xref: /linux/include/rdma/ib_verbs.h (revision f4915455dcf07c4f237d6160a4b6adb0575d2909)
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/dma-mapping.h>
45 #include <linux/kref.h>
46 #include <linux/list.h>
47 #include <linux/rwsem.h>
48 #include <linux/workqueue.h>
49 #include <linux/irq_poll.h>
50 #include <uapi/linux/if_ether.h>
51 #include <net/ipv6.h>
52 #include <net/ip.h>
53 #include <linux/string.h>
54 #include <linux/slab.h>
55 #include <linux/netdevice.h>
56 #include <linux/refcount.h>
57 #include <linux/if_link.h>
58 #include <linux/atomic.h>
59 #include <linux/mmu_notifier.h>
60 #include <linux/uaccess.h>
61 #include <linux/cgroup_rdma.h>
62 #include <linux/irqflags.h>
63 #include <linux/preempt.h>
64 #include <uapi/rdma/ib_user_verbs.h>
65 #include <rdma/rdma_counter.h>
66 #include <rdma/restrack.h>
67 #include <rdma/signature.h>
68 #include <uapi/rdma/rdma_user_ioctl.h>
69 #include <uapi/rdma/ib_user_ioctl_verbs.h>
70 
71 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
72 
73 struct ib_umem_odp;
74 
75 extern struct workqueue_struct *ib_wq;
76 extern struct workqueue_struct *ib_comp_wq;
77 extern struct workqueue_struct *ib_comp_unbound_wq;
78 
79 __printf(3, 4) __cold
80 void ibdev_printk(const char *level, const struct ib_device *ibdev,
81 		  const char *format, ...);
82 __printf(2, 3) __cold
83 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
84 __printf(2, 3) __cold
85 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
86 __printf(2, 3) __cold
87 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
88 __printf(2, 3) __cold
89 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
90 __printf(2, 3) __cold
91 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
92 __printf(2, 3) __cold
93 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
94 __printf(2, 3) __cold
95 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
96 
97 #if defined(CONFIG_DYNAMIC_DEBUG)
98 #define ibdev_dbg(__dev, format, args...)                       \
99 	dynamic_ibdev_dbg(__dev, format, ##args)
100 #elif defined(DEBUG)
101 #define ibdev_dbg(__dev, format, args...)                       \
102 	ibdev_printk(KERN_DEBUG, __dev, format, ##args)
103 #else
104 __printf(2, 3) __cold
105 static inline
106 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
107 #endif
108 
109 union ib_gid {
110 	u8	raw[16];
111 	struct {
112 		__be64	subnet_prefix;
113 		__be64	interface_id;
114 	} global;
115 };
116 
117 extern union ib_gid zgid;
118 
119 enum ib_gid_type {
120 	/* If link layer is Ethernet, this is RoCE V1 */
121 	IB_GID_TYPE_IB        = 0,
122 	IB_GID_TYPE_ROCE      = 0,
123 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
124 	IB_GID_TYPE_SIZE
125 };
126 
127 #define ROCE_V2_UDP_DPORT      4791
128 struct ib_gid_attr {
129 	struct net_device __rcu	*ndev;
130 	struct ib_device	*device;
131 	union ib_gid		gid;
132 	enum ib_gid_type	gid_type;
133 	u16			index;
134 	u8			port_num;
135 };
136 
137 enum {
138 	/* set the local administered indication */
139 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
140 };
141 
142 enum rdma_transport_type {
143 	RDMA_TRANSPORT_IB,
144 	RDMA_TRANSPORT_IWARP,
145 	RDMA_TRANSPORT_USNIC,
146 	RDMA_TRANSPORT_USNIC_UDP,
147 	RDMA_TRANSPORT_UNSPECIFIED,
148 };
149 
150 enum rdma_protocol_type {
151 	RDMA_PROTOCOL_IB,
152 	RDMA_PROTOCOL_IBOE,
153 	RDMA_PROTOCOL_IWARP,
154 	RDMA_PROTOCOL_USNIC_UDP
155 };
156 
157 __attribute_const__ enum rdma_transport_type
158 rdma_node_get_transport(unsigned int node_type);
159 
160 enum rdma_network_type {
161 	RDMA_NETWORK_IB,
162 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
163 	RDMA_NETWORK_IPV4,
164 	RDMA_NETWORK_IPV6
165 };
166 
167 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
168 {
169 	if (network_type == RDMA_NETWORK_IPV4 ||
170 	    network_type == RDMA_NETWORK_IPV6)
171 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
172 
173 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
174 	return IB_GID_TYPE_IB;
175 }
176 
177 static inline enum rdma_network_type
178 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
179 {
180 	if (attr->gid_type == IB_GID_TYPE_IB)
181 		return RDMA_NETWORK_IB;
182 
183 	if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
184 		return RDMA_NETWORK_IPV4;
185 	else
186 		return RDMA_NETWORK_IPV6;
187 }
188 
189 enum rdma_link_layer {
190 	IB_LINK_LAYER_UNSPECIFIED,
191 	IB_LINK_LAYER_INFINIBAND,
192 	IB_LINK_LAYER_ETHERNET,
193 };
194 
195 enum ib_device_cap_flags {
196 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
197 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
198 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
199 	IB_DEVICE_RAW_MULTI			= (1 << 3),
200 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
201 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
202 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
203 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
204 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
205 	/* Not in use, former INIT_TYPE		= (1 << 9),*/
206 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
207 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
208 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
209 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
210 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
211 
212 	/*
213 	 * This device supports a per-device lkey or stag that can be
214 	 * used without performing a memory registration for the local
215 	 * memory.  Note that ULPs should never check this flag, but
216 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
217 	 * which will always contain a usable lkey.
218 	 */
219 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
220 	/* Reserved, old SEND_W_INV		= (1 << 16),*/
221 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
222 	/*
223 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
224 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
225 	 * messages and can verify the validity of checksum for
226 	 * incoming messages.  Setting this flag implies that the
227 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
228 	 */
229 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
230 	IB_DEVICE_UD_TSO			= (1 << 19),
231 	IB_DEVICE_XRC				= (1 << 20),
232 
233 	/*
234 	 * This device supports the IB "base memory management extension",
235 	 * which includes support for fast registrations (IB_WR_REG_MR,
236 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
237 	 * also be set by any iWarp device which must support FRs to comply
238 	 * to the iWarp verbs spec.  iWarp devices also support the
239 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
240 	 * stag.
241 	 */
242 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
243 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
244 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
245 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
246 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
247 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
248 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
249 	/*
250 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
251 	 * support execution of WQEs that involve synchronization
252 	 * of I/O operations with single completion queue managed
253 	 * by hardware.
254 	 */
255 	IB_DEVICE_CROSS_CHANNEL			= (1 << 27),
256 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
257 	IB_DEVICE_INTEGRITY_HANDOVER		= (1 << 30),
258 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
259 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
260 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
261 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
262 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
263 	IB_DEVICE_RDMA_NETDEV_OPA_VNIC		= (1ULL << 35),
264 	/* The device supports padding incoming writes to cacheline. */
265 	IB_DEVICE_PCI_WRITE_END_PADDING		= (1ULL << 36),
266 	IB_DEVICE_ALLOW_USER_UNREG		= (1ULL << 37),
267 };
268 
269 enum ib_atomic_cap {
270 	IB_ATOMIC_NONE,
271 	IB_ATOMIC_HCA,
272 	IB_ATOMIC_GLOB
273 };
274 
275 enum ib_odp_general_cap_bits {
276 	IB_ODP_SUPPORT		= 1 << 0,
277 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
278 };
279 
280 enum ib_odp_transport_cap_bits {
281 	IB_ODP_SUPPORT_SEND	= 1 << 0,
282 	IB_ODP_SUPPORT_RECV	= 1 << 1,
283 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
284 	IB_ODP_SUPPORT_READ	= 1 << 3,
285 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
286 	IB_ODP_SUPPORT_SRQ_RECV	= 1 << 5,
287 };
288 
289 struct ib_odp_caps {
290 	uint64_t general_caps;
291 	struct {
292 		uint32_t  rc_odp_caps;
293 		uint32_t  uc_odp_caps;
294 		uint32_t  ud_odp_caps;
295 		uint32_t  xrc_odp_caps;
296 	} per_transport_caps;
297 };
298 
299 struct ib_rss_caps {
300 	/* Corresponding bit will be set if qp type from
301 	 * 'enum ib_qp_type' is supported, e.g.
302 	 * supported_qpts |= 1 << IB_QPT_UD
303 	 */
304 	u32 supported_qpts;
305 	u32 max_rwq_indirection_tables;
306 	u32 max_rwq_indirection_table_size;
307 };
308 
309 enum ib_tm_cap_flags {
310 	/*  Support tag matching with rendezvous offload for RC transport */
311 	IB_TM_CAP_RNDV_RC = 1 << 0,
312 };
313 
314 struct ib_tm_caps {
315 	/* Max size of RNDV header */
316 	u32 max_rndv_hdr_size;
317 	/* Max number of entries in tag matching list */
318 	u32 max_num_tags;
319 	/* From enum ib_tm_cap_flags */
320 	u32 flags;
321 	/* Max number of outstanding list operations */
322 	u32 max_ops;
323 	/* Max number of SGE in tag matching entry */
324 	u32 max_sge;
325 };
326 
327 struct ib_cq_init_attr {
328 	unsigned int	cqe;
329 	int		comp_vector;
330 	u32		flags;
331 };
332 
333 enum ib_cq_attr_mask {
334 	IB_CQ_MODERATE = 1 << 0,
335 };
336 
337 struct ib_cq_caps {
338 	u16     max_cq_moderation_count;
339 	u16     max_cq_moderation_period;
340 };
341 
342 struct ib_dm_mr_attr {
343 	u64		length;
344 	u64		offset;
345 	u32		access_flags;
346 };
347 
348 struct ib_dm_alloc_attr {
349 	u64	length;
350 	u32	alignment;
351 	u32	flags;
352 };
353 
354 struct ib_device_attr {
355 	u64			fw_ver;
356 	__be64			sys_image_guid;
357 	u64			max_mr_size;
358 	u64			page_size_cap;
359 	u32			vendor_id;
360 	u32			vendor_part_id;
361 	u32			hw_ver;
362 	int			max_qp;
363 	int			max_qp_wr;
364 	u64			device_cap_flags;
365 	int			max_send_sge;
366 	int			max_recv_sge;
367 	int			max_sge_rd;
368 	int			max_cq;
369 	int			max_cqe;
370 	int			max_mr;
371 	int			max_pd;
372 	int			max_qp_rd_atom;
373 	int			max_ee_rd_atom;
374 	int			max_res_rd_atom;
375 	int			max_qp_init_rd_atom;
376 	int			max_ee_init_rd_atom;
377 	enum ib_atomic_cap	atomic_cap;
378 	enum ib_atomic_cap	masked_atomic_cap;
379 	int			max_ee;
380 	int			max_rdd;
381 	int			max_mw;
382 	int			max_raw_ipv6_qp;
383 	int			max_raw_ethy_qp;
384 	int			max_mcast_grp;
385 	int			max_mcast_qp_attach;
386 	int			max_total_mcast_qp_attach;
387 	int			max_ah;
388 	int			max_fmr;
389 	int			max_map_per_fmr;
390 	int			max_srq;
391 	int			max_srq_wr;
392 	int			max_srq_sge;
393 	unsigned int		max_fast_reg_page_list_len;
394 	unsigned int		max_pi_fast_reg_page_list_len;
395 	u16			max_pkeys;
396 	u8			local_ca_ack_delay;
397 	int			sig_prot_cap;
398 	int			sig_guard_cap;
399 	struct ib_odp_caps	odp_caps;
400 	uint64_t		timestamp_mask;
401 	uint64_t		hca_core_clock; /* in KHZ */
402 	struct ib_rss_caps	rss_caps;
403 	u32			max_wq_type_rq;
404 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
405 	struct ib_tm_caps	tm_caps;
406 	struct ib_cq_caps       cq_caps;
407 	u64			max_dm_size;
408 };
409 
410 enum ib_mtu {
411 	IB_MTU_256  = 1,
412 	IB_MTU_512  = 2,
413 	IB_MTU_1024 = 3,
414 	IB_MTU_2048 = 4,
415 	IB_MTU_4096 = 5
416 };
417 
418 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
419 {
420 	switch (mtu) {
421 	case IB_MTU_256:  return  256;
422 	case IB_MTU_512:  return  512;
423 	case IB_MTU_1024: return 1024;
424 	case IB_MTU_2048: return 2048;
425 	case IB_MTU_4096: return 4096;
426 	default: 	  return -1;
427 	}
428 }
429 
430 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
431 {
432 	if (mtu >= 4096)
433 		return IB_MTU_4096;
434 	else if (mtu >= 2048)
435 		return IB_MTU_2048;
436 	else if (mtu >= 1024)
437 		return IB_MTU_1024;
438 	else if (mtu >= 512)
439 		return IB_MTU_512;
440 	else
441 		return IB_MTU_256;
442 }
443 
444 enum ib_port_state {
445 	IB_PORT_NOP		= 0,
446 	IB_PORT_DOWN		= 1,
447 	IB_PORT_INIT		= 2,
448 	IB_PORT_ARMED		= 3,
449 	IB_PORT_ACTIVE		= 4,
450 	IB_PORT_ACTIVE_DEFER	= 5
451 };
452 
453 enum ib_port_width {
454 	IB_WIDTH_1X	= 1,
455 	IB_WIDTH_2X	= 16,
456 	IB_WIDTH_4X	= 2,
457 	IB_WIDTH_8X	= 4,
458 	IB_WIDTH_12X	= 8
459 };
460 
461 static inline int ib_width_enum_to_int(enum ib_port_width width)
462 {
463 	switch (width) {
464 	case IB_WIDTH_1X:  return  1;
465 	case IB_WIDTH_2X:  return  2;
466 	case IB_WIDTH_4X:  return  4;
467 	case IB_WIDTH_8X:  return  8;
468 	case IB_WIDTH_12X: return 12;
469 	default: 	  return -1;
470 	}
471 }
472 
473 enum ib_port_speed {
474 	IB_SPEED_SDR	= 1,
475 	IB_SPEED_DDR	= 2,
476 	IB_SPEED_QDR	= 4,
477 	IB_SPEED_FDR10	= 8,
478 	IB_SPEED_FDR	= 16,
479 	IB_SPEED_EDR	= 32,
480 	IB_SPEED_HDR	= 64
481 };
482 
483 /**
484  * struct rdma_hw_stats
485  * @lock - Mutex to protect parallel write access to lifespan and values
486  *    of counters, which are 64bits and not guaranteeed to be written
487  *    atomicaly on 32bits systems.
488  * @timestamp - Used by the core code to track when the last update was
489  * @lifespan - Used by the core code to determine how old the counters
490  *   should be before being updated again.  Stored in jiffies, defaults
491  *   to 10 milliseconds, drivers can override the default be specifying
492  *   their own value during their allocation routine.
493  * @name - Array of pointers to static names used for the counters in
494  *   directory.
495  * @num_counters - How many hardware counters there are.  If name is
496  *   shorter than this number, a kernel oops will result.  Driver authors
497  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
498  *   in their code to prevent this.
499  * @value - Array of u64 counters that are accessed by the sysfs code and
500  *   filled in by the drivers get_stats routine
501  */
502 struct rdma_hw_stats {
503 	struct mutex	lock; /* Protect lifespan and values[] */
504 	unsigned long	timestamp;
505 	unsigned long	lifespan;
506 	const char * const *names;
507 	int		num_counters;
508 	u64		value[];
509 };
510 
511 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
512 /**
513  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
514  *   for drivers.
515  * @names - Array of static const char *
516  * @num_counters - How many elements in array
517  * @lifespan - How many milliseconds between updates
518  */
519 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
520 		const char * const *names, int num_counters,
521 		unsigned long lifespan)
522 {
523 	struct rdma_hw_stats *stats;
524 
525 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
526 			GFP_KERNEL);
527 	if (!stats)
528 		return NULL;
529 	stats->names = names;
530 	stats->num_counters = num_counters;
531 	stats->lifespan = msecs_to_jiffies(lifespan);
532 
533 	return stats;
534 }
535 
536 
537 /* Define bits for the various functionality this port needs to be supported by
538  * the core.
