xref: /linux/include/rdma/ib_verbs.h (revision 1a7a05e88fa0e4e168f83585d1bb1937197a9745)
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 <uapi/rdma/ib_user_verbs.h>
63 #include <rdma/restrack.h>
64 #include <uapi/rdma/rdma_user_ioctl.h>
65 #include <uapi/rdma/ib_user_ioctl_verbs.h>
66 
67 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
68 
69 struct ib_umem_odp;
70 
71 extern struct workqueue_struct *ib_wq;
72 extern struct workqueue_struct *ib_comp_wq;
73 extern struct workqueue_struct *ib_comp_unbound_wq;
74 
75 union ib_gid {
76 	u8	raw[16];
77 	struct {
78 		__be64	subnet_prefix;
79 		__be64	interface_id;
80 	} global;
81 };
82 
83 extern union ib_gid zgid;
84 
85 enum ib_gid_type {
86 	/* If link layer is Ethernet, this is RoCE V1 */
87 	IB_GID_TYPE_IB        = 0,
88 	IB_GID_TYPE_ROCE      = 0,
89 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
90 	IB_GID_TYPE_SIZE
91 };
92 
93 #define ROCE_V2_UDP_DPORT      4791
94 struct ib_gid_attr {
95 	struct net_device	*ndev;
96 	struct ib_device	*device;
97 	union ib_gid		gid;
98 	enum ib_gid_type	gid_type;
99 	u16			index;
100 	u8			port_num;
101 };
102 
103 enum rdma_node_type {
104 	/* IB values map to NodeInfo:NodeType. */
105 	RDMA_NODE_IB_CA 	= 1,
106 	RDMA_NODE_IB_SWITCH,
107 	RDMA_NODE_IB_ROUTER,
108 	RDMA_NODE_RNIC,
109 	RDMA_NODE_USNIC,
110 	RDMA_NODE_USNIC_UDP,
111 };
112 
113 enum {
114 	/* set the local administered indication */
115 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
116 };
117 
118 enum rdma_transport_type {
119 	RDMA_TRANSPORT_IB,
120 	RDMA_TRANSPORT_IWARP,
121 	RDMA_TRANSPORT_USNIC,
122 	RDMA_TRANSPORT_USNIC_UDP
123 };
124 
125 enum rdma_protocol_type {
126 	RDMA_PROTOCOL_IB,
127 	RDMA_PROTOCOL_IBOE,
128 	RDMA_PROTOCOL_IWARP,
129 	RDMA_PROTOCOL_USNIC_UDP
130 };
131 
132 __attribute_const__ enum rdma_transport_type
133 rdma_node_get_transport(enum rdma_node_type node_type);
134 
135 enum rdma_network_type {
136 	RDMA_NETWORK_IB,
137 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
138 	RDMA_NETWORK_IPV4,
139 	RDMA_NETWORK_IPV6
140 };
141 
142 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
143 {
144 	if (network_type == RDMA_NETWORK_IPV4 ||
145 	    network_type == RDMA_NETWORK_IPV6)
146 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
147 
148 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
149 	return IB_GID_TYPE_IB;
150 }
151 
152 static inline enum rdma_network_type
153 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
154 {
155 	if (attr->gid_type == IB_GID_TYPE_IB)
156 		return RDMA_NETWORK_IB;
157 
158 	if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
159 		return RDMA_NETWORK_IPV4;
160 	else
161 		return RDMA_NETWORK_IPV6;
162 }
163 
164 enum rdma_link_layer {
165 	IB_LINK_LAYER_UNSPECIFIED,
166 	IB_LINK_LAYER_INFINIBAND,
167 	IB_LINK_LAYER_ETHERNET,
168 };
169 
170 enum ib_device_cap_flags {
171 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
172 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
173 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
174 	IB_DEVICE_RAW_MULTI			= (1 << 3),
175 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
176 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
177 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
178 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
179 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
180 	/* Not in use, former INIT_TYPE		= (1 << 9),*/
181 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
182 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
183 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
184 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
185 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
186 
187 	/*
188 	 * This device supports a per-device lkey or stag that can be
189 	 * used without performing a memory registration for the local
190 	 * memory.  Note that ULPs should never check this flag, but
191 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
192 	 * which will always contain a usable lkey.
193 	 */
194 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
195 	/* Reserved, old SEND_W_INV		= (1 << 16),*/
196 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
197 	/*
198 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
199 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
200 	 * messages and can verify the validity of checksum for
201 	 * incoming messages.  Setting this flag implies that the
202 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
203 	 */
204 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
205 	IB_DEVICE_UD_TSO			= (1 << 19),
206 	IB_DEVICE_XRC				= (1 << 20),
207 
208 	/*
209 	 * This device supports the IB "base memory management extension",
210 	 * which includes support for fast registrations (IB_WR_REG_MR,
211 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
212 	 * also be set by any iWarp device which must support FRs to comply
213 	 * to the iWarp verbs spec.  iWarp devices also support the
214 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
215 	 * stag.
216 	 */
217 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
218 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
219 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
220 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
221 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
222 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
223 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
224 	/*
225 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
226 	 * support execution of WQEs that involve synchronization
227 	 * of I/O operations with single completion queue managed
228 	 * by hardware.
229 	 */
230 	IB_DEVICE_CROSS_CHANNEL			= (1 << 27),
231 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
232 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
233 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
234 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
235 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
236 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
237 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
238 	IB_DEVICE_RDMA_NETDEV_OPA_VNIC		= (1ULL << 35),
239 	/* The device supports padding incoming writes to cacheline. */
240 	IB_DEVICE_PCI_WRITE_END_PADDING		= (1ULL << 36),
241 };
242 
243 enum ib_signature_prot_cap {
244 	IB_PROT_T10DIF_TYPE_1 = 1,
245 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
246 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
247 };
248 
249 enum ib_signature_guard_cap {
250 	IB_GUARD_T10DIF_CRC	= 1,
251 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
252 };
253 
254 enum ib_atomic_cap {
255 	IB_ATOMIC_NONE,
256 	IB_ATOMIC_HCA,
257 	IB_ATOMIC_GLOB
258 };
259 
260 enum ib_odp_general_cap_bits {
261 	IB_ODP_SUPPORT		= 1 << 0,
262 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
263 };
264 
265 enum ib_odp_transport_cap_bits {
266 	IB_ODP_SUPPORT_SEND	= 1 << 0,
267 	IB_ODP_SUPPORT_RECV	= 1 << 1,
268 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
269 	IB_ODP_SUPPORT_READ	= 1 << 3,
270 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
271 	IB_ODP_SUPPORT_SRQ_RECV	= 1 << 5,
272 };
273 
274 struct ib_odp_caps {
275 	uint64_t general_caps;
276 	struct {
277 		uint32_t  rc_odp_caps;
278 		uint32_t  uc_odp_caps;
279 		uint32_t  ud_odp_caps;
280 		uint32_t  xrc_odp_caps;
281 	} per_transport_caps;
282 };
283 
284 struct ib_rss_caps {
285 	/* Corresponding bit will be set if qp type from
286 	 * 'enum ib_qp_type' is supported, e.g.
287 	 * supported_qpts |= 1 << IB_QPT_UD
288 	 */
289 	u32 supported_qpts;
290 	u32 max_rwq_indirection_tables;
291 	u32 max_rwq_indirection_table_size;
292 };
293 
294 enum ib_tm_cap_flags {
295 	/*  Support tag matching on RC transport */
296 	IB_TM_CAP_RC		    = 1 << 0,
297 };
298 
299 struct ib_tm_caps {
300 	/* Max size of RNDV header */
301 	u32 max_rndv_hdr_size;
302 	/* Max number of entries in tag matching list */
303 	u32 max_num_tags;
304 	/* From enum ib_tm_cap_flags */
305 	u32 flags;
306 	/* Max number of outstanding list operations */
307 	u32 max_ops;
308 	/* Max number of SGE in tag matching entry */
309 	u32 max_sge;
310 };
311 
312 struct ib_cq_init_attr {
313 	unsigned int	cqe;
314 	int		comp_vector;
315 	u32		flags;
316 };
317 
318 enum ib_cq_attr_mask {
319 	IB_CQ_MODERATE = 1 << 0,
320 };
321 
322 struct ib_cq_caps {
323 	u16     max_cq_moderation_count;
324 	u16     max_cq_moderation_period;
325 };
326 
327 struct ib_dm_mr_attr {
328 	u64		length;
329 	u64		offset;
330 	u32		access_flags;
331 };
332 
333 struct ib_dm_alloc_attr {
334 	u64	length;
335 	u32	alignment;
336 	u32	flags;
337 };
338 
339 struct ib_device_attr {
340 	u64			fw_ver;
341 	__be64			sys_image_guid;
342 	u64			max_mr_size;
343 	u64			page_size_cap;
344 	u32			vendor_id;
345 	u32			vendor_part_id;
346 	u32			hw_ver;
347 	int			max_qp;
348 	int			max_qp_wr;
349 	u64			device_cap_flags;
350 	int			max_send_sge;
351 	int			max_recv_sge;
352 	int			max_sge_rd;
353 	int			max_cq;
354 	int			max_cqe;
355 	int			max_mr;
356 	int			max_pd;
357 	int			max_qp_rd_atom;
358 	int			max_ee_rd_atom;
359 	int			max_res_rd_atom;
360 	int			max_qp_init_rd_atom;
361 	int			max_ee_init_rd_atom;
362 	enum ib_atomic_cap	atomic_cap;
363 	enum ib_atomic_cap	masked_atomic_cap;
364 	int			max_ee;
365 	int			max_rdd;
366 	int			max_mw;
367 	int			max_raw_ipv6_qp;
368 	int			max_raw_ethy_qp;
369 	int			max_mcast_grp;
370 	int			max_mcast_qp_attach;
371 	int			max_total_mcast_qp_attach;
372 	int			max_ah;
373 	int			max_fmr;
374 	int			max_map_per_fmr;
375 	int			max_srq;
376 	int			max_srq_wr;
377 	int			max_srq_sge;
378 	unsigned int		max_fast_reg_page_list_len;
379 	u16			max_pkeys;
380 	u8			local_ca_ack_delay;
381 	int			sig_prot_cap;
382 	int			sig_guard_cap;
383 	struct ib_odp_caps	odp_caps;
384 	uint64_t		timestamp_mask;
385 	uint64_t		hca_core_clock; /* in KHZ */
386 	struct ib_rss_caps	rss_caps;
387 	u32			max_wq_type_rq;
388 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
389 	struct ib_tm_caps	tm_caps;
390 	struct ib_cq_caps       cq_caps;
391 	u64			max_dm_size;
392 };
393 
394 enum ib_mtu {
395 	IB_MTU_256  = 1,
396 	IB_MTU_512  = 2,
397 	IB_MTU_1024 = 3,
398 	IB_MTU_2048 = 4,
399 	IB_MTU_4096 = 5
400 };
401 
402 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
403 {
404 	switch (mtu) {
405 	case IB_MTU_256:  return  256;
406 	case IB_MTU_512:  return  512;
407 	case IB_MTU_1024: return 1024;
408 	case IB_MTU_2048: return 2048;
409 	case IB_MTU_4096: return 4096;
410 	default: 	  return -1;
411 	}
412 }
413 
414 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
415 {
416 	if (mtu >= 4096)
417 		return IB_MTU_4096;
418 	else if (mtu >= 2048)
419 		return IB_MTU_2048;
420 	else if (mtu >= 1024)
421 		return IB_MTU_1024;
422 	else if (mtu >= 512)
423 		return IB_MTU_512;
424 	else
425 		return IB_MTU_256;
426 }
427 
428 enum ib_port_state {
429 	IB_PORT_NOP		= 0,
430 	IB_PORT_DOWN		= 1,
431 	IB_PORT_INIT		= 2,
432 	IB_PORT_ARMED		= 3,
433 	IB_PORT_ACTIVE		= 4,
434 	IB_PORT_ACTIVE_DEFER	= 5
435 };
436 
437 enum ib_port_width {
438 	IB_WIDTH_1X	= 1,
439 	IB_WIDTH_2X	= 16,
440 	IB_WIDTH_4X	= 2,
441 	IB_WIDTH_8X	= 4,
442 	IB_WIDTH_12X	= 8
443 };
444 
445 static inline int ib_width_enum_to_int(enum ib_port_width width)
446 {
447 	switch (width) {
448 	case IB_WIDTH_1X:  return  1;
449 	case IB_WIDTH_2X:  return  2;
450 	case IB_WIDTH_4X:  return  4;
451 	case IB_WIDTH_8X:  return  8;
452 	case IB_WIDTH_12X: return 12;
453 	default: 	  return -1;
454 	}
455 }
456 
457 enum ib_port_speed {
458 	IB_SPEED_SDR	= 1,
459 	IB_SPEED_DDR	= 2,
460 	IB_SPEED_QDR	= 4,
461 	IB_SPEED_FDR10	= 8,
462 	IB_SPEED_FDR	= 16,
463 	IB_SPEED_EDR	= 32,
464 	IB_SPEED_HDR	= 64
465 };
466 
467 /**
468  * struct rdma_hw_stats
469  * @lock - Mutex to protect parallel write access to lifespan and values
470  *    of counters, which are 64bits and not guaranteeed to be written
471  *    atomicaly on 32bits systems.
472  * @timestamp - Used by the core code to track when the last update was
473  * @lifespan - Used by the core code to determine how old the counters
474  *   should be before being updated again.  Stored in jiffies, defaults
475  *   to 10 milliseconds, drivers can override the default be specifying
476  *   their own value during their allocation routine.
477  * @name - Array of pointers to static names used for the counters in
478  *   directory.
479  * @num_counters - How many hardware counters there are.  If name is
480  *   shorter than this number, a kernel oops will result.  Driver authors
481  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
482  *   in their code to prevent this.
483  * @value - Array of u64 counters that are accessed by the sysfs code and
484  *   filled in by the drivers get_stats routine
485  */
486 struct rdma_hw_stats {
487 	struct mutex	lock; /* Protect lifespan and values[] */
488 	unsigned long	timestamp;
489 	unsigned long	lifespan;
490 	const char * const *names;
491 	int		num_counters;
492 	u64		value[];
493 };
494 
495 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
496 /**
497  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
498  *   for drivers.
499  * @names - Array of static const char *
500  * @num_counters - How many elements in array
501  * @lifespan - How many milliseconds between updates
502  */
503 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
504 		const char * const *names, int num_counters,
505 		unsigned long lifespan)
506 {
507 	struct rdma_hw_stats *stats;
508 
509 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
510 			GFP_KERNEL);
511 	if (!stats)
512 		return NULL;
513 	stats->names = names;
514 	stats->num_counters = num_counters;
515 	stats->lifespan = msecs_to_jiffies(lifespan);
516 
517 	return stats;
518 }
519 
520 
521 /* Define bits for the various functionality this port needs to be supported by
522  * the core.
