
1007 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1008 {
1009 	return srq_type == IB_SRQT_XRC ||
1010 	       srq_type == IB_SRQT_TM;
1011 }
1012 
1013 enum ib_srq_attr_mask {
1014 	IB_SRQ_MAX_WR	= 1 << 0,
1015 	IB_SRQ_LIMIT	= 1 << 1,
1016 };
1017 
1018 struct ib_srq_attr {
1019 	u32	max_wr;
1020 	u32	max_sge;
1021 	u32	srq_limit;
1022 };
1023 
1024 struct ib_srq_init_attr {
1025 	void		      (*event_handler)(struct ib_event *, void *);
1026 	void		       *srq_context;
1027 	struct ib_srq_attr	attr;
1028 	enum ib_srq_type	srq_type;
1029 
1030 	struct {
1031 		struct ib_cq   *cq;
1032 		union {
1033 			struct {
1034 				struct ib_xrcd *xrcd;
1035 			} xrc;
1036 
1037 			struct {
1038 				u32		max_num_tags;
1039 			} tag_matching;
1040 		};
1041 	} ext;
1042 };
1043 
1044 struct ib_qp_cap {
1045 	u32	max_send_wr;
1046 	u32	max_recv_wr;
1047 	u32	max_send_sge;
1048 	u32	max_recv_sge;
1049 	u32	max_inline_data;
1050 
1051 	/*
1052 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1053 	 * ib_create_qp() will calculate the right amount of neededed WRs
1054 	 * and MRs based on this.
1055 	 */
1056 	u32	max_rdma_ctxs;
1057 };
1058 
1059 enum ib_sig_type {
1060 	IB_SIGNAL_ALL_WR,
1061 	IB_SIGNAL_REQ_WR
1062 };
1063 
1064 enum ib_qp_type {
1065 	/*
1066 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1067 	 * here (and in that order) since the MAD layer uses them as
1068 	 * indices into a 2-entry table.
1069 	 */
1070 	IB_QPT_SMI,
1071 	IB_QPT_GSI,
1072 
1073 	IB_QPT_RC,
1074 	IB_QPT_UC,
1075 	IB_QPT_UD,
1076 	IB_QPT_RAW_IPV6,
1077 	IB_QPT_RAW_ETHERTYPE,
1078 	IB_QPT_RAW_PACKET = 8,
1079 	IB_QPT_XRC_INI = 9,
1080 	IB_QPT_XRC_TGT,
1081 	IB_QPT_MAX,
1082 	IB_QPT_DRIVER = 0xFF,
1083 	/* Reserve a range for qp types internal to the low level driver.
1084 	 * These qp types will not be visible at the IB core layer, so the
1085 	 * IB_QPT_MAX usages should not be affected in the core layer
1086 	 */
1087 	IB_QPT_RESERVED1 = 0x1000,
1088 	IB_QPT_RESERVED2,
1089 	IB_QPT_RESERVED3,
1090 	IB_QPT_RESERVED4,
1091 	IB_QPT_RESERVED5,
1092 	IB_QPT_RESERVED6,
1093 	IB_QPT_RESERVED7,
1094 	IB_QPT_RESERVED8,
1095 	IB_QPT_RESERVED9,
1096 	IB_QPT_RESERVED10,
1097 };
1098 
1099 enum ib_qp_create_flags {
1100 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1101 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1102 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1103 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1104 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1105 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1106 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1107 	/* FREE					= 1 << 7, */
1108 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1109 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1110 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1111 	IB_QP_CREATE_PCI_WRITE_END_PADDING	= 1 << 11,
1112 	/* reserve bits 26-31 for low level drivers' internal use */
1113 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1114 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1115 };
1116 
1117 /*
1118  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1119  * callback to destroy the passed in QP.
1120  */
1121 
1122 struct ib_qp_init_attr {
1123 	void                  (*event_handler)(struct ib_event *, void *);
1124 	void		       *qp_context;
1125 	struct ib_cq	       *send_cq;
1126 	struct ib_cq	       *recv_cq;
1127 	struct ib_srq	       *srq;
1128 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1129 	struct ib_qp_cap	cap;
1130 	enum ib_sig_type	sq_sig_type;
1131 	enum ib_qp_type		qp_type;
1132 	enum ib_qp_create_flags	create_flags;
1133 
1134 	/*
1135 	 * Only needed for special QP types, or when using the RW API.
1136 	 */
1137 	u8			port_num;
1138 	struct ib_rwq_ind_table *rwq_ind_tbl;
1139 	u32			source_qpn;
1140 };
1141 
1142 struct ib_qp_open_attr {
1143 	void                  (*event_handler)(struct ib_event *, void *);
1144 	void		       *qp_context;
1145 	u32			qp_num;
1146 	enum ib_qp_type		qp_type;
1147 };
1148 
1149 enum ib_rnr_timeout {
1150 	IB_RNR_TIMER_655_36 =  0,
1151 	IB_RNR_TIMER_000_01 =  1,
1152 	IB_RNR_TIMER_000_02 =  2,
1153 	IB_RNR_TIMER_000_03 =  3,
1154 	IB_RNR_TIMER_000_04 =  4,
1155 	IB_RNR_TIMER_000_06 =  5,
1156 	IB_RNR_TIMER_000_08 =  6,
1157 	IB_RNR_TIMER_000_12 =  7,
1158 	IB_RNR_TIMER_000_16 =  8,
1159 	IB_RNR_TIMER_000_24 =  9,
1160 	IB_RNR_TIMER_000_32 = 10,
1161 	IB_RNR_TIMER_000_48 = 11,
1162 	IB_RNR_TIMER_000_64 = 12,
1163 	IB_RNR_TIMER_000_96 = 13,
1164 	IB_RNR_TIMER_001_28 = 14,
1165 	IB_RNR_TIMER_001_92 = 15,
1166 	IB_RNR_TIMER_002_56 = 16,
1167 	IB_RNR_TIMER_003_84 = 17,
1168 	IB_RNR_TIMER_005_12 = 18,
1169 	IB_RNR_TIMER_007_68 = 19,
1170 	IB_RNR_TIMER_010_24 = 20,
1171 	IB_RNR_TIMER_015_36 = 21,
1172 	IB_RNR_TIMER_020_48 = 22,
1173 	IB_RNR_TIMER_030_72 = 23,
1174 	IB_RNR_TIMER_040_96 = 24,
1175 	IB_RNR_TIMER_061_44 = 25,
1176 	IB_RNR_TIMER_081_92 = 26,
1177 	IB_RNR_TIMER_122_88 = 27,
1178 	IB_RNR_TIMER_163_84 = 28,
1179 	IB_RNR_TIMER_245_76 = 29,
1180 	IB_RNR_TIMER_327_68 = 30,
1181 	IB_RNR_TIMER_491_52 = 31
1182 };
1183 
1184 enum ib_qp_attr_mask {
1185 	IB_QP_STATE			= 1,
1186 	IB_QP_CUR_STATE			= (1<<1),
1187 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1188 	IB_QP_ACCESS_FLAGS		= (1<<3),
1189 	IB_QP_PKEY_INDEX		= (1<<4),
1190 	IB_QP_PORT			= (1<<5),
1191 	IB_QP_QKEY			= (1<<6),
1192 	IB_QP_AV			= (1<<7),
1193 	IB_QP_PATH_MTU			= (1<<8),
1194 	IB_QP_TIMEOUT			= (1<<9),
1195 	IB_QP_RETRY_CNT			= (1<<10),
1196 	IB_QP_RNR_RETRY			= (1<<11),
1197 	IB_QP_RQ_PSN			= (1<<12),
1198 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1199 	IB_QP_ALT_PATH			= (1<<14),
1200 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1201 	IB_QP_SQ_PSN			= (1<<16),
1202 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1203 	IB_QP_PATH_MIG_STATE		= (1<<18),
1204 	IB_QP_CAP			= (1<<19),
1205 	IB_QP_DEST_QPN			= (1<<20),
1206 	IB_QP_RESERVED1			= (1<<21),
1207 	IB_QP_RESERVED2			= (1<<22),
1208 	IB_QP_RESERVED3			= (1<<23),
1209 	IB_QP_RESERVED4			= (1<<24),
1210 	IB_QP_RATE_LIMIT		= (1<<25),
1211 };
1212 
1213 enum ib_qp_state {
1214 	IB_QPS_RESET,
1215 	IB_QPS_INIT,
1216 	IB_QPS_RTR,
1217 	IB_QPS_RTS,
1218 	IB_QPS_SQD,
1219 	IB_QPS_SQE,
1220 	IB_QPS_ERR
1221 };
1222 
1223 enum ib_mig_state {
1224 	IB_MIG_MIGRATED,
1225 	IB_MIG_REARM,
1226 	IB_MIG_ARMED
1227 };
1228 
1229 enum ib_mw_type {
1230 	IB_MW_TYPE_1 = 1,
1231 	IB_MW_TYPE_2 = 2
1232 };
1233 
1234 struct ib_qp_attr {
1235 	enum ib_qp_state	qp_state;
1236 	enum ib_qp_state	cur_qp_state;
1237 	enum ib_mtu		path_mtu;
1238 	enum ib_mig_state	path_mig_state;
1239 	u32			qkey;
1240 	u32			rq_psn;
1241 	u32			sq_psn;
1242 	u32			dest_qp_num;
1243 	int			qp_access_flags;
1244 	struct ib_qp_cap	cap;
1245 	struct rdma_ah_attr	ah_attr;
1246 	struct rdma_ah_attr	alt_ah_attr;
1247 	u16			pkey_index;
1248 	u16			alt_pkey_index;
1249 	u8			en_sqd_async_notify;
1250 	u8			sq_draining;
1251 	u8			max_rd_atomic;
1252 	u8			max_dest_rd_atomic;
1253 	u8			min_rnr_timer;
1254 	u8			port_num;
1255 	u8			timeout;
1256 	u8			retry_cnt;
1257 	u8			rnr_retry;
1258 	u8			alt_port_num;
1259 	u8			alt_timeout;
1260 	u32			rate_limit;
1261 };
1262 
1263 enum ib_wr_opcode {
1264 	IB_WR_RDMA_WRITE,
1265 	IB_WR_RDMA_WRITE_WITH_IMM,
1266 	IB_WR_SEND,
1267 	IB_WR_SEND_WITH_IMM,
1268 	IB_WR_RDMA_READ,
1269 	IB_WR_ATOMIC_CMP_AND_SWP,
1270 	IB_WR_ATOMIC_FETCH_AND_ADD,
1271 	IB_WR_LSO,
1272 	IB_WR_SEND_WITH_INV,
1273 	IB_WR_RDMA_READ_WITH_INV,
1274 	IB_WR_LOCAL_INV,
1275 	IB_WR_REG_MR,
1276 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1277 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1278 	IB_WR_REG_SIG_MR,
1279 	/* reserve values for low level drivers' internal use.
1280 	 * These values will not be used at all in the ib core layer.