539  */
540 /* Management                           0x00000FFF */
541 #define RDMA_CORE_CAP_IB_MAD            0x00000001
542 #define RDMA_CORE_CAP_IB_SMI            0x00000002
543 #define RDMA_CORE_CAP_IB_CM             0x00000004
544 #define RDMA_CORE_CAP_IW_CM             0x00000008
545 #define RDMA_CORE_CAP_IB_SA             0x00000010
546 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
547 
548 /* Address format                       0x000FF000 */
549 #define RDMA_CORE_CAP_AF_IB             0x00001000
550 #define RDMA_CORE_CAP_ETH_AH            0x00002000
551 #define RDMA_CORE_CAP_OPA_AH            0x00004000
552 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
553 
554 /* Protocol                             0xFFF00000 */
555 #define RDMA_CORE_CAP_PROT_IB           0x00100000
556 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
557 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
558 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
559 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
560 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
561 
562 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
563 					| RDMA_CORE_CAP_PROT_ROCE     \
564 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
565 
566 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
567 					| RDMA_CORE_CAP_IB_MAD \
568 					| RDMA_CORE_CAP_IB_SMI \
569 					| RDMA_CORE_CAP_IB_CM  \
570 					| RDMA_CORE_CAP_IB_SA  \
571 					| RDMA_CORE_CAP_AF_IB)
572 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
573 					| RDMA_CORE_CAP_IB_MAD  \
574 					| RDMA_CORE_CAP_IB_CM   \
575 					| RDMA_CORE_CAP_AF_IB   \
576 					| RDMA_CORE_CAP_ETH_AH)
577 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
578 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
579 					| RDMA_CORE_CAP_IB_MAD  \
580 					| RDMA_CORE_CAP_IB_CM   \
581 					| RDMA_CORE_CAP_AF_IB   \
582 					| RDMA_CORE_CAP_ETH_AH)
583 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
584 					| RDMA_CORE_CAP_IW_CM)
585 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
586 					| RDMA_CORE_CAP_OPA_MAD)
587 
588 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
589 
590 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
591 
592 struct ib_port_attr {
593 	u64			subnet_prefix;
594 	enum ib_port_state	state;
595 	enum ib_mtu		max_mtu;
596 	enum ib_mtu		active_mtu;
597 	int			gid_tbl_len;
598 	unsigned int		ip_gids:1;
599 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
600 	u32			port_cap_flags;
601 	u32			max_msg_sz;
602 	u32			bad_pkey_cntr;
603 	u32			qkey_viol_cntr;
604 	u16			pkey_tbl_len;
605 	u32			sm_lid;
606 	u32			lid;
607 	u8			lmc;
608 	u8			max_vl_num;
609 	u8			sm_sl;
610 	u8			subnet_timeout;
611 	u8			init_type_reply;
612 	u8			active_width;
613 	u8			active_speed;
614 	u8                      phys_state;
615 	u16			port_cap_flags2;
616 };
617 
618 enum ib_device_modify_flags {
619 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
620 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
621 };
622 
623 #define IB_DEVICE_NODE_DESC_MAX 64
624 
625 struct ib_device_modify {
626 	u64	sys_image_guid;
627 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
628 };
629 
630 enum ib_port_modify_flags {
631 	IB_PORT_SHUTDOWN		= 1,
632 	IB_PORT_INIT_TYPE		= (1<<2),
633 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
634 	IB_PORT_OPA_MASK_CHG		= (1<<4)
635 };
636 
637 struct ib_port_modify {
638 	u32	set_port_cap_mask;
639 	u32	clr_port_cap_mask;
640 	u8	init_type;
641 };
642 
643 enum ib_event_type {
644 	IB_EVENT_CQ_ERR,
645 	IB_EVENT_QP_FATAL,
646 	IB_EVENT_QP_REQ_ERR,
647 	IB_EVENT_QP_ACCESS_ERR,
648 	IB_EVENT_COMM_EST,
649 	IB_EVENT_SQ_DRAINED,
650 	IB_EVENT_PATH_MIG,
651 	IB_EVENT_PATH_MIG_ERR,
652 	IB_EVENT_DEVICE_FATAL,
653 	IB_EVENT_PORT_ACTIVE,
654 	IB_EVENT_PORT_ERR,
655 	IB_EVENT_LID_CHANGE,
656 	IB_EVENT_PKEY_CHANGE,
657 	IB_EVENT_SM_CHANGE,
658 	IB_EVENT_SRQ_ERR,
659 	IB_EVENT_SRQ_LIMIT_REACHED,
660 	IB_EVENT_QP_LAST_WQE_REACHED,
661 	IB_EVENT_CLIENT_REREGISTER,
662 	IB_EVENT_GID_CHANGE,
663 	IB_EVENT_WQ_FATAL,
664 };
665 
666 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
667 
668 struct ib_event {
669 	struct ib_device	*device;
670 	union {
671 		struct ib_cq	*cq;
672 		struct ib_qp	*qp;
673 		struct ib_srq	*srq;
674 		struct ib_wq	*wq;
675 		u8		port_num;
676 	} element;
677 	enum ib_event_type	event;
678 };
679 
680 struct ib_event_handler {
681 	struct ib_device *device;
682 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
683 	struct list_head  list;
684 };
685 
686 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
687 	do {							\
688 		(_ptr)->device  = _device;			\
689 		(_ptr)->handler = _handler;			\
690 		INIT_LIST_HEAD(&(_ptr)->list);			\
691 	} while (0)
692 
693 struct ib_global_route {
694 	const struct ib_gid_attr *sgid_attr;
695 	union ib_gid	dgid;
696 	u32		flow_label;
697 	u8		sgid_index;
698 	u8		hop_limit;
699 	u8		traffic_class;
700 };
701 
702 struct ib_grh {
703 	__be32		version_tclass_flow;
704 	__be16		paylen;
705 	u8		next_hdr;
706 	u8		hop_limit;
707 	union ib_gid	sgid;
708 	union ib_gid	dgid;
709 };
710 
711 union rdma_network_hdr {
712 	struct ib_grh ibgrh;
713 	struct {
714 		/* The IB spec states that if it's IPv4, the header
715 		 * is located in the last 20 bytes of the header.
716 		 */
717 		u8		reserved[20];
718 		struct iphdr	roce4grh;
719 	};
720 };
721 
722 #define IB_QPN_MASK		0xFFFFFF
723 
724 enum {
725 	IB_MULTICAST_QPN = 0xffffff
726 };
727 
728 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
729 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
730 
731 enum ib_ah_flags {
732 	IB_AH_GRH	= 1
733 };
734 
735 enum ib_rate {
736 	IB_RATE_PORT_CURRENT = 0,
737 	IB_RATE_2_5_GBPS = 2,
738 	IB_RATE_5_GBPS   = 5,
739 	IB_RATE_10_GBPS  = 3,
740 	IB_RATE_20_GBPS  = 6,
741 	IB_RATE_30_GBPS  = 4,
742 	IB_RATE_40_GBPS  = 7,
743 	IB_RATE_60_GBPS  = 8,
744 	IB_RATE_80_GBPS  = 9,
745 	IB_RATE_120_GBPS = 10,
746 	IB_RATE_14_GBPS  = 11,
747 	IB_RATE_56_GBPS  = 12,
748 	IB_RATE_112_GBPS = 13,
749 	IB_RATE_168_GBPS = 14,
750 	IB_RATE_25_GBPS  = 15,
751 	IB_RATE_100_GBPS = 16,
752 	IB_RATE_200_GBPS = 17,
753 	IB_RATE_300_GBPS = 18,
754 	IB_RATE_28_GBPS  = 19,
755 	IB_RATE_50_GBPS  = 20,
756 	IB_RATE_400_GBPS = 21,
757 	IB_RATE_600_GBPS = 22,
758 };
759 
760 /**
761  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
762  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
763  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
764  * @rate: rate to convert.
765  */
766 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
767 
768 /**
769  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
770  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
771  * @rate: rate to convert.
772  */
773 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
774 
775 
776 /**
777  * enum ib_mr_type - memory region type
778  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
779  *                            normal registration
780  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
781  *                            register any arbitrary sg lists (without
782  *                            the normal mr constraints - see
783  *                            ib_map_mr_sg)
784  * @IB_MR_TYPE_DM:            memory region that is used for device
785  *                            memory registration
786  * @IB_MR_TYPE_USER:          memory region that is used for the user-space
787  *                            application
788  * @IB_MR_TYPE_DMA:           memory region that is used for DMA operations
789  *                            without address translations (VA=PA)
790  * @IB_MR_TYPE_INTEGRITY:     memory region that is used for
791  *                            data integrity operations
792  */
793 enum ib_mr_type {
794 	IB_MR_TYPE_MEM_REG,
795 	IB_MR_TYPE_SG_GAPS,
796 	IB_MR_TYPE_DM,
797 	IB_MR_TYPE_USER,
798 	IB_MR_TYPE_DMA,
799 	IB_MR_TYPE_INTEGRITY,
800 };
801 
802 enum ib_mr_status_check {
803 	IB_MR_CHECK_SIG_STATUS = 1,
804 };
805 
806 /**
807  * struct ib_mr_status - Memory region status container
808  *
809  * @fail_status: Bitmask of MR checks status. For each
810  *     failed check a corresponding status bit is set.
811  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
812  *     failure.
813  */
814 struct ib_mr_status {
815 	u32		    fail_status;
816 	struct ib_sig_err   sig_err;
817 };
818 
819 /**
820  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
821  * enum.
822  * @mult: multiple to convert.
823  */
824 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
825 
826 enum rdma_ah_attr_type {
827 	RDMA_AH_ATTR_TYPE_UNDEFINED,
828 	RDMA_AH_ATTR_TYPE_IB,
829 	RDMA_AH_ATTR_TYPE_ROCE,
830 	RDMA_AH_ATTR_TYPE_OPA,
831 };
832 
833 struct ib_ah_attr {
834 	u16			dlid;
835 	u8			src_path_bits;
836 };
837 
838 struct roce_ah_attr {
839 	u8			dmac[ETH_ALEN];
840 };
841 
842 struct opa_ah_attr {
843 	u32			dlid;
844 	u8			src_path_bits;
845 	bool			make_grd;
846 };
847 
848 struct rdma_ah_attr {
849 	struct ib_global_route	grh;
850 	u8			sl;
851 	u8			static_rate;
852 	u8			port_num;
853 	u8			ah_flags;
854 	enum rdma_ah_attr_type type;
855 	union {
856 		struct ib_ah_attr ib;
857 		struct roce_ah_attr roce;
858 		struct opa_ah_attr opa;
859 	};
860 };
861 
862 enum ib_wc_status {
863 	IB_WC_SUCCESS,
864 	IB_WC_LOC_LEN_ERR,
865 	IB_WC_LOC_QP_OP_ERR,
866 	IB_WC_LOC_EEC_OP_ERR,
867 	IB_WC_LOC_PROT_ERR,
868 	IB_WC_WR_FLUSH_ERR,
869 	IB_WC_MW_BIND_ERR,
870 	IB_WC_BAD_RESP_ERR,
871 	IB_WC_LOC_ACCESS_ERR,
872 	IB_WC_REM_INV_REQ_ERR,
873 	IB_WC_REM_ACCESS_ERR,
874 	IB_WC_REM_OP_ERR,
875 	IB_WC_RETRY_EXC_ERR,
876 	IB_WC_RNR_RETRY_EXC_ERR,
877 	IB_WC_LOC_RDD_VIOL_ERR,
878 	IB_WC_REM_INV_RD_REQ_ERR,
879 	IB_WC_REM_ABORT_ERR,
880 	IB_WC_INV_EECN_ERR,
881 	IB_WC_INV_EEC_STATE_ERR,
882 	IB_WC_FATAL_ERR,
883 	IB_WC_RESP_TIMEOUT_ERR,
884 	IB_WC_GENERAL_ERR
885 };
886 
887 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
888 
889 enum ib_wc_opcode {
890 	IB_WC_SEND,
891 	IB_WC_RDMA_WRITE,
892 	IB_WC_RDMA_READ,
893 	IB_WC_COMP_SWAP,
894 	IB_WC_FETCH_ADD,
895 	IB_WC_LSO,
896 	IB_WC_LOCAL_INV,
897 	IB_WC_REG_MR,
898 	IB_WC_MASKED_COMP_SWAP,
899 	IB_WC_MASKED_FETCH_ADD,
900 /*
901  * Set value of IB_WC_RECV so consumers can test if a completion is a
902  * receive by testing (opcode & IB_WC_RECV).
903  */
904 	IB_WC_RECV			= 1 << 7,
905 	IB_WC_RECV_RDMA_WITH_IMM
906 };
907 
908 enum ib_wc_flags {
909 	IB_WC_GRH		= 1,
910 	IB_WC_WITH_IMM		= (1<<1),
911 	IB_WC_WITH_INVALIDATE	= (1<<2),
912 	IB_WC_IP_CSUM_OK	= (1<<3),
913 	IB_WC_WITH_SMAC		= (1<<4),
914 	IB_WC_WITH_VLAN		= (1<<5),
915 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
916 };
917 
918 struct ib_wc {
919 	union {
920 		u64		wr_id;
921 		struct ib_cqe	*wr_cqe;
922 	};
923 	enum ib_wc_status	status;
924 	enum ib_wc_opcode	opcode;
925 	u32			vendor_err;
926 	u32			byte_len;
927 	struct ib_qp	       *qp;
928 	union {
929 		__be32		imm_data;
930 		u32		invalidate_rkey;
931 	} ex;
932 	u32			src_qp;
933 	u32			slid;
934 	int			wc_flags;
935 	u16			pkey_index;
936 	u8			sl;
937 	u8			dlid_path_bits;
938 	u8			port_num;	/* valid only for DR SMPs on switches */
939 	u8			smac[ETH_ALEN];
940 	u16			vlan_id;
941 	u8			network_hdr_type;
942 };
943 
944 enum ib_cq_notify_flags {
945 	IB_CQ_SOLICITED			= 1 << 0,
946 	IB_CQ_NEXT_COMP			= 1 << 1,
947 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
948 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
949 };
950 
951 enum ib_srq_type {
952 	IB_SRQT_BASIC,
953 	IB_SRQT_XRC,
954 	IB_SRQT_TM,
955 };
956 
957 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
958 {
959 	return srq_type == IB_SRQT_XRC ||
960 	       srq_type == IB_SRQT_TM;
961 }
962 
963 enum ib_srq_attr_mask {
964 	IB_SRQ_MAX_WR	= 1 << 0,
965 	IB_SRQ_LIMIT	= 1 << 1,
966 };
967 
968 struct ib_srq_attr {
969 	u32	max_wr;
970 	u32	max_sge;
971 	u32	srq_limit;
972 };
973 
974 struct ib_srq_init_attr {
975 	void		      (*event_handler)(struct ib_event *, void *);
976 	void		       *srq_context;
977 	struct ib_srq_attr	attr;
978 	enum ib_srq_type	srq_type;
979 
980 	struct {
981 		struct ib_cq   *cq;
982 		union {
983 			struct {
984 				struct ib_xrcd *xrcd;
985 			} xrc;
986 
987 			struct {
988 				u32		max_num_tags;
989 			} tag_matching;
990 		};
991 	} ext;
992 };
993 
994 struct ib_qp_cap {
995 	u32	max_send_wr;
996 	u32	max_recv_wr;
997 	u32	max_send_sge;
998 	u32	max_recv_sge;
999 	u32	max_inline_data;
1000 
1001 	/*
1002 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1003 	 * ib_create_qp() will calculate the right amount of neededed WRs
1004 	 * and MRs based on this.
1005 	 */
1006 	u32	max_rdma_ctxs;
1007 };
1008 
1009 enum ib_sig_type {
1010 	IB_SIGNAL_ALL_WR,
1011 	IB_SIGNAL_REQ_WR
1012 };
1013 
1014 enum ib_qp_type {
1015 	/*
1016 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1017 	 * here (and in that order) since the MAD layer uses them as
1018 	 * indices into a 2-entry table.
1019 	 */
1020 	IB_QPT_SMI,
1021 	IB_QPT_GSI,
1022 
1023 	IB_QPT_RC,
1024 	IB_QPT_UC,
1025 	IB_QPT_UD,
1026 	IB_QPT_RAW_IPV6,
1027 	IB_QPT_RAW_ETHERTYPE,
1028 	IB_QPT_RAW_PACKET = 8,
1029 	IB_QPT_XRC_INI = 9,
1030 	IB_QPT_XRC_TGT,
1031 	IB_QPT_MAX,
1032 	IB_QPT_DRIVER = 0xFF,
1033 	/* Reserve a range for qp types internal to the low level driver.
1034 	 * These qp types will not be visible at the IB core layer, so the
1035 	 * IB_QPT_MAX usages should not be affected in the core layer
1036 	 */
1037 	IB_QPT_RESERVED1 = 0x1000,
1038 	IB_QPT_RESERVED2,
1039 	IB_QPT_RESERVED3,
1040 	IB_QPT_RESERVED4,
1041 	IB_QPT_RESERVED5,
1042 	IB_QPT_RESERVED6,
1043 	IB_QPT_RESERVED7,
1044 	IB_QPT_RESERVED8,
1045 	IB_QPT_RESERVED9,
1046 	IB_QPT_RESERVED10,
1047 };
1048 
1049 enum ib_qp_create_flags {
1050 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1051 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1052 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1053 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1054 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1055 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1056 	IB_QP_CREATE_INTEGRITY_EN		= 1 << 6,
1057 	/* FREE					= 1 << 7, */
1058 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1059 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1060 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1061 	IB_QP_CREATE_PCI_WRITE_END_PADDING	= 1 << 11,
1062 	/* reserve bits 26-31 for low level drivers' internal use */
1063 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1064 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1065 };
1066 
1067 /*
1068  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1069  * callback to destroy the passed in QP.
1070  */
1071 
1072 struct ib_qp_init_attr {
1073 	/* Consumer's event_handler callback must not block */
1074 	void                  (*event_handler)(struct ib_event *, void *);
1075 
1076 	void		       *qp_context;
1077 	struct ib_cq	       *send_cq;
1078 	struct ib_cq	       *recv_cq;
1079 	struct ib_srq	       *srq;
1080 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1081 	struct ib_qp_cap	cap;
1082 	enum ib_sig_type	sq_sig_type;
1083 	enum ib_qp_type		qp_type;
1084 	u32			create_flags;
1085 
1086 	/*
1087 	 * Only needed for special QP types, or when using the RW API.