523  */
524 /* Management                           0x00000FFF */
525 #define RDMA_CORE_CAP_IB_MAD            0x00000001
526 #define RDMA_CORE_CAP_IB_SMI            0x00000002
527 #define RDMA_CORE_CAP_IB_CM             0x00000004
528 #define RDMA_CORE_CAP_IW_CM             0x00000008
529 #define RDMA_CORE_CAP_IB_SA             0x00000010
530 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
531 
532 /* Address format                       0x000FF000 */
533 #define RDMA_CORE_CAP_AF_IB             0x00001000
534 #define RDMA_CORE_CAP_ETH_AH            0x00002000
535 #define RDMA_CORE_CAP_OPA_AH            0x00004000
536 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
537 
538 /* Protocol                             0xFFF00000 */
539 #define RDMA_CORE_CAP_PROT_IB           0x00100000
540 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
541 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
542 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
543 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
544 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
545 
546 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
547 					| RDMA_CORE_CAP_PROT_ROCE     \
548 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
549 
550 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
551 					| RDMA_CORE_CAP_IB_MAD \
552 					| RDMA_CORE_CAP_IB_SMI \
553 					| RDMA_CORE_CAP_IB_CM  \
554 					| RDMA_CORE_CAP_IB_SA  \
555 					| RDMA_CORE_CAP_AF_IB)
556 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
557 					| RDMA_CORE_CAP_IB_MAD  \
558 					| RDMA_CORE_CAP_IB_CM   \
559 					| RDMA_CORE_CAP_AF_IB   \
560 					| RDMA_CORE_CAP_ETH_AH)
561 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
562 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
563 					| RDMA_CORE_CAP_IB_MAD  \
564 					| RDMA_CORE_CAP_IB_CM   \
565 					| RDMA_CORE_CAP_AF_IB   \
566 					| RDMA_CORE_CAP_ETH_AH)
567 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
568 					| RDMA_CORE_CAP_IW_CM)
569 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
570 					| RDMA_CORE_CAP_OPA_MAD)
571 
572 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
573 
574 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
575 
576 struct ib_port_attr {
577 	u64			subnet_prefix;
578 	enum ib_port_state	state;
579 	enum ib_mtu		max_mtu;
580 	enum ib_mtu		active_mtu;
581 	int			gid_tbl_len;
582 	unsigned int		ip_gids:1;
583 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
584 	u32			port_cap_flags;
585 	u32			max_msg_sz;
586 	u32			bad_pkey_cntr;
587 	u32			qkey_viol_cntr;
588 	u16			pkey_tbl_len;
589 	u32			sm_lid;
590 	u32			lid;
591 	u8			lmc;
592 	u8			max_vl_num;
593 	u8			sm_sl;
594 	u8			subnet_timeout;
595 	u8			init_type_reply;
596 	u8			active_width;
597 	u8			active_speed;
598 	u8                      phys_state;
599 	u16			port_cap_flags2;
600 };
601 
602 enum ib_device_modify_flags {
603 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
604 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
605 };
606 
607 #define IB_DEVICE_NODE_DESC_MAX 64
608 
609 struct ib_device_modify {
610 	u64	sys_image_guid;
611 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
612 };
613 
614 enum ib_port_modify_flags {
615 	IB_PORT_SHUTDOWN		= 1,
616 	IB_PORT_INIT_TYPE		= (1<<2),
617 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
618 	IB_PORT_OPA_MASK_CHG		= (1<<4)
619 };
620 
621 struct ib_port_modify {
622 	u32	set_port_cap_mask;
623 	u32	clr_port_cap_mask;
624 	u8	init_type;
625 };
626 
627 enum ib_event_type {
628 	IB_EVENT_CQ_ERR,
629 	IB_EVENT_QP_FATAL,
630 	IB_EVENT_QP_REQ_ERR,
631 	IB_EVENT_QP_ACCESS_ERR,
632 	IB_EVENT_COMM_EST,
633 	IB_EVENT_SQ_DRAINED,
634 	IB_EVENT_PATH_MIG,
635 	IB_EVENT_PATH_MIG_ERR,
636 	IB_EVENT_DEVICE_FATAL,
637 	IB_EVENT_PORT_ACTIVE,
638 	IB_EVENT_PORT_ERR,
639 	IB_EVENT_LID_CHANGE,
640 	IB_EVENT_PKEY_CHANGE,
641 	IB_EVENT_SM_CHANGE,
642 	IB_EVENT_SRQ_ERR,
643 	IB_EVENT_SRQ_LIMIT_REACHED,
644 	IB_EVENT_QP_LAST_WQE_REACHED,
645 	IB_EVENT_CLIENT_REREGISTER,
646 	IB_EVENT_GID_CHANGE,
647 	IB_EVENT_WQ_FATAL,
648 };
649 
650 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
651 
652 struct ib_event {
653 	struct ib_device	*device;
654 	union {
655 		struct ib_cq	*cq;
656 		struct ib_qp	*qp;
657 		struct ib_srq	*srq;
658 		struct ib_wq	*wq;
659 		u8		port_num;
660 	} element;
661 	enum ib_event_type	event;
662 };
663 
664 struct ib_event_handler {
665 	struct ib_device *device;
666 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
667 	struct list_head  list;
668 };
669 
670 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
671 	do {							\
672 		(_ptr)->device  = _device;			\
673 		(_ptr)->handler = _handler;			\
674 		INIT_LIST_HEAD(&(_ptr)->list);			\
675 	} while (0)
676 
677 struct ib_global_route {
678 	const struct ib_gid_attr *sgid_attr;
679 	union ib_gid	dgid;
680 	u32		flow_label;
681 	u8		sgid_index;
682 	u8		hop_limit;
683 	u8		traffic_class;
684 };
685 
686 struct ib_grh {
687 	__be32		version_tclass_flow;
688 	__be16		paylen;
689 	u8		next_hdr;
690 	u8		hop_limit;
691 	union ib_gid	sgid;
692 	union ib_gid	dgid;
693 };
694 
695 union rdma_network_hdr {
696 	struct ib_grh ibgrh;
697 	struct {
698 		/* The IB spec states that if it's IPv4, the header
699 		 * is located in the last 20 bytes of the header.
700 		 */
701 		u8		reserved[20];
702 		struct iphdr	roce4grh;
703 	};
704 };
705 
706 #define IB_QPN_MASK		0xFFFFFF
707 
708 enum {
709 	IB_MULTICAST_QPN = 0xffffff
710 };
711 
712 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
713 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
714 
715 enum ib_ah_flags {
716 	IB_AH_GRH	= 1
717 };
718 
719 enum ib_rate {
720 	IB_RATE_PORT_CURRENT = 0,
721 	IB_RATE_2_5_GBPS = 2,
722 	IB_RATE_5_GBPS   = 5,
723 	IB_RATE_10_GBPS  = 3,
724 	IB_RATE_20_GBPS  = 6,
725 	IB_RATE_30_GBPS  = 4,
726 	IB_RATE_40_GBPS  = 7,
727 	IB_RATE_60_GBPS  = 8,
728 	IB_RATE_80_GBPS  = 9,
729 	IB_RATE_120_GBPS = 10,
730 	IB_RATE_14_GBPS  = 11,
731 	IB_RATE_56_GBPS  = 12,
732 	IB_RATE_112_GBPS = 13,
733 	IB_RATE_168_GBPS = 14,
734 	IB_RATE_25_GBPS  = 15,
735 	IB_RATE_100_GBPS = 16,
736 	IB_RATE_200_GBPS = 17,
737 	IB_RATE_300_GBPS = 18,
738 	IB_RATE_28_GBPS  = 19,
739 	IB_RATE_50_GBPS  = 20,
740 	IB_RATE_400_GBPS = 21,
741 	IB_RATE_600_GBPS = 22,
742 };
743 
744 /**
745  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
746  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
747  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
748  * @rate: rate to convert.
749  */
750 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
751 
752 /**
753  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
754  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
755  * @rate: rate to convert.
756  */
757 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
758 
759 
760 /**
761  * enum ib_mr_type - memory region type
762  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
763  *                            normal registration
764  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
765  *                            signature operations (data-integrity
766  *                            capable regions)
767  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
768  *                            register any arbitrary sg lists (without
769  *                            the normal mr constraints - see
770  *                            ib_map_mr_sg)
771  */
772 enum ib_mr_type {
773 	IB_MR_TYPE_MEM_REG,
774 	IB_MR_TYPE_SIGNATURE,
775 	IB_MR_TYPE_SG_GAPS,
776 };
777 
778 /**
779  * Signature types
780  * IB_SIG_TYPE_NONE: Unprotected.
781  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
782  */
783 enum ib_signature_type {
784 	IB_SIG_TYPE_NONE,
785 	IB_SIG_TYPE_T10_DIF,
786 };
787 
788 /**
789  * Signature T10-DIF block-guard types
790  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
791  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
792  */
793 enum ib_t10_dif_bg_type {
794 	IB_T10DIF_CRC,
795 	IB_T10DIF_CSUM
796 };
797 
798 /**
799  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
800  *     domain.
801  * @bg_type: T10-DIF block guard type (CRC|CSUM)
802  * @pi_interval: protection information interval.
803  * @bg: seed of guard computation.
804  * @app_tag: application tag of guard block
805  * @ref_tag: initial guard block reference tag.
806  * @ref_remap: Indicate wethear the reftag increments each block
807  * @app_escape: Indicate to skip block check if apptag=0xffff
808  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
809  * @apptag_check_mask: check bitmask of application tag.
810  */
811 struct ib_t10_dif_domain {
812 	enum ib_t10_dif_bg_type bg_type;
813 	u16			pi_interval;
814 	u16			bg;
815 	u16			app_tag;
816 	u32			ref_tag;
817 	bool			ref_remap;
818 	bool			app_escape;
819 	bool			ref_escape;
820 	u16			apptag_check_mask;
821 };
822 
823 /**
824  * struct ib_sig_domain - Parameters for signature domain
825  * @sig_type: specific signauture type
826  * @sig: union of all signature domain attributes that may
827  *     be used to set domain layout.
828  */
829 struct ib_sig_domain {
830 	enum ib_signature_type sig_type;
831 	union {
832 		struct ib_t10_dif_domain dif;
833 	} sig;
834 };
835 
836 /**
837  * struct ib_sig_attrs - Parameters for signature handover operation
838  * @check_mask: bitmask for signature byte check (8 bytes)
839  * @mem: memory domain layout desciptor.
840  * @wire: wire domain layout desciptor.
841  */
842 struct ib_sig_attrs {
843 	u8			check_mask;
844 	struct ib_sig_domain	mem;
845 	struct ib_sig_domain	wire;
846 };
847 
848 enum ib_sig_err_type {
849 	IB_SIG_BAD_GUARD,
850 	IB_SIG_BAD_REFTAG,
851 	IB_SIG_BAD_APPTAG,
852 };
853 
854 /**
855  * Signature check masks (8 bytes in total) according to the T10-PI standard:
856  *  -------- -------- ------------
857  * | GUARD  | APPTAG |   REFTAG   |
858  * |  2B    |  2B    |    4B      |
859  *  -------- -------- ------------
860  */
861 enum {
862 	IB_SIG_CHECK_GUARD	= 0xc0,
863 	IB_SIG_CHECK_APPTAG	= 0x30,
864 	IB_SIG_CHECK_REFTAG	= 0x0f,
865 };
866 
867 /**
868  * struct ib_sig_err - signature error descriptor
869  */
870 struct ib_sig_err {
871 	enum ib_sig_err_type	err_type;
872 	u32			expected;
873 	u32			actual;
874 	u64			sig_err_offset;
875 	u32			key;
876 };
877 
878 enum ib_mr_status_check {
879 	IB_MR_CHECK_SIG_STATUS = 1,
880 };
881 
882 /**
883  * struct ib_mr_status - Memory region status container
884  *
885  * @fail_status: Bitmask of MR checks status. For each
886  *     failed check a corresponding status bit is set.
887  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
888  *     failure.
889  */
890 struct ib_mr_status {
891 	u32		    fail_status;
892 	struct ib_sig_err   sig_err;
893 };
894 
895 /**
896  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
897  * enum.
898  * @mult: multiple to convert.
899  */
900 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
901 
902 enum rdma_ah_attr_type {
903 	RDMA_AH_ATTR_TYPE_UNDEFINED,
904 	RDMA_AH_ATTR_TYPE_IB,
905 	RDMA_AH_ATTR_TYPE_ROCE,
906 	RDMA_AH_ATTR_TYPE_OPA,
907 };
908 
909 struct ib_ah_attr {
910 	u16			dlid;
911 	u8			src_path_bits;
912 };
913 
914 struct roce_ah_attr {
915 	u8			dmac[ETH_ALEN];
916 };
917 
918 struct opa_ah_attr {
919 	u32			dlid;
920 	u8			src_path_bits;
921 	bool			make_grd;
922 };
923 
924 struct rdma_ah_attr {
925 	struct ib_global_route	grh;
926 	u8			sl;
927 	u8			static_rate;
928 	u8			port_num;
929 	u8			ah_flags;
930 	enum rdma_ah_attr_type type;
931 	union {
932 		struct ib_ah_attr ib;
933 		struct roce_ah_attr roce;
934 		struct opa_ah_attr opa;
935 	};
936 };
937 
938 enum ib_wc_status {
939 	IB_WC_SUCCESS,
940 	IB_WC_LOC_LEN_ERR,
941 	IB_WC_LOC_QP_OP_ERR,
942 	IB_WC_LOC_EEC_OP_ERR,
943 	IB_WC_LOC_PROT_ERR,
944 	IB_WC_WR_FLUSH_ERR,
945 	IB_WC_MW_BIND_ERR,
946 	IB_WC_BAD_RESP_ERR,
947 	IB_WC_LOC_ACCESS_ERR,
948 	IB_WC_REM_INV_REQ_ERR,
949 	IB_WC_REM_ACCESS_ERR,
950 	IB_WC_REM_OP_ERR,
951 	IB_WC_RETRY_EXC_ERR,
952 	IB_WC_RNR_RETRY_EXC_ERR,
953 	IB_WC_LOC_RDD_VIOL_ERR,
954 	IB_WC_REM_INV_RD_REQ_ERR,
955 	IB_WC_REM_ABORT_ERR,
956 	IB_WC_INV_EECN_ERR,
957 	IB_WC_INV_EEC_STATE_ERR,
958 	IB_WC_FATAL_ERR,
959 	IB_WC_RESP_TIMEOUT_ERR,
960 	IB_WC_GENERAL_ERR
961 };
962 
963 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
964 
965 enum ib_wc_opcode {
966 	IB_WC_SEND,
967 	IB_WC_RDMA_WRITE,
968 	IB_WC_RDMA_READ,
969 	IB_WC_COMP_SWAP,
970 	IB_WC_FETCH_ADD,
971 	IB_WC_LSO,
972 	IB_WC_LOCAL_INV,
973 	IB_WC_REG_MR,
974 	IB_WC_MASKED_COMP_SWAP,
975 	IB_WC_MASKED_FETCH_ADD,
976 /*
977  * Set value of IB_WC_RECV so consumers can test if a completion is a
978  * receive by testing (opcode & IB_WC_RECV).