1281 	 */
1282 	IB_WR_RESERVED1 = 0xf0,
1283 	IB_WR_RESERVED2,
1284 	IB_WR_RESERVED3,
1285 	IB_WR_RESERVED4,
1286 	IB_WR_RESERVED5,
1287 	IB_WR_RESERVED6,
1288 	IB_WR_RESERVED7,
1289 	IB_WR_RESERVED8,
1290 	IB_WR_RESERVED9,
1291 	IB_WR_RESERVED10,
1292 };
1293 
1294 enum ib_send_flags {
1295 	IB_SEND_FENCE		= 1,
1296 	IB_SEND_SIGNALED	= (1<<1),
1297 	IB_SEND_SOLICITED	= (1<<2),
1298 	IB_SEND_INLINE		= (1<<3),
1299 	IB_SEND_IP_CSUM		= (1<<4),
1300 
1301 	/* reserve bits 26-31 for low level drivers' internal use */
1302 	IB_SEND_RESERVED_START	= (1 << 26),
1303 	IB_SEND_RESERVED_END	= (1 << 31),
1304 };
1305 
1306 struct ib_sge {
1307 	u64	addr;
1308 	u32	length;
1309 	u32	lkey;
1310 };
1311 
1312 struct ib_cqe {
1313 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1314 };
1315 
1316 struct ib_send_wr {
1317 	struct ib_send_wr      *next;
1318 	union {
1319 		u64		wr_id;
1320 		struct ib_cqe	*wr_cqe;
1321 	};
1322 	struct ib_sge	       *sg_list;
1323 	int			num_sge;
1324 	enum ib_wr_opcode	opcode;
1325 	int			send_flags;
1326 	union {
1327 		__be32		imm_data;
1328 		u32		invalidate_rkey;
1329 	} ex;
1330 };
1331 
1332 struct ib_rdma_wr {
1333 	struct ib_send_wr	wr;
1334 	u64			remote_addr;
1335 	u32			rkey;
1336 };
1337 
1338 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1339 {
1340 	return container_of(wr, struct ib_rdma_wr, wr);
1341 }
1342 
1343 struct ib_atomic_wr {
1344 	struct ib_send_wr	wr;
1345 	u64			remote_addr;
1346 	u64			compare_add;
1347 	u64			swap;
1348 	u64			compare_add_mask;
1349 	u64			swap_mask;
1350 	u32			rkey;
1351 };
1352 
1353 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1354 {
1355 	return container_of(wr, struct ib_atomic_wr, wr);
1356 }
1357 
1358 struct ib_ud_wr {
1359 	struct ib_send_wr	wr;
1360 	struct ib_ah		*ah;
1361 	void			*header;
1362 	int			hlen;
1363 	int			mss;
1364 	u32			remote_qpn;
1365 	u32			remote_qkey;
1366 	u16			pkey_index; /* valid for GSI only */
1367 	u8			port_num;   /* valid for DR SMPs on switch only */
1368 };
1369 
1370 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1371 {
1372 	return container_of(wr, struct ib_ud_wr, wr);
1373 }
1374 
1375 struct ib_reg_wr {
1376 	struct ib_send_wr	wr;
1377 	struct ib_mr		*mr;
1378 	u32			key;
1379 	int			access;
1380 };
1381 
1382 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1383 {
1384 	return container_of(wr, struct ib_reg_wr, wr);
1385 }
1386 
1387 struct ib_sig_handover_wr {
1388 	struct ib_send_wr	wr;
1389 	struct ib_sig_attrs    *sig_attrs;
1390 	struct ib_mr	       *sig_mr;
1391 	int			access_flags;
1392 	struct ib_sge	       *prot;
1393 };
1394 
1395 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1396 {
1397 	return container_of(wr, struct ib_sig_handover_wr, wr);
1398 }
1399 
1400 struct ib_recv_wr {
1401 	struct ib_recv_wr      *next;
1402 	union {
1403 		u64		wr_id;
1404 		struct ib_cqe	*wr_cqe;
1405 	};
1406 	struct ib_sge	       *sg_list;
1407 	int			num_sge;
1408 };
1409 
1410 enum ib_access_flags {
1411 	IB_ACCESS_LOCAL_WRITE	= 1,
1412 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1413 	IB_ACCESS_REMOTE_READ	= (1<<2),
1414 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1415 	IB_ACCESS_MW_BIND	= (1<<4),
1416 	IB_ZERO_BASED		= (1<<5),
1417 	IB_ACCESS_ON_DEMAND     = (1<<6),
1418 	IB_ACCESS_HUGETLB	= (1<<7),
1419 };
1420 
1421 /*
1422  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1423  * are hidden here instead of a uapi header!
1424  */
1425 enum ib_mr_rereg_flags {
1426 	IB_MR_REREG_TRANS	= 1,
1427 	IB_MR_REREG_PD		= (1<<1),
1428 	IB_MR_REREG_ACCESS	= (1<<2),
1429 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1430 };
1431 
1432 struct ib_fmr_attr {
1433 	int	max_pages;
1434 	int	max_maps;
1435 	u8	page_shift;
1436 };
1437 
1438 struct ib_umem;
1439 
1440 enum rdma_remove_reason {
1441 	/* Userspace requested uobject deletion. Call could fail */
1442 	RDMA_REMOVE_DESTROY,
1443 	/* Context deletion. This call should delete the actual object itself */
1444 	RDMA_REMOVE_CLOSE,
1445 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1446 	RDMA_REMOVE_DRIVER_REMOVE,
1447 	/* Context is being cleaned-up, but commit was just completed */
1448 	RDMA_REMOVE_DURING_CLEANUP,
1449 };
1450 
1451 struct ib_rdmacg_object {
1452 #ifdef CONFIG_CGROUP_RDMA
1453 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1454 #endif
1455 };
1456 
1457 struct ib_ucontext {
1458 	struct ib_device       *device;
1459 	struct ib_uverbs_file  *ufile;
1460 	int			closing;
1461 
1462 	/* locking the uobjects_list */
1463 	struct mutex		uobjects_lock;
1464 	struct list_head	uobjects;
1465 	/* protects cleanup process from other actions */
1466 	struct rw_semaphore	cleanup_rwsem;
1467 	enum rdma_remove_reason cleanup_reason;
1468 
1469 	struct pid             *tgid;
1470 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1471 	struct rb_root_cached   umem_tree;
1472 	/*
1473 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1474 	 * mmu notifiers registration.
1475 	 */
1476 	struct rw_semaphore	umem_rwsem;
1477 	void (*invalidate_range)(struct ib_umem *umem,
1478 				 unsigned long start, unsigned long end);
1479 
1480 	struct mmu_notifier	mn;
1481 	atomic_t		notifier_count;
1482 	/* A list of umems that don't have private mmu notifier counters yet. */
1483 	struct list_head	no_private_counters;
1484 	int                     odp_mrs_count;
1485 #endif
1486 
1487 	struct ib_rdmacg_object	cg_obj;
1488 };
1489 
1490 struct ib_uobject {
1491 	u64			user_handle;	/* handle given to us by userspace */
1492 	struct ib_ucontext     *context;	/* associated user context */
1493 	void		       *object;		/* containing object */
1494 	struct list_head	list;		/* link to context's list */
1495 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1496 	int			id;		/* index into kernel idr */
1497 	struct kref		ref;
1498 	atomic_t		usecnt;		/* protects exclusive access */
1499 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1500 
1501 	const struct uverbs_obj_type *type;
1502 };
1503 
1504 struct ib_uobject_file {
1505 	struct ib_uobject	uobj;
1506 	/* ufile contains the lock between context release and file close */
1507 	struct ib_uverbs_file	*ufile;
1508 };
1509 
1510 struct ib_udata {
1511 	const void __user *inbuf;
1512 	void __user *outbuf;
1513 	size_t       inlen;
1514 	size_t       outlen;
1515 };
1516 
1517 struct ib_pd {
1518 	u32			local_dma_lkey;
1519 	u32			flags;
1520 	struct ib_device       *device;
1521 	struct ib_uobject      *uobject;
1522 	atomic_t          	usecnt; /* count all resources */
1523 
1524 	u32			unsafe_global_rkey;
1525 
1526 	/*
1527 	 * Implementation details of the RDMA core, don't use in drivers:
1528 	 */
1529 	struct ib_mr	       *__internal_mr;
1530 	struct rdma_restrack_entry res;
1531 };
1532 
1533 struct ib_xrcd {
1534 	struct ib_device       *device;
1535 	atomic_t		usecnt; /* count all exposed resources */
1536 	struct inode	       *inode;
1537 
1538 	struct mutex		tgt_qp_mutex;
1539 	struct list_head	tgt_qp_list;
1540 };
1541 
1542 struct ib_ah {
1543 	struct ib_device	*device;
1544 	struct ib_pd		*pd;
1545 	struct ib_uobject	*uobject;
1546 	enum rdma_ah_attr_type	type;
1547 };
1548 
1549 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1550 
1551 enum ib_poll_context {
1552 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1553 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1554 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1555 };
1556 
1557 struct ib_cq {
1558 	struct ib_device       *device;
1559 	struct ib_uobject      *uobject;
1560 	ib_comp_handler   	comp_handler;
1561 	void                  (*event_handler)(struct ib_event *, void *);
1562 	void                   *cq_context;
1563 	int               	cqe;
1564 	atomic_t          	usecnt; /* count number of work queues */
1565 	enum ib_poll_context	poll_ctx;
1566 	struct ib_wc		*wc;
1567 	union {
1568 		struct irq_poll		iop;
1569 		struct work_struct	work;
1570 	};
1571 	/*
1572 	 * Implementation details of the RDMA core, don't use in drivers:
1573 	 */
1574 	struct rdma_restrack_entry res;
1575 };
1576 
1577 struct ib_srq {
1578 	struct ib_device       *device;
1579 	struct ib_pd	       *pd;
1580 	struct ib_uobject      *uobject;
1581 	void		      (*event_handler)(struct ib_event *, void *);
1582 	void		       *srq_context;
1583 	enum ib_srq_type	srq_type;
1584 	atomic_t		usecnt;
1585 
1586 	struct {
1587 		struct ib_cq   *cq;
1588 		union {
1589 			struct {
1590 				struct ib_xrcd *xrcd;
1591 				u32		srq_num;
1592 			} xrc;
1593 		};
1594 	} ext;
1595 };
1596 
1597 enum ib_raw_packet_caps {
1598 	/* Strip cvlan from incoming packet and report it in the matching work
1599 	 * completion is supported.
1600 	 */
1601 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1602 	/* Scatter FCS field of an incoming packet to host memory is supported.
1603 	 */
1604 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1605 	/* Checksum offloads are supported (for both send and receive). */
1606 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1607 	/* When a packet is received for an RQ with no receive WQEs, the
1608 	 * packet processing is delayed.
1609 	 */
1610 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1611 };
1612 
1613 enum ib_wq_type {
1614 	IB_WQT_RQ
1615 };
1616 
1617 enum ib_wq_state {
1618 	IB_WQS_RESET,
1619 	IB_WQS_RDY,
1620 	IB_WQS_ERR
1621 };
1622 
1623 struct ib_wq {
1624 	struct ib_device       *device;
1625 	struct ib_uobject      *uobject;
1626 	void		    *wq_context;
1627 	void		    (*event_handler)(struct ib_event *, void *);
1628 	struct ib_pd	       *pd;
1629 	struct ib_cq	       *cq;
1630 	u32		wq_num;
1631 	enum ib_wq_state       state;
1632 	enum ib_wq_type	wq_type;
1633 	atomic_t		usecnt;
1634 };
1635 
1636 enum ib_wq_flags {
1637 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1638 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1639 	IB_WQ_FLAGS_DELAY_DROP		= 1 << 2,
1640 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1641 };
1642 
1643 struct ib_wq_init_attr {
1644 	void		       *wq_context;
1645 	enum ib_wq_type	wq_type;
1646 	u32		max_wr;
1647 	u32		max_sge;
1648 	struct	ib_cq	       *cq;
1649 	void		    (*event_handler)(struct ib_event *, void *);
1650 	u32		create_flags; /* Use enum ib_wq_flags */
1651 };
1652 
1653 enum ib_wq_attr_mask {
1654 	IB_WQ_STATE		= 1 << 0,
1655 	IB_WQ_CUR_STATE		= 1 << 1,
1656 	IB_WQ_FLAGS		= 1 << 2,
1657 };
1658 
1659 struct ib_wq_attr {
1660 	enum	ib_wq_state	wq_state;
1661 	enum	ib_wq_state	curr_wq_state;
1662 	u32			flags; /* Use enum ib_wq_flags */
1663 	u32			flags_mask; /* Use enum ib_wq_flags */
1664 };
1665 
1666 struct ib_rwq_ind_table {
1667 	struct ib_device	*device;
1668 	struct ib_uobject      *uobject;
1669 	atomic_t		usecnt;
1670 	u32		ind_tbl_num;
1671 	u32		log_ind_tbl_size;
1672 	struct ib_wq	**ind_tbl;
1673 };
1674 
1675 struct ib_rwq_ind_table_init_attr {
1676 	u32		log_ind_tbl_size;
1677 	/* Each entry is a pointer to Receive Work Queue */
1678 	struct ib_wq	**ind_tbl;
1679 };
1680 
1681 enum port_pkey_state {
1682 	IB_PORT_PKEY_NOT_VALID = 0,
1683 	IB_PORT_PKEY_VALID = 1,
1684 	IB_PORT_PKEY_LISTED = 2,
1685 };
1686 
1687 struct ib_qp_security;
1688 
1689 struct ib_port_pkey {
1690 	enum port_pkey_state	state;
1691 	u16			pkey_index;
1692 	u8			port_num;
1693 	struct list_head	qp_list;
1694 	struct list_head	to_error_list;
1695 	struct ib_qp_security  *sec;
1696 };
1697 
1698 struct ib_ports_pkeys {
1699 	struct ib_port_pkey	main;
1700 	struct ib_port_pkey	alt;
1701 };
1702 
1703 struct ib_qp_security {
1704 	struct ib_qp	       *qp;
1705 	struct ib_device       *dev;
1706 	/* Hold this mutex when changing port and pkey settings. */
1707 	struct mutex		mutex;
1708 	struct ib_ports_pkeys  *ports_pkeys;
1709 	/* A list of all open shared QP handles.  Required to enforce security
1710 	 * properly for all users of a shared QP.