1088 	 */
1089 	u8			port_num;
1090 	struct ib_rwq_ind_table *rwq_ind_tbl;
1091 	u32			source_qpn;
1092 };
1093 
1094 struct ib_qp_open_attr {
1095 	void                  (*event_handler)(struct ib_event *, void *);
1096 	void		       *qp_context;
1097 	u32			qp_num;
1098 	enum ib_qp_type		qp_type;
1099 };
1100 
1101 enum ib_rnr_timeout {
1102 	IB_RNR_TIMER_655_36 =  0,
1103 	IB_RNR_TIMER_000_01 =  1,
1104 	IB_RNR_TIMER_000_02 =  2,
1105 	IB_RNR_TIMER_000_03 =  3,
1106 	IB_RNR_TIMER_000_04 =  4,
1107 	IB_RNR_TIMER_000_06 =  5,
1108 	IB_RNR_TIMER_000_08 =  6,
1109 	IB_RNR_TIMER_000_12 =  7,
1110 	IB_RNR_TIMER_000_16 =  8,
1111 	IB_RNR_TIMER_000_24 =  9,
1112 	IB_RNR_TIMER_000_32 = 10,
1113 	IB_RNR_TIMER_000_48 = 11,
1114 	IB_RNR_TIMER_000_64 = 12,
1115 	IB_RNR_TIMER_000_96 = 13,
1116 	IB_RNR_TIMER_001_28 = 14,
1117 	IB_RNR_TIMER_001_92 = 15,
1118 	IB_RNR_TIMER_002_56 = 16,
1119 	IB_RNR_TIMER_003_84 = 17,
1120 	IB_RNR_TIMER_005_12 = 18,
1121 	IB_RNR_TIMER_007_68 = 19,
1122 	IB_RNR_TIMER_010_24 = 20,
1123 	IB_RNR_TIMER_015_36 = 21,
1124 	IB_RNR_TIMER_020_48 = 22,
1125 	IB_RNR_TIMER_030_72 = 23,
1126 	IB_RNR_TIMER_040_96 = 24,
1127 	IB_RNR_TIMER_061_44 = 25,
1128 	IB_RNR_TIMER_081_92 = 26,
1129 	IB_RNR_TIMER_122_88 = 27,
1130 	IB_RNR_TIMER_163_84 = 28,
1131 	IB_RNR_TIMER_245_76 = 29,
1132 	IB_RNR_TIMER_327_68 = 30,
1133 	IB_RNR_TIMER_491_52 = 31
1134 };
1135 
1136 enum ib_qp_attr_mask {
1137 	IB_QP_STATE			= 1,
1138 	IB_QP_CUR_STATE			= (1<<1),
1139 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1140 	IB_QP_ACCESS_FLAGS		= (1<<3),
1141 	IB_QP_PKEY_INDEX		= (1<<4),
1142 	IB_QP_PORT			= (1<<5),
1143 	IB_QP_QKEY			= (1<<6),
1144 	IB_QP_AV			= (1<<7),
1145 	IB_QP_PATH_MTU			= (1<<8),
1146 	IB_QP_TIMEOUT			= (1<<9),
1147 	IB_QP_RETRY_CNT			= (1<<10),
1148 	IB_QP_RNR_RETRY			= (1<<11),
1149 	IB_QP_RQ_PSN			= (1<<12),
1150 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1151 	IB_QP_ALT_PATH			= (1<<14),
1152 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1153 	IB_QP_SQ_PSN			= (1<<16),
1154 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1155 	IB_QP_PATH_MIG_STATE		= (1<<18),
1156 	IB_QP_CAP			= (1<<19),
1157 	IB_QP_DEST_QPN			= (1<<20),
1158 	IB_QP_RESERVED1			= (1<<21),
1159 	IB_QP_RESERVED2			= (1<<22),
1160 	IB_QP_RESERVED3			= (1<<23),
1161 	IB_QP_RESERVED4			= (1<<24),
1162 	IB_QP_RATE_LIMIT		= (1<<25),
1163 };
1164 
1165 enum ib_qp_state {
1166 	IB_QPS_RESET,
1167 	IB_QPS_INIT,
1168 	IB_QPS_RTR,
1169 	IB_QPS_RTS,
1170 	IB_QPS_SQD,
1171 	IB_QPS_SQE,
1172 	IB_QPS_ERR
1173 };
1174 
1175 enum ib_mig_state {
1176 	IB_MIG_MIGRATED,
1177 	IB_MIG_REARM,
1178 	IB_MIG_ARMED
1179 };
1180 
1181 enum ib_mw_type {
1182 	IB_MW_TYPE_1 = 1,
1183 	IB_MW_TYPE_2 = 2
1184 };
1185 
1186 struct ib_qp_attr {
1187 	enum ib_qp_state	qp_state;
1188 	enum ib_qp_state	cur_qp_state;
1189 	enum ib_mtu		path_mtu;
1190 	enum ib_mig_state	path_mig_state;
1191 	u32			qkey;
1192 	u32			rq_psn;
1193 	u32			sq_psn;
1194 	u32			dest_qp_num;
1195 	int			qp_access_flags;
1196 	struct ib_qp_cap	cap;
1197 	struct rdma_ah_attr	ah_attr;
1198 	struct rdma_ah_attr	alt_ah_attr;
1199 	u16			pkey_index;
1200 	u16			alt_pkey_index;
1201 	u8			en_sqd_async_notify;
1202 	u8			sq_draining;
1203 	u8			max_rd_atomic;
1204 	u8			max_dest_rd_atomic;
1205 	u8			min_rnr_timer;
1206 	u8			port_num;
1207 	u8			timeout;
1208 	u8			retry_cnt;
1209 	u8			rnr_retry;
1210 	u8			alt_port_num;
1211 	u8			alt_timeout;
1212 	u32			rate_limit;
1213 };
1214 
1215 enum ib_wr_opcode {
1216 	/* These are shared with userspace */
1217 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1218 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1219 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1220 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1221 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1222 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1223 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1224 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1225 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1226 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1227 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1228 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1229 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1230 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1231 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1232 
1233 	/* These are kernel only and can not be issued by userspace */
1234 	IB_WR_REG_MR = 0x20,
1235 	IB_WR_REG_MR_INTEGRITY,
1236 
1237 	/* reserve values for low level drivers' internal use.
1238 	 * These values will not be used at all in the ib core layer.
1239 	 */
1240 	IB_WR_RESERVED1 = 0xf0,
1241 	IB_WR_RESERVED2,
1242 	IB_WR_RESERVED3,
1243 	IB_WR_RESERVED4,
1244 	IB_WR_RESERVED5,
1245 	IB_WR_RESERVED6,
1246 	IB_WR_RESERVED7,
1247 	IB_WR_RESERVED8,
1248 	IB_WR_RESERVED9,
1249 	IB_WR_RESERVED10,
1250 };
1251 
1252 enum ib_send_flags {
1253 	IB_SEND_FENCE		= 1,
1254 	IB_SEND_SIGNALED	= (1<<1),
1255 	IB_SEND_SOLICITED	= (1<<2),
1256 	IB_SEND_INLINE		= (1<<3),
1257 	IB_SEND_IP_CSUM		= (1<<4),
1258 
1259 	/* reserve bits 26-31 for low level drivers' internal use */
1260 	IB_SEND_RESERVED_START	= (1 << 26),
1261 	IB_SEND_RESERVED_END	= (1 << 31),
1262 };
1263 
1264 struct ib_sge {
1265 	u64	addr;
1266 	u32	length;
1267 	u32	lkey;
1268 };
1269 
1270 struct ib_cqe {
1271 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1272 };
1273 
1274 struct ib_send_wr {
1275 	struct ib_send_wr      *next;
1276 	union {
1277 		u64		wr_id;
1278 		struct ib_cqe	*wr_cqe;
1279 	};
1280 	struct ib_sge	       *sg_list;
1281 	int			num_sge;
1282 	enum ib_wr_opcode	opcode;
1283 	int			send_flags;
1284 	union {
1285 		__be32		imm_data;
1286 		u32		invalidate_rkey;
1287 	} ex;
1288 };
1289 
1290 struct ib_rdma_wr {
1291 	struct ib_send_wr	wr;
1292 	u64			remote_addr;
1293 	u32			rkey;
1294 };
1295 
1296 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1297 {
1298 	return container_of(wr, struct ib_rdma_wr, wr);
1299 }
1300 
1301 struct ib_atomic_wr {
1302 	struct ib_send_wr	wr;
1303 	u64			remote_addr;
1304 	u64			compare_add;
1305 	u64			swap;
1306 	u64			compare_add_mask;
1307 	u64			swap_mask;
1308 	u32			rkey;
1309 };
1310 
1311 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1312 {
1313 	return container_of(wr, struct ib_atomic_wr, wr);
1314 }
1315 
1316 struct ib_ud_wr {
1317 	struct ib_send_wr	wr;
1318 	struct ib_ah		*ah;
1319 	void			*header;
1320 	int			hlen;
1321 	int			mss;
1322 	u32			remote_qpn;
1323 	u32			remote_qkey;
1324 	u16			pkey_index; /* valid for GSI only */
1325 	u8			port_num;   /* valid for DR SMPs on switch only */
1326 };
1327 
1328 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1329 {
1330 	return container_of(wr, struct ib_ud_wr, wr);
1331 }
1332 
1333 struct ib_reg_wr {
1334 	struct ib_send_wr	wr;
1335 	struct ib_mr		*mr;
1336 	u32			key;
1337 	int			access;
1338 };
1339 
1340 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1341 {
1342 	return container_of(wr, struct ib_reg_wr, wr);
1343 }
1344 
1345 struct ib_recv_wr {
1346 	struct ib_recv_wr      *next;
1347 	union {
1348 		u64		wr_id;
1349 		struct ib_cqe	*wr_cqe;
1350 	};
1351 	struct ib_sge	       *sg_list;
1352 	int			num_sge;
1353 };
1354 
1355 enum ib_access_flags {
1356 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1357 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1358 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1359 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1360 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1361 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1362 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1363 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1364 
1365 	IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1366 };
1367 
1368 /*
1369  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1370  * are hidden here instead of a uapi header!
1371  */
1372 enum ib_mr_rereg_flags {
1373 	IB_MR_REREG_TRANS	= 1,
1374 	IB_MR_REREG_PD		= (1<<1),
1375 	IB_MR_REREG_ACCESS	= (1<<2),
1376 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1377 };
1378 
1379 struct ib_fmr_attr {
1380 	int	max_pages;
1381 	int	max_maps;
1382 	u8	page_shift;
1383 };
1384 
1385 struct ib_umem;
1386 
1387 enum rdma_remove_reason {
1388 	/*
1389 	 * Userspace requested uobject deletion or initial try
1390 	 * to remove uobject via cleanup. Call could fail
1391 	 */
1392 	RDMA_REMOVE_DESTROY,
1393 	/* Context deletion. This call should delete the actual object itself */
1394 	RDMA_REMOVE_CLOSE,
1395 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1396 	RDMA_REMOVE_DRIVER_REMOVE,
1397 	/* uobj is being cleaned-up before being committed */
1398 	RDMA_REMOVE_ABORT,
1399 };
1400 
1401 struct ib_rdmacg_object {
1402 #ifdef CONFIG_CGROUP_RDMA
1403 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1404 #endif
1405 };
1406 
1407 struct ib_ucontext {
1408 	struct ib_device       *device;
1409 	struct ib_uverbs_file  *ufile;
1410 	/*
1411 	 * 'closing' can be read by the driver only during a destroy callback,
1412 	 * it is set when we are closing the file descriptor and indicates
1413 	 * that mm_sem may be locked.
1414 	 */
1415 	bool closing;
1416 
1417 	bool cleanup_retryable;
1418 
1419 	void (*invalidate_range)(struct ib_umem_odp *umem_odp,
1420 				 unsigned long start, unsigned long end);
1421 	struct mutex per_mm_list_lock;
1422 	struct list_head per_mm_list;
1423 
1424 	struct ib_rdmacg_object	cg_obj;
1425 	/*
1426 	 * Implementation details of the RDMA core, don't use in drivers:
1427 	 */
1428 	struct rdma_restrack_entry res;
1429 };
1430 
1431 struct ib_uobject {
1432 	u64			user_handle;	/* handle given to us by userspace */
1433 	/* ufile & ucontext owning this object */
1434 	struct ib_uverbs_file  *ufile;
1435 	/* FIXME, save memory: ufile->context == context */
1436 	struct ib_ucontext     *context;	/* associated user context */
1437 	void		       *object;		/* containing object */
1438 	struct list_head	list;		/* link to context's list */
1439 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1440 	int			id;		/* index into kernel idr */
1441 	struct kref		ref;
1442 	atomic_t		usecnt;		/* protects exclusive access */
1443 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1444 
1445 	const struct uverbs_api_object *uapi_object;
1446 };
1447 
1448 struct ib_udata {
1449 	const void __user *inbuf;
1450 	void __user *outbuf;
1451 	size_t       inlen;
1452 	size_t       outlen;
1453 };
1454 
1455 struct ib_pd {
1456 	u32			local_dma_lkey;
1457 	u32			flags;
1458 	struct ib_device       *device;
1459 	struct ib_uobject      *uobject;
1460 	atomic_t          	usecnt; /* count all resources */
1461 
1462 	u32			unsafe_global_rkey;
1463 
1464 	/*
1465 	 * Implementation details of the RDMA core, don't use in drivers:
1466 	 */
1467 	struct ib_mr	       *__internal_mr;
1468 	struct rdma_restrack_entry res;
1469 };
1470 
1471 struct ib_xrcd {
1472 	struct ib_device       *device;
1473 	atomic_t		usecnt; /* count all exposed resources */
1474 	struct inode	       *inode;
1475 
1476 	struct mutex		tgt_qp_mutex;
1477 	struct list_head	tgt_qp_list;
1478 };
1479 
1480 struct ib_ah {
1481 	struct ib_device	*device;
1482 	struct ib_pd		*pd;
1483 	struct ib_uobject	*uobject;
1484 	const struct ib_gid_attr *sgid_attr;
1485 	enum rdma_ah_attr_type	type;
1486 };
1487 
1488 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1489 
1490 enum ib_poll_context {
1491 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1492 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1493 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1494 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1495 };
1496 
1497 struct ib_cq {
1498 	struct ib_device       *device;
1499 	struct ib_uobject      *uobject;
1500 	ib_comp_handler   	comp_handler;
1501 	void                  (*event_handler)(struct ib_event *, void *);
1502 	void                   *cq_context;
1503 	int               	cqe;
1504 	atomic_t          	usecnt; /* count number of work queues */
1505 	enum ib_poll_context	poll_ctx;
1506 	struct ib_wc		*wc;
1507 	union {
1508 		struct irq_poll		iop;
1509 		struct work_struct	work;
1510 	};
1511 	struct workqueue_struct *comp_wq;
1512 	/*
1513 	 * Implementation details of the RDMA core, don't use in drivers:
1514 	 */
1515 	struct rdma_restrack_entry res;
1516 };
1517 
1518 struct ib_srq {
1519 	struct ib_device       *device;
1520 	struct ib_pd	       *pd;
1521 	struct ib_uobject      *uobject;
1522 	void		      (*event_handler)(struct ib_event *, void *);
1523 	void		       *srq_context;
1524 	enum ib_srq_type	srq_type;
1525 	atomic_t		usecnt;
1526 
1527 	struct {
1528 		struct ib_cq   *cq;
1529 		union {
1530 			struct {
1531 				struct ib_xrcd *xrcd;
1532 				u32		srq_num;
1533 			} xrc;
1534 		};
1535 	} ext;
1536 };
1537 
1538 enum ib_raw_packet_caps {
1539 	/* Strip cvlan from incoming packet and report it in the matching work
1540 	 * completion is supported.
1541 	 */
1542 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1543 	/* Scatter FCS field of an incoming packet to host memory is supported.
1544 	 */
1545 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1546 	/* Checksum offloads are supported (for both send and receive). */
1547 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1548 	/* When a packet is received for an RQ with no receive WQEs, the
1549 	 * packet processing is delayed.
1550 	 */
1551 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1552 };
1553 
1554 enum ib_wq_type {
1555 	IB_WQT_RQ
1556 };
1557 
1558 enum ib_wq_state {
1559 	IB_WQS_RESET,
1560 	IB_WQS_RDY,
1561 	IB_WQS_ERR
1562 };
1563 
1564 struct ib_wq {
1565 	struct ib_device       *device;
1566 	struct ib_uobject      *uobject;
1567 	void		    *wq_context;
1568 	void		    (*event_handler)(struct ib_event *, void *);
1569 	struct ib_pd	       *pd;
1570 	struct ib_cq	       *cq;
1571 	u32		wq_num;
1572 	enum ib_wq_state       state;
1573 	enum ib_wq_type	wq_type;
1574 	atomic_t		usecnt;
1575 };
1576 
1577 enum ib_wq_flags {
1578 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1579 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1580 	IB_WQ_FLAGS_DELAY_DROP		= 1 << 2,
1581 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1582 };
1583 
1584 struct ib_wq_init_attr {
1585 	void		       *wq_context;
1586 	enum ib_wq_type	wq_type;
1587 	u32		max_wr;
1588 	u32		max_sge;
1589 	struct	ib_cq	       *cq;
1590 	void		    (*event_handler)(struct ib_event *, void *);
1591 	u32		create_flags; /* Use enum ib_wq_flags */
1592 };
1593 
1594 enum ib_wq_attr_mask {
1595 	IB_WQ_STATE		= 1 << 0,
1596 	IB_WQ_CUR_STATE		= 1 << 1,
1597 	IB_WQ_FLAGS		= 1 << 2,
1598 };
1599 
1600 struct ib_wq_attr {
1601 	enum	ib_wq_state	wq_state;
1602 	enum	ib_wq_state	curr_wq_state;
1603 	u32			flags; /* Use enum ib_wq_flags */
1604 	u32			flags_mask; /* Use enum ib_wq_flags */
1605 };
1606 
1607 struct ib_rwq_ind_table {
1608 	struct ib_device	*device;
1609 	struct ib_uobject      *uobject;
1610 	atomic_t		usecnt;
1611 	u32		ind_tbl_num;
1612 	u32		log_ind_tbl_size;
1613 	struct ib_wq	**ind_tbl;
1614 };
1615 
1616 struct ib_rwq_ind_table_init_attr {
1617 	u32		log_ind_tbl_size;
1618 	/* Each entry is a pointer to Receive Work Queue */
1619 	struct ib_wq	**ind_tbl;
1620 };
1621 
1622 enum port_pkey_state {
1623 	IB_PORT_PKEY_NOT_VALID = 0,
1624 	IB_PORT_PKEY_VALID = 1,
1625 	IB_PORT_PKEY_LISTED = 2,
1626 };
1627 
1628 struct ib_qp_security;
1629 
1630 struct ib_port_pkey {
1631 	enum port_pkey_state	state;
1632 	u16			pkey_index;
1633 	u8			port_num;
1634 	struct list_head	qp_list;
1635 	struct list_head	to_error_list;
1636 	struct ib_qp_security  *sec;
1637 };
1638 
1639 struct ib_ports_pkeys {
1640 	struct ib_port_pkey	main;
1641 	struct ib_port_pkey	alt;
1642 };
1643 
1644 struct ib_qp_security {
1645 	struct ib_qp	       *qp;
1646 	struct ib_device       *dev;
1647 	/* Hold this mutex when changing port and pkey settings. */
1648 	struct mutex		mutex;
1649 	struct ib_ports_pkeys  *ports_pkeys;
1650 	/* A list of all open shared QP handles.  Required to enforce security
1651 	 * properly for all users of a shared QP.