979  */
980 	IB_WC_RECV			= 1 << 7,
981 	IB_WC_RECV_RDMA_WITH_IMM
982 };
983 
984 enum ib_wc_flags {
985 	IB_WC_GRH		= 1,
986 	IB_WC_WITH_IMM		= (1<<1),
987 	IB_WC_WITH_INVALIDATE	= (1<<2),
988 	IB_WC_IP_CSUM_OK	= (1<<3),
989 	IB_WC_WITH_SMAC		= (1<<4),
990 	IB_WC_WITH_VLAN		= (1<<5),
991 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
992 };
993 
994 struct ib_wc {
995 	union {
996 		u64		wr_id;
997 		struct ib_cqe	*wr_cqe;
998 	};
999 	enum ib_wc_status	status;
1000 	enum ib_wc_opcode	opcode;
1001 	u32			vendor_err;
1002 	u32			byte_len;
1003 	struct ib_qp	       *qp;
1004 	union {
1005 		__be32		imm_data;
1006 		u32		invalidate_rkey;
1007 	} ex;
1008 	u32			src_qp;
1009 	u32			slid;
1010 	int			wc_flags;
1011 	u16			pkey_index;
1012 	u8			sl;
1013 	u8			dlid_path_bits;
1014 	u8			port_num;	/* valid only for DR SMPs on switches */
1015 	u8			smac[ETH_ALEN];
1016 	u16			vlan_id;
1017 	u8			network_hdr_type;
1018 };
1019 
1020 enum ib_cq_notify_flags {
1021 	IB_CQ_SOLICITED			= 1 << 0,
1022 	IB_CQ_NEXT_COMP			= 1 << 1,
1023 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1024 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
1025 };
1026 
1027 enum ib_srq_type {
1028 	IB_SRQT_BASIC,
1029 	IB_SRQT_XRC,
1030 	IB_SRQT_TM,
1031 };
1032 
1033 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1034 {
1035 	return srq_type == IB_SRQT_XRC ||
1036 	       srq_type == IB_SRQT_TM;
1037 }
1038 
1039 enum ib_srq_attr_mask {
1040 	IB_SRQ_MAX_WR	= 1 << 0,
1041 	IB_SRQ_LIMIT	= 1 << 1,
1042 };
1043 
1044 struct ib_srq_attr {
1045 	u32	max_wr;
1046 	u32	max_sge;
1047 	u32	srq_limit;
1048 };
1049 
1050 struct ib_srq_init_attr {
1051 	void		      (*event_handler)(struct ib_event *, void *);
1052 	void		       *srq_context;
1053 	struct ib_srq_attr	attr;
1054 	enum ib_srq_type	srq_type;
1055 
1056 	struct {
1057 		struct ib_cq   *cq;
1058 		union {
1059 			struct {
1060 				struct ib_xrcd *xrcd;
1061 			} xrc;
1062 
1063 			struct {
1064 				u32		max_num_tags;
1065 			} tag_matching;
1066 		};
1067 	} ext;
1068 };
1069 
1070 struct ib_qp_cap {
1071 	u32	max_send_wr;
1072 	u32	max_recv_wr;
1073 	u32	max_send_sge;
1074 	u32	max_recv_sge;
1075 	u32	max_inline_data;
1076 
1077 	/*
1078 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1079 	 * ib_create_qp() will calculate the right amount of neededed WRs
1080 	 * and MRs based on this.
1081 	 */
1082 	u32	max_rdma_ctxs;
1083 };
1084 
1085 enum ib_sig_type {
1086 	IB_SIGNAL_ALL_WR,
1087 	IB_SIGNAL_REQ_WR
1088 };
1089 
1090 enum ib_qp_type {
1091 	/*
1092 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1093 	 * here (and in that order) since the MAD layer uses them as
1094 	 * indices into a 2-entry table.
1095 	 */
1096 	IB_QPT_SMI,
1097 	IB_QPT_GSI,
1098 
1099 	IB_QPT_RC,
1100 	IB_QPT_UC,
1101 	IB_QPT_UD,
1102 	IB_QPT_RAW_IPV6,
1103 	IB_QPT_RAW_ETHERTYPE,
1104 	IB_QPT_RAW_PACKET = 8,
1105 	IB_QPT_XRC_INI = 9,
1106 	IB_QPT_XRC_TGT,
1107 	IB_QPT_MAX,
1108 	IB_QPT_DRIVER = 0xFF,
1109 	/* Reserve a range for qp types internal to the low level driver.
1110 	 * These qp types will not be visible at the IB core layer, so the
1111 	 * IB_QPT_MAX usages should not be affected in the core layer
1112 	 */
1113 	IB_QPT_RESERVED1 = 0x1000,
1114 	IB_QPT_RESERVED2,
1115 	IB_QPT_RESERVED3,
1116 	IB_QPT_RESERVED4,
1117 	IB_QPT_RESERVED5,
1118 	IB_QPT_RESERVED6,
1119 	IB_QPT_RESERVED7,
1120 	IB_QPT_RESERVED8,
1121 	IB_QPT_RESERVED9,
1122 	IB_QPT_RESERVED10,
1123 };
1124 
1125 enum ib_qp_create_flags {
1126 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1127 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1128 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1129 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1130 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1131 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1132 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1133 	/* FREE					= 1 << 7, */
1134 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1135 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1136 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1137 	IB_QP_CREATE_PCI_WRITE_END_PADDING	= 1 << 11,
1138 	/* reserve bits 26-31 for low level drivers' internal use */
1139 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1140 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1141 };
1142 
1143 /*
1144  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1145  * callback to destroy the passed in QP.
1146  */
1147 
1148 struct ib_qp_init_attr {
1149 	/* Consumer's event_handler callback must not block */
1150 	void                  (*event_handler)(struct ib_event *, void *);
1151 
1152 	void		       *qp_context;
1153 	struct ib_cq	       *send_cq;
1154 	struct ib_cq	       *recv_cq;
1155 	struct ib_srq	       *srq;
1156 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1157 	struct ib_qp_cap	cap;
1158 	enum ib_sig_type	sq_sig_type;
1159 	enum ib_qp_type		qp_type;
1160 	u32			create_flags;
1161 
1162 	/*
1163 	 * Only needed for special QP types, or when using the RW API.
1164 	 */
1165 	u8			port_num;
1166 	struct ib_rwq_ind_table *rwq_ind_tbl;
1167 	u32			source_qpn;
1168 };
1169 
1170 struct ib_qp_open_attr {
1171 	void                  (*event_handler)(struct ib_event *, void *);
1172 	void		       *qp_context;
1173 	u32			qp_num;
1174 	enum ib_qp_type		qp_type;
1175 };
1176 
1177 enum ib_rnr_timeout {
1178 	IB_RNR_TIMER_655_36 =  0,
1179 	IB_RNR_TIMER_000_01 =  1,
1180 	IB_RNR_TIMER_000_02 =  2,
1181 	IB_RNR_TIMER_000_03 =  3,
1182 	IB_RNR_TIMER_000_04 =  4,
1183 	IB_RNR_TIMER_000_06 =  5,
1184 	IB_RNR_TIMER_000_08 =  6,
1185 	IB_RNR_TIMER_000_12 =  7,
1186 	IB_RNR_TIMER_000_16 =  8,
1187 	IB_RNR_TIMER_000_24 =  9,
1188 	IB_RNR_TIMER_000_32 = 10,
1189 	IB_RNR_TIMER_000_48 = 11,
1190 	IB_RNR_TIMER_000_64 = 12,
1191 	IB_RNR_TIMER_000_96 = 13,
1192 	IB_RNR_TIMER_001_28 = 14,
1193 	IB_RNR_TIMER_001_92 = 15,
1194 	IB_RNR_TIMER_002_56 = 16,
1195 	IB_RNR_TIMER_003_84 = 17,
1196 	IB_RNR_TIMER_005_12 = 18,
1197 	IB_RNR_TIMER_007_68 = 19,
1198 	IB_RNR_TIMER_010_24 = 20,
1199 	IB_RNR_TIMER_015_36 = 21,
1200 	IB_RNR_TIMER_020_48 = 22,
1201 	IB_RNR_TIMER_030_72 = 23,
1202 	IB_RNR_TIMER_040_96 = 24,
1203 	IB_RNR_TIMER_061_44 = 25,
1204 	IB_RNR_TIMER_081_92 = 26,
1205 	IB_RNR_TIMER_122_88 = 27,
1206 	IB_RNR_TIMER_163_84 = 28,
1207 	IB_RNR_TIMER_245_76 = 29,
1208 	IB_RNR_TIMER_327_68 = 30,
1209 	IB_RNR_TIMER_491_52 = 31
1210 };
1211 
1212 enum ib_qp_attr_mask {
1213 	IB_QP_STATE			= 1,
1214 	IB_QP_CUR_STATE			= (1<<1),
1215 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1216 	IB_QP_ACCESS_FLAGS		= (1<<3),
1217 	IB_QP_PKEY_INDEX		= (1<<4),
1218 	IB_QP_PORT			= (1<<5),
1219 	IB_QP_QKEY			= (1<<6),
1220 	IB_QP_AV			= (1<<7),
1221 	IB_QP_PATH_MTU			= (1<<8),
1222 	IB_QP_TIMEOUT			= (1<<9),
1223 	IB_QP_RETRY_CNT			= (1<<10),
1224 	IB_QP_RNR_RETRY			= (1<<11),
1225 	IB_QP_RQ_PSN			= (1<<12),
1226 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1227 	IB_QP_ALT_PATH			= (1<<14),
1228 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1229 	IB_QP_SQ_PSN			= (1<<16),
1230 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1231 	IB_QP_PATH_MIG_STATE		= (1<<18),
1232 	IB_QP_CAP			= (1<<19),
1233 	IB_QP_DEST_QPN			= (1<<20),
1234 	IB_QP_RESERVED1			= (1<<21),
1235 	IB_QP_RESERVED2			= (1<<22),
1236 	IB_QP_RESERVED3			= (1<<23),
1237 	IB_QP_RESERVED4			= (1<<24),
1238 	IB_QP_RATE_LIMIT		= (1<<25),
1239 };
1240 
1241 enum ib_qp_state {
1242 	IB_QPS_RESET,
1243 	IB_QPS_INIT,
1244 	IB_QPS_RTR,
1245 	IB_QPS_RTS,
1246 	IB_QPS_SQD,
1247 	IB_QPS_SQE,
1248 	IB_QPS_ERR
1249 };
1250 
1251 enum ib_mig_state {
1252 	IB_MIG_MIGRATED,
1253 	IB_MIG_REARM,
1254 	IB_MIG_ARMED
1255 };
1256 
1257 enum ib_mw_type {
1258 	IB_MW_TYPE_1 = 1,
1259 	IB_MW_TYPE_2 = 2
1260 };
1261 
1262 struct ib_qp_attr {
1263 	enum ib_qp_state	qp_state;
1264 	enum ib_qp_state	cur_qp_state;
1265 	enum ib_mtu		path_mtu;
1266 	enum ib_mig_state	path_mig_state;
1267 	u32			qkey;
1268 	u32			rq_psn;
1269 	u32			sq_psn;
1270 	u32			dest_qp_num;
1271 	int			qp_access_flags;
1272 	struct ib_qp_cap	cap;
1273 	struct rdma_ah_attr	ah_attr;
1274 	struct rdma_ah_attr	alt_ah_attr;
1275 	u16			pkey_index;
1276 	u16			alt_pkey_index;
1277 	u8			en_sqd_async_notify;
1278 	u8			sq_draining;
1279 	u8			max_rd_atomic;
1280 	u8			max_dest_rd_atomic;
1281 	u8			min_rnr_timer;
1282 	u8			port_num;
1283 	u8			timeout;
1284 	u8			retry_cnt;
1285 	u8			rnr_retry;
1286 	u8			alt_port_num;
1287 	u8			alt_timeout;
1288 	u32			rate_limit;
1289 };
1290 
1291 enum ib_wr_opcode {
1292 	/* These are shared with userspace */
1293 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1294 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1295 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1296 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1297 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1298 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1299 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1300 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1301 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1302 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1303 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1304 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1305 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1306 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1307 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1308 
1309 	/* These are kernel only and can not be issued by userspace */
1310 	IB_WR_REG_MR = 0x20,
1311 	IB_WR_REG_SIG_MR,
1312 
1313 	/* reserve values for low level drivers' internal use.
1314 	 * These values will not be used at all in the ib core layer.
1315 	 */
1316 	IB_WR_RESERVED1 = 0xf0,
1317 	IB_WR_RESERVED2,
1318 	IB_WR_RESERVED3,
1319 	IB_WR_RESERVED4,
1320 	IB_WR_RESERVED5,
1321 	IB_WR_RESERVED6,
1322 	IB_WR_RESERVED7,
1323 	IB_WR_RESERVED8,
1324 	IB_WR_RESERVED9,
1325 	IB_WR_RESERVED10,
1326 };
1327 
1328 enum ib_send_flags {
1329 	IB_SEND_FENCE		= 1,
1330 	IB_SEND_SIGNALED	= (1<<1),
1331 	IB_SEND_SOLICITED	= (1<<2),
1332 	IB_SEND_INLINE		= (1<<3),
1333 	IB_SEND_IP_CSUM		= (1<<4),
1334 
1335 	/* reserve bits 26-31 for low level drivers' internal use */
1336 	IB_SEND_RESERVED_START	= (1 << 26),
1337 	IB_SEND_RESERVED_END	= (1 << 31),
1338 };
1339 
1340 struct ib_sge {
1341 	u64	addr;
1342 	u32	length;
1343 	u32	lkey;
1344 };
1345 
1346 struct ib_cqe {
1347 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1348 };
1349 
1350 struct ib_send_wr {
1351 	struct ib_send_wr      *next;
1352 	union {
1353 		u64		wr_id;
1354 		struct ib_cqe	*wr_cqe;
1355 	};
1356 	struct ib_sge	       *sg_list;
1357 	int			num_sge;
1358 	enum ib_wr_opcode	opcode;
1359 	int			send_flags;
1360 	union {
1361 		__be32		imm_data;
1362 		u32		invalidate_rkey;
1363 	} ex;
1364 };
1365 
1366 struct ib_rdma_wr {
1367 	struct ib_send_wr	wr;
1368 	u64			remote_addr;
1369 	u32			rkey;
1370 };
1371 
1372 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1373 {
1374 	return container_of(wr, struct ib_rdma_wr, wr);
1375 }
1376 
1377 struct ib_atomic_wr {
1378 	struct ib_send_wr	wr;
1379 	u64			remote_addr;
1380 	u64			compare_add;
1381 	u64			swap;
1382 	u64			compare_add_mask;
1383 	u64			swap_mask;
1384 	u32			rkey;
1385 };
1386 
1387 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1388 {
1389 	return container_of(wr, struct ib_atomic_wr, wr);
1390 }
1391 
1392 struct ib_ud_wr {
1393 	struct ib_send_wr	wr;
1394 	struct ib_ah		*ah;
1395 	void			*header;
1396 	int			hlen;
1397 	int			mss;
1398 	u32			remote_qpn;
1399 	u32			remote_qkey;
1400 	u16			pkey_index; /* valid for GSI only */
1401 	u8			port_num;   /* valid for DR SMPs on switch only */
1402 };
1403 
1404 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1405 {
1406 	return container_of(wr, struct ib_ud_wr, wr);
1407 }
1408 
1409 struct ib_reg_wr {
1410 	struct ib_send_wr	wr;
1411 	struct ib_mr		*mr;
1412 	u32			key;
1413 	int			access;
1414 };
1415 
1416 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1417 {
1418 	return container_of(wr, struct ib_reg_wr, wr);
1419 }
1420 
1421 struct ib_sig_handover_wr {
1422 	struct ib_send_wr	wr;
1423 	struct ib_sig_attrs    *sig_attrs;
1424 	struct ib_mr	       *sig_mr;
1425 	int			access_flags;
1426 	struct ib_sge	       *prot;
1427 };
1428 
1429 static inline const struct ib_sig_handover_wr *
1430 sig_handover_wr(const struct ib_send_wr *wr)
1431 {
1432 	return container_of(wr, struct ib_sig_handover_wr, wr);
1433 }
1434 
1435 struct ib_recv_wr {
1436 	struct ib_recv_wr      *next;
1437 	union {
1438 		u64		wr_id;
1439 		struct ib_cqe	*wr_cqe;
1440 	};
1441 	struct ib_sge	       *sg_list;
1442 	int			num_sge;
1443 };
1444 
1445 enum ib_access_flags {
1446 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1447 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1448 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1449 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1450 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1451 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1452 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1453 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1454 
1455 	IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1456 };
1457 
1458 /*
1459  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1460  * are hidden here instead of a uapi header!