1711 	 */
1712 	struct list_head        shared_qp_list;
1713 	void                   *security;
1714 	bool			destroying;
1715 	atomic_t		error_list_count;
1716 	struct completion	error_complete;
1717 	int			error_comps_pending;
1718 };
1719 
1720 /*
1721  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1722  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1723  */
1724 struct ib_qp {
1725 	struct ib_device       *device;
1726 	struct ib_pd	       *pd;
1727 	struct ib_cq	       *send_cq;
1728 	struct ib_cq	       *recv_cq;
1729 	spinlock_t		mr_lock;
1730 	int			mrs_used;
1731 	struct list_head	rdma_mrs;
1732 	struct list_head	sig_mrs;
1733 	struct ib_srq	       *srq;
1734 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1735 	struct list_head	xrcd_list;
1736 
1737 	/* count times opened, mcast attaches, flow attaches */
1738 	atomic_t		usecnt;
1739 	struct list_head	open_list;
1740 	struct ib_qp           *real_qp;
1741 	struct ib_uobject      *uobject;
1742 	void                  (*event_handler)(struct ib_event *, void *);
1743 	void		       *qp_context;
1744 	u32			qp_num;
1745 	u32			max_write_sge;
1746 	u32			max_read_sge;
1747 	enum ib_qp_type		qp_type;
1748 	struct ib_rwq_ind_table *rwq_ind_tbl;
1749 	struct ib_qp_security  *qp_sec;
1750 	u8			port;
1751 
1752 	/*
1753 	 * Implementation details of the RDMA core, don't use in drivers:
1754 	 */
1755 	struct rdma_restrack_entry     res;
1756 };
1757 
1758 struct ib_mr {
1759 	struct ib_device  *device;
1760 	struct ib_pd	  *pd;
1761 	u32		   lkey;
1762 	u32		   rkey;
1763 	u64		   iova;
1764 	u64		   length;
1765 	unsigned int	   page_size;
1766 	bool		   need_inval;
1767 	union {
1768 		struct ib_uobject	*uobject;	/* user */
1769 		struct list_head	qp_entry;	/* FR */
1770 	};
1771 };
1772 
1773 struct ib_mw {
1774 	struct ib_device	*device;
1775 	struct ib_pd		*pd;
1776 	struct ib_uobject	*uobject;
1777 	u32			rkey;
1778 	enum ib_mw_type         type;
1779 };
1780 
1781 struct ib_fmr {
1782 	struct ib_device	*device;
1783 	struct ib_pd		*pd;
1784 	struct list_head	list;
1785 	u32			lkey;
1786 	u32			rkey;
1787 };
1788 
1789 /* Supported steering options */
1790 enum ib_flow_attr_type {
1791 	/* steering according to rule specifications */
1792 	IB_FLOW_ATTR_NORMAL		= 0x0,
1793 	/* default unicast and multicast rule -
1794 	 * receive all Eth traffic which isn't steered to any QP
1795 	 */
1796 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1797 	/* default multicast rule -
1798 	 * receive all Eth multicast traffic which isn't steered to any QP
1799 	 */
1800 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1801 	/* sniffer rule - receive all port traffic */
1802 	IB_FLOW_ATTR_SNIFFER		= 0x3
1803 };
1804 
1805 /* Supported steering header types */
1806 enum ib_flow_spec_type {
1807 	/* L2 headers*/
1808 	IB_FLOW_SPEC_ETH		= 0x20,
1809 	IB_FLOW_SPEC_IB			= 0x22,
1810 	/* L3 header*/
1811 	IB_FLOW_SPEC_IPV4		= 0x30,
1812 	IB_FLOW_SPEC_IPV6		= 0x31,
1813 	/* L4 headers*/
1814 	IB_FLOW_SPEC_TCP		= 0x40,
1815 	IB_FLOW_SPEC_UDP		= 0x41,
1816 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1817 	IB_FLOW_SPEC_INNER		= 0x100,
1818 	/* Actions */
1819 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1820 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1821 };
1822 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1823 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1824 
1825 /* Flow steering rule priority is set according to it's domain.
1826  * Lower domain value means higher priority.
1827  */
1828 enum ib_flow_domain {
1829 	IB_FLOW_DOMAIN_USER,
1830 	IB_FLOW_DOMAIN_ETHTOOL,
1831 	IB_FLOW_DOMAIN_RFS,
1832 	IB_FLOW_DOMAIN_NIC,
1833 	IB_FLOW_DOMAIN_NUM /* Must be last */
1834 };
1835 
1836 enum ib_flow_flags {
1837 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1838 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1839 };
1840 
1841 struct ib_flow_eth_filter {
1842 	u8	dst_mac[6];
1843 	u8	src_mac[6];
1844 	__be16	ether_type;
1845 	__be16	vlan_tag;
1846 	/* Must be last */
1847 	u8	real_sz[0];
1848 };
1849 
1850 struct ib_flow_spec_eth {
1851 	u32			  type;
1852 	u16			  size;
1853 	struct ib_flow_eth_filter val;
1854 	struct ib_flow_eth_filter mask;
1855 };
1856 
1857 struct ib_flow_ib_filter {
1858 	__be16 dlid;
1859 	__u8   sl;
1860 	/* Must be last */
1861 	u8	real_sz[0];
1862 };
1863 
1864 struct ib_flow_spec_ib {
1865 	u32			 type;
1866 	u16			 size;
1867 	struct ib_flow_ib_filter val;
1868 	struct ib_flow_ib_filter mask;
1869 };
1870 
1871 /* IPv4 header flags */
1872 enum ib_ipv4_flags {
1873 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1874 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1875 				    last have this flag set */
1876 };
1877 
1878 struct ib_flow_ipv4_filter {
1879 	__be32	src_ip;
1880 	__be32	dst_ip;
1881 	u8	proto;
1882 	u8	tos;
1883 	u8	ttl;
1884 	u8	flags;
1885 	/* Must be last */
1886 	u8	real_sz[0];
1887 };
1888 
1889 struct ib_flow_spec_ipv4 {
1890 	u32			   type;
1891 	u16			   size;
1892 	struct ib_flow_ipv4_filter val;
1893 	struct ib_flow_ipv4_filter mask;
1894 };
1895 
1896 struct ib_flow_ipv6_filter {
1897 	u8	src_ip[16];
1898 	u8	dst_ip[16];
1899 	__be32	flow_label;
1900 	u8	next_hdr;
1901 	u8	traffic_class;
1902 	u8	hop_limit;
1903 	/* Must be last */
1904 	u8	real_sz[0];
1905 };
1906 
1907 struct ib_flow_spec_ipv6 {
1908 	u32			   type;
1909 	u16			   size;
1910 	struct ib_flow_ipv6_filter val;
1911 	struct ib_flow_ipv6_filter mask;
1912 };
1913 
1914 struct ib_flow_tcp_udp_filter {
1915 	__be16	dst_port;
1916 	__be16	src_port;
1917 	/* Must be last */
1918 	u8	real_sz[0];
1919 };
1920 
1921 struct ib_flow_spec_tcp_udp {
1922 	u32			      type;
1923 	u16			      size;
1924 	struct ib_flow_tcp_udp_filter val;
1925 	struct ib_flow_tcp_udp_filter mask;
1926 };
1927 
1928 struct ib_flow_tunnel_filter {
1929 	__be32	tunnel_id;
1930 	u8	real_sz[0];
1931 };
1932 
1933 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1934  * the tunnel_id from val has the vni value
1935  */
1936 struct ib_flow_spec_tunnel {
1937 	u32			      type;
1938 	u16			      size;
1939 	struct ib_flow_tunnel_filter  val;
1940 	struct ib_flow_tunnel_filter  mask;
1941 };
1942 
1943 struct ib_flow_spec_action_tag {
1944 	enum ib_flow_spec_type	      type;
1945 	u16			      size;
1946 	u32                           tag_id;
1947 };
1948 
1949 struct ib_flow_spec_action_drop {
1950 	enum ib_flow_spec_type	      type;
1951 	u16			      size;
1952 };
1953 
1954 union ib_flow_spec {
1955 	struct {
1956 		u32			type;
1957 		u16			size;
1958 	};
1959 	struct ib_flow_spec_eth		eth;
1960 	struct ib_flow_spec_ib		ib;
1961 	struct ib_flow_spec_ipv4        ipv4;
1962 	struct ib_flow_spec_tcp_udp	tcp_udp;
1963 	struct ib_flow_spec_ipv6        ipv6;
1964 	struct ib_flow_spec_tunnel      tunnel;
1965 	struct ib_flow_spec_action_tag  flow_tag;
1966 	struct ib_flow_spec_action_drop drop;
1967 };
1968 
1969 struct ib_flow_attr {
1970 	enum ib_flow_attr_type type;
1971 	u16	     size;
1972 	u16	     priority;
1973 	u32	     flags;
1974 	u8	     num_of_specs;
1975 	u8	     port;
1976 	/* Following are the optional layers according to user request
1977 	 * struct ib_flow_spec_xxx
1978 	 * struct ib_flow_spec_yyy
1979 	 */
1980 };
1981 
1982 struct ib_flow {
1983 	struct ib_qp		*qp;
1984 	struct ib_uobject	*uobject;
1985 };
1986 
1987 struct ib_mad_hdr;
1988 struct ib_grh;
1989 
1990 enum ib_process_mad_flags {
1991 	IB_MAD_IGNORE_MKEY	= 1,
1992 	IB_MAD_IGNORE_BKEY	= 2,
1993 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1994 };
1995 
1996 enum ib_mad_result {
1997 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1998 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1999 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2000 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2001 };
2002 
2003 struct ib_port_cache {
2004 	u64		      subnet_prefix;
2005 	struct ib_pkey_cache  *pkey;
2006 	struct ib_gid_table   *gid;
2007 	u8                     lmc;
2008 	enum ib_port_state     port_state;
2009 };
2010 
2011 struct ib_cache {
2012 	rwlock_t                lock;
2013 	struct ib_event_handler event_handler;
2014 	struct ib_port_cache   *ports;
2015 };
2016 
2017 struct iw_cm_verbs;
2018 
2019 struct ib_port_immutable {
2020 	int                           pkey_tbl_len;
2021 	int                           gid_tbl_len;
2022 	u32                           core_cap_flags;
2023 	u32                           max_mad_size;
2024 };
2025 
2026 /* rdma netdev type - specifies protocol type */
2027 enum rdma_netdev_t {
2028 	RDMA_NETDEV_OPA_VNIC,
2029 	RDMA_NETDEV_IPOIB,
2030 };
2031 
2032 /**
2033  * struct rdma_netdev - rdma netdev
2034  * For cases where netstack interfacing is required.