1652 	 */
1653 	struct list_head        shared_qp_list;
1654 	void                   *security;
1655 	bool			destroying;
1656 	atomic_t		error_list_count;
1657 	struct completion	error_complete;
1658 	int			error_comps_pending;
1659 };
1660 
1661 /*
1662  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1663  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1664  */
1665 struct ib_qp {
1666 	struct ib_device       *device;
1667 	struct ib_pd	       *pd;
1668 	struct ib_cq	       *send_cq;
1669 	struct ib_cq	       *recv_cq;
1670 	spinlock_t		mr_lock;
1671 	int			mrs_used;
1672 	struct list_head	rdma_mrs;
1673 	struct list_head	sig_mrs;
1674 	struct ib_srq	       *srq;
1675 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1676 	struct list_head	xrcd_list;
1677 
1678 	/* count times opened, mcast attaches, flow attaches */
1679 	atomic_t		usecnt;
1680 	struct list_head	open_list;
1681 	struct ib_qp           *real_qp;
1682 	struct ib_uobject      *uobject;
1683 	void                  (*event_handler)(struct ib_event *, void *);
1684 	void		       *qp_context;
1685 	/* sgid_attrs associated with the AV's */
1686 	const struct ib_gid_attr *av_sgid_attr;
1687 	const struct ib_gid_attr *alt_path_sgid_attr;
1688 	u32			qp_num;
1689 	u32			max_write_sge;
1690 	u32			max_read_sge;
1691 	enum ib_qp_type		qp_type;
1692 	struct ib_rwq_ind_table *rwq_ind_tbl;
1693 	struct ib_qp_security  *qp_sec;
1694 	u8			port;
1695 
1696 	bool			integrity_en;
1697 	/*
1698 	 * Implementation details of the RDMA core, don't use in drivers:
1699 	 */
1700 	struct rdma_restrack_entry     res;
1701 
1702 	/* The counter the qp is bind to */
1703 	struct rdma_counter    *counter;
1704 };
1705 
1706 struct ib_dm {
1707 	struct ib_device  *device;
1708 	u32		   length;
1709 	u32		   flags;
1710 	struct ib_uobject *uobject;
1711 	atomic_t	   usecnt;
1712 };
1713 
1714 struct ib_mr {
1715 	struct ib_device  *device;
1716 	struct ib_pd	  *pd;
1717 	u32		   lkey;
1718 	u32		   rkey;
1719 	u64		   iova;
1720 	u64		   length;
1721 	unsigned int	   page_size;
1722 	enum ib_mr_type	   type;
1723 	bool		   need_inval;
1724 	union {
1725 		struct ib_uobject	*uobject;	/* user */
1726 		struct list_head	qp_entry;	/* FR */
1727 	};
1728 
1729 	struct ib_dm      *dm;
1730 	struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1731 	/*
1732 	 * Implementation details of the RDMA core, don't use in drivers:
1733 	 */
1734 	struct rdma_restrack_entry res;
1735 };
1736 
1737 struct ib_mw {
1738 	struct ib_device	*device;
1739 	struct ib_pd		*pd;
1740 	struct ib_uobject	*uobject;
1741 	u32			rkey;
1742 	enum ib_mw_type         type;
1743 };
1744 
1745 struct ib_fmr {
1746 	struct ib_device	*device;
1747 	struct ib_pd		*pd;
1748 	struct list_head	list;
1749 	u32			lkey;
1750 	u32			rkey;
1751 };
1752 
1753 /* Supported steering options */
1754 enum ib_flow_attr_type {
1755 	/* steering according to rule specifications */
1756 	IB_FLOW_ATTR_NORMAL		= 0x0,
1757 	/* default unicast and multicast rule -
1758 	 * receive all Eth traffic which isn't steered to any QP
1759 	 */
1760 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1761 	/* default multicast rule -
1762 	 * receive all Eth multicast traffic which isn't steered to any QP
1763 	 */
1764 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1765 	/* sniffer rule - receive all port traffic */
1766 	IB_FLOW_ATTR_SNIFFER		= 0x3
1767 };
1768 
1769 /* Supported steering header types */
1770 enum ib_flow_spec_type {
1771 	/* L2 headers*/
1772 	IB_FLOW_SPEC_ETH		= 0x20,
1773 	IB_FLOW_SPEC_IB			= 0x22,
1774 	/* L3 header*/
1775 	IB_FLOW_SPEC_IPV4		= 0x30,
1776 	IB_FLOW_SPEC_IPV6		= 0x31,
1777 	IB_FLOW_SPEC_ESP                = 0x34,
1778 	/* L4 headers*/
1779 	IB_FLOW_SPEC_TCP		= 0x40,
1780 	IB_FLOW_SPEC_UDP		= 0x41,
1781 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1782 	IB_FLOW_SPEC_GRE		= 0x51,
1783 	IB_FLOW_SPEC_MPLS		= 0x60,
1784 	IB_FLOW_SPEC_INNER		= 0x100,
1785 	/* Actions */
1786 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1787 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1788 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1789 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1790 };
1791 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1792 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1793 
1794 /* Flow steering rule priority is set according to it's domain.
1795  * Lower domain value means higher priority.
1796  */
1797 enum ib_flow_domain {
1798 	IB_FLOW_DOMAIN_USER,
1799 	IB_FLOW_DOMAIN_ETHTOOL,
1800 	IB_FLOW_DOMAIN_RFS,
1801 	IB_FLOW_DOMAIN_NIC,
1802 	IB_FLOW_DOMAIN_NUM /* Must be last */
1803 };
1804 
1805 enum ib_flow_flags {
1806 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1807 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1808 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1809 };
1810 
1811 struct ib_flow_eth_filter {
1812 	u8	dst_mac[6];
1813 	u8	src_mac[6];
1814 	__be16	ether_type;
1815 	__be16	vlan_tag;
1816 	/* Must be last */
1817 	u8	real_sz[0];
1818 };
1819 
1820 struct ib_flow_spec_eth {
1821 	u32			  type;
1822 	u16			  size;
1823 	struct ib_flow_eth_filter val;
1824 	struct ib_flow_eth_filter mask;
1825 };
1826 
1827 struct ib_flow_ib_filter {
1828 	__be16 dlid;
1829 	__u8   sl;
1830 	/* Must be last */
1831 	u8	real_sz[0];
1832 };
1833 
1834 struct ib_flow_spec_ib {
1835 	u32			 type;
1836 	u16			 size;
1837 	struct ib_flow_ib_filter val;
1838 	struct ib_flow_ib_filter mask;
1839 };
1840 
1841 /* IPv4 header flags */
1842 enum ib_ipv4_flags {
1843 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1844 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1845 				    last have this flag set */
1846 };
1847 
1848 struct ib_flow_ipv4_filter {
1849 	__be32	src_ip;
1850 	__be32	dst_ip;
1851 	u8	proto;
1852 	u8	tos;
1853 	u8	ttl;
1854 	u8	flags;
1855 	/* Must be last */
1856 	u8	real_sz[0];
1857 };
1858 
1859 struct ib_flow_spec_ipv4 {
1860 	u32			   type;
1861 	u16			   size;
1862 	struct ib_flow_ipv4_filter val;
1863 	struct ib_flow_ipv4_filter mask;
1864 };
1865 
1866 struct ib_flow_ipv6_filter {
1867 	u8	src_ip[16];
1868 	u8	dst_ip[16];
1869 	__be32	flow_label;
1870 	u8	next_hdr;
1871 	u8	traffic_class;
1872 	u8	hop_limit;
1873 	/* Must be last */
1874 	u8	real_sz[0];
1875 };
1876 
1877 struct ib_flow_spec_ipv6 {
1878 	u32			   type;
1879 	u16			   size;
1880 	struct ib_flow_ipv6_filter val;
1881 	struct ib_flow_ipv6_filter mask;
1882 };
1883 
1884 struct ib_flow_tcp_udp_filter {
1885 	__be16	dst_port;
1886 	__be16	src_port;
1887 	/* Must be last */
1888 	u8	real_sz[0];
1889 };
1890 
1891 struct ib_flow_spec_tcp_udp {
1892 	u32			      type;
1893 	u16			      size;
1894 	struct ib_flow_tcp_udp_filter val;
1895 	struct ib_flow_tcp_udp_filter mask;
1896 };
1897 
1898 struct ib_flow_tunnel_filter {
1899 	__be32	tunnel_id;
1900 	u8	real_sz[0];
1901 };
1902 
1903 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1904  * the tunnel_id from val has the vni value
1905  */
1906 struct ib_flow_spec_tunnel {
1907 	u32			      type;
1908 	u16			      size;
1909 	struct ib_flow_tunnel_filter  val;
1910 	struct ib_flow_tunnel_filter  mask;
1911 };
1912 
1913 struct ib_flow_esp_filter {
1914 	__be32	spi;
1915 	__be32  seq;
1916 	/* Must be last */
1917 	u8	real_sz[0];
1918 };
1919 
1920 struct ib_flow_spec_esp {
1921 	u32                           type;
1922 	u16			      size;
1923 	struct ib_flow_esp_filter     val;
1924 	struct ib_flow_esp_filter     mask;
1925 };
1926 
1927 struct ib_flow_gre_filter {
1928 	__be16 c_ks_res0_ver;
1929 	__be16 protocol;
1930 	__be32 key;
1931 	/* Must be last */
1932 	u8	real_sz[0];
1933 };
1934 
1935 struct ib_flow_spec_gre {
1936 	u32                           type;
1937 	u16			      size;
1938 	struct ib_flow_gre_filter     val;
1939 	struct ib_flow_gre_filter     mask;
1940 };
1941 
1942 struct ib_flow_mpls_filter {
1943 	__be32 tag;
1944 	/* Must be last */
1945 	u8	real_sz[0];
1946 };
1947 
1948 struct ib_flow_spec_mpls {
1949 	u32                           type;
1950 	u16			      size;
1951 	struct ib_flow_mpls_filter     val;
1952 	struct ib_flow_mpls_filter     mask;
1953 };
1954 
1955 struct ib_flow_spec_action_tag {
1956 	enum ib_flow_spec_type	      type;
1957 	u16			      size;
1958 	u32                           tag_id;
1959 };
1960 
1961 struct ib_flow_spec_action_drop {
1962 	enum ib_flow_spec_type	      type;
1963 	u16			      size;
1964 };
1965 
1966 struct ib_flow_spec_action_handle {
1967 	enum ib_flow_spec_type	      type;
1968 	u16			      size;
1969 	struct ib_flow_action	     *act;
1970 };
1971 
1972 enum ib_counters_description {
1973 	IB_COUNTER_PACKETS,
1974 	IB_COUNTER_BYTES,
1975 };
1976 
1977 struct ib_flow_spec_action_count {
1978 	enum ib_flow_spec_type type;
1979 	u16 size;
1980 	struct ib_counters *counters;
1981 };
1982 
1983 union ib_flow_spec {
1984 	struct {
1985 		u32			type;
1986 		u16			size;
1987 	};
1988 	struct ib_flow_spec_eth		eth;
1989 	struct ib_flow_spec_ib		ib;
1990 	struct ib_flow_spec_ipv4        ipv4;
1991 	struct ib_flow_spec_tcp_udp	tcp_udp;
1992 	struct ib_flow_spec_ipv6        ipv6;
1993 	struct ib_flow_spec_tunnel      tunnel;
1994 	struct ib_flow_spec_esp		esp;
1995 	struct ib_flow_spec_gre		gre;
1996 	struct ib_flow_spec_mpls	mpls;
1997 	struct ib_flow_spec_action_tag  flow_tag;
1998 	struct ib_flow_spec_action_drop drop;
1999 	struct ib_flow_spec_action_handle action;
2000 	struct ib_flow_spec_action_count flow_count;
2001 };
2002 
2003 struct ib_flow_attr {
2004 	enum ib_flow_attr_type type;
2005 	u16	     size;
2006 	u16	     priority;
2007 	u32	     flags;
2008 	u8	     num_of_specs;
2009 	u8	     port;
2010 	union ib_flow_spec flows[];
2011 };
2012 
2013 struct ib_flow {
2014 	struct ib_qp		*qp;
2015 	struct ib_device	*device;
2016 	struct ib_uobject	*uobject;
2017 };
2018 
2019 enum ib_flow_action_type {
2020 	IB_FLOW_ACTION_UNSPECIFIED,
2021 	IB_FLOW_ACTION_ESP = 1,
2022 };
2023 
2024 struct ib_flow_action_attrs_esp_keymats {
2025 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2026 	union {
2027 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2028 	} keymat;
2029 };
2030 
2031 struct ib_flow_action_attrs_esp_replays {
2032 	enum ib_uverbs_flow_action_esp_replay			protocol;
2033 	union {
2034 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2035 	} replay;
2036 };
2037 
2038 enum ib_flow_action_attrs_esp_flags {
2039 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2040 	 * This is done in order to share the same flags between user-space and
2041 	 * kernel and spare an unnecessary translation.
2042 	 */
2043 
2044 	/* Kernel flags */
2045 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2046 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2047 };
2048 
2049 struct ib_flow_spec_list {
2050 	struct ib_flow_spec_list	*next;
2051 	union ib_flow_spec		spec;
2052 };
2053 
2054 struct ib_flow_action_attrs_esp {
2055 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2056 	struct ib_flow_action_attrs_esp_replays		*replay;
2057 	struct ib_flow_spec_list			*encap;
2058 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2059 	 * Value of 0 is a valid value.
2060 	 */
2061 	u32						esn;
2062 	u32						spi;
2063 	u32						seq;
2064 	u32						tfc_pad;
2065 	/* Use enum ib_flow_action_attrs_esp_flags */
2066 	u64						flags;
2067 	u64						hard_limit_pkts;
2068 };
2069 
2070 struct ib_flow_action {
2071 	struct ib_device		*device;
2072 	struct ib_uobject		*uobject;
2073 	enum ib_flow_action_type	type;
2074 	atomic_t			usecnt;
2075 };
2076 
2077 struct ib_mad_hdr;
2078 struct ib_grh;
2079 
2080 enum ib_process_mad_flags {
2081 	IB_MAD_IGNORE_MKEY	= 1,
2082 	IB_MAD_IGNORE_BKEY	= 2,
2083 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2084 };
2085 
2086 enum ib_mad_result {
2087 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2088 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2089 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2090 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2091 };
2092 
2093 struct ib_port_cache {
2094 	u64		      subnet_prefix;
2095 	struct ib_pkey_cache  *pkey;
2096 	struct ib_gid_table   *gid;
2097 	u8                     lmc;
2098 	enum ib_port_state     port_state;
2099 };
2100 
2101 struct ib_cache {
2102 	rwlock_t                lock;
2103 	struct ib_event_handler event_handler;
2104 };
2105 
2106 struct ib_port_immutable {
2107 	int                           pkey_tbl_len;
2108 	int                           gid_tbl_len;
2109 	u32                           core_cap_flags;
2110 	u32                           max_mad_size;
2111 };
2112 
2113 struct ib_port_data {
2114 	struct ib_device *ib_dev;
2115 
2116 	struct ib_port_immutable immutable;
2117 
2118 	spinlock_t pkey_list_lock;
2119 	struct list_head pkey_list;
2120 
2121 	struct ib_port_cache cache;
2122 
2123 	spinlock_t netdev_lock;
2124 	struct net_device __rcu *netdev;
2125 	struct hlist_node ndev_hash_link;
2126 	struct rdma_port_counter port_counter;
2127 	struct rdma_hw_stats *hw_stats;
2128 };
2129 
2130 /* rdma netdev type - specifies protocol type */
2131 enum rdma_netdev_t {
2132 	RDMA_NETDEV_OPA_VNIC,
2133 	RDMA_NETDEV_IPOIB,
2134 };
2135 
2136 /**
2137  * struct rdma_netdev - rdma netdev
2138  * For cases where netstack interfacing is required.
2139  */
2140 struct rdma_netdev {
2141 	void              *clnt_priv;
2142 	struct ib_device  *hca;
2143 	u8                 port_num;
2144 
2145 	/*
2146 	 * cleanup function must be specified.
2147 	 * FIXME: This is only used for OPA_VNIC and that usage should be
2148 	 * removed too.
2149 	 */
2150 	void (*free_rdma_netdev)(struct net_device *netdev);
2151 
2152 	/* control functions */
2153 	void (*set_id)(struct net_device *netdev, int id);
2154 	/* send packet */
2155 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2156 		    struct ib_ah *address, u32 dqpn);
2157 	/* multicast */
2158 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2159 			    union ib_gid *gid, u16 mlid,
2160 			    int set_qkey, u32 qkey);
2161 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2162 			    union ib_gid *gid, u16 mlid);
2163 };
2164 
2165 struct rdma_netdev_alloc_params {
2166 	size_t sizeof_priv;
2167 	unsigned int txqs;
2168 	unsigned int rxqs;
2169 	void *param;
2170 
2171 	int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2172 				      struct net_device *netdev, void *param);
2173 };
2174 
2175 struct ib_counters {
2176 	struct ib_device	*device;
2177 	struct ib_uobject	*uobject;
2178 	/* num of objects attached */
2179 	atomic_t	usecnt;
2180 };
2181 
2182 struct ib_counters_read_attr {
2183 	u64	*counters_buff;
2184 	u32	ncounters;
2185 	u32	flags; /* use enum ib_read_counters_flags */
2186 };
2187 
2188 struct uverbs_attr_bundle;
2189 struct iw_cm_id;
2190 struct iw_cm_conn_param;
2191 
2192 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member)                      \
2193 	.size_##ib_struct =                                                    \
2194 		(sizeof(struct drv_struct) +                                   \
2195 		 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) +      \
2196 		 BUILD_BUG_ON_ZERO(                                            \
2197 			 !__same_type(((struct drv_struct *)NULL)->member,     \
2198 				      struct ib_struct)))
2199 
2200 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp)                         \
2201 	((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2202 
2203 #define rdma_zalloc_drv_obj(ib_dev, ib_type)                                   \
2204 	rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2205 
2206 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2207 
2208 /**
2209  * struct ib_device_ops - InfiniBand device operations
2210  * This structure defines all the InfiniBand device operations, providers will
2211  * need to define the supported operations, otherwise they will be set to null.
2212  */
2213 struct ib_device_ops {
2214 	struct module *owner;
2215 	enum rdma_driver_id driver_id;
2216 	u32 uverbs_abi_ver;
2217 	unsigned int uverbs_no_driver_id_binding:1;
2218 
2219 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2220 			 const struct ib_send_wr **bad_send_wr);
2221 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2222 			 const struct ib_recv_wr **bad_recv_wr);
2223 	void (*drain_rq)(struct ib_qp *qp);
2224 	void (*drain_sq)(struct ib_qp *qp);
2225 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2226 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2227 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2228 	int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2229 	int (*post_srq_recv)(struct ib_srq *srq,
2230 			     const struct ib_recv_wr *recv_wr,
2231 			     const struct ib_recv_wr **bad_recv_wr);
2232 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2233 			   u8 port_num, const struct ib_wc *in_wc,
2234 			   const struct ib_grh *in_grh,
2235 			   const struct ib_mad_hdr *in_mad, size_t in_mad_size,
2236 			   struct ib_mad_hdr *out_mad, size_t *out_mad_size,
2237 			   u16 *out_mad_pkey_index);
2238 	int (*query_device)(struct ib_device *device,
2239 			    struct ib_device_attr *device_attr,
2240 			    struct ib_udata *udata);
2241 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2242 			     struct ib_device_modify *device_modify);
2243 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2244 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2245 						     int comp_vector);
2246 	int (*query_port)(struct ib_device *device, u8 port_num,
2247 			  struct ib_port_attr *port_attr);
2248 	int (*modify_port)(struct ib_device *device, u8 port_num,
2249 			   int port_modify_mask,
2250 			   struct ib_port_modify *port_modify);
2251 	/**
2252 	 * The following mandatory functions are used only at device
2253 	 * registration.  Keep functions such as these at the end of this
2254 	 * structure to avoid cache line misses when accessing struct ib_device
2255 	 * in fast paths.
2256 	 */
2257 	int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2258 				  struct ib_port_immutable *immutable);
2259 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2260 					       u8 port_num);
2261 	/**
2262 	 * When calling get_netdev, the HW vendor's driver should return the
2263 	 * net device of device @device at port @port_num or NULL if such
2264 	 * a net device doesn't exist. The vendor driver should call dev_hold
2265 	 * on this net device. The HW vendor's device driver must guarantee
2266 	 * that this function returns NULL before the net device has finished
2267 	 * NETDEV_UNREGISTER state.