1461  */
1462 enum ib_mr_rereg_flags {
1463 	IB_MR_REREG_TRANS	= 1,
1464 	IB_MR_REREG_PD		= (1<<1),
1465 	IB_MR_REREG_ACCESS	= (1<<2),
1466 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1467 };
1468 
1469 struct ib_fmr_attr {
1470 	int	max_pages;
1471 	int	max_maps;
1472 	u8	page_shift;
1473 };
1474 
1475 struct ib_umem;
1476 
1477 enum rdma_remove_reason {
1478 	/*
1479 	 * Userspace requested uobject deletion or initial try
1480 	 * to remove uobject via cleanup. Call could fail
1481 	 */
1482 	RDMA_REMOVE_DESTROY,
1483 	/* Context deletion. This call should delete the actual object itself */
1484 	RDMA_REMOVE_CLOSE,
1485 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1486 	RDMA_REMOVE_DRIVER_REMOVE,
1487 	/* uobj is being cleaned-up before being committed */
1488 	RDMA_REMOVE_ABORT,
1489 };
1490 
1491 struct ib_rdmacg_object {
1492 #ifdef CONFIG_CGROUP_RDMA
1493 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1494 #endif
1495 };
1496 
1497 struct ib_ucontext {
1498 	struct ib_device       *device;
1499 	struct ib_uverbs_file  *ufile;
1500 	/*
1501 	 * 'closing' can be read by the driver only during a destroy callback,
1502 	 * it is set when we are closing the file descriptor and indicates
1503 	 * that mm_sem may be locked.
1504 	 */
1505 	bool closing;
1506 
1507 	bool cleanup_retryable;
1508 
1509 	void (*invalidate_range)(struct ib_umem_odp *umem_odp,
1510 				 unsigned long start, unsigned long end);
1511 	struct mutex per_mm_list_lock;
1512 	struct list_head per_mm_list;
1513 
1514 	struct ib_rdmacg_object	cg_obj;
1515 	/*
1516 	 * Implementation details of the RDMA core, don't use in drivers:
1517 	 */
1518 	struct rdma_restrack_entry res;
1519 };
1520 
1521 struct ib_uobject {
1522 	u64			user_handle;	/* handle given to us by userspace */
1523 	/* ufile & ucontext owning this object */
1524 	struct ib_uverbs_file  *ufile;
1525 	/* FIXME, save memory: ufile->context == context */
1526 	struct ib_ucontext     *context;	/* associated user context */
1527 	void		       *object;		/* containing object */
1528 	struct list_head	list;		/* link to context's list */
1529 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1530 	int			id;		/* index into kernel idr */
1531 	struct kref		ref;
1532 	atomic_t		usecnt;		/* protects exclusive access */
1533 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1534 
1535 	const struct uverbs_api_object *uapi_object;
1536 };
1537 
1538 struct ib_udata {
1539 	const void __user *inbuf;
1540 	void __user *outbuf;
1541 	size_t       inlen;
1542 	size_t       outlen;
1543 };
1544 
1545 struct ib_pd {
1546 	u32			local_dma_lkey;
1547 	u32			flags;
1548 	struct ib_device       *device;
1549 	struct ib_uobject      *uobject;
1550 	atomic_t          	usecnt; /* count all resources */
1551 
1552 	u32			unsafe_global_rkey;
1553 
1554 	/*
1555 	 * Implementation details of the RDMA core, don't use in drivers:
1556 	 */
1557 	struct ib_mr	       *__internal_mr;
1558 	struct rdma_restrack_entry res;
1559 };
1560 
1561 struct ib_xrcd {
1562 	struct ib_device       *device;
1563 	atomic_t		usecnt; /* count all exposed resources */
1564 	struct inode	       *inode;
1565 
1566 	struct mutex		tgt_qp_mutex;
1567 	struct list_head	tgt_qp_list;
1568 };
1569 
1570 struct ib_ah {
1571 	struct ib_device	*device;
1572 	struct ib_pd		*pd;
1573 	struct ib_uobject	*uobject;
1574 	const struct ib_gid_attr *sgid_attr;
1575 	enum rdma_ah_attr_type	type;
1576 };
1577 
1578 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1579 
1580 enum ib_poll_context {
1581 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1582 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1583 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1584 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1585 };
1586 
1587 struct ib_cq {
1588 	struct ib_device       *device;
1589 	struct ib_uobject      *uobject;
1590 	ib_comp_handler   	comp_handler;
1591 	void                  (*event_handler)(struct ib_event *, void *);
1592 	void                   *cq_context;
1593 	int               	cqe;
1594 	atomic_t          	usecnt; /* count number of work queues */
1595 	enum ib_poll_context	poll_ctx;
1596 	struct ib_wc		*wc;
1597 	union {
1598 		struct irq_poll		iop;
1599 		struct work_struct	work;
1600 	};
1601 	struct workqueue_struct *comp_wq;
1602 	/*
1603 	 * Implementation details of the RDMA core, don't use in drivers:
1604 	 */
1605 	struct rdma_restrack_entry res;
1606 };
1607 
1608 struct ib_srq {
1609 	struct ib_device       *device;
1610 	struct ib_pd	       *pd;
1611 	struct ib_uobject      *uobject;
1612 	void		      (*event_handler)(struct ib_event *, void *);
1613 	void		       *srq_context;
1614 	enum ib_srq_type	srq_type;
1615 	atomic_t		usecnt;
1616 
1617 	struct {
1618 		struct ib_cq   *cq;
1619 		union {
1620 			struct {
1621 				struct ib_xrcd *xrcd;
1622 				u32		srq_num;
1623 			} xrc;
1624 		};
1625 	} ext;
1626 };
1627 
1628 enum ib_raw_packet_caps {
1629 	/* Strip cvlan from incoming packet and report it in the matching work
1630 	 * completion is supported.
1631 	 */
1632 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1633 	/* Scatter FCS field of an incoming packet to host memory is supported.
1634 	 */
1635 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1636 	/* Checksum offloads are supported (for both send and receive). */
1637 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1638 	/* When a packet is received for an RQ with no receive WQEs, the
1639 	 * packet processing is delayed.
1640 	 */
1641 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1642 };
1643 
1644 enum ib_wq_type {
1645 	IB_WQT_RQ
1646 };
1647 
1648 enum ib_wq_state {
1649 	IB_WQS_RESET,
1650 	IB_WQS_RDY,
1651 	IB_WQS_ERR
1652 };
1653 
1654 struct ib_wq {
1655 	struct ib_device       *device;
1656 	struct ib_uobject      *uobject;
1657 	void		    *wq_context;
1658 	void		    (*event_handler)(struct ib_event *, void *);
1659 	struct ib_pd	       *pd;
1660 	struct ib_cq	       *cq;
1661 	u32		wq_num;
1662 	enum ib_wq_state       state;
1663 	enum ib_wq_type	wq_type;
1664 	atomic_t		usecnt;
1665 };
1666 
1667 enum ib_wq_flags {
1668 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1669 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1670 	IB_WQ_FLAGS_DELAY_DROP		= 1 << 2,
1671 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1672 };
1673 
1674 struct ib_wq_init_attr {
1675 	void		       *wq_context;
1676 	enum ib_wq_type	wq_type;
1677 	u32		max_wr;
1678 	u32		max_sge;
1679 	struct	ib_cq	       *cq;
1680 	void		    (*event_handler)(struct ib_event *, void *);
1681 	u32		create_flags; /* Use enum ib_wq_flags */
1682 };
1683 
1684 enum ib_wq_attr_mask {
1685 	IB_WQ_STATE		= 1 << 0,
1686 	IB_WQ_CUR_STATE		= 1 << 1,
1687 	IB_WQ_FLAGS		= 1 << 2,
1688 };
1689 
1690 struct ib_wq_attr {
1691 	enum	ib_wq_state	wq_state;
1692 	enum	ib_wq_state	curr_wq_state;
1693 	u32			flags; /* Use enum ib_wq_flags */
1694 	u32			flags_mask; /* Use enum ib_wq_flags */
1695 };
1696 
1697 struct ib_rwq_ind_table {
1698 	struct ib_device	*device;
1699 	struct ib_uobject      *uobject;
1700 	atomic_t		usecnt;
1701 	u32		ind_tbl_num;
1702 	u32		log_ind_tbl_size;
1703 	struct ib_wq	**ind_tbl;
1704 };
1705 
1706 struct ib_rwq_ind_table_init_attr {
1707 	u32		log_ind_tbl_size;
1708 	/* Each entry is a pointer to Receive Work Queue */
1709 	struct ib_wq	**ind_tbl;
1710 };
1711 
1712 enum port_pkey_state {
1713 	IB_PORT_PKEY_NOT_VALID = 0,
1714 	IB_PORT_PKEY_VALID = 1,
1715 	IB_PORT_PKEY_LISTED = 2,
1716 };
1717 
1718 struct ib_qp_security;
1719 
1720 struct ib_port_pkey {
1721 	enum port_pkey_state	state;
1722 	u16			pkey_index;
1723 	u8			port_num;
1724 	struct list_head	qp_list;
1725 	struct list_head	to_error_list;
1726 	struct ib_qp_security  *sec;
1727 };
1728 
1729 struct ib_ports_pkeys {
1730 	struct ib_port_pkey	main;
1731 	struct ib_port_pkey	alt;
1732 };
1733 
1734 struct ib_qp_security {
1735 	struct ib_qp	       *qp;
1736 	struct ib_device       *dev;
1737 	/* Hold this mutex when changing port and pkey settings. */
1738 	struct mutex		mutex;
1739 	struct ib_ports_pkeys  *ports_pkeys;
1740 	/* A list of all open shared QP handles.  Required to enforce security
1741 	 * properly for all users of a shared QP.
1742 	 */
1743 	struct list_head        shared_qp_list;
1744 	void                   *security;
1745 	bool			destroying;
1746 	atomic_t		error_list_count;
1747 	struct completion	error_complete;
1748 	int			error_comps_pending;
1749 };
1750 
1751 /*
1752  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1753  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1754  */
1755 struct ib_qp {
1756 	struct ib_device       *device;
1757 	struct ib_pd	       *pd;
1758 	struct ib_cq	       *send_cq;
1759 	struct ib_cq	       *recv_cq;
1760 	spinlock_t		mr_lock;
1761 	int			mrs_used;
1762 	struct list_head	rdma_mrs;
1763 	struct list_head	sig_mrs;
1764 	struct ib_srq	       *srq;
1765 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1766 	struct list_head	xrcd_list;
1767 
1768 	/* count times opened, mcast attaches, flow attaches */
1769 	atomic_t		usecnt;
1770 	struct list_head	open_list;
1771 	struct ib_qp           *real_qp;
1772 	struct ib_uobject      *uobject;
1773 	void                  (*event_handler)(struct ib_event *, void *);
1774 	void		       *qp_context;
1775 	/* sgid_attrs associated with the AV's */
1776 	const struct ib_gid_attr *av_sgid_attr;
1777 	const struct ib_gid_attr *alt_path_sgid_attr;
1778 	u32			qp_num;
1779 	u32			max_write_sge;
1780 	u32			max_read_sge;
1781 	enum ib_qp_type		qp_type;
1782 	struct ib_rwq_ind_table *rwq_ind_tbl;
1783 	struct ib_qp_security  *qp_sec;
1784 	u8			port;
1785 
1786 	/*
1787 	 * Implementation details of the RDMA core, don't use in drivers:
1788 	 */
1789 	struct rdma_restrack_entry     res;
1790 };
1791 
1792 struct ib_dm {
1793 	struct ib_device  *device;
1794 	u32		   length;
1795 	u32		   flags;
1796 	struct ib_uobject *uobject;
1797 	atomic_t	   usecnt;
1798 };
1799 
1800 struct ib_mr {
1801 	struct ib_device  *device;
1802 	struct ib_pd	  *pd;
1803 	u32		   lkey;
1804 	u32		   rkey;
1805 	u64		   iova;
1806 	u64		   length;
1807 	unsigned int	   page_size;
1808 	bool		   need_inval;
1809 	union {
1810 		struct ib_uobject	*uobject;	/* user */
1811 		struct list_head	qp_entry;	/* FR */
1812 	};
1813 
1814 	struct ib_dm      *dm;
1815 
1816 	/*
1817 	 * Implementation details of the RDMA core, don't use in drivers:
1818 	 */
1819 	struct rdma_restrack_entry res;
1820 };
1821 
1822 struct ib_mw {
1823 	struct ib_device	*device;
1824 	struct ib_pd		*pd;
1825 	struct ib_uobject	*uobject;
1826 	u32			rkey;
1827 	enum ib_mw_type         type;
1828 };
1829 
1830 struct ib_fmr {
1831 	struct ib_device	*device;
1832 	struct ib_pd		*pd;
1833 	struct list_head	list;
1834 	u32			lkey;
1835 	u32			rkey;
1836 };
1837 
1838 /* Supported steering options */
1839 enum ib_flow_attr_type {
1840 	/* steering according to rule specifications */
1841 	IB_FLOW_ATTR_NORMAL		= 0x0,
1842 	/* default unicast and multicast rule -
1843 	 * receive all Eth traffic which isn't steered to any QP
1844 	 */
1845 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1846 	/* default multicast rule -
1847 	 * receive all Eth multicast traffic which isn't steered to any QP
1848 	 */
1849 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1850 	/* sniffer rule - receive all port traffic */
1851 	IB_FLOW_ATTR_SNIFFER		= 0x3
1852 };
1853 
1854 /* Supported steering header types */
1855 enum ib_flow_spec_type {
1856 	/* L2 headers*/
1857 	IB_FLOW_SPEC_ETH		= 0x20,
1858 	IB_FLOW_SPEC_IB			= 0x22,
1859 	/* L3 header*/
1860 	IB_FLOW_SPEC_IPV4		= 0x30,
1861 	IB_FLOW_SPEC_IPV6		= 0x31,
1862 	IB_FLOW_SPEC_ESP                = 0x34,
1863 	/* L4 headers*/
1864 	IB_FLOW_SPEC_TCP		= 0x40,
1865 	IB_FLOW_SPEC_UDP		= 0x41,
1866 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1867 	IB_FLOW_SPEC_GRE		= 0x51,
1868 	IB_FLOW_SPEC_MPLS		= 0x60,
1869 	IB_FLOW_SPEC_INNER		= 0x100,
1870 	/* Actions */
1871 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1872 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1873 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1874 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1875 };
1876 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1877 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1878 
1879 /* Flow steering rule priority is set according to it's domain.