2035  */
2036 struct rdma_netdev {
2037 	void              *clnt_priv;
2038 	struct ib_device  *hca;
2039 	u8                 port_num;
2040 
2041 	/* cleanup function must be specified */
2042 	void (*free_rdma_netdev)(struct net_device *netdev);
2043 
2044 	/* control functions */
2045 	void (*set_id)(struct net_device *netdev, int id);
2046 	/* send packet */
2047 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2048 		    struct ib_ah *address, u32 dqpn);
2049 	/* multicast */
2050 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2051 			    union ib_gid *gid, u16 mlid,
2052 			    int set_qkey, u32 qkey);
2053 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2054 			    union ib_gid *gid, u16 mlid);
2055 };
2056 
2057 struct ib_port_pkey_list {
2058 	/* Lock to hold while modifying the list. */
2059 	spinlock_t                    list_lock;
2060 	struct list_head              pkey_list;
2061 };
2062 
2063 struct ib_device {
2064 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2065 	struct device                *dma_device;
2066 
2067 	char                          name[IB_DEVICE_NAME_MAX];
2068 
2069 	struct list_head              event_handler_list;
2070 	spinlock_t                    event_handler_lock;
2071 
2072 	spinlock_t                    client_data_lock;
2073 	struct list_head              core_list;
2074 	/* Access to the client_data_list is protected by the client_data_lock
2075 	 * spinlock and the lists_rwsem read-write semaphore */
2076 	struct list_head              client_data_list;
2077 
2078 	struct ib_cache               cache;
2079 	/**
2080 	 * port_immutable is indexed by port number
2081 	 */
2082 	struct ib_port_immutable     *port_immutable;
2083 
2084 	int			      num_comp_vectors;
2085 
2086 	struct ib_port_pkey_list     *port_pkey_list;
2087 
2088 	struct iw_cm_verbs	     *iwcm;
2089 
2090 	/**
2091 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2092 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2093 	 *   core when the device is removed.  A lifespan of -1 in the return
2094 	 *   struct tells the core to set a default lifespan.
2095 	 */
2096 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
2097 						     u8 port_num);
2098 	/**
2099 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2100 	 * @index - The index in the value array we wish to have updated, or
2101 	 *   num_counters if we want all stats updated
2102 	 * Return codes -
2103 	 *   < 0 - Error, no counters updated
2104 	 *   index - Updated the single counter pointed to by index
2105 	 *   num_counters - Updated all counters (will reset the timestamp
2106 	 *     and prevent further calls for lifespan milliseconds)
2107 	 * Drivers are allowed to update all counters in leiu of just the
2108 	 *   one given in index at their option
2109 	 */
2110 	int		           (*get_hw_stats)(struct ib_device *device,
2111 						   struct rdma_hw_stats *stats,
2112 						   u8 port, int index);
2113 	int		           (*query_device)(struct ib_device *device,
2114 						   struct ib_device_attr *device_attr,
2115 						   struct ib_udata *udata);
2116 	int		           (*query_port)(struct ib_device *device,
2117 						 u8 port_num,
2118 						 struct ib_port_attr *port_attr);
2119 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
2120 						     u8 port_num);
2121 	/* When calling get_netdev, the HW vendor's driver should return the
2122 	 * net device of device @device at port @port_num or NULL if such
2123 	 * a net device doesn't exist. The vendor driver should call dev_hold
2124 	 * on this net device. The HW vendor's device driver must guarantee
2125 	 * that this function returns NULL before the net device has finished
2126 	 * NETDEV_UNREGISTER state.
2127 	 */
2128 	struct net_device	  *(*get_netdev)(struct ib_device *device,
2129 						 u8 port_num);
2130 	int		           (*query_gid)(struct ib_device *device,
2131 						u8 port_num, int index,
2132 						union ib_gid *gid);
2133 	/* When calling add_gid, the HW vendor's driver should
2134 	 * add the gid of device @device at gid index @index of
2135 	 * port @port_num to be @gid. Meta-info of that gid (for example,
2136 	 * the network device related to this gid is available
2137 	 * at @attr. @context allows the HW vendor driver to store extra
2138 	 * information together with a GID entry. The HW vendor may allocate
2139 	 * memory to contain this information and store it in @context when a
2140 	 * new GID entry is written to. Params are consistent until the next
2141 	 * call of add_gid or delete_gid. The function should return 0 on
2142 	 * success or error otherwise. The function could be called
2143 	 * concurrently for different ports. This function is only called
2144 	 * when roce_gid_table is used.
2145 	 */
2146 	int		           (*add_gid)(struct ib_device *device,
2147 					      u8 port_num,
2148 					      unsigned int index,
2149 					      const union ib_gid *gid,
2150 					      const struct ib_gid_attr *attr,
2151 					      void **context);
2152 	/* When calling del_gid, the HW vendor's driver should delete the
2153 	 * gid of device @device at gid index @index of port @port_num.
2154 	 * Upon the deletion of a GID entry, the HW vendor must free any
2155 	 * allocated memory. The caller will clear @context afterwards.
2156 	 * This function is only called when roce_gid_table is used.
2157 	 */
2158 	int		           (*del_gid)(struct ib_device *device,
2159 					      u8 port_num,
2160 					      unsigned int index,
2161 					      void **context);
2162 	int		           (*query_pkey)(struct ib_device *device,
2163 						 u8 port_num, u16 index, u16 *pkey);
2164 	int		           (*modify_device)(struct ib_device *device,
2165 						    int device_modify_mask,
2166 						    struct ib_device_modify *device_modify);
2167 	int		           (*modify_port)(struct ib_device *device,
2168 						  u8 port_num, int port_modify_mask,
2169 						  struct ib_port_modify *port_modify);
2170 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
2171 						     struct ib_udata *udata);
2172 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
2173 	int                        (*mmap)(struct ib_ucontext *context,
2174 					   struct vm_area_struct *vma);
2175 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
2176 					       struct ib_ucontext *context,
2177 					       struct ib_udata *udata);
2178 	int                        (*dealloc_pd)(struct ib_pd *pd);
2179 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
2180 						struct rdma_ah_attr *ah_attr,
2181 						struct ib_udata *udata);
2182 	int                        (*modify_ah)(struct ib_ah *ah,
2183 						struct rdma_ah_attr *ah_attr);
2184 	int                        (*query_ah)(struct ib_ah *ah,
2185 					       struct rdma_ah_attr *ah_attr);
2186 	int                        (*destroy_ah)(struct ib_ah *ah);
2187 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
2188 						 struct ib_srq_init_attr *srq_init_attr,
2189 						 struct ib_udata *udata);
2190 	int                        (*modify_srq)(struct ib_srq *srq,
2191 						 struct ib_srq_attr *srq_attr,
2192 						 enum ib_srq_attr_mask srq_attr_mask,
2193 						 struct ib_udata *udata);
2194 	int                        (*query_srq)(struct ib_srq *srq,
2195 						struct ib_srq_attr *srq_attr);
2196 	int                        (*destroy_srq)(struct ib_srq *srq);
2197 	int                        (*post_srq_recv)(struct ib_srq *srq,
2198 						    struct ib_recv_wr *recv_wr,
2199 						    struct ib_recv_wr **bad_recv_wr);
2200 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
2201 						struct ib_qp_init_attr *qp_init_attr,
2202 						struct ib_udata *udata);
2203 	int                        (*modify_qp)(struct ib_qp *qp,
2204 						struct ib_qp_attr *qp_attr,
2205 						int qp_attr_mask,
2206 						struct ib_udata *udata);
2207 	int                        (*query_qp)(struct ib_qp *qp,
2208 					       struct ib_qp_attr *qp_attr,
2209 					       int qp_attr_mask,
2210 					       struct ib_qp_init_attr *qp_init_attr);
2211 	int                        (*destroy_qp)(struct ib_qp *qp);
2212 	int                        (*post_send)(struct ib_qp *qp,
2213 						struct ib_send_wr *send_wr,
2214 						struct ib_send_wr **bad_send_wr);
2215 	int                        (*post_recv)(struct ib_qp *qp,
2216 						struct ib_recv_wr *recv_wr,
2217 						struct ib_recv_wr **bad_recv_wr);
2218 	struct ib_cq *             (*create_cq)(struct ib_device *device,
2219 						const struct ib_cq_init_attr *attr,
2220 						struct ib_ucontext *context,
2221 						struct ib_udata *udata);
2222 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2223 						u16 cq_period);
2224 	int                        (*destroy_cq)(struct ib_cq *cq);
2225 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
2226 						struct ib_udata *udata);
2227 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
2228 					      struct ib_wc *wc);
2229 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2230 	int                        (*req_notify_cq)(struct ib_cq *cq,
2231 						    enum ib_cq_notify_flags flags);
2232 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
2233 						      int wc_cnt);
2234 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
2235 						 int mr_access_flags);
2236 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
2237 						  u64 start, u64 length,
2238 						  u64 virt_addr,
2239 						  int mr_access_flags,
2240 						  struct ib_udata *udata);
2241 	int			   (*rereg_user_mr)(struct ib_mr *mr,
2242 						    int flags,
2243 						    u64 start, u64 length,
2244 						    u64 virt_addr,
2245 						    int mr_access_flags,
2246 						    struct ib_pd *pd,
2247 						    struct ib_udata *udata);
2248 	int                        (*dereg_mr)(struct ib_mr *mr);
2249 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2250 					       enum ib_mr_type mr_type,
2251 					       u32 max_num_sg);
2252 	int                        (*map_mr_sg)(struct ib_mr *mr,
2253 						struct scatterlist *sg,
2254 						int sg_nents,
2255 						unsigned int *sg_offset);
2256 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2257 					       enum ib_mw_type type,
2258 					       struct ib_udata *udata);
2259 	int                        (*dealloc_mw)(struct ib_mw *mw);
2260 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2261 						int mr_access_flags,
2262 						struct ib_fmr_attr *fmr_attr);
2263 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2264 						   u64 *page_list, int list_len,
2265 						   u64 iova);
2266 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2267 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2268 	int                        (*attach_mcast)(struct ib_qp *qp,
2269 						   union ib_gid *gid,
2270 						   u16 lid);
2271 	int                        (*detach_mcast)(struct ib_qp *qp,
2272 						   union ib_gid *gid,
2273 						   u16 lid);
2274 	int                        (*process_mad)(struct ib_device *device,
2275 						  int process_mad_flags,
2276 						  u8 port_num,
2277 						  const struct ib_wc *in_wc,
2278 						  const struct ib_grh *in_grh,
2279 						  const struct ib_mad_hdr *in_mad,
2280 						  size_t in_mad_size,
2281 						  struct ib_mad_hdr *out_mad,
2282 						  size_t *out_mad_size,
2283 						  u16 *out_mad_pkey_index);
2284 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2285 						 struct ib_ucontext *ucontext,
2286 						 struct ib_udata *udata);
2287 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2288 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2289 						  struct ib_flow_attr
2290 						  *flow_attr,
2291 						  int domain);
2292 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2293 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2294 						      struct ib_mr_status *mr_status);
2295 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2296 	void			   (*drain_rq)(struct ib_qp *qp);
2297 	void			   (*drain_sq)(struct ib_qp *qp);
2298 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2299 							int state);
2300 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2301 						   struct ifla_vf_info *ivf);
2302 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2303 						   struct ifla_vf_stats *stats);
2304 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2305 						  int type);
2306 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2307 						struct ib_wq_init_attr *init_attr,
2308 						struct ib_udata *udata);
2309 	int			   (*destroy_wq)(struct ib_wq *wq);
2310 	int			   (*modify_wq)(struct ib_wq *wq,
2311 						struct ib_wq_attr *attr,
2312 						u32 wq_attr_mask,
2313 						struct ib_udata *udata);
2314 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2315 							   struct ib_rwq_ind_table_init_attr *init_attr,
2316 							   struct ib_udata *udata);
2317 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2318 	/**
2319 	 * rdma netdev operation
2320 	 *
2321 	 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2322 	 * doesn't support the specified rdma netdev type.