2268 	 */
2269 	struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2270 	/**
2271 	 * rdma netdev operation
2272 	 *
2273 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2274 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2275 	 */
2276 	struct net_device *(*alloc_rdma_netdev)(
2277 		struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2278 		const char *name, unsigned char name_assign_type,
2279 		void (*setup)(struct net_device *));
2280 
2281 	int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2282 				      enum rdma_netdev_t type,
2283 				      struct rdma_netdev_alloc_params *params);
2284 	/**
2285 	 * query_gid should be return GID value for @device, when @port_num
2286 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2287 	 * is RoCE link layer.
2288 	 */
2289 	int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2290 			 union ib_gid *gid);
2291 	/**
2292 	 * When calling add_gid, the HW vendor's driver should add the gid
2293 	 * of device of port at gid index available at @attr. Meta-info of
2294 	 * that gid (for example, the network device related to this gid) is
2295 	 * available at @attr. @context allows the HW vendor driver to store
2296 	 * extra information together with a GID entry. The HW vendor driver may
2297 	 * allocate memory to contain this information and store it in @context
2298 	 * when a new GID entry is written to. Params are consistent until the
2299 	 * next call of add_gid or delete_gid. The function should return 0 on
2300 	 * success or error otherwise. The function could be called
2301 	 * concurrently for different ports. This function is only called when
2302 	 * roce_gid_table is used.
2303 	 */
2304 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2305 	/**
2306 	 * When calling del_gid, the HW vendor's driver should delete the
2307 	 * gid of device @device at gid index gid_index of port port_num
2308 	 * available in @attr.
2309 	 * Upon the deletion of a GID entry, the HW vendor must free any
2310 	 * allocated memory. The caller will clear @context afterwards.
2311 	 * This function is only called when roce_gid_table is used.
2312 	 */
2313 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2314 	int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2315 			  u16 *pkey);
2316 	int (*alloc_ucontext)(struct ib_ucontext *context,
2317 			      struct ib_udata *udata);
2318 	void (*dealloc_ucontext)(struct ib_ucontext *context);
2319 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2320 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2321 	int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2322 	void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2323 	int (*create_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr,
2324 			 u32 flags, struct ib_udata *udata);
2325 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2326 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2327 	void (*destroy_ah)(struct ib_ah *ah, u32 flags);
2328 	int (*create_srq)(struct ib_srq *srq,
2329 			  struct ib_srq_init_attr *srq_init_attr,
2330 			  struct ib_udata *udata);
2331 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2332 			  enum ib_srq_attr_mask srq_attr_mask,
2333 			  struct ib_udata *udata);
2334 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2335 	void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2336 	struct ib_qp *(*create_qp)(struct ib_pd *pd,
2337 				   struct ib_qp_init_attr *qp_init_attr,
2338 				   struct ib_udata *udata);
2339 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2340 			 int qp_attr_mask, struct ib_udata *udata);
2341 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2342 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2343 	int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2344 	int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2345 			 struct ib_udata *udata);
2346 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2347 	void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2348 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2349 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2350 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2351 				     u64 virt_addr, int mr_access_flags,
2352 				     struct ib_udata *udata);
2353 	int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2354 			     u64 virt_addr, int mr_access_flags,
2355 			     struct ib_pd *pd, struct ib_udata *udata);
2356 	int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2357 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2358 				  u32 max_num_sg, struct ib_udata *udata);
2359 	struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2360 					    u32 max_num_data_sg,
2361 					    u32 max_num_meta_sg);
2362 	int (*advise_mr)(struct ib_pd *pd,
2363 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2364 			 struct ib_sge *sg_list, u32 num_sge,
2365 			 struct uverbs_attr_bundle *attrs);
2366 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2367 			 unsigned int *sg_offset);
2368 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2369 			       struct ib_mr_status *mr_status);
2370 	struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2371 				  struct ib_udata *udata);
2372 	int (*dealloc_mw)(struct ib_mw *mw);
2373 	struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2374 				    struct ib_fmr_attr *fmr_attr);
2375 	int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2376 			    u64 iova);
2377 	int (*unmap_fmr)(struct list_head *fmr_list);
2378 	int (*dealloc_fmr)(struct ib_fmr *fmr);
2379 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2380 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2381 	struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2382 				      struct ib_udata *udata);
2383 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2384 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2385 				       struct ib_flow_attr *flow_attr,
2386 				       int domain, struct ib_udata *udata);
2387 	int (*destroy_flow)(struct ib_flow *flow_id);
2388 	struct ib_flow_action *(*create_flow_action_esp)(
2389 		struct ib_device *device,
2390 		const struct ib_flow_action_attrs_esp *attr,
2391 		struct uverbs_attr_bundle *attrs);
2392 	int (*destroy_flow_action)(struct ib_flow_action *action);
2393 	int (*modify_flow_action_esp)(
2394 		struct ib_flow_action *action,
2395 		const struct ib_flow_action_attrs_esp *attr,
2396 		struct uverbs_attr_bundle *attrs);
2397 	int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2398 				 int state);
2399 	int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2400 			     struct ifla_vf_info *ivf);
2401 	int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2402 			    struct ifla_vf_stats *stats);
2403 	int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2404 			   int type);
2405 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2406 				   struct ib_wq_init_attr *init_attr,
2407 				   struct ib_udata *udata);
2408 	void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2409 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2410 			 u32 wq_attr_mask, struct ib_udata *udata);
2411 	struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2412 		struct ib_device *device,
2413 		struct ib_rwq_ind_table_init_attr *init_attr,
2414 		struct ib_udata *udata);
2415 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2416 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2417 				  struct ib_ucontext *context,
2418 				  struct ib_dm_alloc_attr *attr,
2419 				  struct uverbs_attr_bundle *attrs);
2420 	int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2421 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2422 				   struct ib_dm_mr_attr *attr,
2423 				   struct uverbs_attr_bundle *attrs);
2424 	struct ib_counters *(*create_counters)(
2425 		struct ib_device *device, struct uverbs_attr_bundle *attrs);
2426 	int (*destroy_counters)(struct ib_counters *counters);
2427 	int (*read_counters)(struct ib_counters *counters,
2428 			     struct ib_counters_read_attr *counters_read_attr,
2429 			     struct uverbs_attr_bundle *attrs);
2430 	int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2431 			    int data_sg_nents, unsigned int *data_sg_offset,
2432 			    struct scatterlist *meta_sg, int meta_sg_nents,
2433 			    unsigned int *meta_sg_offset);
2434 
2435 	/**
2436 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2437 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2438 	 *   core when the device is removed.  A lifespan of -1 in the return
2439 	 *   struct tells the core to set a default lifespan.
2440 	 */
2441 	struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2442 						u8 port_num);
2443 	/**
2444 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2445 	 * @index - The index in the value array we wish to have updated, or
2446 	 *   num_counters if we want all stats updated
2447 	 * Return codes -
2448 	 *   < 0 - Error, no counters updated
2449 	 *   index - Updated the single counter pointed to by index
2450 	 *   num_counters - Updated all counters (will reset the timestamp
2451 	 *     and prevent further calls for lifespan milliseconds)
2452 	 * Drivers are allowed to update all counters in leiu of just the
2453 	 *   one given in index at their option
2454 	 */
2455 	int (*get_hw_stats)(struct ib_device *device,
2456 			    struct rdma_hw_stats *stats, u8 port, int index);
2457 	/*
2458 	 * This function is called once for each port when a ib device is
2459 	 * registered.
2460 	 */
2461 	int (*init_port)(struct ib_device *device, u8 port_num,
2462 			 struct kobject *port_sysfs);
2463 	/**
2464 	 * Allows rdma drivers to add their own restrack attributes.
2465 	 */
2466 	int (*fill_res_entry)(struct sk_buff *msg,
2467 			      struct rdma_restrack_entry *entry);
2468 
2469 	/* Device lifecycle callbacks */
2470 	/*
2471 	 * Called after the device becomes registered, before clients are
2472 	 * attached
2473 	 */
2474 	int (*enable_driver)(struct ib_device *dev);
2475 	/*
2476 	 * This is called as part of ib_dealloc_device().
2477 	 */
2478 	void (*dealloc_driver)(struct ib_device *dev);
2479 
2480 	/* iWarp CM callbacks */
2481 	void (*iw_add_ref)(struct ib_qp *qp);
2482 	void (*iw_rem_ref)(struct ib_qp *qp);
2483 	struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2484 	int (*iw_connect)(struct iw_cm_id *cm_id,
2485 			  struct iw_cm_conn_param *conn_param);
2486 	int (*iw_accept)(struct iw_cm_id *cm_id,
2487 			 struct iw_cm_conn_param *conn_param);
2488 	int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2489 			 u8 pdata_len);
2490 	int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2491 	int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2492 	/**
2493 	 * counter_bind_qp - Bind a QP to a counter.
2494 	 * @counter - The counter to be bound. If counter->id is zero then
2495 	 *   the driver needs to allocate a new counter and set counter->id
2496 	 */
2497 	int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2498 	/**
2499 	 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2500 	 *   counter and bind it onto the default one
2501 	 */
2502 	int (*counter_unbind_qp)(struct ib_qp *qp);
2503 	/**
2504 	 * counter_dealloc -De-allocate the hw counter
2505 	 */
2506 	int (*counter_dealloc)(struct rdma_counter *counter);
2507 	/**
2508 	 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2509 	 * the driver initialized data.
2510 	 */
2511 	struct rdma_hw_stats *(*counter_alloc_stats)(
2512 		struct rdma_counter *counter);
2513 	/**
2514 	 * counter_update_stats - Query the stats value of this counter
2515 	 */
2516 	int (*counter_update_stats)(struct rdma_counter *counter);
2517 
2518 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
2519 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
2520 	DECLARE_RDMA_OBJ_SIZE(ib_pd);
2521 	DECLARE_RDMA_OBJ_SIZE(ib_srq);
2522 	DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2523 };
2524 
2525 struct ib_core_device {
2526 	/* device must be the first element in structure until,
2527 	 * union of ib_core_device and device exists in ib_device.
2528 	 */
2529 	struct device dev;
2530 	possible_net_t rdma_net;
2531 	struct kobject *ports_kobj;
2532 	struct list_head port_list;
2533 	struct ib_device *owner; /* reach back to owner ib_device */
2534 };
2535 
2536 struct rdma_restrack_root;
2537 struct ib_device {
2538 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2539 	struct device                *dma_device;
2540 	struct ib_device_ops	     ops;
2541 	char                          name[IB_DEVICE_NAME_MAX];
2542 	struct rcu_head rcu_head;
2543 
2544 	struct list_head              event_handler_list;
2545 	spinlock_t                    event_handler_lock;
2546 
2547 	struct rw_semaphore	      client_data_rwsem;
2548 	struct xarray                 client_data;
2549 	struct mutex                  unregistration_lock;
2550 
2551 	struct ib_cache               cache;
2552 	/**
2553 	 * port_data is indexed by port number
2554 	 */
2555 	struct ib_port_data *port_data;
2556 
2557 	int			      num_comp_vectors;
2558 
2559 	union {
2560 		struct device		dev;
2561 		struct ib_core_device	coredev;
2562 	};
2563 
2564 	/* First group for device attributes,
2565 	 * Second group for driver provided attributes (optional).
2566 	 * It is NULL terminated array.
2567 	 */
2568 	const struct attribute_group	*groups[3];
2569 
2570 	u64			     uverbs_cmd_mask;
2571 	u64			     uverbs_ex_cmd_mask;
2572 
2573 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2574 	__be64			     node_guid;
2575 	u32			     local_dma_lkey;
2576 	u16                          is_switch:1;
2577 	/* Indicates kernel verbs support, should not be used in drivers */
2578 	u16                          kverbs_provider:1;
2579 	u8                           node_type;
2580 	u8                           phys_port_cnt;
2581 	struct ib_device_attr        attrs;
2582 	struct attribute_group	     *hw_stats_ag;
2583 	struct rdma_hw_stats         *hw_stats;
2584 
2585 #ifdef CONFIG_CGROUP_RDMA
2586 	struct rdmacg_device         cg_device;
2587 #endif
2588 
2589 	u32                          index;
2590 	struct rdma_restrack_root *res;
2591 
2592 	const struct uapi_definition   *driver_def;
2593 
2594 	/*
2595 	 * Positive refcount indicates that the device is currently
2596 	 * registered and cannot be unregistered.
2597 	 */
2598 	refcount_t refcount;
2599 	struct completion unreg_completion;
2600 	struct work_struct unregistration_work;
2601 
2602 	const struct rdma_link_ops *link_ops;
2603 
2604 	/* Protects compat_devs xarray modifications */
2605 	struct mutex compat_devs_mutex;
2606 	/* Maintains compat devices for each net namespace */
2607 	struct xarray compat_devs;
2608 
2609 	/* Used by iWarp CM */
2610 	char iw_ifname[IFNAMSIZ];
2611 	u32 iw_driver_flags;
2612 };
2613 
2614 struct ib_client_nl_info;
2615 struct ib_client {
2616 	const char *name;
2617 	void (*add)   (struct ib_device *);
2618 	void (*remove)(struct ib_device *, void *client_data);
2619 	void (*rename)(struct ib_device *dev, void *client_data);
2620 	int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2621 			   struct ib_client_nl_info *res);
2622 	int (*get_global_nl_info)(struct ib_client_nl_info *res);
2623 
2624 	/* Returns the net_dev belonging to this ib_client and matching the
2625 	 * given parameters.
2626 	 * @dev:	 An RDMA device that the net_dev use for communication.
2627 	 * @port:	 A physical port number on the RDMA device.
2628 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2629 	 * @gid:	 A GID that the net_dev uses to communicate.
2630 	 * @addr:	 An IP address the net_dev is configured with.
2631 	 * @client_data: The device's client data set by ib_set_client_data().
2632 	 *
2633 	 * An ib_client that implements a net_dev on top of RDMA devices
2634 	 * (such as IP over IB) should implement this callback, allowing the
2635 	 * rdma_cm module to find the right net_dev for a given request.
2636 	 *
2637 	 * The caller is responsible for calling dev_put on the returned
2638 	 * netdev. */
2639 	struct net_device *(*get_net_dev_by_params)(
2640 			struct ib_device *dev,
2641 			u8 port,
2642 			u16 pkey,
2643 			const union ib_gid *gid,
2644 			const struct sockaddr *addr,
2645 			void *client_data);
2646 	struct list_head list;
2647 	u32 client_id;
2648 
2649 	/* kverbs are not required by the client */
2650 	u8 no_kverbs_req:1;
2651 };
2652 
2653 /*
2654  * IB block DMA iterator
2655  *
2656  * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2657  * to a HW supported page size.
2658  */
2659 struct ib_block_iter {
2660 	/* internal states */
2661 	struct scatterlist *__sg;	/* sg holding the current aligned block */
2662 	dma_addr_t __dma_addr;		/* unaligned DMA address of this block */
2663 	unsigned int __sg_nents;	/* number of SG entries */
2664 	unsigned int __sg_advance;	/* number of bytes to advance in sg in next step */
2665 	unsigned int __pg_bit;		/* alignment of current block */
2666 };
2667 
2668 struct ib_device *_ib_alloc_device(size_t size);
2669 #define ib_alloc_device(drv_struct, member)                                    \
2670 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2671 				      BUILD_BUG_ON_ZERO(offsetof(              \
2672 					      struct drv_struct, member))),    \
2673 		     struct drv_struct, member)
2674 
2675 void ib_dealloc_device(struct ib_device *device);
2676 
2677 void ib_get_device_fw_str(struct ib_device *device, char *str);
2678 
2679 int ib_register_device(struct ib_device *device, const char *name);
2680 void ib_unregister_device(struct ib_device *device);
2681 void ib_unregister_driver(enum rdma_driver_id driver_id);
2682 void ib_unregister_device_and_put(struct ib_device *device);
2683 void ib_unregister_device_queued(struct ib_device *ib_dev);
2684 
2685 int ib_register_client   (struct ib_client *client);
2686 void ib_unregister_client(struct ib_client *client);
2687 
2688 void __rdma_block_iter_start(struct ib_block_iter *biter,
2689 			     struct scatterlist *sglist,
2690 			     unsigned int nents,
2691 			     unsigned long pgsz);
2692 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2693 
2694 /**
2695  * rdma_block_iter_dma_address - get the aligned dma address of the current
2696  * block held by the block iterator.
2697  * @biter: block iterator holding the memory block
2698  */
2699 static inline dma_addr_t
2700 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2701 {
2702 	return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2703 }
2704 
2705 /**
2706  * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2707  * @sglist: sglist to iterate over
2708  * @biter: block iterator holding the memory block
2709  * @nents: maximum number of sg entries to iterate over
2710  * @pgsz: best HW supported page size to use
2711  *
2712  * Callers may use rdma_block_iter_dma_address() to get each
2713  * blocks aligned DMA address.
2714  */
2715 #define rdma_for_each_block(sglist, biter, nents, pgsz)		\
2716 	for (__rdma_block_iter_start(biter, sglist, nents,	\
2717 				     pgsz);			\
2718 	     __rdma_block_iter_next(biter);)
2719 
2720 /**
2721  * ib_get_client_data - Get IB client context
2722  * @device:Device to get context for
2723  * @client:Client to get context for
2724  *
2725  * ib_get_client_data() returns the client context data set with
2726  * ib_set_client_data(). This can only be called while the client is
2727  * registered to the device, once the ib_client remove() callback returns this
2728  * cannot be called.
2729  */
2730 static inline void *ib_get_client_data(struct ib_device *device,
2731 				       struct ib_client *client)
2732 {
2733 	return xa_load(&device->client_data, client->client_id);
2734 }
2735 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2736 			 void *data);
2737 void ib_set_device_ops(struct ib_device *device,
2738 		       const struct ib_device_ops *ops);
2739 
2740 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2741 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2742 		      unsigned long pfn, unsigned long size, pgprot_t prot);
2743 #else
2744 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
2745 				    struct vm_area_struct *vma,
2746 				    unsigned long pfn, unsigned long size,
2747 				    pgprot_t prot)
2748 {
2749 	return -EINVAL;
2750 }
2751 #endif
2752 
2753 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2754 {
2755 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2756 }
2757 
2758 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2759 {
2760 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2761 }
2762 
2763 static inline bool ib_is_buffer_cleared(const void __user *p,
2764 					size_t len)
2765 {
2766 	bool ret;
2767 	u8 *buf;
2768 
2769 	if (len > USHRT_MAX)
2770 		return false;
2771 
2772 	buf = memdup_user(p, len);
2773 	if (IS_ERR(buf))
2774 		return false;
2775 
2776 	ret = !memchr_inv(buf, 0, len);
2777 	kfree(buf);
2778 	return ret;
2779 }
2780 
2781 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2782 				       size_t offset,
2783 				       size_t len)
2784 {
2785 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
2786 }
2787 
2788 /**
2789  * ib_is_destroy_retryable - Check whether the uobject destruction
2790  * is retryable.