1880  * Lower domain value means higher priority.
1881  */
1882 enum ib_flow_domain {
1883 	IB_FLOW_DOMAIN_USER,
1884 	IB_FLOW_DOMAIN_ETHTOOL,
1885 	IB_FLOW_DOMAIN_RFS,
1886 	IB_FLOW_DOMAIN_NIC,
1887 	IB_FLOW_DOMAIN_NUM /* Must be last */
1888 };
1889 
1890 enum ib_flow_flags {
1891 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1892 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1893 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1894 };
1895 
1896 struct ib_flow_eth_filter {
1897 	u8	dst_mac[6];
1898 	u8	src_mac[6];
1899 	__be16	ether_type;
1900 	__be16	vlan_tag;
1901 	/* Must be last */
1902 	u8	real_sz[0];
1903 };
1904 
1905 struct ib_flow_spec_eth {
1906 	u32			  type;
1907 	u16			  size;
1908 	struct ib_flow_eth_filter val;
1909 	struct ib_flow_eth_filter mask;
1910 };
1911 
1912 struct ib_flow_ib_filter {
1913 	__be16 dlid;
1914 	__u8   sl;
1915 	/* Must be last */
1916 	u8	real_sz[0];
1917 };
1918 
1919 struct ib_flow_spec_ib {
1920 	u32			 type;
1921 	u16			 size;
1922 	struct ib_flow_ib_filter val;
1923 	struct ib_flow_ib_filter mask;
1924 };
1925 
1926 /* IPv4 header flags */
1927 enum ib_ipv4_flags {
1928 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1929 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1930 				    last have this flag set */
1931 };
1932 
1933 struct ib_flow_ipv4_filter {
1934 	__be32	src_ip;
1935 	__be32	dst_ip;
1936 	u8	proto;
1937 	u8	tos;
1938 	u8	ttl;
1939 	u8	flags;
1940 	/* Must be last */
1941 	u8	real_sz[0];
1942 };
1943 
1944 struct ib_flow_spec_ipv4 {
1945 	u32			   type;
1946 	u16			   size;
1947 	struct ib_flow_ipv4_filter val;
1948 	struct ib_flow_ipv4_filter mask;
1949 };
1950 
1951 struct ib_flow_ipv6_filter {
1952 	u8	src_ip[16];
1953 	u8	dst_ip[16];
1954 	__be32	flow_label;
1955 	u8	next_hdr;
1956 	u8	traffic_class;
1957 	u8	hop_limit;
1958 	/* Must be last */
1959 	u8	real_sz[0];
1960 };
1961 
1962 struct ib_flow_spec_ipv6 {
1963 	u32			   type;
1964 	u16			   size;
1965 	struct ib_flow_ipv6_filter val;
1966 	struct ib_flow_ipv6_filter mask;
1967 };
1968 
1969 struct ib_flow_tcp_udp_filter {
1970 	__be16	dst_port;
1971 	__be16	src_port;
1972 	/* Must be last */
1973 	u8	real_sz[0];
1974 };
1975 
1976 struct ib_flow_spec_tcp_udp {
1977 	u32			      type;
1978 	u16			      size;
1979 	struct ib_flow_tcp_udp_filter val;
1980 	struct ib_flow_tcp_udp_filter mask;
1981 };
1982 
1983 struct ib_flow_tunnel_filter {
1984 	__be32	tunnel_id;
1985 	u8	real_sz[0];
1986 };
1987 
1988 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1989  * the tunnel_id from val has the vni value
1990  */
1991 struct ib_flow_spec_tunnel {
1992 	u32			      type;
1993 	u16			      size;
1994 	struct ib_flow_tunnel_filter  val;
1995 	struct ib_flow_tunnel_filter  mask;
1996 };
1997 
1998 struct ib_flow_esp_filter {
1999 	__be32	spi;
2000 	__be32  seq;
2001 	/* Must be last */
2002 	u8	real_sz[0];
2003 };
2004 
2005 struct ib_flow_spec_esp {
2006 	u32                           type;
2007 	u16			      size;
2008 	struct ib_flow_esp_filter     val;
2009 	struct ib_flow_esp_filter     mask;
2010 };
2011 
2012 struct ib_flow_gre_filter {
2013 	__be16 c_ks_res0_ver;
2014 	__be16 protocol;
2015 	__be32 key;
2016 	/* Must be last */
2017 	u8	real_sz[0];
2018 };
2019 
2020 struct ib_flow_spec_gre {
2021 	u32                           type;
2022 	u16			      size;
2023 	struct ib_flow_gre_filter     val;
2024 	struct ib_flow_gre_filter     mask;
2025 };
2026 
2027 struct ib_flow_mpls_filter {
2028 	__be32 tag;
2029 	/* Must be last */
2030 	u8	real_sz[0];
2031 };
2032 
2033 struct ib_flow_spec_mpls {
2034 	u32                           type;
2035 	u16			      size;
2036 	struct ib_flow_mpls_filter     val;
2037 	struct ib_flow_mpls_filter     mask;
2038 };
2039 
2040 struct ib_flow_spec_action_tag {
2041 	enum ib_flow_spec_type	      type;
2042 	u16			      size;
2043 	u32                           tag_id;
2044 };
2045 
2046 struct ib_flow_spec_action_drop {
2047 	enum ib_flow_spec_type	      type;
2048 	u16			      size;
2049 };
2050 
2051 struct ib_flow_spec_action_handle {
2052 	enum ib_flow_spec_type	      type;
2053 	u16			      size;
2054 	struct ib_flow_action	     *act;
2055 };
2056 
2057 enum ib_counters_description {
2058 	IB_COUNTER_PACKETS,
2059 	IB_COUNTER_BYTES,
2060 };
2061 
2062 struct ib_flow_spec_action_count {
2063 	enum ib_flow_spec_type type;
2064 	u16 size;
2065 	struct ib_counters *counters;
2066 };
2067 
2068 union ib_flow_spec {
2069 	struct {
2070 		u32			type;
2071 		u16			size;
2072 	};
2073 	struct ib_flow_spec_eth		eth;
2074 	struct ib_flow_spec_ib		ib;
2075 	struct ib_flow_spec_ipv4        ipv4;
2076 	struct ib_flow_spec_tcp_udp	tcp_udp;
2077 	struct ib_flow_spec_ipv6        ipv6;
2078 	struct ib_flow_spec_tunnel      tunnel;
2079 	struct ib_flow_spec_esp		esp;
2080 	struct ib_flow_spec_gre		gre;
2081 	struct ib_flow_spec_mpls	mpls;
2082 	struct ib_flow_spec_action_tag  flow_tag;
2083 	struct ib_flow_spec_action_drop drop;
2084 	struct ib_flow_spec_action_handle action;
2085 	struct ib_flow_spec_action_count flow_count;
2086 };
2087 
2088 struct ib_flow_attr {
2089 	enum ib_flow_attr_type type;
2090 	u16	     size;
2091 	u16	     priority;
2092 	u32	     flags;
2093 	u8	     num_of_specs;
2094 	u8	     port;
2095 	union ib_flow_spec flows[];
2096 };
2097 
2098 struct ib_flow {
2099 	struct ib_qp		*qp;
2100 	struct ib_device	*device;
2101 	struct ib_uobject	*uobject;
2102 };
2103 
2104 enum ib_flow_action_type {
2105 	IB_FLOW_ACTION_UNSPECIFIED,
2106 	IB_FLOW_ACTION_ESP = 1,
2107 };
2108 
2109 struct ib_flow_action_attrs_esp_keymats {
2110 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2111 	union {
2112 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2113 	} keymat;
2114 };
2115 
2116 struct ib_flow_action_attrs_esp_replays {
2117 	enum ib_uverbs_flow_action_esp_replay			protocol;
2118 	union {
2119 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2120 	} replay;
2121 };
2122 
2123 enum ib_flow_action_attrs_esp_flags {
2124 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2125 	 * This is done in order to share the same flags between user-space and
2126 	 * kernel and spare an unnecessary translation.
2127 	 */
2128 
2129 	/* Kernel flags */
2130 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2131 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2132 };
2133 
2134 struct ib_flow_spec_list {
2135 	struct ib_flow_spec_list	*next;
2136 	union ib_flow_spec		spec;
2137 };
2138 
2139 struct ib_flow_action_attrs_esp {
2140 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2141 	struct ib_flow_action_attrs_esp_replays		*replay;
2142 	struct ib_flow_spec_list			*encap;
2143 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2144 	 * Value of 0 is a valid value.
2145 	 */
2146 	u32						esn;
2147 	u32						spi;
2148 	u32						seq;
2149 	u32						tfc_pad;
2150 	/* Use enum ib_flow_action_attrs_esp_flags */
2151 	u64						flags;
2152 	u64						hard_limit_pkts;
2153 };
2154 
2155 struct ib_flow_action {
2156 	struct ib_device		*device;
2157 	struct ib_uobject		*uobject;
2158 	enum ib_flow_action_type	type;
2159 	atomic_t			usecnt;
2160 };
2161 
2162 struct ib_mad_hdr;
2163 struct ib_grh;
2164 
2165 enum ib_process_mad_flags {
2166 	IB_MAD_IGNORE_MKEY	= 1,
2167 	IB_MAD_IGNORE_BKEY	= 2,
2168 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2169 };
2170 
2171 enum ib_mad_result {
2172 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2173 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2174 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2175 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2176 };
2177 
2178 struct ib_port_cache {
2179 	u64		      subnet_prefix;
2180 	struct ib_pkey_cache  *pkey;
2181 	struct ib_gid_table   *gid;
2182 	u8                     lmc;
2183 	enum ib_port_state     port_state;
2184 };
2185 
2186 struct ib_cache {
2187 	rwlock_t                lock;
2188 	struct ib_event_handler event_handler;
2189 	struct ib_port_cache   *ports;
2190 };
2191 
2192 struct iw_cm_verbs;
2193 
2194 struct ib_port_immutable {
2195 	int                           pkey_tbl_len;
2196 	int                           gid_tbl_len;
2197 	u32                           core_cap_flags;
2198 	u32                           max_mad_size;
2199 };
2200 
2201 /* rdma netdev type - specifies protocol type */
2202 enum rdma_netdev_t {
2203 	RDMA_NETDEV_OPA_VNIC,
2204 	RDMA_NETDEV_IPOIB,
2205 };
2206 
2207 /**
2208  * struct rdma_netdev - rdma netdev
2209  * For cases where netstack interfacing is required.
2210  */
2211 struct rdma_netdev {
2212 	void              *clnt_priv;
2213 	struct ib_device  *hca;
2214 	u8                 port_num;
2215 
2216 	/*
2217 	 * cleanup function must be specified.
2218 	 * FIXME: This is only used for OPA_VNIC and that usage should be
2219 	 * removed too.
2220 	 */
2221 	void (*free_rdma_netdev)(struct net_device *netdev);
2222 
2223 	/* control functions */
2224 	void (*set_id)(struct net_device *netdev, int id);
2225 	/* send packet */
2226 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2227 		    struct ib_ah *address, u32 dqpn);
2228 	/* multicast */
2229 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2230 			    union ib_gid *gid, u16 mlid,
2231 			    int set_qkey, u32 qkey);
2232 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2233 			    union ib_gid *gid, u16 mlid);
2234 };
2235 
2236 struct rdma_netdev_alloc_params {
2237 	size_t sizeof_priv;
2238 	unsigned int txqs;
2239 	unsigned int rxqs;
2240 	void *param;
2241 
2242 	int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2243 				      struct net_device *netdev, void *param);
2244 };
2245 
2246 struct ib_port_pkey_list {
2247 	/* Lock to hold while modifying the list. */
2248 	spinlock_t                    list_lock;
2249 	struct list_head              pkey_list;
2250 };
2251 
2252 struct ib_counters {
2253 	struct ib_device	*device;
2254 	struct ib_uobject	*uobject;
2255 	/* num of objects attached */
2256 	atomic_t	usecnt;
2257 };
2258 
2259 struct ib_counters_read_attr {
2260 	u64	*counters_buff;
2261 	u32	ncounters;
2262 	u32	flags; /* use enum ib_read_counters_flags */
2263 };
2264 
2265 struct uverbs_attr_bundle;
2266 
2267 /**
2268  * struct ib_device_ops - InfiniBand device operations
2269  * This structure defines all the InfiniBand device operations, providers will
2270  * need to define the supported operations, otherwise they will be set to null.
2271  */
2272 struct ib_device_ops {
2273 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2274 			 const struct ib_send_wr **bad_send_wr);
2275 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2276 			 const struct ib_recv_wr **bad_recv_wr);
2277 	void (*drain_rq)(struct ib_qp *qp);
2278 	void (*drain_sq)(struct ib_qp *qp);
2279 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2280 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2281 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2282 	int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2283 	int (*post_srq_recv)(struct ib_srq *srq,
2284 			     const struct ib_recv_wr *recv_wr,
2285 			     const struct ib_recv_wr **bad_recv_wr);
2286 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2287 			   u8 port_num, const struct ib_wc *in_wc,
2288 			   const struct ib_grh *in_grh,
2289 			   const struct ib_mad_hdr *in_mad, size_t in_mad_size,
2290 			   struct ib_mad_hdr *out_mad, size_t *out_mad_size,
2291 			   u16 *out_mad_pkey_index);
2292 	int (*query_device)(struct ib_device *device,
2293 			    struct ib_device_attr *device_attr,
2294 			    struct ib_udata *udata);
2295 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2296 			     struct ib_device_modify *device_modify);
2297 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2298 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2299 						     int comp_vector);
2300 	int (*query_port)(struct ib_device *device, u8 port_num,
2301 			  struct ib_port_attr *port_attr);
2302 	int (*modify_port)(struct ib_device *device, u8 port_num,
2303 			   int port_modify_mask,
2304 			   struct ib_port_modify *port_modify);
2305 	/**
2306 	 * The following mandatory functions are used only at device
2307 	 * registration.  Keep functions such as these at the end of this
2308 	 * structure to avoid cache line misses when accessing struct ib_device
2309 	 * in fast paths.