2323 	 */
2324 	struct net_device *(*alloc_rdma_netdev)(
2325 					struct ib_device *device,
2326 					u8 port_num,
2327 					enum rdma_netdev_t type,
2328 					const char *name,
2329 					unsigned char name_assign_type,
2330 					void (*setup)(struct net_device *));
2331 
2332 	struct module               *owner;
2333 	struct device                dev;
2334 	struct kobject               *ports_parent;
2335 	struct list_head             port_list;
2336 
2337 	enum {
2338 		IB_DEV_UNINITIALIZED,
2339 		IB_DEV_REGISTERED,
2340 		IB_DEV_UNREGISTERED
2341 	}                            reg_state;
2342 
2343 	int			     uverbs_abi_ver;
2344 	u64			     uverbs_cmd_mask;
2345 	u64			     uverbs_ex_cmd_mask;
2346 
2347 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2348 	__be64			     node_guid;
2349 	u32			     local_dma_lkey;
2350 	u16                          is_switch:1;
2351 	u8                           node_type;
2352 	u8                           phys_port_cnt;
2353 	struct ib_device_attr        attrs;
2354 	struct attribute_group	     *hw_stats_ag;
2355 	struct rdma_hw_stats         *hw_stats;
2356 
2357 #ifdef CONFIG_CGROUP_RDMA
2358 	struct rdmacg_device         cg_device;
2359 #endif
2360 
2361 	u32                          index;
2362 	/*
2363 	 * Implementation details of the RDMA core, don't use in drivers
2364 	 */
2365 	struct rdma_restrack_root     res;
2366 
2367 	/**
2368 	 * The following mandatory functions are used only at device
2369 	 * registration.  Keep functions such as these at the end of this
2370 	 * structure to avoid cache line misses when accessing struct ib_device
2371 	 * in fast paths.
2372 	 */
2373 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2374 	void (*get_dev_fw_str)(struct ib_device *, char *str);
2375 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2376 						     int comp_vector);
2377 
2378 	struct uverbs_root_spec		*specs_root;
2379 };
2380 
2381 struct ib_client {
2382 	char  *name;
2383 	void (*add)   (struct ib_device *);
2384 	void (*remove)(struct ib_device *, void *client_data);
2385 
2386 	/* Returns the net_dev belonging to this ib_client and matching the
2387 	 * given parameters.
2388 	 * @dev:	 An RDMA device that the net_dev use for communication.
2389 	 * @port:	 A physical port number on the RDMA device.
2390 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2391 	 * @gid:	 A GID that the net_dev uses to communicate.
2392 	 * @addr:	 An IP address the net_dev is configured with.
2393 	 * @client_data: The device's client data set by ib_set_client_data().
2394 	 *
2395 	 * An ib_client that implements a net_dev on top of RDMA devices
2396 	 * (such as IP over IB) should implement this callback, allowing the
2397 	 * rdma_cm module to find the right net_dev for a given request.
2398 	 *
2399 	 * The caller is responsible for calling dev_put on the returned
2400 	 * netdev. */
2401 	struct net_device *(*get_net_dev_by_params)(
2402 			struct ib_device *dev,
2403 			u8 port,
2404 			u16 pkey,
2405 			const union ib_gid *gid,
2406 			const struct sockaddr *addr,
2407 			void *client_data);
2408 	struct list_head list;
2409 };
2410 
2411 struct ib_device *ib_alloc_device(size_t size);
2412 void ib_dealloc_device(struct ib_device *device);
2413 
2414 void ib_get_device_fw_str(struct ib_device *device, char *str);
2415 
2416 int ib_register_device(struct ib_device *device,
2417 		       int (*port_callback)(struct ib_device *,
2418 					    u8, struct kobject *));
2419 void ib_unregister_device(struct ib_device *device);
2420 
2421 int ib_register_client   (struct ib_client *client);
2422 void ib_unregister_client(struct ib_client *client);
2423 
2424 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2425 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2426 			 void *data);
2427 
2428 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2429 {
2430 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2431 }
2432 
2433 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2434 {
2435 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2436 }
2437 
2438 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2439 				       size_t offset,
2440 				       size_t len)
2441 {
2442 	const void __user *p = udata->inbuf + offset;
2443 	bool ret;
2444 	u8 *buf;
2445 
2446 	if (len > USHRT_MAX)
2447 		return false;
2448 
2449 	buf = memdup_user(p, len);
2450 	if (IS_ERR(buf))
2451 		return false;
2452 
2453 	ret = !memchr_inv(buf, 0, len);
2454 	kfree(buf);
2455 	return ret;
2456 }
2457 
2458 /**
2459  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2460  * contains all required attributes and no attributes not allowed for
2461  * the given QP state transition.
2462  * @cur_state: Current QP state
2463  * @next_state: Next QP state
2464  * @type: QP type
2465  * @mask: Mask of supplied QP attributes
2466  * @ll : link layer of port
2467  *
2468  * This function is a helper function that a low-level driver's
2469  * modify_qp method can use to validate the consumer's input.  It
2470  * checks that cur_state and next_state are valid QP states, that a
2471  * transition from cur_state to next_state is allowed by the IB spec,
2472  * and that the attribute mask supplied is allowed for the transition.
2473  */
2474 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2475 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2476 		       enum rdma_link_layer ll);
2477 
2478 void ib_register_event_handler(struct ib_event_handler *event_handler);
2479 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2480 void ib_dispatch_event(struct ib_event *event);
2481 
2482 int ib_query_port(struct ib_device *device,
2483 		  u8 port_num, struct ib_port_attr *port_attr);
2484 
2485 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2486 					       u8 port_num);
2487 
2488 /**
2489  * rdma_cap_ib_switch - Check if the device is IB switch
2490  * @device: Device to check
2491  *
2492  * Device driver is responsible for setting is_switch bit on
2493  * in ib_device structure at init time.
2494  *
2495  * Return: true if the device is IB switch.
2496  */
2497 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2498 {
2499 	return device->is_switch;
2500 }
2501 
2502 /**
2503  * rdma_start_port - Return the first valid port number for the device
2504  * specified
2505  *
2506  * @device: Device to be checked
2507  *
2508  * Return start port number
2509  */
2510 static inline u8 rdma_start_port(const struct ib_device *device)
2511 {
2512 	return rdma_cap_ib_switch(device) ? 0 : 1;
2513 }
2514 
2515 /**
2516  * rdma_end_port - Return the last valid port number for the device
2517  * specified
2518  *
2519  * @device: Device to be checked
2520  *
2521  * Return last port number
2522  */
2523 static inline u8 rdma_end_port(const struct ib_device *device)
2524 {
2525 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2526 }
2527 
2528 static inline int rdma_is_port_valid(const struct ib_device *device,
2529 				     unsigned int port)
2530 {
2531 	return (port >= rdma_start_port(device) &&
2532 		port <= rdma_end_port(device));
2533 }
2534 
2535 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2536 {
2537 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2538 }
2539 
2540 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2541 {
2542 	return device->port_immutable[port_num].core_cap_flags &
2543 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2544 }
2545 
2546 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2547 {
2548 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2549 }
2550 
2551 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2552 {
2553 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2554 }
2555 
2556 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2557 {
2558 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2559 }
2560 
2561 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2562 {
2563 	return rdma_protocol_ib(device, port_num) ||
2564 		rdma_protocol_roce(device, port_num);
2565 }
2566 
2567 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2568 {
2569 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2570 }
2571 
2572 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2573 {
2574 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2575 }
2576 
2577 /**
2578  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2579  * Management Datagrams.
2580  * @device: Device to check
2581  * @port_num: Port number to check
2582  *
2583  * Management Datagrams (MAD) are a required part of the InfiniBand
2584  * specification and are supported on all InfiniBand devices.  A slightly
2585  * extended version are also supported on OPA interfaces.
2586  *
2587  * Return: true if the port supports sending/receiving of MAD packets.
2588  */
2589 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2590 {
2591 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2592 }
2593 
2594 /**
2595  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2596  * Management Datagrams.
2597  * @device: Device to check
2598  * @port_num: Port number to check
2599  *
2600  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2601  * datagrams with their own versions.  These OPA MADs share many but not all of
2602  * the characteristics of InfiniBand MADs.
2603  *
2604  * OPA MADs differ in the following ways:
2605  *
2606  *    1) MADs are variable size up to 2K
2607  *       IBTA defined MADs remain fixed at 256 bytes
2608  *    2) OPA SMPs must carry valid PKeys
2609  *    3) OPA SMP packets are a different format
2610  *
2611  * Return: true if the port supports OPA MAD packet formats.
2612  */
2613 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2614 {
2615 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2616 		== RDMA_CORE_CAP_OPA_MAD;
2617 }
2618 
2619 /**
2620  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2621  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2622  * @device: Device to check
2623  * @port_num: Port number to check
2624  *
2625  * Each InfiniBand node is required to provide a Subnet Management Agent
2626  * that the subnet manager can access.  Prior to the fabric being fully
2627  * configured by the subnet manager, the SMA is accessed via a well known
2628  * interface called the Subnet Management Interface (SMI).  This interface
2629  * uses directed route packets to communicate with the SM to get around the
2630  * chicken and egg problem of the SM needing to know what's on the fabric
2631  * in order to configure the fabric, and needing to configure the fabric in
2632  * order to send packets to the devices on the fabric.  These directed
2633  * route packets do not need the fabric fully configured in order to reach
2634  * their destination.  The SMI is the only method allowed to send
2635  * directed route packets on an InfiniBand fabric.
2636  *
2637  * Return: true if the port provides an SMI.
2638  */
2639 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2640 {
2641 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2642 }
2643 
2644 /**
2645  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2646  * Communication Manager.
2647  * @device: Device to check
2648  * @port_num: Port number to check
2649  *
2650  * The InfiniBand Communication Manager is one of many pre-defined General
2651  * Service Agents (GSA) that are accessed via the General Service
2652  * Interface (GSI).  It's role is to facilitate establishment of connections
2653  * between nodes as well as other management related tasks for established
2654  * connections.
2655  *
2656  * Return: true if the port supports an IB CM (this does not guarantee that
2657  * a CM is actually running however).
2658  */
2659 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2660 {
2661 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2662 }
2663 
2664 /**
2665  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2666  * Communication Manager.
2667  * @device: Device to check
2668  * @port_num: Port number to check
2669  *
2670  * Similar to above, but specific to iWARP connections which have a different
2671  * managment protocol than InfiniBand.
2672  *
2673  * Return: true if the port supports an iWARP CM (this does not guarantee that
2674  * a CM is actually running however).
2675  */
2676 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2677 {
2678 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2679 }
2680 
2681 /**
2682  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2683  * Subnet Administration.
2684  * @device: Device to check
2685  * @port_num: Port number to check
2686  *
2687  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2688  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2689  * fabrics, devices should resolve routes to other hosts by contacting the
2690  * SA to query the proper route.
2691  *
2692  * Return: true if the port should act as a client to the fabric Subnet
2693  * Administration interface.  This does not imply that the SA service is
2694  * running locally.
2695  */
2696 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2697 {
2698 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2699 }
2700 
2701 /**
2702  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2703  * Multicast.
2704  * @device: Device to check
2705  * @port_num: Port number to check
2706  *
2707  * InfiniBand multicast registration is more complex than normal IPv4 or
2708  * IPv6 multicast registration.  Each Host Channel Adapter must register
2709  * with the Subnet Manager when it wishes to join a multicast group.  It
2710  * should do so only once regardless of how many queue pairs it subscribes
2711  * to this group.  And it should leave the group only after all queue pairs
2712  * attached to the group have been detached.
2713  *
2714  * Return: true if the port must undertake the additional adminstrative
2715  * overhead of registering/unregistering with the SM and tracking of the
2716  * total number of queue pairs attached to the multicast group.
2717  */
2718 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2719 {
2720 	return rdma_cap_ib_sa(device, port_num);
2721 }
2722 
2723 /**
2724  * rdma_cap_af_ib - Check if the port of device has the capability
2725  * Native Infiniband Address.
2726  * @device: Device to check
2727  * @port_num: Port number to check
2728  *
2729  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2730  * GID.  RoCE uses a different mechanism, but still generates a GID via
2731  * a prescribed mechanism and port specific data.