2791  * @ret: The initial destruction return code
2792  * @why: remove reason
2793  * @uobj: The uobject that is destroyed
2794  *
2795  * This function is a helper function that IB layer and low-level drivers
2796  * can use to consider whether the destruction of the given uobject is
2797  * retry-able.
2798  * It checks the original return code, if it wasn't success the destruction
2799  * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2800  * the remove reason. (i.e. why).
2801  * Must be called with the object locked for destroy.
2802  */
2803 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2804 					   struct ib_uobject *uobj)
2805 {
2806 	return ret && (why == RDMA_REMOVE_DESTROY ||
2807 		       uobj->context->cleanup_retryable);
2808 }
2809 
2810 /**
2811  * ib_destroy_usecnt - Called during destruction to check the usecnt
2812  * @usecnt: The usecnt atomic
2813  * @why: remove reason
2814  * @uobj: The uobject that is destroyed
2815  *
2816  * Non-zero usecnts will block destruction unless destruction was triggered by
2817  * a ucontext cleanup.
2818  */
2819 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2820 				    enum rdma_remove_reason why,
2821 				    struct ib_uobject *uobj)
2822 {
2823 	if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2824 		return -EBUSY;
2825 	return 0;
2826 }
2827 
2828 /**
2829  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2830  * contains all required attributes and no attributes not allowed for
2831  * the given QP state transition.
2832  * @cur_state: Current QP state
2833  * @next_state: Next QP state
2834  * @type: QP type
2835  * @mask: Mask of supplied QP attributes
2836  *
2837  * This function is a helper function that a low-level driver's
2838  * modify_qp method can use to validate the consumer's input.  It
2839  * checks that cur_state and next_state are valid QP states, that a
2840  * transition from cur_state to next_state is allowed by the IB spec,
2841  * and that the attribute mask supplied is allowed for the transition.
2842  */
2843 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2844 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
2845 
2846 void ib_register_event_handler(struct ib_event_handler *event_handler);
2847 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2848 void ib_dispatch_event(struct ib_event *event);
2849 
2850 int ib_query_port(struct ib_device *device,
2851 		  u8 port_num, struct ib_port_attr *port_attr);
2852 
2853 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2854 					       u8 port_num);
2855 
2856 /**
2857  * rdma_cap_ib_switch - Check if the device is IB switch
2858  * @device: Device to check
2859  *
2860  * Device driver is responsible for setting is_switch bit on
2861  * in ib_device structure at init time.
2862  *
2863  * Return: true if the device is IB switch.
2864  */
2865 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2866 {
2867 	return device->is_switch;
2868 }
2869 
2870 /**
2871  * rdma_start_port - Return the first valid port number for the device
2872  * specified
2873  *
2874  * @device: Device to be checked
2875  *
2876  * Return start port number
2877  */
2878 static inline u8 rdma_start_port(const struct ib_device *device)
2879 {
2880 	return rdma_cap_ib_switch(device) ? 0 : 1;
2881 }
2882 
2883 /**
2884  * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2885  * @device - The struct ib_device * to iterate over
2886  * @iter - The unsigned int to store the port number
2887  */
2888 #define rdma_for_each_port(device, iter)                                       \
2889 	for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type(   \
2890 						     unsigned int, iter)));    \
2891 	     iter <= rdma_end_port(device); (iter)++)
2892 
2893 /**
2894  * rdma_end_port - Return the last valid port number for the device
2895  * specified
2896  *
2897  * @device: Device to be checked
2898  *
2899  * Return last port number
2900  */
2901 static inline u8 rdma_end_port(const struct ib_device *device)
2902 {
2903 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2904 }
2905 
2906 static inline int rdma_is_port_valid(const struct ib_device *device,
2907 				     unsigned int port)
2908 {
2909 	return (port >= rdma_start_port(device) &&
2910 		port <= rdma_end_port(device));
2911 }
2912 
2913 static inline bool rdma_is_grh_required(const struct ib_device *device,
2914 					u8 port_num)
2915 {
2916 	return device->port_data[port_num].immutable.core_cap_flags &
2917 	       RDMA_CORE_PORT_IB_GRH_REQUIRED;
2918 }
2919 
2920 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2921 {
2922 	return device->port_data[port_num].immutable.core_cap_flags &
2923 	       RDMA_CORE_CAP_PROT_IB;
2924 }
2925 
2926 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2927 {
2928 	return device->port_data[port_num].immutable.core_cap_flags &
2929 	       (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2930 }
2931 
2932 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2933 {
2934 	return device->port_data[port_num].immutable.core_cap_flags &
2935 	       RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2936 }
2937 
2938 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2939 {
2940 	return device->port_data[port_num].immutable.core_cap_flags &
2941 	       RDMA_CORE_CAP_PROT_ROCE;
2942 }
2943 
2944 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2945 {
2946 	return device->port_data[port_num].immutable.core_cap_flags &
2947 	       RDMA_CORE_CAP_PROT_IWARP;
2948 }
2949 
2950 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2951 {
2952 	return rdma_protocol_ib(device, port_num) ||
2953 		rdma_protocol_roce(device, port_num);
2954 }
2955 
2956 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2957 {
2958 	return device->port_data[port_num].immutable.core_cap_flags &
2959 	       RDMA_CORE_CAP_PROT_RAW_PACKET;
2960 }
2961 
2962 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2963 {
2964 	return device->port_data[port_num].immutable.core_cap_flags &
2965 	       RDMA_CORE_CAP_PROT_USNIC;
2966 }
2967 
2968 /**
2969  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2970  * Management Datagrams.
2971  * @device: Device to check
2972  * @port_num: Port number to check
2973  *
2974  * Management Datagrams (MAD) are a required part of the InfiniBand
2975  * specification and are supported on all InfiniBand devices.  A slightly
2976  * extended version are also supported on OPA interfaces.
2977  *
2978  * Return: true if the port supports sending/receiving of MAD packets.
2979  */
2980 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2981 {
2982 	return device->port_data[port_num].immutable.core_cap_flags &
2983 	       RDMA_CORE_CAP_IB_MAD;
2984 }
2985 
2986 /**
2987  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2988  * Management Datagrams.
2989  * @device: Device to check
2990  * @port_num: Port number to check
2991  *
2992  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2993  * datagrams with their own versions.  These OPA MADs share many but not all of
2994  * the characteristics of InfiniBand MADs.
2995  *
2996  * OPA MADs differ in the following ways:
2997  *
2998  *    1) MADs are variable size up to 2K
2999  *       IBTA defined MADs remain fixed at 256 bytes
3000  *    2) OPA SMPs must carry valid PKeys
3001  *    3) OPA SMP packets are a different format
3002  *
3003  * Return: true if the port supports OPA MAD packet formats.
3004  */
3005 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3006 {
3007 	return device->port_data[port_num].immutable.core_cap_flags &
3008 		RDMA_CORE_CAP_OPA_MAD;
3009 }
3010 
3011 /**
3012  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3013  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3014  * @device: Device to check
3015  * @port_num: Port number to check
3016  *
3017  * Each InfiniBand node is required to provide a Subnet Management Agent
3018  * that the subnet manager can access.  Prior to the fabric being fully
3019  * configured by the subnet manager, the SMA is accessed via a well known
3020  * interface called the Subnet Management Interface (SMI).  This interface
3021  * uses directed route packets to communicate with the SM to get around the
3022  * chicken and egg problem of the SM needing to know what's on the fabric
3023  * in order to configure the fabric, and needing to configure the fabric in
3024  * order to send packets to the devices on the fabric.  These directed
3025  * route packets do not need the fabric fully configured in order to reach
3026  * their destination.  The SMI is the only method allowed to send
3027  * directed route packets on an InfiniBand fabric.
3028  *
3029  * Return: true if the port provides an SMI.
3030  */
3031 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3032 {
3033 	return device->port_data[port_num].immutable.core_cap_flags &
3034 	       RDMA_CORE_CAP_IB_SMI;
3035 }
3036 
3037 /**
3038  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3039  * Communication Manager.
3040  * @device: Device to check
3041  * @port_num: Port number to check
3042  *
3043  * The InfiniBand Communication Manager is one of many pre-defined General
3044  * Service Agents (GSA) that are accessed via the General Service
3045  * Interface (GSI).  It's role is to facilitate establishment of connections
3046  * between nodes as well as other management related tasks for established
3047  * connections.
3048  *
3049  * Return: true if the port supports an IB CM (this does not guarantee that
3050  * a CM is actually running however).
3051  */
3052 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3053 {
3054 	return device->port_data[port_num].immutable.core_cap_flags &
3055 	       RDMA_CORE_CAP_IB_CM;
3056 }
3057 
3058 /**
3059  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3060  * Communication Manager.
3061  * @device: Device to check
3062  * @port_num: Port number to check
3063  *
3064  * Similar to above, but specific to iWARP connections which have a different
3065  * managment protocol than InfiniBand.
3066  *
3067  * Return: true if the port supports an iWARP CM (this does not guarantee that
3068  * a CM is actually running however).
3069  */
3070 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3071 {
3072 	return device->port_data[port_num].immutable.core_cap_flags &
3073 	       RDMA_CORE_CAP_IW_CM;
3074 }
3075 
3076 /**
3077  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3078  * Subnet Administration.
3079  * @device: Device to check
3080  * @port_num: Port number to check
3081  *
3082  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3083  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
3084  * fabrics, devices should resolve routes to other hosts by contacting the
3085  * SA to query the proper route.
3086  *
3087  * Return: true if the port should act as a client to the fabric Subnet
3088  * Administration interface.  This does not imply that the SA service is
3089  * running locally.
3090  */
3091 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3092 {
3093 	return device->port_data[port_num].immutable.core_cap_flags &
3094 	       RDMA_CORE_CAP_IB_SA;
3095 }
3096 
3097 /**
3098  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3099  * Multicast.
3100  * @device: Device to check
3101  * @port_num: Port number to check
3102  *
3103  * InfiniBand multicast registration is more complex than normal IPv4 or
3104  * IPv6 multicast registration.  Each Host Channel Adapter must register
3105  * with the Subnet Manager when it wishes to join a multicast group.  It
3106  * should do so only once regardless of how many queue pairs it subscribes
3107  * to this group.  And it should leave the group only after all queue pairs
3108  * attached to the group have been detached.
3109  *
3110  * Return: true if the port must undertake the additional adminstrative
3111  * overhead of registering/unregistering with the SM and tracking of the
3112  * total number of queue pairs attached to the multicast group.
3113  */
3114 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3115 {
3116 	return rdma_cap_ib_sa(device, port_num);
3117 }
3118 
3119 /**
3120  * rdma_cap_af_ib - Check if the port of device has the capability
3121  * Native Infiniband Address.
3122  * @device: Device to check
3123  * @port_num: Port number to check
3124  *
3125  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3126  * GID.  RoCE uses a different mechanism, but still generates a GID via
3127  * a prescribed mechanism and port specific data.
3128  *
3129  * Return: true if the port uses a GID address to identify devices on the
3130  * network.
3131  */
3132 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3133 {
3134 	return device->port_data[port_num].immutable.core_cap_flags &
3135 	       RDMA_CORE_CAP_AF_IB;
3136 }
3137 
3138 /**
3139  * rdma_cap_eth_ah - Check if the port of device has the capability
3140  * Ethernet Address Handle.
3141  * @device: Device to check
3142  * @port_num: Port number to check
3143  *
3144  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3145  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3146  * port.  Normally, packet headers are generated by the sending host
3147  * adapter, but when sending connectionless datagrams, we must manually
3148  * inject the proper headers for the fabric we are communicating over.
3149  *
3150  * Return: true if we are running as a RoCE port and must force the
3151  * addition of a Global Route Header built from our Ethernet Address
3152  * Handle into our header list for connectionless packets.
3153  */
3154 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3155 {
3156 	return device->port_data[port_num].immutable.core_cap_flags &
3157 	       RDMA_CORE_CAP_ETH_AH;
3158 }
3159 
3160 /**
3161  * rdma_cap_opa_ah - Check if the port of device supports
3162  * OPA Address handles
3163  * @device: Device to check
3164  * @port_num: Port number to check
3165  *
3166  * Return: true if we are running on an OPA device which supports
3167  * the extended OPA addressing.
3168  */
3169 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3170 {
3171 	return (device->port_data[port_num].immutable.core_cap_flags &
3172 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3173 }
3174 
3175 /**
3176  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3177  *
3178  * @device: Device
3179  * @port_num: Port number
3180  *
3181  * This MAD size includes the MAD headers and MAD payload.  No other headers
3182  * are included.
3183  *
3184  * Return the max MAD size required by the Port.  Will return 0 if the port
3185  * does not support MADs
3186  */
3187 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3188 {
3189 	return device->port_data[port_num].immutable.max_mad_size;
3190 }
3191 
3192 /**
3193  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3194  * @device: Device to check
3195  * @port_num: Port number to check
3196  *
3197  * RoCE GID table mechanism manages the various GIDs for a device.
3198  *
3199  * NOTE: if allocating the port's GID table has failed, this call will still
3200  * return true, but any RoCE GID table API will fail.
3201  *
3202  * Return: true if the port uses RoCE GID table mechanism in order to manage
3203  * its GIDs.
3204  */
3205 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3206 					   u8 port_num)
3207 {
3208 	return rdma_protocol_roce(device, port_num) &&
3209 		device->ops.add_gid && device->ops.del_gid;
3210 }
3211 
3212 /*
3213  * Check if the device supports READ W/ INVALIDATE.
3214  */
3215 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3216 {
3217 	/*
3218 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3219 	 * has support for it yet.
3220 	 */
3221 	return rdma_protocol_iwarp(dev, port_num);
3222 }
3223 
3224 /**
3225  * rdma_find_pg_bit - Find page bit given address and HW supported page sizes
3226  *
3227  * @addr: address
3228  * @pgsz_bitmap: bitmap of HW supported page sizes
3229  */
3230 static inline unsigned int rdma_find_pg_bit(unsigned long addr,
3231 					    unsigned long pgsz_bitmap)
3232 {
3233 	unsigned long align;
3234 	unsigned long pgsz;
3235 
3236 	align = addr & -addr;
3237 
3238 	/* Find page bit such that addr is aligned to the highest supported
3239 	 * HW page size
3240 	 */
3241 	pgsz = pgsz_bitmap & ~(-align << 1);
3242 	if (!pgsz)
3243 		return __ffs(pgsz_bitmap);
3244 
3245 	return __fls(pgsz);
3246 }
3247 
3248 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3249 			 int state);
3250 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3251 		     struct ifla_vf_info *info);
3252 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3253 		    struct ifla_vf_stats *stats);
3254 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3255 		   int type);
3256 
3257 int ib_query_pkey(struct ib_device *device,
3258 		  u8 port_num, u16 index, u16 *pkey);
3259 
3260 int ib_modify_device(struct ib_device *device,
3261 		     int device_modify_mask,
3262 		     struct ib_device_modify *device_modify);
3263 
3264 int ib_modify_port(struct ib_device *device,
3265 		   u8 port_num, int port_modify_mask,
3266 		   struct ib_port_modify *port_modify);
3267 
3268 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3269 		u8 *port_num, u16 *index);
3270 
3271 int ib_find_pkey(struct ib_device *device,
3272 		 u8 port_num, u16 pkey, u16 *index);
3273 
3274 enum ib_pd_flags {
3275 	/*
3276 	 * Create a memory registration for all memory in the system and place
3277 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3278 	 * ULPs to avoid the overhead of dynamic MRs.
3279 	 *
3280 	 * This flag is generally considered unsafe and must only be used in
3281 	 * extremly trusted environments.  Every use of it will log a warning
3282 	 * in the kernel log.
3283 	 */
3284 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3285 };
3286 
3287 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3288 		const char *caller);
3289 
3290 #define ib_alloc_pd(device, flags) \
3291 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3292 
3293 /**
3294  * ib_dealloc_pd_user - Deallocate kernel/user PD
3295  * @pd: The protection domain
3296  * @udata: Valid user data or NULL for kernel objects
3297  */
3298 void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3299 
3300 /**
3301  * ib_dealloc_pd - Deallocate kernel PD
3302  * @pd: The protection domain
3303  *
3304  * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3305  */
3306 static inline void ib_dealloc_pd(struct ib_pd *pd)
3307 {
3308 	ib_dealloc_pd_user(pd, NULL);
3309 }
3310 
3311 enum rdma_create_ah_flags {
3312 	/* In a sleepable context */
3313 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3314 };
3315 
3316 /**
3317  * rdma_create_ah - Creates an address handle for the given address vector.
3318  * @pd: The protection domain associated with the address handle.
3319  * @ah_attr: The attributes of the address vector.
3320  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3321  *
3322  * The address handle is used to reference a local or global destination
3323  * in all UD QP post sends.
3324  */
3325 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3326 			     u32 flags);
3327 
3328 /**
3329  * rdma_create_user_ah - Creates an address handle for the given address vector.
3330  * It resolves destination mac address for ah attribute of RoCE type.
3331  * @pd: The protection domain associated with the address handle.
3332  * @ah_attr: The attributes of the address vector.
3333  * @udata: pointer to user's input output buffer information need by
3334  *         provider driver.
3335  *
3336  * It returns 0 on success and returns appropriate error code on error.
3337  * The address handle is used to reference a local or global destination
3338  * in all UD QP post sends.
3339  */
3340 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3341 				  struct rdma_ah_attr *ah_attr,
3342 				  struct ib_udata *udata);
3343 /**
3344  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3345  *   work completion.
3346  * @hdr: the L3 header to parse
3347  * @net_type: type of header to parse
3348  * @sgid: place to store source gid
3349  * @dgid: place to store destination gid
3350  */
3351 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3352 			      enum rdma_network_type net_type,
3353 			      union ib_gid *sgid, union ib_gid *dgid);
3354 
3355 /**
3356  * ib_get_rdma_header_version - Get the header version
3357  * @hdr: the L3 header to parse
3358  */
3359 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3360 
3361 /**
3362  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3363  *   work completion.
3364  * @device: Device on which the received message arrived.
3365  * @port_num: Port on which the received message arrived.
3366  * @wc: Work completion associated with the received message.
3367  * @grh: References the received global route header.  This parameter is
3368  *   ignored unless the work completion indicates that the GRH is valid.
3369  * @ah_attr: Returned attributes that can be used when creating an address
3370  *   handle for replying to the message.
3371  * When ib_init_ah_attr_from_wc() returns success,
3372  * (a) for IB link layer it optionally contains a reference to SGID attribute
3373  * when GRH is present for IB link layer.
3374  * (b) for RoCE link layer it contains a reference to SGID attribute.
3375  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3376  * attributes which are initialized using ib_init_ah_attr_from_wc().