2310 	 */
2311 	int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2312 				  struct ib_port_immutable *immutable);
2313 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2314 					       u8 port_num);
2315 	/**
2316 	 * When calling get_netdev, the HW vendor's driver should return the
2317 	 * net device of device @device at port @port_num or NULL if such
2318 	 * a net device doesn't exist. The vendor driver should call dev_hold
2319 	 * on this net device. The HW vendor's device driver must guarantee
2320 	 * that this function returns NULL before the net device has finished
2321 	 * NETDEV_UNREGISTER state.
2322 	 */
2323 	struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2324 	/**
2325 	 * rdma netdev operation
2326 	 *
2327 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2328 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2329 	 */
2330 	struct net_device *(*alloc_rdma_netdev)(
2331 		struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2332 		const char *name, unsigned char name_assign_type,
2333 		void (*setup)(struct net_device *));
2334 
2335 	int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2336 				      enum rdma_netdev_t type,
2337 				      struct rdma_netdev_alloc_params *params);
2338 	/**
2339 	 * query_gid should be return GID value for @device, when @port_num
2340 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2341 	 * is RoCE link layer.
2342 	 */
2343 	int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2344 			 union ib_gid *gid);
2345 	/**
2346 	 * When calling add_gid, the HW vendor's driver should add the gid
2347 	 * of device of port at gid index available at @attr. Meta-info of
2348 	 * that gid (for example, the network device related to this gid) is
2349 	 * available at @attr. @context allows the HW vendor driver to store
2350 	 * extra information together with a GID entry. The HW vendor driver may
2351 	 * allocate memory to contain this information and store it in @context
2352 	 * when a new GID entry is written to. Params are consistent until the
2353 	 * next call of add_gid or delete_gid. The function should return 0 on
2354 	 * success or error otherwise. The function could be called
2355 	 * concurrently for different ports. This function is only called when
2356 	 * roce_gid_table is used.
2357 	 */
2358 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2359 	/**
2360 	 * When calling del_gid, the HW vendor's driver should delete the
2361 	 * gid of device @device at gid index gid_index of port port_num
2362 	 * available in @attr.
2363 	 * Upon the deletion of a GID entry, the HW vendor must free any
2364 	 * allocated memory. The caller will clear @context afterwards.
2365 	 * This function is only called when roce_gid_table is used.
2366 	 */
2367 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2368 	int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2369 			  u16 *pkey);
2370 	struct ib_ucontext *(*alloc_ucontext)(struct ib_device *device,
2371 					      struct ib_udata *udata);
2372 	int (*dealloc_ucontext)(struct ib_ucontext *context);
2373 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2374 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2375 	struct ib_pd *(*alloc_pd)(struct ib_device *device,
2376 				  struct ib_ucontext *context,
2377 				  struct ib_udata *udata);
2378 	int (*dealloc_pd)(struct ib_pd *pd);
2379 	struct ib_ah *(*create_ah)(struct ib_pd *pd,
2380 				   struct rdma_ah_attr *ah_attr, u32 flags,
2381 				   struct ib_udata *udata);
2382 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2383 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2384 	int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2385 	struct ib_srq *(*create_srq)(struct ib_pd *pd,
2386 				     struct ib_srq_init_attr *srq_init_attr,
2387 				     struct ib_udata *udata);
2388 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2389 			  enum ib_srq_attr_mask srq_attr_mask,
2390 			  struct ib_udata *udata);
2391 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2392 	int (*destroy_srq)(struct ib_srq *srq);
2393 	struct ib_qp *(*create_qp)(struct ib_pd *pd,
2394 				   struct ib_qp_init_attr *qp_init_attr,
2395 				   struct ib_udata *udata);
2396 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2397 			 int qp_attr_mask, struct ib_udata *udata);
2398 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2399 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2400 	int (*destroy_qp)(struct ib_qp *qp);
2401 	struct ib_cq *(*create_cq)(struct ib_device *device,
2402 				   const struct ib_cq_init_attr *attr,
2403 				   struct ib_ucontext *context,
2404 				   struct ib_udata *udata);
2405 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2406 	int (*destroy_cq)(struct ib_cq *cq);
2407 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2408 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2409 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2410 				     u64 virt_addr, int mr_access_flags,
2411 				     struct ib_udata *udata);
2412 	int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2413 			     u64 virt_addr, int mr_access_flags,
2414 			     struct ib_pd *pd, struct ib_udata *udata);
2415 	int (*dereg_mr)(struct ib_mr *mr);
2416 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2417 				  u32 max_num_sg);
2418 	int (*advise_mr)(struct ib_pd *pd,
2419 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2420 			 struct ib_sge *sg_list, u32 num_sge,
2421 			 struct uverbs_attr_bundle *attrs);
2422 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2423 			 unsigned int *sg_offset);
2424 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2425 			       struct ib_mr_status *mr_status);
2426 	struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2427 				  struct ib_udata *udata);
2428 	int (*dealloc_mw)(struct ib_mw *mw);
2429 	struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2430 				    struct ib_fmr_attr *fmr_attr);
2431 	int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2432 			    u64 iova);
2433 	int (*unmap_fmr)(struct list_head *fmr_list);
2434 	int (*dealloc_fmr)(struct ib_fmr *fmr);
2435 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2436 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2437 	struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2438 				      struct ib_ucontext *ucontext,
2439 				      struct ib_udata *udata);
2440 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2441 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2442 				       struct ib_flow_attr *flow_attr,
2443 				       int domain, struct ib_udata *udata);
2444 	int (*destroy_flow)(struct ib_flow *flow_id);
2445 	struct ib_flow_action *(*create_flow_action_esp)(
2446 		struct ib_device *device,
2447 		const struct ib_flow_action_attrs_esp *attr,
2448 		struct uverbs_attr_bundle *attrs);
2449 	int (*destroy_flow_action)(struct ib_flow_action *action);
2450 	int (*modify_flow_action_esp)(
2451 		struct ib_flow_action *action,
2452 		const struct ib_flow_action_attrs_esp *attr,
2453 		struct uverbs_attr_bundle *attrs);
2454 	int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2455 				 int state);
2456 	int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2457 			     struct ifla_vf_info *ivf);
2458 	int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2459 			    struct ifla_vf_stats *stats);
2460 	int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2461 			   int type);
2462 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2463 				   struct ib_wq_init_attr *init_attr,
2464 				   struct ib_udata *udata);
2465 	int (*destroy_wq)(struct ib_wq *wq);
2466 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2467 			 u32 wq_attr_mask, struct ib_udata *udata);
2468 	struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2469 		struct ib_device *device,
2470 		struct ib_rwq_ind_table_init_attr *init_attr,
2471 		struct ib_udata *udata);
2472 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2473 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2474 				  struct ib_ucontext *context,
2475 				  struct ib_dm_alloc_attr *attr,
2476 				  struct uverbs_attr_bundle *attrs);
2477 	int (*dealloc_dm)(struct ib_dm *dm);
2478 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2479 				   struct ib_dm_mr_attr *attr,
2480 				   struct uverbs_attr_bundle *attrs);
2481 	struct ib_counters *(*create_counters)(
2482 		struct ib_device *device, struct uverbs_attr_bundle *attrs);
2483 	int (*destroy_counters)(struct ib_counters *counters);
2484 	int (*read_counters)(struct ib_counters *counters,
2485 			     struct ib_counters_read_attr *counters_read_attr,
2486 			     struct uverbs_attr_bundle *attrs);
2487 	/**
2488 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2489 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2490 	 *   core when the device is removed.  A lifespan of -1 in the return
2491 	 *   struct tells the core to set a default lifespan.
2492 	 */
2493 	struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2494 						u8 port_num);
2495 	/**
2496 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2497 	 * @index - The index in the value array we wish to have updated, or
2498 	 *   num_counters if we want all stats updated
2499 	 * Return codes -
2500 	 *   < 0 - Error, no counters updated
2501 	 *   index - Updated the single counter pointed to by index
2502 	 *   num_counters - Updated all counters (will reset the timestamp
2503 	 *     and prevent further calls for lifespan milliseconds)
2504 	 * Drivers are allowed to update all counters in leiu of just the
2505 	 *   one given in index at their option
2506 	 */
2507 	int (*get_hw_stats)(struct ib_device *device,
2508 			    struct rdma_hw_stats *stats, u8 port, int index);
2509 	/*
2510 	 * This function is called once for each port when a ib device is
2511 	 * registered.
2512 	 */
2513 	int (*init_port)(struct ib_device *device, u8 port_num,
2514 			 struct kobject *port_sysfs);
2515 	/**
2516 	 * Allows rdma drivers to add their own restrack attributes.
2517 	 */
2518 	int (*fill_res_entry)(struct sk_buff *msg,
2519 			      struct rdma_restrack_entry *entry);
2520 };
2521 
2522 struct ib_device {
2523 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2524 	struct device                *dma_device;
2525 	struct ib_device_ops	     ops;
2526 	char                          name[IB_DEVICE_NAME_MAX];
2527 
2528 	struct list_head              event_handler_list;
2529 	spinlock_t                    event_handler_lock;
2530 
2531 	rwlock_t			client_data_lock;
2532 	struct list_head              core_list;
2533 	/* Access to the client_data_list is protected by the client_data_lock
2534 	 * rwlock and the lists_rwsem read-write semaphore
2535 	 */
2536 	struct list_head              client_data_list;
2537 
2538 	struct ib_cache               cache;
2539 	/**
2540 	 * port_immutable is indexed by port number
2541 	 */
2542 	struct ib_port_immutable     *port_immutable;
2543 
2544 	int			      num_comp_vectors;
2545 
2546 	struct ib_port_pkey_list     *port_pkey_list;
2547 
2548 	struct iw_cm_verbs	     *iwcm;
2549 
2550 	struct module               *owner;
2551 	struct device                dev;
2552 	/* First group for device attributes,
2553 	 * Second group for driver provided attributes (optional).
2554 	 * It is NULL terminated array.
2555 	 */
2556 	const struct attribute_group	*groups[3];
2557 
2558 	struct kobject			*ports_kobj;
2559 	struct list_head             port_list;
2560 
2561 	enum {
2562 		IB_DEV_UNINITIALIZED,
2563 		IB_DEV_REGISTERED,
2564 		IB_DEV_UNREGISTERED
2565 	}                            reg_state;
2566 
2567 	int			     uverbs_abi_ver;
2568 	u64			     uverbs_cmd_mask;
2569 	u64			     uverbs_ex_cmd_mask;
2570 
2571 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2572 	__be64			     node_guid;
2573 	u32			     local_dma_lkey;
2574 	u16                          is_switch:1;
2575 	/* Indicates kernel verbs support, should not be used in drivers */
2576 	u16                          kverbs_provider:1;
2577 	u8                           node_type;
2578 	u8                           phys_port_cnt;
2579 	struct ib_device_attr        attrs;
2580 	struct attribute_group	     *hw_stats_ag;
2581 	struct rdma_hw_stats         *hw_stats;
2582 
2583 #ifdef CONFIG_CGROUP_RDMA
2584 	struct rdmacg_device         cg_device;
2585 #endif
2586 
2587 	u32                          index;
2588 	/*
2589 	 * Implementation details of the RDMA core, don't use in drivers
2590 	 */
2591 	struct rdma_restrack_root     res;
2592 
2593 	const struct uapi_definition   *driver_def;
2594 	enum rdma_driver_id		driver_id;
2595 
2596 	/*
2597 	 * Positive refcount indicates that the device is currently
2598 	 * registered and cannot be unregistered.
2599 	 */
2600 	refcount_t refcount;
2601 	struct completion unreg_completion;
2602 };
2603 
2604 struct ib_client {
2605 	char  *name;
2606 	void (*add)   (struct ib_device *);
2607 	void (*remove)(struct ib_device *, void *client_data);
2608 
2609 	/* Returns the net_dev belonging to this ib_client and matching the
2610 	 * given parameters.
2611 	 * @dev:	 An RDMA device that the net_dev use for communication.
2612 	 * @port:	 A physical port number on the RDMA device.
2613 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2614 	 * @gid:	 A GID that the net_dev uses to communicate.
2615 	 * @addr:	 An IP address the net_dev is configured with.
2616 	 * @client_data: The device's client data set by ib_set_client_data().
2617 	 *
2618 	 * An ib_client that implements a net_dev on top of RDMA devices
2619 	 * (such as IP over IB) should implement this callback, allowing the
2620 	 * rdma_cm module to find the right net_dev for a given request.
2621 	 *
2622 	 * The caller is responsible for calling dev_put on the returned
2623 	 * netdev. */
2624 	struct net_device *(*get_net_dev_by_params)(
2625 			struct ib_device *dev,
2626 			u8 port,
2627 			u16 pkey,
2628 			const union ib_gid *gid,
2629 			const struct sockaddr *addr,
2630 			void *client_data);
2631 	struct list_head list;
2632 
2633 	/* kverbs are not required by the client */
2634 	u8 no_kverbs_req:1;
2635 };
2636 
2637 struct ib_device *_ib_alloc_device(size_t size);
2638 #define ib_alloc_device(drv_struct, member)                                    \
2639 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2640 				      BUILD_BUG_ON_ZERO(offsetof(              \
2641 					      struct drv_struct, member))),    \
2642 		     struct drv_struct, member)
2643 
2644 void ib_dealloc_device(struct ib_device *device);
2645 
2646 void ib_get_device_fw_str(struct ib_device *device, char *str);
2647 
2648 int ib_register_device(struct ib_device *device, const char *name);
2649 void ib_unregister_device(struct ib_device *device);
2650 
2651 int ib_register_client   (struct ib_client *client);
2652 void ib_unregister_client(struct ib_client *client);
2653 
2654 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2655 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2656 			 void *data);
2657 void ib_set_device_ops(struct ib_device *device,
2658 		       const struct ib_device_ops *ops);
2659 
2660 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2661 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2662 		      unsigned long pfn, unsigned long size, pgprot_t prot);
2663 int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2664 			struct vm_area_struct *vma, struct page *page,
2665 			unsigned long size);
2666 #else
2667 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
2668 				    struct vm_area_struct *vma,
2669 				    unsigned long pfn, unsigned long size,
2670 				    pgprot_t prot)
2671 {
2672 	return -EINVAL;
2673 }
2674 static inline int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2675 				struct vm_area_struct *vma, struct page *page,
2676 				unsigned long size)
2677 {
2678 	return -EINVAL;
2679 }
2680 #endif
2681 
2682 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2683 {
2684 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2685 }
2686 
2687 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2688 {
2689 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2690 }
2691 
2692 static inline bool ib_is_buffer_cleared(const void __user *p,
2693 					size_t len)
2694 {
2695 	bool ret;
2696 	u8 *buf;
2697 
2698 	if (len > USHRT_MAX)
2699 		return false;
2700 
2701 	buf = memdup_user(p, len);
2702 	if (IS_ERR(buf))
2703 		return false;
2704 
2705 	ret = !memchr_inv(buf, 0, len);
2706 	kfree(buf);
2707 	return ret;
2708 }
2709 
2710 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2711 				       size_t offset,
2712 				       size_t len)
2713 {
2714 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
2715 }
2716 
2717 /**
2718  * ib_is_destroy_retryable - Check whether the uobject destruction
2719  * is retryable.