2732  *
2733  * Return: true if the port uses a GID address to identify devices on the
2734  * network.
2735  */
2736 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2737 {
2738 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2739 }
2740 
2741 /**
2742  * rdma_cap_eth_ah - Check if the port of device has the capability
2743  * Ethernet Address Handle.
2744  * @device: Device to check
2745  * @port_num: Port number to check
2746  *
2747  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2748  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2749  * port.  Normally, packet headers are generated by the sending host
2750  * adapter, but when sending connectionless datagrams, we must manually
2751  * inject the proper headers for the fabric we are communicating over.
2752  *
2753  * Return: true if we are running as a RoCE port and must force the
2754  * addition of a Global Route Header built from our Ethernet Address
2755  * Handle into our header list for connectionless packets.
2756  */
2757 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2758 {
2759 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2760 }
2761 
2762 /**
2763  * rdma_cap_opa_ah - Check if the port of device supports
2764  * OPA Address handles
2765  * @device: Device to check
2766  * @port_num: Port number to check
2767  *
2768  * Return: true if we are running on an OPA device which supports
2769  * the extended OPA addressing.
2770  */
2771 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
2772 {
2773 	return (device->port_immutable[port_num].core_cap_flags &
2774 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
2775 }
2776 
2777 /**
2778  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2779  *
2780  * @device: Device
2781  * @port_num: Port number
2782  *
2783  * This MAD size includes the MAD headers and MAD payload.  No other headers
2784  * are included.
2785  *
2786  * Return the max MAD size required by the Port.  Will return 0 if the port
2787  * does not support MADs
2788  */
2789 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2790 {
2791 	return device->port_immutable[port_num].max_mad_size;
2792 }
2793 
2794 /**
2795  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2796  * @device: Device to check
2797  * @port_num: Port number to check
2798  *
2799  * RoCE GID table mechanism manages the various GIDs for a device.
2800  *
2801  * NOTE: if allocating the port's GID table has failed, this call will still
2802  * return true, but any RoCE GID table API will fail.
2803  *
2804  * Return: true if the port uses RoCE GID table mechanism in order to manage
2805  * its GIDs.
2806  */
2807 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2808 					   u8 port_num)
2809 {
2810 	return rdma_protocol_roce(device, port_num) &&
2811 		device->add_gid && device->del_gid;
2812 }
2813 
2814 /*
2815  * Check if the device supports READ W/ INVALIDATE.
2816  */
2817 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2818 {
2819 	/*
2820 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2821 	 * has support for it yet.
2822 	 */
2823 	return rdma_protocol_iwarp(dev, port_num);
2824 }
2825 
2826 int ib_query_gid(struct ib_device *device,
2827 		 u8 port_num, int index, union ib_gid *gid,
2828 		 struct ib_gid_attr *attr);
2829 
2830 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2831 			 int state);
2832 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2833 		     struct ifla_vf_info *info);
2834 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2835 		    struct ifla_vf_stats *stats);
2836 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2837 		   int type);
2838 
2839 int ib_query_pkey(struct ib_device *device,
2840 		  u8 port_num, u16 index, u16 *pkey);
2841 
2842 int ib_modify_device(struct ib_device *device,
2843 		     int device_modify_mask,
2844 		     struct ib_device_modify *device_modify);
2845 
2846 int ib_modify_port(struct ib_device *device,
2847 		   u8 port_num, int port_modify_mask,
2848 		   struct ib_port_modify *port_modify);
2849 
2850 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2851 		struct net_device *ndev, u8 *port_num, u16 *index);
2852 
2853 int ib_find_pkey(struct ib_device *device,
2854 		 u8 port_num, u16 pkey, u16 *index);
2855 
2856 enum ib_pd_flags {
2857 	/*
2858 	 * Create a memory registration for all memory in the system and place
2859 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2860 	 * ULPs to avoid the overhead of dynamic MRs.
2861 	 *
2862 	 * This flag is generally considered unsafe and must only be used in
2863 	 * extremly trusted environments.  Every use of it will log a warning
2864 	 * in the kernel log.
2865 	 */
2866 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2867 };
2868 
2869 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2870 		const char *caller);
2871 #define ib_alloc_pd(device, flags) \
2872 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
2873 void ib_dealloc_pd(struct ib_pd *pd);
2874 
2875 /**
2876  * rdma_create_ah - Creates an address handle for the given address vector.
2877  * @pd: The protection domain associated with the address handle.
2878  * @ah_attr: The attributes of the address vector.
2879  *
2880  * The address handle is used to reference a local or global destination
2881  * in all UD QP post sends.
2882  */
2883 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
2884 
2885 /**
2886  * rdma_create_user_ah - Creates an address handle for the given address vector.
2887  * It resolves destination mac address for ah attribute of RoCE type.
2888  * @pd: The protection domain associated with the address handle.
2889  * @ah_attr: The attributes of the address vector.
2890  * @udata: pointer to user's input output buffer information need by
2891  *         provider driver.
2892  *
2893  * It returns 0 on success and returns appropriate error code on error.
2894  * The address handle is used to reference a local or global destination
2895  * in all UD QP post sends.
2896  */
2897 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
2898 				  struct rdma_ah_attr *ah_attr,
2899 				  struct ib_udata *udata);
2900 /**
2901  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2902  *   work completion.
2903  * @hdr: the L3 header to parse
2904  * @net_type: type of header to parse
2905  * @sgid: place to store source gid
2906  * @dgid: place to store destination gid
2907  */
2908 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2909 			      enum rdma_network_type net_type,
2910 			      union ib_gid *sgid, union ib_gid *dgid);
2911 
2912 /**
2913  * ib_get_rdma_header_version - Get the header version
2914  * @hdr: the L3 header to parse
2915  */
2916 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2917 
2918 /**
2919  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
2920  *   work completion.
2921  * @device: Device on which the received message arrived.
2922  * @port_num: Port on which the received message arrived.
2923  * @wc: Work completion associated with the received message.
2924  * @grh: References the received global route header.  This parameter is
2925  *   ignored unless the work completion indicates that the GRH is valid.
2926  * @ah_attr: Returned attributes that can be used when creating an address
2927  *   handle for replying to the message.
2928  */
2929 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
2930 			    const struct ib_wc *wc, const struct ib_grh *grh,
2931 			    struct rdma_ah_attr *ah_attr);
2932 
2933 /**
2934  * ib_create_ah_from_wc - Creates an address handle associated with the
2935  *   sender of the specified work completion.
2936  * @pd: The protection domain associated with the address handle.
2937  * @wc: Work completion information associated with a received message.
2938  * @grh: References the received global route header.  This parameter is
2939  *   ignored unless the work completion indicates that the GRH is valid.
2940  * @port_num: The outbound port number to associate with the address.
2941  *
2942  * The address handle is used to reference a local or global destination
2943  * in all UD QP post sends.
2944  */
2945 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2946 				   const struct ib_grh *grh, u8 port_num);
2947 
2948 /**
2949  * rdma_modify_ah - Modifies the address vector associated with an address
2950  *   handle.
2951  * @ah: The address handle to modify.
2952  * @ah_attr: The new address vector attributes to associate with the
2953  *   address handle.
2954  */
2955 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2956 
2957 /**
2958  * rdma_query_ah - Queries the address vector associated with an address
2959  *   handle.
2960  * @ah: The address handle to query.
2961  * @ah_attr: The address vector attributes associated with the address
2962  *   handle.
2963  */
2964 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2965 
2966 /**
2967  * rdma_destroy_ah - Destroys an address handle.
2968  * @ah: The address handle to destroy.
2969  */
2970 int rdma_destroy_ah(struct ib_ah *ah);
2971 
2972 /**
2973  * ib_create_srq - Creates a SRQ associated with the specified protection
2974  *   domain.
2975  * @pd: The protection domain associated with the SRQ.
2976  * @srq_init_attr: A list of initial attributes required to create the
2977  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2978  *   the actual capabilities of the created SRQ.
2979  *
2980  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2981  * requested size of the SRQ, and set to the actual values allocated
2982  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2983  * will always be at least as large as the requested values.
2984  */
2985 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2986 			     struct ib_srq_init_attr *srq_init_attr);
2987 
2988 /**
2989  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2990  * @srq: The SRQ to modify.
2991  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2992  *   the current values of selected SRQ attributes are returned.
2993  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2994  *   are being modified.
2995  *
2996  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2997  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2998  * the number of receives queued drops below the limit.
2999  */
3000 int ib_modify_srq(struct ib_srq *srq,
3001 		  struct ib_srq_attr *srq_attr,
3002 		  enum ib_srq_attr_mask srq_attr_mask);
3003 
3004 /**
3005  * ib_query_srq - Returns the attribute list and current values for the
3006  *   specified SRQ.
3007  * @srq: The SRQ to query.
3008  * @srq_attr: The attributes of the specified SRQ.
3009  */
3010 int ib_query_srq(struct ib_srq *srq,
3011 		 struct ib_srq_attr *srq_attr);
3012 
3013 /**
3014  * ib_destroy_srq - Destroys the specified SRQ.
3015  * @srq: The SRQ to destroy.
3016  */
3017 int ib_destroy_srq(struct ib_srq *srq);
3018 
3019 /**
3020  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3021  * @srq: The SRQ to post the work request on.
3022  * @recv_wr: A list of work requests to post on the receive queue.
3023  * @bad_recv_wr: On an immediate failure, this parameter will reference
3024  *   the work request that failed to be posted on the QP.
3025  */
3026 static inline int ib_post_srq_recv(struct ib_srq *srq,
3027 				   struct ib_recv_wr *recv_wr,
3028 				   struct ib_recv_wr **bad_recv_wr)
3029 {
3030 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
3031 }
3032 
3033 /**
3034  * ib_create_qp - Creates a QP associated with the specified protection
3035  *   domain.
3036  * @pd: The protection domain associated with the QP.
3037  * @qp_init_attr: A list of initial attributes required to create the
3038  *   QP.  If QP creation succeeds, then the attributes are updated to
3039  *   the actual capabilities of the created QP.
3040  */
3041 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3042 			   struct ib_qp_init_attr *qp_init_attr);
3043 
3044 /**
3045  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3046  * @qp: The QP to modify.
3047  * @attr: On input, specifies the QP attributes to modify.  On output,
3048  *   the current values of selected QP attributes are returned.
3049  * @attr_mask: A bit-mask used to specify which attributes of the QP
3050  *   are being modified.
3051  * @udata: pointer to user's input output buffer information
3052  *   are being modified.
3053  * It returns 0 on success and returns appropriate error code on error.
3054  */
3055 int ib_modify_qp_with_udata(struct ib_qp *qp,
3056 			    struct ib_qp_attr *attr,
3057 			    int attr_mask,
3058 			    struct ib_udata *udata);
3059 
3060 /**
3061  * ib_modify_qp - Modifies the attributes for the specified QP and then
3062  *   transitions the QP to the given state.
3063  * @qp: The QP to modify.
3064  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3065  *   the current values of selected QP attributes are returned.
3066  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3067  *   are being modified.
3068  */
3069 int ib_modify_qp(struct ib_qp *qp,
3070 		 struct ib_qp_attr *qp_attr,
3071 		 int qp_attr_mask);
3072 
3073 /**
3074  * ib_query_qp - Returns the attribute list and current values for the
3075  *   specified QP.
3076  * @qp: The QP to query.
3077  * @qp_attr: The attributes of the specified QP.
3078  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3079  * @qp_init_attr: Additional attributes of the selected QP.
3080  *
3081  * The qp_attr_mask may be used to limit the query to gathering only the
3082  * selected attributes.
3083  */
3084 int ib_query_qp(struct ib_qp *qp,
3085 		struct ib_qp_attr *qp_attr,
3086 		int qp_attr_mask,
3087 		struct ib_qp_init_attr *qp_init_attr);
3088 
3089 /**
3090  * ib_destroy_qp - Destroys the specified QP.
3091  * @qp: The QP to destroy.