3377  *
3378  */
3379 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3380 			    const struct ib_wc *wc, const struct ib_grh *grh,
3381 			    struct rdma_ah_attr *ah_attr);
3382 
3383 /**
3384  * ib_create_ah_from_wc - Creates an address handle associated with the
3385  *   sender of the specified work completion.
3386  * @pd: The protection domain associated with the address handle.
3387  * @wc: Work completion information associated with a received message.
3388  * @grh: References the received global route header.  This parameter is
3389  *   ignored unless the work completion indicates that the GRH is valid.
3390  * @port_num: The outbound port number to associate with the address.
3391  *
3392  * The address handle is used to reference a local or global destination
3393  * in all UD QP post sends.
3394  */
3395 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3396 				   const struct ib_grh *grh, u8 port_num);
3397 
3398 /**
3399  * rdma_modify_ah - Modifies the address vector associated with an address
3400  *   handle.
3401  * @ah: The address handle to modify.
3402  * @ah_attr: The new address vector attributes to associate with the
3403  *   address handle.
3404  */
3405 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3406 
3407 /**
3408  * rdma_query_ah - Queries the address vector associated with an address
3409  *   handle.
3410  * @ah: The address handle to query.
3411  * @ah_attr: The address vector attributes associated with the address
3412  *   handle.
3413  */
3414 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3415 
3416 enum rdma_destroy_ah_flags {
3417 	/* In a sleepable context */
3418 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3419 };
3420 
3421 /**
3422  * rdma_destroy_ah_user - Destroys an address handle.
3423  * @ah: The address handle to destroy.
3424  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3425  * @udata: Valid user data or NULL for kernel objects
3426  */
3427 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3428 
3429 /**
3430  * rdma_destroy_ah - Destroys an kernel address handle.
3431  * @ah: The address handle to destroy.
3432  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3433  *
3434  * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3435  */
3436 static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3437 {
3438 	return rdma_destroy_ah_user(ah, flags, NULL);
3439 }
3440 
3441 /**
3442  * ib_create_srq - Creates a SRQ associated with the specified protection
3443  *   domain.
3444  * @pd: The protection domain associated with the SRQ.
3445  * @srq_init_attr: A list of initial attributes required to create the
3446  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
3447  *   the actual capabilities of the created SRQ.
3448  *
3449  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3450  * requested size of the SRQ, and set to the actual values allocated
3451  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
3452  * will always be at least as large as the requested values.
3453  */
3454 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3455 			     struct ib_srq_init_attr *srq_init_attr);
3456 
3457 /**
3458  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3459  * @srq: The SRQ to modify.
3460  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3461  *   the current values of selected SRQ attributes are returned.
3462  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3463  *   are being modified.
3464  *
3465  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3466  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3467  * the number of receives queued drops below the limit.
3468  */
3469 int ib_modify_srq(struct ib_srq *srq,
3470 		  struct ib_srq_attr *srq_attr,
3471 		  enum ib_srq_attr_mask srq_attr_mask);
3472 
3473 /**
3474  * ib_query_srq - Returns the attribute list and current values for the
3475  *   specified SRQ.
3476  * @srq: The SRQ to query.
3477  * @srq_attr: The attributes of the specified SRQ.
3478  */
3479 int ib_query_srq(struct ib_srq *srq,
3480 		 struct ib_srq_attr *srq_attr);
3481 
3482 /**
3483  * ib_destroy_srq_user - Destroys the specified SRQ.
3484  * @srq: The SRQ to destroy.
3485  * @udata: Valid user data or NULL for kernel objects
3486  */
3487 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3488 
3489 /**
3490  * ib_destroy_srq - Destroys the specified kernel SRQ.
3491  * @srq: The SRQ to destroy.
3492  *
3493  * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3494  */
3495 static inline int ib_destroy_srq(struct ib_srq *srq)
3496 {
3497 	return ib_destroy_srq_user(srq, NULL);
3498 }
3499 
3500 /**
3501  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3502  * @srq: The SRQ to post the work request on.
3503  * @recv_wr: A list of work requests to post on the receive queue.
3504  * @bad_recv_wr: On an immediate failure, this parameter will reference
3505  *   the work request that failed to be posted on the QP.
3506  */
3507 static inline int ib_post_srq_recv(struct ib_srq *srq,
3508 				   const struct ib_recv_wr *recv_wr,
3509 				   const struct ib_recv_wr **bad_recv_wr)
3510 {
3511 	const struct ib_recv_wr *dummy;
3512 
3513 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3514 					      bad_recv_wr ? : &dummy);
3515 }
3516 
3517 /**
3518  * ib_create_qp_user - Creates a QP associated with the specified protection
3519  *   domain.
3520  * @pd: The protection domain associated with the QP.
3521  * @qp_init_attr: A list of initial attributes required to create the
3522  *   QP.  If QP creation succeeds, then the attributes are updated to
3523  *   the actual capabilities of the created QP.
3524  * @udata: Valid user data or NULL for kernel objects
3525  */
3526 struct ib_qp *ib_create_qp_user(struct ib_pd *pd,
3527 				struct ib_qp_init_attr *qp_init_attr,
3528 				struct ib_udata *udata);
3529 
3530 /**
3531  * ib_create_qp - Creates a kernel QP associated with the specified protection
3532  *   domain.
3533  * @pd: The protection domain associated with the QP.
3534  * @qp_init_attr: A list of initial attributes required to create the
3535  *   QP.  If QP creation succeeds, then the attributes are updated to
3536  *   the actual capabilities of the created QP.
3537  * @udata: Valid user data or NULL for kernel objects
3538  *
3539  * NOTE: for user qp use ib_create_qp_user with valid udata!
3540  */
3541 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3542 					 struct ib_qp_init_attr *qp_init_attr)
3543 {
3544 	return ib_create_qp_user(pd, qp_init_attr, NULL);
3545 }
3546 
3547 /**
3548  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3549  * @qp: The QP to modify.
3550  * @attr: On input, specifies the QP attributes to modify.  On output,
3551  *   the current values of selected QP attributes are returned.
3552  * @attr_mask: A bit-mask used to specify which attributes of the QP
3553  *   are being modified.
3554  * @udata: pointer to user's input output buffer information
3555  *   are being modified.
3556  * It returns 0 on success and returns appropriate error code on error.
3557  */
3558 int ib_modify_qp_with_udata(struct ib_qp *qp,
3559 			    struct ib_qp_attr *attr,
3560 			    int attr_mask,
3561 			    struct ib_udata *udata);
3562 
3563 /**
3564  * ib_modify_qp - Modifies the attributes for the specified QP and then
3565  *   transitions the QP to the given state.
3566  * @qp: The QP to modify.
3567  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3568  *   the current values of selected QP attributes are returned.
3569  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3570  *   are being modified.
3571  */
3572 int ib_modify_qp(struct ib_qp *qp,
3573 		 struct ib_qp_attr *qp_attr,
3574 		 int qp_attr_mask);
3575 
3576 /**
3577  * ib_query_qp - Returns the attribute list and current values for the
3578  *   specified QP.
3579  * @qp: The QP to query.
3580  * @qp_attr: The attributes of the specified QP.
3581  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3582  * @qp_init_attr: Additional attributes of the selected QP.
3583  *
3584  * The qp_attr_mask may be used to limit the query to gathering only the
3585  * selected attributes.
3586  */
3587 int ib_query_qp(struct ib_qp *qp,
3588 		struct ib_qp_attr *qp_attr,
3589 		int qp_attr_mask,
3590 		struct ib_qp_init_attr *qp_init_attr);
3591 
3592 /**
3593  * ib_destroy_qp - Destroys the specified QP.
3594  * @qp: The QP to destroy.
3595  * @udata: Valid udata or NULL for kernel objects
3596  */
3597 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3598 
3599 /**
3600  * ib_destroy_qp - Destroys the specified kernel QP.
3601  * @qp: The QP to destroy.
3602  *
3603  * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3604  */
3605 static inline int ib_destroy_qp(struct ib_qp *qp)
3606 {
3607 	return ib_destroy_qp_user(qp, NULL);
3608 }
3609 
3610 /**
3611  * ib_open_qp - Obtain a reference to an existing sharable QP.
3612  * @xrcd - XRC domain
3613  * @qp_open_attr: Attributes identifying the QP to open.
3614  *
3615  * Returns a reference to a sharable QP.
3616  */
3617 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3618 			 struct ib_qp_open_attr *qp_open_attr);
3619 
3620 /**
3621  * ib_close_qp - Release an external reference to a QP.
3622  * @qp: The QP handle to release
3623  *
3624  * The opened QP handle is released by the caller.  The underlying
3625  * shared QP is not destroyed until all internal references are released.
3626  */
3627 int ib_close_qp(struct ib_qp *qp);
3628 
3629 /**
3630  * ib_post_send - Posts a list of work requests to the send queue of
3631  *   the specified QP.
3632  * @qp: The QP to post the work request on.
3633  * @send_wr: A list of work requests to post on the send queue.
3634  * @bad_send_wr: On an immediate failure, this parameter will reference
3635  *   the work request that failed to be posted on the QP.
3636  *
3637  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3638  * error is returned, the QP state shall not be affected,
3639  * ib_post_send() will return an immediate error after queueing any
3640  * earlier work requests in the list.
3641  */
3642 static inline int ib_post_send(struct ib_qp *qp,
3643 			       const struct ib_send_wr *send_wr,
3644 			       const struct ib_send_wr **bad_send_wr)
3645 {
3646 	const struct ib_send_wr *dummy;
3647 
3648 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3649 }
3650 
3651 /**
3652  * ib_post_recv - Posts a list of work requests to the receive queue of
3653  *   the specified QP.
3654  * @qp: The QP to post the work request on.
3655  * @recv_wr: A list of work requests to post on the receive queue.
3656  * @bad_recv_wr: On an immediate failure, this parameter will reference
3657  *   the work request that failed to be posted on the QP.
3658  */
3659 static inline int ib_post_recv(struct ib_qp *qp,
3660 			       const struct ib_recv_wr *recv_wr,
3661 			       const struct ib_recv_wr **bad_recv_wr)
3662 {
3663 	const struct ib_recv_wr *dummy;
3664 
3665 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3666 }
3667 
3668 struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
3669 				 int nr_cqe, int comp_vector,
3670 				 enum ib_poll_context poll_ctx,
3671 				 const char *caller, struct ib_udata *udata);
3672 
3673 /**
3674  * ib_alloc_cq_user: Allocate kernel/user CQ
3675  * @dev: The IB device
3676  * @private: Private data attached to the CQE
3677  * @nr_cqe: Number of CQEs in the CQ
3678  * @comp_vector: Completion vector used for the IRQs
3679  * @poll_ctx: Context used for polling the CQ
3680  * @udata: Valid user data or NULL for kernel objects
3681  */
3682 static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
3683 					     void *private, int nr_cqe,
3684 					     int comp_vector,
3685 					     enum ib_poll_context poll_ctx,
3686 					     struct ib_udata *udata)
3687 {
3688 	return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3689 				  KBUILD_MODNAME, udata);
3690 }
3691 
3692 /**
3693  * ib_alloc_cq: Allocate kernel CQ
3694  * @dev: The IB device
3695  * @private: Private data attached to the CQE
3696  * @nr_cqe: Number of CQEs in the CQ
3697  * @comp_vector: Completion vector used for the IRQs
3698  * @poll_ctx: Context used for polling the CQ
3699  *
3700  * NOTE: for user cq use ib_alloc_cq_user with valid udata!
3701  */
3702 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3703 					int nr_cqe, int comp_vector,
3704 					enum ib_poll_context poll_ctx)
3705 {
3706 	return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3707 				NULL);
3708 }
3709 
3710 /**
3711  * ib_free_cq_user - Free kernel/user CQ
3712  * @cq: The CQ to free
3713  * @udata: Valid user data or NULL for kernel objects
3714  */
3715 void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3716 
3717 /**
3718  * ib_free_cq - Free kernel CQ
3719  * @cq: The CQ to free
3720  *
3721  * NOTE: for user cq use ib_free_cq_user with valid udata!
3722  */
3723 static inline void ib_free_cq(struct ib_cq *cq)
3724 {
3725 	ib_free_cq_user(cq, NULL);
3726 }
3727 
3728 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3729 
3730 /**
3731  * ib_create_cq - Creates a CQ on the specified device.
3732  * @device: The device on which to create the CQ.
3733  * @comp_handler: A user-specified callback that is invoked when a
3734  *   completion event occurs on the CQ.
3735  * @event_handler: A user-specified callback that is invoked when an
3736  *   asynchronous event not associated with a completion occurs on the CQ.
3737  * @cq_context: Context associated with the CQ returned to the user via
3738  *   the associated completion and event handlers.
3739  * @cq_attr: The attributes the CQ should be created upon.
3740  *
3741  * Users can examine the cq structure to determine the actual CQ size.
3742  */
3743 struct ib_cq *__ib_create_cq(struct ib_device *device,
3744 			     ib_comp_handler comp_handler,
3745 			     void (*event_handler)(struct ib_event *, void *),
3746 			     void *cq_context,
3747 			     const struct ib_cq_init_attr *cq_attr,
3748 			     const char *caller);
3749 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3750 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3751 
3752 /**
3753  * ib_resize_cq - Modifies the capacity of the CQ.
3754  * @cq: The CQ to resize.
3755  * @cqe: The minimum size of the CQ.
3756  *
3757  * Users can examine the cq structure to determine the actual CQ size.
3758  */
3759 int ib_resize_cq(struct ib_cq *cq, int cqe);
3760 
3761 /**
3762  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3763  * @cq: The CQ to modify.
3764  * @cq_count: number of CQEs that will trigger an event
3765  * @cq_period: max period of time in usec before triggering an event
3766  *
3767  */
3768 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3769 
3770 /**
3771  * ib_destroy_cq_user - Destroys the specified CQ.
3772  * @cq: The CQ to destroy.
3773  * @udata: Valid user data or NULL for kernel objects
3774  */
3775 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3776 
3777 /**
3778  * ib_destroy_cq - Destroys the specified kernel CQ.
3779  * @cq: The CQ to destroy.
3780  *
3781  * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3782  */
3783 static inline void ib_destroy_cq(struct ib_cq *cq)
3784 {
3785 	ib_destroy_cq_user(cq, NULL);
3786 }
3787 
3788 /**
3789  * ib_poll_cq - poll a CQ for completion(s)
3790  * @cq:the CQ being polled
3791  * @num_entries:maximum number of completions to return
3792  * @wc:array of at least @num_entries &struct ib_wc where completions
3793  *   will be returned
3794  *
3795  * Poll a CQ for (possibly multiple) completions.  If the return value
3796  * is < 0, an error occurred.  If the return value is >= 0, it is the
3797  * number of completions returned.  If the return value is
3798  * non-negative and < num_entries, then the CQ was emptied.
3799  */
3800 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3801 			     struct ib_wc *wc)
3802 {
3803 	return cq->device->ops.poll_cq(cq, num_entries, wc);
3804 }
3805 
3806 /**
3807  * ib_req_notify_cq - Request completion notification on a CQ.
3808  * @cq: The CQ to generate an event for.
3809  * @flags:
3810  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3811  *   to request an event on the next solicited event or next work
3812  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3813  *   may also be |ed in to request a hint about missed events, as
3814  *   described below.
3815  *
3816  * Return Value:
3817  *    < 0 means an error occurred while requesting notification
3818  *   == 0 means notification was requested successfully, and if
3819  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3820  *        were missed and it is safe to wait for another event.  In
3821  *        this case is it guaranteed that any work completions added
3822  *        to the CQ since the last CQ poll will trigger a completion
3823  *        notification event.
3824  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3825  *        in.  It means that the consumer must poll the CQ again to
3826  *        make sure it is empty to avoid missing an event because of a
3827  *        race between requesting notification and an entry being
3828  *        added to the CQ.  This return value means it is possible
3829  *        (but not guaranteed) that a work completion has been added
3830  *        to the CQ since the last poll without triggering a
3831  *        completion notification event.
3832  */
3833 static inline int ib_req_notify_cq(struct ib_cq *cq,
3834 				   enum ib_cq_notify_flags flags)
3835 {
3836 	return cq->device->ops.req_notify_cq(cq, flags);
3837 }
3838 
3839 /**
3840  * ib_req_ncomp_notif - Request completion notification when there are
3841  *   at least the specified number of unreaped completions on the CQ.
3842  * @cq: The CQ to generate an event for.
3843  * @wc_cnt: The number of unreaped completions that should be on the
3844  *   CQ before an event is generated.
3845  */
3846 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3847 {
3848 	return cq->device->ops.req_ncomp_notif ?
3849 		cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3850 		-ENOSYS;
3851 }
3852 
3853 /**
3854  * ib_dma_mapping_error - check a DMA addr for error
3855  * @dev: The device for which the dma_addr was created
3856  * @dma_addr: The DMA address to check
3857  */
3858 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3859 {
3860 	return dma_mapping_error(dev->dma_device, dma_addr);
3861 }
3862 
3863 /**
3864  * ib_dma_map_single - Map a kernel virtual address to DMA address
3865  * @dev: The device for which the dma_addr is to be created
3866  * @cpu_addr: The kernel virtual address
3867  * @size: The size of the region in bytes
3868  * @direction: The direction of the DMA
3869  */
3870 static inline u64 ib_dma_map_single(struct ib_device *dev,
3871 				    void *cpu_addr, size_t size,
3872 				    enum dma_data_direction direction)
3873 {
3874 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3875 }
3876 
3877 /**
3878  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3879  * @dev: The device for which the DMA address was created
3880  * @addr: The DMA address
3881  * @size: The size of the region in bytes
3882  * @direction: The direction of the DMA
3883  */
3884 static inline void ib_dma_unmap_single(struct ib_device *dev,
3885 				       u64 addr, size_t size,
3886 				       enum dma_data_direction direction)
3887 {
3888 	dma_unmap_single(dev->dma_device, addr, size, direction);
3889 }
3890 
3891 /**
3892  * ib_dma_map_page - Map a physical page to DMA address
3893  * @dev: The device for which the dma_addr is to be created
3894  * @page: The page to be mapped
3895  * @offset: The offset within the page
3896  * @size: The size of the region in bytes
3897  * @direction: The direction of the DMA
3898  */
3899 static inline u64 ib_dma_map_page(struct ib_device *dev,
3900 				  struct page *page,
3901 				  unsigned long offset,
3902 				  size_t size,
3903 					 enum dma_data_direction direction)
3904 {
3905 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3906 }
3907 
3908 /**
3909  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3910  * @dev: The device for which the DMA address was created
3911  * @addr: The DMA address
3912  * @size: The size of the region in bytes
3913  * @direction: The direction of the DMA
3914  */
3915 static inline void ib_dma_unmap_page(struct ib_device *dev,
3916 				     u64 addr, size_t size,
3917 				     enum dma_data_direction direction)
3918 {
3919 	dma_unmap_page(dev->dma_device, addr, size, direction);
3920 }
3921 
3922 /**
3923  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3924  * @dev: The device for which the DMA addresses are to be created
3925  * @sg: The array of scatter/gather entries
3926  * @nents: The number of scatter/gather entries
3927  * @direction: The direction of the DMA
3928  */
3929 static inline int ib_dma_map_sg(struct ib_device *dev,
3930 				struct scatterlist *sg, int nents,
3931 				enum dma_data_direction direction)
3932 {
3933 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3934 }
3935 
3936 /**
3937  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3938  * @dev: The device for which the DMA addresses were created
3939  * @sg: The array of scatter/gather entries
3940  * @nents: The number of scatter/gather entries
3941  * @direction: The direction of the DMA
3942  */
3943 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3944 				   struct scatterlist *sg, int nents,
3945 				   enum dma_data_direction direction)
3946 {
3947 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
3948 }
3949 
3950 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3951 				      struct scatterlist *sg, int nents,
3952 				      enum dma_data_direction direction,
3953 				      unsigned long dma_attrs)
3954 {
3955 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3956 				dma_attrs);
3957 }
3958 
3959 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3960 					 struct scatterlist *sg, int nents,
3961 					 enum dma_data_direction direction,
3962 					 unsigned long dma_attrs)
3963 {
3964 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3965 }
3966 
3967 /**
3968  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
3969  * @dev: The device to query
3970  *
3971  * The returned value represents a size in bytes.