2720  * @ret: The initial destruction return code
2721  * @why: remove reason
2722  * @uobj: The uobject that is destroyed
2723  *
2724  * This function is a helper function that IB layer and low-level drivers
2725  * can use to consider whether the destruction of the given uobject is
2726  * retry-able.
2727  * It checks the original return code, if it wasn't success the destruction
2728  * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2729  * the remove reason. (i.e. why).
2730  * Must be called with the object locked for destroy.
2731  */
2732 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2733 					   struct ib_uobject *uobj)
2734 {
2735 	return ret && (why == RDMA_REMOVE_DESTROY ||
2736 		       uobj->context->cleanup_retryable);
2737 }
2738 
2739 /**
2740  * ib_destroy_usecnt - Called during destruction to check the usecnt
2741  * @usecnt: The usecnt atomic
2742  * @why: remove reason
2743  * @uobj: The uobject that is destroyed
2744  *
2745  * Non-zero usecnts will block destruction unless destruction was triggered by
2746  * a ucontext cleanup.
2747  */
2748 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2749 				    enum rdma_remove_reason why,
2750 				    struct ib_uobject *uobj)
2751 {
2752 	if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2753 		return -EBUSY;
2754 	return 0;
2755 }
2756 
2757 /**
2758  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2759  * contains all required attributes and no attributes not allowed for
2760  * the given QP state transition.
2761  * @cur_state: Current QP state
2762  * @next_state: Next QP state
2763  * @type: QP type
2764  * @mask: Mask of supplied QP attributes
2765  *
2766  * This function is a helper function that a low-level driver's
2767  * modify_qp method can use to validate the consumer's input.  It
2768  * checks that cur_state and next_state are valid QP states, that a
2769  * transition from cur_state to next_state is allowed by the IB spec,
2770  * and that the attribute mask supplied is allowed for the transition.
2771  */
2772 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2773 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
2774 
2775 void ib_register_event_handler(struct ib_event_handler *event_handler);
2776 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2777 void ib_dispatch_event(struct ib_event *event);
2778 
2779 int ib_query_port(struct ib_device *device,
2780 		  u8 port_num, struct ib_port_attr *port_attr);
2781 
2782 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2783 					       u8 port_num);
2784 
2785 /**
2786  * rdma_cap_ib_switch - Check if the device is IB switch
2787  * @device: Device to check
2788  *
2789  * Device driver is responsible for setting is_switch bit on
2790  * in ib_device structure at init time.
2791  *
2792  * Return: true if the device is IB switch.
2793  */
2794 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2795 {
2796 	return device->is_switch;
2797 }
2798 
2799 /**
2800  * rdma_start_port - Return the first valid port number for the device
2801  * specified
2802  *
2803  * @device: Device to be checked
2804  *
2805  * Return start port number
2806  */
2807 static inline u8 rdma_start_port(const struct ib_device *device)
2808 {
2809 	return rdma_cap_ib_switch(device) ? 0 : 1;
2810 }
2811 
2812 /**
2813  * rdma_end_port - Return the last valid port number for the device
2814  * specified
2815  *
2816  * @device: Device to be checked
2817  *
2818  * Return last port number
2819  */
2820 static inline u8 rdma_end_port(const struct ib_device *device)
2821 {
2822 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2823 }
2824 
2825 static inline int rdma_is_port_valid(const struct ib_device *device,
2826 				     unsigned int port)
2827 {
2828 	return (port >= rdma_start_port(device) &&
2829 		port <= rdma_end_port(device));
2830 }
2831 
2832 static inline bool rdma_is_grh_required(const struct ib_device *device,
2833 					u8 port_num)
2834 {
2835 	return device->port_immutable[port_num].core_cap_flags &
2836 		RDMA_CORE_PORT_IB_GRH_REQUIRED;
2837 }
2838 
2839 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2840 {
2841 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2842 }
2843 
2844 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2845 {
2846 	return device->port_immutable[port_num].core_cap_flags &
2847 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2848 }
2849 
2850 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2851 {
2852 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2853 }
2854 
2855 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2856 {
2857 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2858 }
2859 
2860 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2861 {
2862 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2863 }
2864 
2865 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2866 {
2867 	return rdma_protocol_ib(device, port_num) ||
2868 		rdma_protocol_roce(device, port_num);
2869 }
2870 
2871 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2872 {
2873 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2874 }
2875 
2876 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2877 {
2878 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2879 }
2880 
2881 /**
2882  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2883  * Management Datagrams.
2884  * @device: Device to check
2885  * @port_num: Port number to check
2886  *
2887  * Management Datagrams (MAD) are a required part of the InfiniBand
2888  * specification and are supported on all InfiniBand devices.  A slightly
2889  * extended version are also supported on OPA interfaces.
2890  *
2891  * Return: true if the port supports sending/receiving of MAD packets.
2892  */
2893 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2894 {
2895 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2896 }
2897 
2898 /**
2899  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2900  * Management Datagrams.
2901  * @device: Device to check
2902  * @port_num: Port number to check
2903  *
2904  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2905  * datagrams with their own versions.  These OPA MADs share many but not all of
2906  * the characteristics of InfiniBand MADs.
2907  *
2908  * OPA MADs differ in the following ways:
2909  *
2910  *    1) MADs are variable size up to 2K
2911  *       IBTA defined MADs remain fixed at 256 bytes
2912  *    2) OPA SMPs must carry valid PKeys
2913  *    3) OPA SMP packets are a different format
2914  *
2915  * Return: true if the port supports OPA MAD packet formats.
2916  */
2917 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2918 {
2919 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2920 		== RDMA_CORE_CAP_OPA_MAD;
2921 }
2922 
2923 /**
2924  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2925  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2926  * @device: Device to check
2927  * @port_num: Port number to check
2928  *
2929  * Each InfiniBand node is required to provide a Subnet Management Agent
2930  * that the subnet manager can access.  Prior to the fabric being fully
2931  * configured by the subnet manager, the SMA is accessed via a well known
2932  * interface called the Subnet Management Interface (SMI).  This interface
2933  * uses directed route packets to communicate with the SM to get around the
2934  * chicken and egg problem of the SM needing to know what's on the fabric
2935  * in order to configure the fabric, and needing to configure the fabric in
2936  * order to send packets to the devices on the fabric.  These directed
2937  * route packets do not need the fabric fully configured in order to reach
2938  * their destination.  The SMI is the only method allowed to send
2939  * directed route packets on an InfiniBand fabric.
2940  *
2941  * Return: true if the port provides an SMI.
2942  */
2943 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2944 {
2945 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2946 }
2947 
2948 /**
2949  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2950  * Communication Manager.
2951  * @device: Device to check
2952  * @port_num: Port number to check
2953  *
2954  * The InfiniBand Communication Manager is one of many pre-defined General
2955  * Service Agents (GSA) that are accessed via the General Service
2956  * Interface (GSI).  It's role is to facilitate establishment of connections
2957  * between nodes as well as other management related tasks for established
2958  * connections.
2959  *
2960  * Return: true if the port supports an IB CM (this does not guarantee that
2961  * a CM is actually running however).
2962  */
2963 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2964 {
2965 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2966 }
2967 
2968 /**
2969  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2970  * Communication Manager.
2971  * @device: Device to check
2972  * @port_num: Port number to check
2973  *
2974  * Similar to above, but specific to iWARP connections which have a different
2975  * managment protocol than InfiniBand.
2976  *
2977  * Return: true if the port supports an iWARP CM (this does not guarantee that
2978  * a CM is actually running however).
2979  */
2980 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2981 {
2982 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2983 }
2984 
2985 /**
2986  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2987  * Subnet Administration.
2988  * @device: Device to check
2989  * @port_num: Port number to check
2990  *
2991  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2992  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2993  * fabrics, devices should resolve routes to other hosts by contacting the
2994  * SA to query the proper route.
2995  *
2996  * Return: true if the port should act as a client to the fabric Subnet
2997  * Administration interface.  This does not imply that the SA service is
2998  * running locally.
2999  */
3000 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3001 {
3002 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
3003 }
3004 
3005 /**
3006  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3007  * Multicast.
3008  * @device: Device to check
3009  * @port_num: Port number to check
3010  *
3011  * InfiniBand multicast registration is more complex than normal IPv4 or
3012  * IPv6 multicast registration.  Each Host Channel Adapter must register
3013  * with the Subnet Manager when it wishes to join a multicast group.  It
3014  * should do so only once regardless of how many queue pairs it subscribes
3015  * to this group.  And it should leave the group only after all queue pairs
3016  * attached to the group have been detached.
3017  *
3018  * Return: true if the port must undertake the additional adminstrative
3019  * overhead of registering/unregistering with the SM and tracking of the
3020  * total number of queue pairs attached to the multicast group.
3021  */
3022 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3023 {
3024 	return rdma_cap_ib_sa(device, port_num);
3025 }
3026 
3027 /**
3028  * rdma_cap_af_ib - Check if the port of device has the capability
3029  * Native Infiniband Address.
3030  * @device: Device to check
3031  * @port_num: Port number to check
3032  *
3033  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3034  * GID.  RoCE uses a different mechanism, but still generates a GID via
3035  * a prescribed mechanism and port specific data.
3036  *
3037  * Return: true if the port uses a GID address to identify devices on the
3038  * network.
3039  */
3040 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3041 {
3042 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
3043 }
3044 
3045 /**
3046  * rdma_cap_eth_ah - Check if the port of device has the capability
3047  * Ethernet Address Handle.
3048  * @device: Device to check
3049  * @port_num: Port number to check
3050  *
3051  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3052  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3053  * port.  Normally, packet headers are generated by the sending host
3054  * adapter, but when sending connectionless datagrams, we must manually
3055  * inject the proper headers for the fabric we are communicating over.
3056  *
3057  * Return: true if we are running as a RoCE port and must force the
3058  * addition of a Global Route Header built from our Ethernet Address
3059  * Handle into our header list for connectionless packets.
3060  */
3061 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3062 {
3063 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
3064 }
3065 
3066 /**
3067  * rdma_cap_opa_ah - Check if the port of device supports
3068  * OPA Address handles
3069  * @device: Device to check
3070  * @port_num: Port number to check
3071  *
3072  * Return: true if we are running on an OPA device which supports
3073  * the extended OPA addressing.
3074  */
3075 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3076 {
3077 	return (device->port_immutable[port_num].core_cap_flags &
3078 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3079 }
3080 
3081 /**
3082  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3083  *
3084  * @device: Device
3085  * @port_num: Port number
3086  *
3087  * This MAD size includes the MAD headers and MAD payload.  No other headers
3088  * are included.
3089  *
3090  * Return the max MAD size required by the Port.  Will return 0 if the port
3091  * does not support MADs
3092  */
3093 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3094 {
3095 	return device->port_immutable[port_num].max_mad_size;
3096 }
3097 
3098 /**
3099  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3100  * @device: Device to check
3101  * @port_num: Port number to check
3102  *
3103  * RoCE GID table mechanism manages the various GIDs for a device.
3104  *
3105  * NOTE: if allocating the port's GID table has failed, this call will still
3106  * return true, but any RoCE GID table API will fail.
3107  *
3108  * Return: true if the port uses RoCE GID table mechanism in order to manage
3109  * its GIDs.
3110  */
3111 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3112 					   u8 port_num)
3113 {
3114 	return rdma_protocol_roce(device, port_num) &&
3115 		device->ops.add_gid && device->ops.del_gid;
3116 }
3117 
3118 /*
3119  * Check if the device supports READ W/ INVALIDATE.
3120  */
3121 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3122 {
3123 	/*
3124 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3125 	 * has support for it yet.
3126 	 */
3127 	return rdma_protocol_iwarp(dev, port_num);
3128 }
3129 
3130 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3131 			 int state);
3132 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3133 		     struct ifla_vf_info *info);
3134 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3135 		    struct ifla_vf_stats *stats);
3136 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3137 		   int type);
3138 
3139 int ib_query_pkey(struct ib_device *device,
3140 		  u8 port_num, u16 index, u16 *pkey);
3141 
3142 int ib_modify_device(struct ib_device *device,
3143 		     int device_modify_mask,
3144 		     struct ib_device_modify *device_modify);
3145 
3146 int ib_modify_port(struct ib_device *device,
3147 		   u8 port_num, int port_modify_mask,
3148 		   struct ib_port_modify *port_modify);
3149 
3150 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3151 		u8 *port_num, u16 *index);
3152 
3153 int ib_find_pkey(struct ib_device *device,
3154 		 u8 port_num, u16 pkey, u16 *index);
3155 
3156 enum ib_pd_flags {
3157 	/*
3158 	 * Create a memory registration for all memory in the system and place
3159 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3160 	 * ULPs to avoid the overhead of dynamic MRs.
3161 	 *
3162 	 * This flag is generally considered unsafe and must only be used in
3163 	 * extremly trusted environments.  Every use of it will log a warning
3164 	 * in the kernel log.
3165 	 */
3166 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3167 };
3168 
3169 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3170 		const char *caller);
3171 #define ib_alloc_pd(device, flags) \
3172 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3173 void ib_dealloc_pd(struct ib_pd *pd);
3174 
3175 enum rdma_create_ah_flags {
3176 	/* In a sleepable context */
3177 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3178 };
3179 
3180 /**
3181  * rdma_create_ah - Creates an address handle for the given address vector.
3182  * @pd: The protection domain associated with the address handle.
3183  * @ah_attr: The attributes of the address vector.
3184  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3185  *
3186  * The address handle is used to reference a local or global destination
3187  * in all UD QP post sends.
3188  */
3189 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3190 			     u32 flags);
3191 
3192 /**
3193  * rdma_create_user_ah - Creates an address handle for the given address vector.
3194  * It resolves destination mac address for ah attribute of RoCE type.
3195  * @pd: The protection domain associated with the address handle.
3196  * @ah_attr: The attributes of the address vector.
3197  * @udata: pointer to user's input output buffer information need by
3198  *         provider driver.