3092  */
3093 int ib_destroy_qp(struct ib_qp *qp);
3094 
3095 /**
3096  * ib_open_qp - Obtain a reference to an existing sharable QP.
3097  * @xrcd - XRC domain
3098  * @qp_open_attr: Attributes identifying the QP to open.
3099  *
3100  * Returns a reference to a sharable QP.
3101  */
3102 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3103 			 struct ib_qp_open_attr *qp_open_attr);
3104 
3105 /**
3106  * ib_close_qp - Release an external reference to a QP.
3107  * @qp: The QP handle to release
3108  *
3109  * The opened QP handle is released by the caller.  The underlying
3110  * shared QP is not destroyed until all internal references are released.
3111  */
3112 int ib_close_qp(struct ib_qp *qp);
3113 
3114 /**
3115  * ib_post_send - Posts a list of work requests to the send queue of
3116  *   the specified QP.
3117  * @qp: The QP to post the work request on.
3118  * @send_wr: A list of work requests to post on the send queue.
3119  * @bad_send_wr: On an immediate failure, this parameter will reference
3120  *   the work request that failed to be posted on the QP.
3121  *
3122  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3123  * error is returned, the QP state shall not be affected,
3124  * ib_post_send() will return an immediate error after queueing any
3125  * earlier work requests in the list.
3126  */
3127 static inline int ib_post_send(struct ib_qp *qp,
3128 			       struct ib_send_wr *send_wr,
3129 			       struct ib_send_wr **bad_send_wr)
3130 {
3131 	return qp->device->post_send(qp, send_wr, bad_send_wr);
3132 }
3133 
3134 /**
3135  * ib_post_recv - Posts a list of work requests to the receive queue of
3136  *   the specified QP.
3137  * @qp: The QP to post the work request on.
3138  * @recv_wr: A list of work requests to post on the receive queue.
3139  * @bad_recv_wr: On an immediate failure, this parameter will reference
3140  *   the work request that failed to be posted on the QP.
3141  */
3142 static inline int ib_post_recv(struct ib_qp *qp,
3143 			       struct ib_recv_wr *recv_wr,
3144 			       struct ib_recv_wr **bad_recv_wr)
3145 {
3146 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
3147 }
3148 
3149 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3150 			    int nr_cqe, int comp_vector,
3151 			    enum ib_poll_context poll_ctx, const char *caller);
3152 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3153 	__ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3154 
3155 void ib_free_cq(struct ib_cq *cq);
3156 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3157 
3158 /**
3159  * ib_create_cq - Creates a CQ on the specified device.
3160  * @device: The device on which to create the CQ.
3161  * @comp_handler: A user-specified callback that is invoked when a
3162  *   completion event occurs on the CQ.
3163  * @event_handler: A user-specified callback that is invoked when an
3164  *   asynchronous event not associated with a completion occurs on the CQ.
3165  * @cq_context: Context associated with the CQ returned to the user via
3166  *   the associated completion and event handlers.
3167  * @cq_attr: The attributes the CQ should be created upon.
3168  *
3169  * Users can examine the cq structure to determine the actual CQ size.
3170  */
3171 struct ib_cq *ib_create_cq(struct ib_device *device,
3172 			   ib_comp_handler comp_handler,
3173 			   void (*event_handler)(struct ib_event *, void *),
3174 			   void *cq_context,
3175 			   const struct ib_cq_init_attr *cq_attr);
3176 
3177 /**
3178  * ib_resize_cq - Modifies the capacity of the CQ.
3179  * @cq: The CQ to resize.
3180  * @cqe: The minimum size of the CQ.
3181  *
3182  * Users can examine the cq structure to determine the actual CQ size.
3183  */
3184 int ib_resize_cq(struct ib_cq *cq, int cqe);
3185 
3186 /**
3187  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3188  * @cq: The CQ to modify.
3189  * @cq_count: number of CQEs that will trigger an event
3190  * @cq_period: max period of time in usec before triggering an event
3191  *
3192  */
3193 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3194 
3195 /**
3196  * ib_destroy_cq - Destroys the specified CQ.
3197  * @cq: The CQ to destroy.
3198  */
3199 int ib_destroy_cq(struct ib_cq *cq);
3200 
3201 /**
3202  * ib_poll_cq - poll a CQ for completion(s)
3203  * @cq:the CQ being polled
3204  * @num_entries:maximum number of completions to return
3205  * @wc:array of at least @num_entries &struct ib_wc where completions
3206  *   will be returned
3207  *
3208  * Poll a CQ for (possibly multiple) completions.  If the return value
3209  * is < 0, an error occurred.  If the return value is >= 0, it is the
3210  * number of completions returned.  If the return value is
3211  * non-negative and < num_entries, then the CQ was emptied.
3212  */
3213 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3214 			     struct ib_wc *wc)
3215 {
3216 	return cq->device->poll_cq(cq, num_entries, wc);
3217 }
3218 
3219 /**
3220  * ib_peek_cq - Returns the number of unreaped completions currently
3221  *   on the specified CQ.
3222  * @cq: The CQ to peek.
3223  * @wc_cnt: A minimum number of unreaped completions to check for.
3224  *
3225  * If the number of unreaped completions is greater than or equal to wc_cnt,
3226  * this function returns wc_cnt, otherwise, it returns the actual number of
3227  * unreaped completions.
3228  */
3229 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
3230 
3231 /**
3232  * ib_req_notify_cq - Request completion notification on a CQ.
3233  * @cq: The CQ to generate an event for.
3234  * @flags:
3235  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3236  *   to request an event on the next solicited event or next work
3237  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3238  *   may also be |ed in to request a hint about missed events, as
3239  *   described below.
3240  *
3241  * Return Value:
3242  *    < 0 means an error occurred while requesting notification
3243  *   == 0 means notification was requested successfully, and if
3244  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3245  *        were missed and it is safe to wait for another event.  In
3246  *        this case is it guaranteed that any work completions added
3247  *        to the CQ since the last CQ poll will trigger a completion
3248  *        notification event.
3249  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3250  *        in.  It means that the consumer must poll the CQ again to
3251  *        make sure it is empty to avoid missing an event because of a
3252  *        race between requesting notification and an entry being
3253  *        added to the CQ.  This return value means it is possible
3254  *        (but not guaranteed) that a work completion has been added
3255  *        to the CQ since the last poll without triggering a
3256  *        completion notification event.
3257  */
3258 static inline int ib_req_notify_cq(struct ib_cq *cq,
3259 				   enum ib_cq_notify_flags flags)
3260 {
3261 	return cq->device->req_notify_cq(cq, flags);
3262 }
3263 
3264 /**
3265  * ib_req_ncomp_notif - Request completion notification when there are
3266  *   at least the specified number of unreaped completions on the CQ.
3267  * @cq: The CQ to generate an event for.
3268  * @wc_cnt: The number of unreaped completions that should be on the
3269  *   CQ before an event is generated.
3270  */
3271 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3272 {
3273 	return cq->device->req_ncomp_notif ?
3274 		cq->device->req_ncomp_notif(cq, wc_cnt) :
3275 		-ENOSYS;
3276 }
3277 
3278 /**
3279  * ib_dma_mapping_error - check a DMA addr for error
3280  * @dev: The device for which the dma_addr was created
3281  * @dma_addr: The DMA address to check
3282  */
3283 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3284 {
3285 	return dma_mapping_error(dev->dma_device, dma_addr);
3286 }
3287 
3288 /**
3289  * ib_dma_map_single - Map a kernel virtual address to DMA address
3290  * @dev: The device for which the dma_addr is to be created
3291  * @cpu_addr: The kernel virtual address
3292  * @size: The size of the region in bytes
3293  * @direction: The direction of the DMA
3294  */
3295 static inline u64 ib_dma_map_single(struct ib_device *dev,
3296 				    void *cpu_addr, size_t size,
3297 				    enum dma_data_direction direction)
3298 {
3299 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3300 }
3301 
3302 /**
3303  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3304  * @dev: The device for which the DMA address was created
3305  * @addr: The DMA address
3306  * @size: The size of the region in bytes
3307  * @direction: The direction of the DMA
3308  */
3309 static inline void ib_dma_unmap_single(struct ib_device *dev,
3310 				       u64 addr, size_t size,
3311 				       enum dma_data_direction direction)
3312 {
3313 	dma_unmap_single(dev->dma_device, addr, size, direction);
3314 }
3315 
3316 /**
3317  * ib_dma_map_page - Map a physical page to DMA address
3318  * @dev: The device for which the dma_addr is to be created
3319  * @page: The page to be mapped
3320  * @offset: The offset within the page
3321  * @size: The size of the region in bytes
3322  * @direction: The direction of the DMA
3323  */
3324 static inline u64 ib_dma_map_page(struct ib_device *dev,
3325 				  struct page *page,
3326 				  unsigned long offset,
3327 				  size_t size,
3328 					 enum dma_data_direction direction)
3329 {
3330 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3331 }
3332 
3333 /**
3334  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3335  * @dev: The device for which the DMA address was created
3336  * @addr: The DMA address
3337  * @size: The size of the region in bytes
3338  * @direction: The direction of the DMA
3339  */
3340 static inline void ib_dma_unmap_page(struct ib_device *dev,
3341 				     u64 addr, size_t size,
3342 				     enum dma_data_direction direction)
3343 {
3344 	dma_unmap_page(dev->dma_device, addr, size, direction);
3345 }
3346 
3347 /**
3348  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3349  * @dev: The device for which the DMA addresses are to be created
3350  * @sg: The array of scatter/gather entries
3351  * @nents: The number of scatter/gather entries
3352  * @direction: The direction of the DMA
3353  */
3354 static inline int ib_dma_map_sg(struct ib_device *dev,
3355 				struct scatterlist *sg, int nents,
3356 				enum dma_data_direction direction)
3357 {
3358 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3359 }
3360 
3361 /**
3362  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3363  * @dev: The device for which the DMA addresses were created
3364  * @sg: The array of scatter/gather entries
3365  * @nents: The number of scatter/gather entries
3366  * @direction: The direction of the DMA
3367  */
3368 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3369 				   struct scatterlist *sg, int nents,
3370 				   enum dma_data_direction direction)
3371 {
3372 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
3373 }
3374 
3375 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3376 				      struct scatterlist *sg, int nents,
3377 				      enum dma_data_direction direction,
3378 				      unsigned long dma_attrs)
3379 {
3380 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3381 				dma_attrs);
3382 }
3383 
3384 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3385 					 struct scatterlist *sg, int nents,
3386 					 enum dma_data_direction direction,
3387 					 unsigned long dma_attrs)
3388 {
3389 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3390 }
3391 /**
3392  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3393  * @dev: The device for which the DMA addresses were created
3394  * @sg: The scatter/gather entry
3395  *
3396  * Note: this function is obsolete. To do: change all occurrences of
3397  * ib_sg_dma_address() into sg_dma_address().
3398  */
3399 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3400 				    struct scatterlist *sg)
3401 {
3402 	return sg_dma_address(sg);
3403 }
3404 
3405 /**
3406  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3407  * @dev: The device for which the DMA addresses were created
3408  * @sg: The scatter/gather entry
3409  *
3410  * Note: this function is obsolete. To do: change all occurrences of
3411  * ib_sg_dma_len() into sg_dma_len().