3972  */
3973 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
3974 {
3975 	struct device_dma_parameters *p = dev->dma_device->dma_parms;
3976 
3977 	return p ? p->max_segment_size : UINT_MAX;
3978 }
3979 
3980 /**
3981  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3982  * @dev: The device for which the DMA address was created
3983  * @addr: The DMA address
3984  * @size: The size of the region in bytes
3985  * @dir: The direction of the DMA
3986  */
3987 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3988 					      u64 addr,
3989 					      size_t size,
3990 					      enum dma_data_direction dir)
3991 {
3992 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3993 }
3994 
3995 /**
3996  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3997  * @dev: The device for which the DMA address was created
3998  * @addr: The DMA address
3999  * @size: The size of the region in bytes
4000  * @dir: The direction of the DMA
4001  */
4002 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4003 						 u64 addr,
4004 						 size_t size,
4005 						 enum dma_data_direction dir)
4006 {
4007 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4008 }
4009 
4010 /**
4011  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4012  * @dev: The device for which the DMA address is requested
4013  * @size: The size of the region to allocate in bytes
4014  * @dma_handle: A pointer for returning the DMA address of the region
4015  * @flag: memory allocator flags
4016  */
4017 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4018 					   size_t size,
4019 					   dma_addr_t *dma_handle,
4020 					   gfp_t flag)
4021 {
4022 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4023 }
4024 
4025 /**
4026  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4027  * @dev: The device for which the DMA addresses were allocated
4028  * @size: The size of the region
4029  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4030  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4031  */
4032 static inline void ib_dma_free_coherent(struct ib_device *dev,
4033 					size_t size, void *cpu_addr,
4034 					dma_addr_t dma_handle)
4035 {
4036 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4037 }
4038 
4039 /**
4040  * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4041  *   HCA translation table.
4042  * @mr: The memory region to deregister.
4043  * @udata: Valid user data or NULL for kernel object
4044  *
4045  * This function can fail, if the memory region has memory windows bound to it.
4046  */
4047 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4048 
4049 /**
4050  * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4051  *   HCA translation table.
4052  * @mr: The memory region to deregister.
4053  *
4054  * This function can fail, if the memory region has memory windows bound to it.
4055  *
4056  * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4057  */
4058 static inline int ib_dereg_mr(struct ib_mr *mr)
4059 {
4060 	return ib_dereg_mr_user(mr, NULL);
4061 }
4062 
4063 struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type,
4064 			       u32 max_num_sg, struct ib_udata *udata);
4065 
4066 static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
4067 					enum ib_mr_type mr_type, u32 max_num_sg)
4068 {
4069 	return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL);
4070 }
4071 
4072 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4073 				    u32 max_num_data_sg,
4074 				    u32 max_num_meta_sg);
4075 
4076 /**
4077  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4078  *   R_Key and L_Key.
4079  * @mr - struct ib_mr pointer to be updated.
4080  * @newkey - new key to be used.
4081  */
4082 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4083 {
4084 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4085 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4086 }
4087 
4088 /**
4089  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4090  * for calculating a new rkey for type 2 memory windows.
4091  * @rkey - the rkey to increment.
4092  */
4093 static inline u32 ib_inc_rkey(u32 rkey)
4094 {
4095 	const u32 mask = 0x000000ff;
4096 	return ((rkey + 1) & mask) | (rkey & ~mask);
4097 }
4098 
4099 /**
4100  * ib_alloc_fmr - Allocates a unmapped fast memory region.
4101  * @pd: The protection domain associated with the unmapped region.
4102  * @mr_access_flags: Specifies the memory access rights.
4103  * @fmr_attr: Attributes of the unmapped region.
4104  *
4105  * A fast memory region must be mapped before it can be used as part of
4106  * a work request.
4107  */
4108 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
4109 			    int mr_access_flags,
4110 			    struct ib_fmr_attr *fmr_attr);
4111 
4112 /**
4113  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
4114  * @fmr: The fast memory region to associate with the pages.
4115  * @page_list: An array of physical pages to map to the fast memory region.
4116  * @list_len: The number of pages in page_list.
4117  * @iova: The I/O virtual address to use with the mapped region.
4118  */
4119 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
4120 				  u64 *page_list, int list_len,
4121 				  u64 iova)
4122 {
4123 	return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
4124 }
4125 
4126 /**
4127  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
4128  * @fmr_list: A linked list of fast memory regions to unmap.
4129  */
4130 int ib_unmap_fmr(struct list_head *fmr_list);
4131 
4132 /**
4133  * ib_dealloc_fmr - Deallocates a fast memory region.
4134  * @fmr: The fast memory region to deallocate.
4135  */
4136 int ib_dealloc_fmr(struct ib_fmr *fmr);
4137 
4138 /**
4139  * ib_attach_mcast - Attaches the specified QP to a multicast group.
4140  * @qp: QP to attach to the multicast group.  The QP must be type
4141  *   IB_QPT_UD.
4142  * @gid: Multicast group GID.
4143  * @lid: Multicast group LID in host byte order.
4144  *
4145  * In order to send and receive multicast packets, subnet
4146  * administration must have created the multicast group and configured
4147  * the fabric appropriately.  The port associated with the specified
4148  * QP must also be a member of the multicast group.
4149  */
4150 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4151 
4152 /**
4153  * ib_detach_mcast - Detaches the specified QP from a multicast group.
4154  * @qp: QP to detach from the multicast group.
4155  * @gid: Multicast group GID.
4156  * @lid: Multicast group LID in host byte order.
4157  */
4158 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4159 
4160 /**
4161  * ib_alloc_xrcd - Allocates an XRC domain.
4162  * @device: The device on which to allocate the XRC domain.
4163  * @caller: Module name for kernel consumers
4164  */
4165 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
4166 #define ib_alloc_xrcd(device) \
4167 	__ib_alloc_xrcd((device), KBUILD_MODNAME)
4168 
4169 /**
4170  * ib_dealloc_xrcd - Deallocates an XRC domain.
4171  * @xrcd: The XRC domain to deallocate.
4172  * @udata: Valid user data or NULL for kernel object
4173  */
4174 int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
4175 
4176 static inline int ib_check_mr_access(int flags)
4177 {
4178 	/*
4179 	 * Local write permission is required if remote write or
4180 	 * remote atomic permission is also requested.
4181 	 */
4182 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4183 	    !(flags & IB_ACCESS_LOCAL_WRITE))
4184 		return -EINVAL;
4185 
4186 	return 0;
4187 }
4188 
4189 static inline bool ib_access_writable(int access_flags)
4190 {
4191 	/*
4192 	 * We have writable memory backing the MR if any of the following
4193 	 * access flags are set.  "Local write" and "remote write" obviously
4194 	 * require write access.  "Remote atomic" can do things like fetch and
4195 	 * add, which will modify memory, and "MW bind" can change permissions
4196 	 * by binding a window.
4197 	 */
4198 	return access_flags &
4199 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
4200 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4201 }
4202 
4203 /**
4204  * ib_check_mr_status: lightweight check of MR status.
4205  *     This routine may provide status checks on a selected
4206  *     ib_mr. first use is for signature status check.
4207  *
4208  * @mr: A memory region.
4209  * @check_mask: Bitmask of which checks to perform from
4210  *     ib_mr_status_check enumeration.
4211  * @mr_status: The container of relevant status checks.
4212  *     failed checks will be indicated in the status bitmask
4213  *     and the relevant info shall be in the error item.
4214  */
4215 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4216 		       struct ib_mr_status *mr_status);
4217 
4218 /**
4219  * ib_device_try_get: Hold a registration lock
4220  * device: The device to lock
4221  *
4222  * A device under an active registration lock cannot become unregistered. It
4223  * is only possible to obtain a registration lock on a device that is fully
4224  * registered, otherwise this function returns false.
4225  *
4226  * The registration lock is only necessary for actions which require the
4227  * device to still be registered. Uses that only require the device pointer to
4228  * be valid should use get_device(&ibdev->dev) to hold the memory.
4229  *
4230  */
4231 static inline bool ib_device_try_get(struct ib_device *dev)
4232 {
4233 	return refcount_inc_not_zero(&dev->refcount);
4234 }
4235 
4236 void ib_device_put(struct ib_device *device);
4237 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4238 					  enum rdma_driver_id driver_id);
4239 struct ib_device *ib_device_get_by_name(const char *name,
4240 					enum rdma_driver_id driver_id);
4241 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4242 					    u16 pkey, const union ib_gid *gid,
4243 					    const struct sockaddr *addr);
4244 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4245 			 unsigned int port);
4246 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4247 
4248 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4249 			   struct ib_wq_init_attr *init_attr);
4250 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
4251 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4252 		 u32 wq_attr_mask);
4253 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4254 						 struct ib_rwq_ind_table_init_attr*
4255 						 wq_ind_table_init_attr);
4256 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4257 
4258 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4259 		 unsigned int *sg_offset, unsigned int page_size);
4260 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4261 		    int data_sg_nents, unsigned int *data_sg_offset,
4262 		    struct scatterlist *meta_sg, int meta_sg_nents,
4263 		    unsigned int *meta_sg_offset, unsigned int page_size);
4264 
4265 static inline int
4266 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4267 		  unsigned int *sg_offset, unsigned int page_size)
4268 {
4269 	int n;
4270 
4271 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4272 	mr->iova = 0;
4273 
4274 	return n;
4275 }
4276 
4277 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4278 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4279 
4280 void ib_drain_rq(struct ib_qp *qp);
4281 void ib_drain_sq(struct ib_qp *qp);
4282 void ib_drain_qp(struct ib_qp *qp);
4283 
4284 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4285 
4286 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4287 {
4288 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4289 		return attr->roce.dmac;
4290 	return NULL;
4291 }
4292 
4293 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4294 {
4295 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4296 		attr->ib.dlid = (u16)dlid;
4297 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4298 		attr->opa.dlid = dlid;
4299 }
4300 
4301 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4302 {
4303 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4304 		return attr->ib.dlid;
4305 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4306 		return attr->opa.dlid;
4307 	return 0;
4308 }
4309 
4310 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4311 {
4312 	attr->sl = sl;
4313 }
4314 
4315 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4316 {
4317 	return attr->sl;
4318 }
4319 
4320 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4321 					 u8 src_path_bits)
4322 {
4323 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4324 		attr->ib.src_path_bits = src_path_bits;
4325 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4326 		attr->opa.src_path_bits = src_path_bits;
4327 }
4328 
4329 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4330 {
4331 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4332 		return attr->ib.src_path_bits;
4333 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4334 		return attr->opa.src_path_bits;
4335 	return 0;
4336 }
4337 
4338 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4339 					bool make_grd)
4340 {
4341 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4342 		attr->opa.make_grd = make_grd;
4343 }
4344 
4345 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4346 {
4347 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4348 		return attr->opa.make_grd;
4349 	return false;
4350 }
4351 
4352 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4353 {
4354 	attr->port_num = port_num;
4355 }
4356 
4357 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4358 {
4359 	return attr->port_num;
4360 }
4361 
4362 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4363 					   u8 static_rate)
4364 {
4365 	attr->static_rate = static_rate;
4366 }
4367 
4368 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4369 {
4370 	return attr->static_rate;
4371 }
4372 
4373 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4374 					enum ib_ah_flags flag)
4375 {
4376 	attr->ah_flags = flag;
4377 }
4378 
4379 static inline enum ib_ah_flags
4380 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4381 {
4382 	return attr->ah_flags;
4383 }
4384 
4385 static inline const struct ib_global_route
4386 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4387 {
4388 	return &attr->grh;
4389 }
4390 
4391 /*To retrieve and modify the grh */
4392 static inline struct ib_global_route
4393 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4394 {
4395 	return &attr->grh;
4396 }
4397 
4398 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4399 {
4400 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4401 
4402 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4403 }
4404 
4405 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4406 					     __be64 prefix)
4407 {
4408 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4409 
4410 	grh->dgid.global.subnet_prefix = prefix;
4411 }
4412 
4413 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4414 					    __be64 if_id)
4415 {
4416 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4417 
4418 	grh->dgid.global.interface_id = if_id;
4419 }
4420 
4421 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4422 				   union ib_gid *dgid, u32 flow_label,
4423 				   u8 sgid_index, u8 hop_limit,
4424 				   u8 traffic_class)
4425 {
4426 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4427 
4428 	attr->ah_flags = IB_AH_GRH;
4429 	if (dgid)
4430 		grh->dgid = *dgid;
4431 	grh->flow_label = flow_label;
4432 	grh->sgid_index = sgid_index;
4433 	grh->hop_limit = hop_limit;
4434 	grh->traffic_class = traffic_class;
4435 	grh->sgid_attr = NULL;
4436 }
4437 
4438 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4439 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4440 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4441 			     const struct ib_gid_attr *sgid_attr);
4442 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4443 		       const struct rdma_ah_attr *src);
4444 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4445 			  const struct rdma_ah_attr *new);
4446 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4447 
4448 /**
4449  * rdma_ah_find_type - Return address handle type.
4450  *
4451  * @dev: Device to be checked
4452  * @port_num: Port number
4453  */
4454 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4455 						       u8 port_num)
4456 {
4457 	if (rdma_protocol_roce(dev, port_num))
4458 		return RDMA_AH_ATTR_TYPE_ROCE;
4459 	if (rdma_protocol_ib(dev, port_num)) {
4460 		if (rdma_cap_opa_ah(dev, port_num))
4461 			return RDMA_AH_ATTR_TYPE_OPA;
4462 		return RDMA_AH_ATTR_TYPE_IB;
4463 	}
4464 
4465 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4466 }
4467 
4468 /**
4469  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4470  *     In the current implementation the only way to get
4471  *     get the 32bit lid is from other sources for OPA.
4472  *     For IB, lids will always be 16bits so cast the
4473  *     value accordingly.
4474  *
4475  * @lid: A 32bit LID
4476  */
4477 static inline u16 ib_lid_cpu16(u32 lid)
4478 {
4479 	WARN_ON_ONCE(lid & 0xFFFF0000);
4480 	return (u16)lid;
4481 }
4482 
4483 /**
4484  * ib_lid_be16 - Return lid in 16bit BE encoding.
4485  *
4486  * @lid: A 32bit LID
4487  */
4488 static inline __be16 ib_lid_be16(u32 lid)
4489 {
4490 	WARN_ON_ONCE(lid & 0xFFFF0000);
4491 	return cpu_to_be16((u16)lid);
4492 }
4493 
4494 /**
4495  * ib_get_vector_affinity - Get the affinity mappings of a given completion
4496  *   vector
4497  * @device:         the rdma device
4498  * @comp_vector:    index of completion vector
4499  *
4500  * Returns NULL on failure, otherwise a corresponding cpu map of the
4501  * completion vector (returns all-cpus map if the device driver doesn't
4502  * implement get_vector_affinity).
4503  */
4504 static inline const struct cpumask *
4505 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4506 {
4507 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4508 	    !device->ops.get_vector_affinity)
4509 		return NULL;
4510 
4511 	return device->ops.get_vector_affinity(device, comp_vector);
4512 
4513 }
4514 
4515 /**
4516  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4517  * and add their gids, as needed, to the relevant RoCE devices.
4518  *
4519  * @device:         the rdma device
4520  */
4521 void rdma_roce_rescan_device(struct ib_device *ibdev);
4522 
4523 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4524 
4525 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4526 
4527 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4528 				     enum rdma_netdev_t type, const char *name,
4529 				     unsigned char name_assign_type,
4530 				     void (*setup)(struct net_device *));
4531 
4532 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4533 		     enum rdma_netdev_t type, const char *name,
4534 		     unsigned char name_assign_type,
4535 		     void (*setup)(struct net_device *),
4536 		     struct net_device *netdev);
4537 
4538 /**
4539  * rdma_set_device_sysfs_group - Set device attributes group to have
4540  *				 driver specific sysfs entries at
4541  *				 for infiniband class.
4542  *
4543  * @device:	device pointer for which attributes to be created
4544  * @group:	Pointer to group which should be added when device
4545  *		is registered with sysfs.
4546  * rdma_set_device_sysfs_group() allows existing drivers to expose one
4547  * group per device to have sysfs attributes.
4548  *
4549  * NOTE: New drivers should not make use of this API; instead new device
4550  * parameter should be exposed via netlink command. This API and mechanism
4551  * exist only for existing drivers.
4552  */
4553 static inline void
4554 rdma_set_device_sysfs_group(struct ib_device *dev,
4555 			    const struct attribute_group *group)
4556 {
4557 	dev->groups[1] = group;
4558 }
4559 
4560 /**
4561  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4562  *
4563  * @device:	device pointer for which ib_device pointer to retrieve
4564  *
4565  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4566  *
4567  */
4568 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4569 {
4570 	struct ib_core_device *coredev =
4571 		container_of(device, struct ib_core_device, dev);
4572 
4573 	return coredev->owner;
4574 }
4575 
4576 /**
4577  * rdma_device_to_drv_device - Helper macro to reach back to driver's
4578  *			       ib_device holder structure from device pointer.
4579  *
4580  * NOTE: New drivers should not make use of this API; This API is only for
4581  * existing drivers who have exposed sysfs entries using
4582  * rdma_set_device_sysfs_group().
4583  */
4584 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
4585 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4586 
4587 bool rdma_dev_access_netns(const struct ib_device *device,
4588 			   const struct net *net);
4589 #endif /* IB_VERBS_H */
4590