3412  */
3413 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3414 					 struct scatterlist *sg)
3415 {
3416 	return sg_dma_len(sg);
3417 }
3418 
3419 /**
3420  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3421  * @dev: The device for which the DMA address was created
3422  * @addr: The DMA address
3423  * @size: The size of the region in bytes
3424  * @dir: The direction of the DMA
3425  */
3426 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3427 					      u64 addr,
3428 					      size_t size,
3429 					      enum dma_data_direction dir)
3430 {
3431 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3432 }
3433 
3434 /**
3435  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3436  * @dev: The device for which the DMA address was created
3437  * @addr: The DMA address
3438  * @size: The size of the region in bytes
3439  * @dir: The direction of the DMA
3440  */
3441 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3442 						 u64 addr,
3443 						 size_t size,
3444 						 enum dma_data_direction dir)
3445 {
3446 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3447 }
3448 
3449 /**
3450  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3451  * @dev: The device for which the DMA address is requested
3452  * @size: The size of the region to allocate in bytes
3453  * @dma_handle: A pointer for returning the DMA address of the region
3454  * @flag: memory allocator flags
3455  */
3456 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3457 					   size_t size,
3458 					   dma_addr_t *dma_handle,
3459 					   gfp_t flag)
3460 {
3461 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3462 }
3463 
3464 /**
3465  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3466  * @dev: The device for which the DMA addresses were allocated
3467  * @size: The size of the region
3468  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3469  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3470  */
3471 static inline void ib_dma_free_coherent(struct ib_device *dev,
3472 					size_t size, void *cpu_addr,
3473 					dma_addr_t dma_handle)
3474 {
3475 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3476 }
3477 
3478 /**
3479  * ib_dereg_mr - Deregisters a memory region and removes it from the
3480  *   HCA translation table.
3481  * @mr: The memory region to deregister.
3482  *
3483  * This function can fail, if the memory region has memory windows bound to it.
3484  */
3485 int ib_dereg_mr(struct ib_mr *mr);
3486 
3487 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3488 			  enum ib_mr_type mr_type,
3489 			  u32 max_num_sg);
3490 
3491 /**
3492  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3493  *   R_Key and L_Key.
3494  * @mr - struct ib_mr pointer to be updated.
3495  * @newkey - new key to be used.
3496  */
3497 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3498 {
3499 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3500 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3501 }
3502 
3503 /**
3504  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3505  * for calculating a new rkey for type 2 memory windows.
3506  * @rkey - the rkey to increment.
3507  */
3508 static inline u32 ib_inc_rkey(u32 rkey)
3509 {
3510 	const u32 mask = 0x000000ff;
3511 	return ((rkey + 1) & mask) | (rkey & ~mask);
3512 }
3513 
3514 /**
3515  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3516  * @pd: The protection domain associated with the unmapped region.
3517  * @mr_access_flags: Specifies the memory access rights.
3518  * @fmr_attr: Attributes of the unmapped region.
3519  *
3520  * A fast memory region must be mapped before it can be used as part of
3521  * a work request.
3522  */
3523 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3524 			    int mr_access_flags,
3525 			    struct ib_fmr_attr *fmr_attr);
3526 
3527 /**
3528  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3529  * @fmr: The fast memory region to associate with the pages.
3530  * @page_list: An array of physical pages to map to the fast memory region.
3531  * @list_len: The number of pages in page_list.
3532  * @iova: The I/O virtual address to use with the mapped region.
3533  */
3534 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3535 				  u64 *page_list, int list_len,
3536 				  u64 iova)
3537 {
3538 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3539 }
3540 
3541 /**
3542  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3543  * @fmr_list: A linked list of fast memory regions to unmap.
3544  */
3545 int ib_unmap_fmr(struct list_head *fmr_list);
3546 
3547 /**
3548  * ib_dealloc_fmr - Deallocates a fast memory region.
3549  * @fmr: The fast memory region to deallocate.
3550  */
3551 int ib_dealloc_fmr(struct ib_fmr *fmr);
3552 
3553 /**
3554  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3555  * @qp: QP to attach to the multicast group.  The QP must be type
3556  *   IB_QPT_UD.
3557  * @gid: Multicast group GID.
3558  * @lid: Multicast group LID in host byte order.
3559  *
3560  * In order to send and receive multicast packets, subnet
3561  * administration must have created the multicast group and configured
3562  * the fabric appropriately.  The port associated with the specified
3563  * QP must also be a member of the multicast group.
3564  */
3565 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3566 
3567 /**
3568  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3569  * @qp: QP to detach from the multicast group.
3570  * @gid: Multicast group GID.
3571  * @lid: Multicast group LID in host byte order.
3572  */
3573 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3574 
3575 /**
3576  * ib_alloc_xrcd - Allocates an XRC domain.
3577  * @device: The device on which to allocate the XRC domain.
3578  * @caller: Module name for kernel consumers
3579  */
3580 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3581 #define ib_alloc_xrcd(device) \
3582 	__ib_alloc_xrcd((device), KBUILD_MODNAME)
3583 
3584 /**
3585  * ib_dealloc_xrcd - Deallocates an XRC domain.
3586  * @xrcd: The XRC domain to deallocate.
3587  */
3588 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3589 
3590 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3591 			       struct ib_flow_attr *flow_attr, int domain);
3592 int ib_destroy_flow(struct ib_flow *flow_id);
3593 
3594 static inline int ib_check_mr_access(int flags)
3595 {
3596 	/*
3597 	 * Local write permission is required if remote write or
3598 	 * remote atomic permission is also requested.
3599 	 */
3600 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3601 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3602 		return -EINVAL;
3603 
3604 	return 0;
3605 }
3606 
3607 /**
3608  * ib_check_mr_status: lightweight check of MR status.
3609  *     This routine may provide status checks on a selected
3610  *     ib_mr. first use is for signature status check.
3611  *
3612  * @mr: A memory region.
3613  * @check_mask: Bitmask of which checks to perform from
3614  *     ib_mr_status_check enumeration.
3615  * @mr_status: The container of relevant status checks.
3616  *     failed checks will be indicated in the status bitmask
3617  *     and the relevant info shall be in the error item.
3618  */
3619 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3620 		       struct ib_mr_status *mr_status);
3621 
3622 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3623 					    u16 pkey, const union ib_gid *gid,
3624 					    const struct sockaddr *addr);
3625 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3626 			   struct ib_wq_init_attr *init_attr);
3627 int ib_destroy_wq(struct ib_wq *wq);
3628 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3629 		 u32 wq_attr_mask);
3630 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3631 						 struct ib_rwq_ind_table_init_attr*
3632 						 wq_ind_table_init_attr);
3633 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3634 
3635 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3636 		 unsigned int *sg_offset, unsigned int page_size);
3637 
3638 static inline int
3639 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3640 		  unsigned int *sg_offset, unsigned int page_size)
3641 {
3642 	int n;
3643 
3644 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3645 	mr->iova = 0;
3646 
3647 	return n;
3648 }
3649 
3650 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3651 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3652 
3653 void ib_drain_rq(struct ib_qp *qp);
3654 void ib_drain_sq(struct ib_qp *qp);
3655 void ib_drain_qp(struct ib_qp *qp);
3656 
3657 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3658 
3659 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3660 {
3661 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3662 		return attr->roce.dmac;
3663 	return NULL;
3664 }
3665 
3666 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3667 {
3668 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3669 		attr->ib.dlid = (u16)dlid;
3670 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3671 		attr->opa.dlid = dlid;
3672 }
3673 
3674 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3675 {
3676 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3677 		return attr->ib.dlid;
3678 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3679 		return attr->opa.dlid;
3680 	return 0;
3681 }
3682 
3683 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3684 {
3685 	attr->sl = sl;
3686 }
3687 
3688 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3689 {
3690 	return attr->sl;
3691 }
3692 
3693 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3694 					 u8 src_path_bits)
3695 {
3696 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3697 		attr->ib.src_path_bits = src_path_bits;
3698 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3699 		attr->opa.src_path_bits = src_path_bits;
3700 }
3701 
3702 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3703 {
3704 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3705 		return attr->ib.src_path_bits;
3706 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3707 		return attr->opa.src_path_bits;
3708 	return 0;
3709 }
3710 
3711 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
3712 					bool make_grd)
3713 {
3714 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3715 		attr->opa.make_grd = make_grd;
3716 }
3717 
3718 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
3719 {
3720 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3721 		return attr->opa.make_grd;
3722 	return false;
3723 }
3724 
3725 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3726 {
3727 	attr->port_num = port_num;
3728 }
3729 
3730 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3731 {
3732 	return attr->port_num;
3733 }
3734 
3735 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3736 					   u8 static_rate)
3737 {
3738 	attr->static_rate = static_rate;
3739 }
3740 
3741 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3742 {
3743 	return attr->static_rate;
3744 }
3745 
3746 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3747 					enum ib_ah_flags flag)
3748 {
3749 	attr->ah_flags = flag;
3750 }
3751 
3752 static inline enum ib_ah_flags
3753 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3754 {
3755 	return attr->ah_flags;
3756 }
3757 
3758 static inline const struct ib_global_route
3759 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
3760 {
3761 	return &attr->grh;
3762 }
3763 
3764 /*To retrieve and modify the grh */
3765 static inline struct ib_global_route
3766 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
3767 {
3768 	return &attr->grh;
3769 }
3770 
3771 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
3772 {
3773 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3774 
3775 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
3776 }
3777 
3778 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
3779 					     __be64 prefix)
3780 {
3781 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3782 
3783 	grh->dgid.global.subnet_prefix = prefix;
3784 }
3785 
3786 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
3787 					    __be64 if_id)
3788 {
3789 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3790 
3791 	grh->dgid.global.interface_id = if_id;
3792 }
3793 
3794 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
3795 				   union ib_gid *dgid, u32 flow_label,
3796 				   u8 sgid_index, u8 hop_limit,
3797 				   u8 traffic_class)
3798 {
3799 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3800 
3801 	attr->ah_flags = IB_AH_GRH;
3802 	if (dgid)
3803 		grh->dgid = *dgid;
3804 	grh->flow_label = flow_label;
3805 	grh->sgid_index = sgid_index;
3806 	grh->hop_limit = hop_limit;
3807 	grh->traffic_class = traffic_class;
3808 }
3809 
3810 /**
3811  * rdma_ah_find_type - Return address handle type.
3812  *
3813  * @dev: Device to be checked
3814  * @port_num: Port number
3815  */
3816 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
3817 						       u8 port_num)
3818 {
3819 	if (rdma_protocol_roce(dev, port_num))
3820 		return RDMA_AH_ATTR_TYPE_ROCE;
3821 	if (rdma_protocol_ib(dev, port_num)) {
3822 		if (rdma_cap_opa_ah(dev, port_num))
3823 			return RDMA_AH_ATTR_TYPE_OPA;
3824 		return RDMA_AH_ATTR_TYPE_IB;
3825 	}
3826 
3827 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
3828 }
3829 
3830 /**
3831  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
3832  *     In the current implementation the only way to get
3833  *     get the 32bit lid is from other sources for OPA.
3834  *     For IB, lids will always be 16bits so cast the
3835  *     value accordingly.
3836  *
3837  * @lid: A 32bit LID
3838  */
3839 static inline u16 ib_lid_cpu16(u32 lid)
3840 {
3841 	WARN_ON_ONCE(lid & 0xFFFF0000);
3842 	return (u16)lid;
3843 }
3844 
3845 /**
3846  * ib_lid_be16 - Return lid in 16bit BE encoding.
3847  *
3848  * @lid: A 32bit LID
3849  */
3850 static inline __be16 ib_lid_be16(u32 lid)
3851 {
3852 	WARN_ON_ONCE(lid & 0xFFFF0000);
3853 	return cpu_to_be16((u16)lid);
3854 }
3855 
3856 /**
3857  * ib_get_vector_affinity - Get the affinity mappings of a given completion
3858  *   vector
3859  * @device:         the rdma device
3860  * @comp_vector:    index of completion vector
3861  *
3862  * Returns NULL on failure, otherwise a corresponding cpu map of the
3863  * completion vector (returns all-cpus map if the device driver doesn't
3864  * implement get_vector_affinity).
3865  */
3866 static inline const struct cpumask *
3867 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
3868 {
3869 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
3870 	    !device->get_vector_affinity)
3871 		return NULL;
3872 
3873 	return device->get_vector_affinity(device, comp_vector);
3874 
3875 }
3876 
3877 /**
3878  * rdma_roce_rescan_device - Rescan all of the network devices in the system
3879  * and add their gids, as needed, to the relevant RoCE devices.
3880  *
3881  * @device:         the rdma device
3882  */
3883 void rdma_roce_rescan_device(struct ib_device *ibdev);
3884 
3885 #endif /* IB_VERBS_H */
3886