3199  *
3200  * It returns 0 on success and returns appropriate error code on error.
3201  * The address handle is used to reference a local or global destination
3202  * in all UD QP post sends.
3203  */
3204 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3205 				  struct rdma_ah_attr *ah_attr,
3206 				  struct ib_udata *udata);
3207 /**
3208  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3209  *   work completion.
3210  * @hdr: the L3 header to parse
3211  * @net_type: type of header to parse
3212  * @sgid: place to store source gid
3213  * @dgid: place to store destination gid
3214  */
3215 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3216 			      enum rdma_network_type net_type,
3217 			      union ib_gid *sgid, union ib_gid *dgid);
3218 
3219 /**
3220  * ib_get_rdma_header_version - Get the header version
3221  * @hdr: the L3 header to parse
3222  */
3223 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3224 
3225 /**
3226  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3227  *   work completion.
3228  * @device: Device on which the received message arrived.
3229  * @port_num: Port on which the received message arrived.
3230  * @wc: Work completion associated with the received message.
3231  * @grh: References the received global route header.  This parameter is
3232  *   ignored unless the work completion indicates that the GRH is valid.
3233  * @ah_attr: Returned attributes that can be used when creating an address
3234  *   handle for replying to the message.
3235  * When ib_init_ah_attr_from_wc() returns success,
3236  * (a) for IB link layer it optionally contains a reference to SGID attribute
3237  * when GRH is present for IB link layer.
3238  * (b) for RoCE link layer it contains a reference to SGID attribute.
3239  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3240  * attributes which are initialized using ib_init_ah_attr_from_wc().
3241  *
3242  */
3243 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3244 			    const struct ib_wc *wc, const struct ib_grh *grh,
3245 			    struct rdma_ah_attr *ah_attr);
3246 
3247 /**
3248  * ib_create_ah_from_wc - Creates an address handle associated with the
3249  *   sender of the specified work completion.
3250  * @pd: The protection domain associated with the address handle.
3251  * @wc: Work completion information associated with a received message.
3252  * @grh: References the received global route header.  This parameter is
3253  *   ignored unless the work completion indicates that the GRH is valid.
3254  * @port_num: The outbound port number to associate with the address.
3255  *
3256  * The address handle is used to reference a local or global destination
3257  * in all UD QP post sends.
3258  */
3259 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3260 				   const struct ib_grh *grh, u8 port_num);
3261 
3262 /**
3263  * rdma_modify_ah - Modifies the address vector associated with an address
3264  *   handle.
3265  * @ah: The address handle to modify.
3266  * @ah_attr: The new address vector attributes to associate with the
3267  *   address handle.
3268  */
3269 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3270 
3271 /**
3272  * rdma_query_ah - Queries the address vector associated with an address
3273  *   handle.
3274  * @ah: The address handle to query.
3275  * @ah_attr: The address vector attributes associated with the address
3276  *   handle.
3277  */
3278 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3279 
3280 enum rdma_destroy_ah_flags {
3281 	/* In a sleepable context */
3282 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3283 };
3284 
3285 /**
3286  * rdma_destroy_ah - Destroys an address handle.
3287  * @ah: The address handle to destroy.
3288  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3289  */
3290 int rdma_destroy_ah(struct ib_ah *ah, u32 flags);
3291 
3292 /**
3293  * ib_create_srq - Creates a SRQ associated with the specified protection
3294  *   domain.
3295  * @pd: The protection domain associated with the SRQ.
3296  * @srq_init_attr: A list of initial attributes required to create the
3297  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
3298  *   the actual capabilities of the created SRQ.
3299  *
3300  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3301  * requested size of the SRQ, and set to the actual values allocated
3302  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
3303  * will always be at least as large as the requested values.
3304  */
3305 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3306 			     struct ib_srq_init_attr *srq_init_attr);
3307 
3308 /**
3309  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3310  * @srq: The SRQ to modify.
3311  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3312  *   the current values of selected SRQ attributes are returned.
3313  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3314  *   are being modified.
3315  *
3316  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3317  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3318  * the number of receives queued drops below the limit.
3319  */
3320 int ib_modify_srq(struct ib_srq *srq,
3321 		  struct ib_srq_attr *srq_attr,
3322 		  enum ib_srq_attr_mask srq_attr_mask);
3323 
3324 /**
3325  * ib_query_srq - Returns the attribute list and current values for the
3326  *   specified SRQ.
3327  * @srq: The SRQ to query.
3328  * @srq_attr: The attributes of the specified SRQ.
3329  */
3330 int ib_query_srq(struct ib_srq *srq,
3331 		 struct ib_srq_attr *srq_attr);
3332 
3333 /**
3334  * ib_destroy_srq - Destroys the specified SRQ.
3335  * @srq: The SRQ to destroy.
3336  */
3337 int ib_destroy_srq(struct ib_srq *srq);
3338 
3339 /**
3340  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3341  * @srq: The SRQ to post the work request on.
3342  * @recv_wr: A list of work requests to post on the receive queue.
3343  * @bad_recv_wr: On an immediate failure, this parameter will reference
3344  *   the work request that failed to be posted on the QP.
3345  */
3346 static inline int ib_post_srq_recv(struct ib_srq *srq,
3347 				   const struct ib_recv_wr *recv_wr,
3348 				   const struct ib_recv_wr **bad_recv_wr)
3349 {
3350 	const struct ib_recv_wr *dummy;
3351 
3352 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3353 					      bad_recv_wr ? : &dummy);
3354 }
3355 
3356 /**
3357  * ib_create_qp - Creates a QP associated with the specified protection
3358  *   domain.
3359  * @pd: The protection domain associated with the QP.
3360  * @qp_init_attr: A list of initial attributes required to create the
3361  *   QP.  If QP creation succeeds, then the attributes are updated to
3362  *   the actual capabilities of the created QP.
3363  */
3364 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3365 			   struct ib_qp_init_attr *qp_init_attr);
3366 
3367 /**
3368  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3369  * @qp: The QP to modify.
3370  * @attr: On input, specifies the QP attributes to modify.  On output,
3371  *   the current values of selected QP attributes are returned.
3372  * @attr_mask: A bit-mask used to specify which attributes of the QP
3373  *   are being modified.
3374  * @udata: pointer to user's input output buffer information
3375  *   are being modified.
3376  * It returns 0 on success and returns appropriate error code on error.
3377  */
3378 int ib_modify_qp_with_udata(struct ib_qp *qp,
3379 			    struct ib_qp_attr *attr,
3380 			    int attr_mask,
3381 			    struct ib_udata *udata);
3382 
3383 /**
3384  * ib_modify_qp - Modifies the attributes for the specified QP and then
3385  *   transitions the QP to the given state.
3386  * @qp: The QP to modify.
3387  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3388  *   the current values of selected QP attributes are returned.
3389  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3390  *   are being modified.
3391  */
3392 int ib_modify_qp(struct ib_qp *qp,
3393 		 struct ib_qp_attr *qp_attr,
3394 		 int qp_attr_mask);
3395 
3396 /**
3397  * ib_query_qp - Returns the attribute list and current values for the
3398  *   specified QP.
3399  * @qp: The QP to query.
3400  * @qp_attr: The attributes of the specified QP.
3401  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3402  * @qp_init_attr: Additional attributes of the selected QP.
3403  *
3404  * The qp_attr_mask may be used to limit the query to gathering only the
3405  * selected attributes.
3406  */
3407 int ib_query_qp(struct ib_qp *qp,
3408 		struct ib_qp_attr *qp_attr,
3409 		int qp_attr_mask,
3410 		struct ib_qp_init_attr *qp_init_attr);
3411 
3412 /**
3413  * ib_destroy_qp - Destroys the specified QP.
3414  * @qp: The QP to destroy.
3415  */
3416 int ib_destroy_qp(struct ib_qp *qp);
3417 
3418 /**
3419  * ib_open_qp - Obtain a reference to an existing sharable QP.
3420  * @xrcd - XRC domain
3421  * @qp_open_attr: Attributes identifying the QP to open.
3422  *
3423  * Returns a reference to a sharable QP.
3424  */
3425 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3426 			 struct ib_qp_open_attr *qp_open_attr);
3427 
3428 /**
3429  * ib_close_qp - Release an external reference to a QP.
3430  * @qp: The QP handle to release
3431  *
3432  * The opened QP handle is released by the caller.  The underlying
3433  * shared QP is not destroyed until all internal references are released.
3434  */
3435 int ib_close_qp(struct ib_qp *qp);
3436 
3437 /**
3438  * ib_post_send - Posts a list of work requests to the send queue of
3439  *   the specified QP.
3440  * @qp: The QP to post the work request on.
3441  * @send_wr: A list of work requests to post on the send queue.
3442  * @bad_send_wr: On an immediate failure, this parameter will reference
3443  *   the work request that failed to be posted on the QP.
3444  *
3445  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3446  * error is returned, the QP state shall not be affected,
3447  * ib_post_send() will return an immediate error after queueing any
3448  * earlier work requests in the list.
3449  */
3450 static inline int ib_post_send(struct ib_qp *qp,
3451 			       const struct ib_send_wr *send_wr,
3452 			       const struct ib_send_wr **bad_send_wr)
3453 {
3454 	const struct ib_send_wr *dummy;
3455 
3456 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3457 }
3458 
3459 /**
3460  * ib_post_recv - Posts a list of work requests to the receive queue of
3461  *   the specified QP.
3462  * @qp: The QP to post the work request on.
3463  * @recv_wr: A list of work requests to post on the receive queue.
3464  * @bad_recv_wr: On an immediate failure, this parameter will reference
3465  *   the work request that failed to be posted on the QP.
3466  */
3467 static inline int ib_post_recv(struct ib_qp *qp,
3468 			       const struct ib_recv_wr *recv_wr,
3469 			       const struct ib_recv_wr **bad_recv_wr)
3470 {
3471 	const struct ib_recv_wr *dummy;
3472 
3473 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3474 }
3475 
3476 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3477 			    int nr_cqe, int comp_vector,
3478 			    enum ib_poll_context poll_ctx, const char *caller);
3479 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3480 	__ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3481 
3482 void ib_free_cq(struct ib_cq *cq);
3483 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3484 
3485 /**
3486  * ib_create_cq - Creates a CQ on the specified device.
3487  * @device: The device on which to create the CQ.
3488  * @comp_handler: A user-specified callback that is invoked when a
3489  *   completion event occurs on the CQ.
3490  * @event_handler: A user-specified callback that is invoked when an
3491  *   asynchronous event not associated with a completion occurs on the CQ.
3492  * @cq_context: Context associated with the CQ returned to the user via
3493  *   the associated completion and event handlers.
3494  * @cq_attr: The attributes the CQ should be created upon.
3495  *
3496  * Users can examine the cq structure to determine the actual CQ size.
3497  */
3498 struct ib_cq *__ib_create_cq(struct ib_device *device,
3499 			     ib_comp_handler comp_handler,
3500 			     void (*event_handler)(struct ib_event *, void *),
3501 			     void *cq_context,
3502 			     const struct ib_cq_init_attr *cq_attr,
3503 			     const char *caller);
3504 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3505 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3506 
3507 /**
3508  * ib_resize_cq - Modifies the capacity of the CQ.
3509  * @cq: The CQ to resize.
3510  * @cqe: The minimum size of the CQ.
3511  *
3512  * Users can examine the cq structure to determine the actual CQ size.
3513  */
3514 int ib_resize_cq(struct ib_cq *cq, int cqe);
3515 
3516 /**
3517  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3518  * @cq: The CQ to modify.
3519  * @cq_count: number of CQEs that will trigger an event
3520  * @cq_period: max period of time in usec before triggering an event
3521  *
3522  */
3523 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3524 
3525 /**
3526  * ib_destroy_cq - Destroys the specified CQ.
3527  * @cq: The CQ to destroy.
3528  */
3529 int ib_destroy_cq(struct ib_cq *cq);
3530 
3531 /**
3532  * ib_poll_cq - poll a CQ for completion(s)
3533  * @cq:the CQ being polled
3534  * @num_entries:maximum number of completions to return
3535  * @wc:array of at least @num_entries &struct ib_wc where completions
3536  *   will be returned
3537  *
3538  * Poll a CQ for (possibly multiple) completions.  If the return value
3539  * is < 0, an error occurred.  If the return value is >= 0, it is the
3540  * number of completions returned.  If the return value is
3541  * non-negative and < num_entries, then the CQ was emptied.
3542  */
3543 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3544 			     struct ib_wc *wc)
3545 {
3546 	return cq->device->ops.poll_cq(cq, num_entries, wc);
3547 }
3548 
3549 /**
3550  * ib_req_notify_cq - Request completion notification on a CQ.
3551  * @cq: The CQ to generate an event for.
3552  * @flags:
3553  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3554  *   to request an event on the next solicited event or next work
3555  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3556  *   may also be |ed in to request a hint about missed events, as
3557  *   described below.
3558  *
3559  * Return Value:
3560  *    < 0 means an error occurred while requesting notification
3561  *   == 0 means notification was requested successfully, and if
3562  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3563  *        were missed and it is safe to wait for another event.  In
3564  *        this case is it guaranteed that any work completions added
3565  *        to the CQ since the last CQ poll will trigger a completion
3566  *        notification event.
3567  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3568  *        in.  It means that the consumer must poll the CQ again to
3569  *        make sure it is empty to avoid missing an event because of a
3570  *        race between requesting notification and an entry being
3571  *        added to the CQ.  This return value means it is possible
3572  *        (but not guaranteed) that a work completion has been added
3573  *        to the CQ since the last poll without triggering a
3574  *        completion notification event.
3575  */
3576 static inline int ib_req_notify_cq(struct ib_cq *cq,
3577 				   enum ib_cq_notify_flags flags)
3578 {
3579 	return cq->device->ops.req_notify_cq(cq, flags);
3580 }
3581 
3582 /**
3583  * ib_req_ncomp_notif - Request completion notification when there are
3584  *   at least the specified number of unreaped completions on the CQ.
3585  * @cq: The CQ to generate an event for.
3586  * @wc_cnt: The number of unreaped completions that should be on the
3587  *   CQ before an event is generated.
3588  */
3589 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3590 {
3591 	return cq->device->ops.req_ncomp_notif ?
3592 		cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3593 		-ENOSYS;
3594 }
3595 
3596 /**
3597  * ib_dma_mapping_error - check a DMA addr for error
3598  * @dev: The device for which the dma_addr was created
3599  * @dma_addr: The DMA address to check
3600  */
3601 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3602 {
3603 	return dma_mapping_error(dev->dma_device, dma_addr);
3604 }
3605 
3606 /**
3607  * ib_dma_map_single - Map a kernel virtual address to DMA address
3608  * @dev: The device for which the dma_addr is to be created
3609  * @cpu_addr: The kernel virtual address
3610  * @size: The size of the region in bytes
3611  * @direction: The direction of the DMA
3612  */
3613 static inline u64 ib_dma_map_single(struct ib_device *dev,
3614 				    void *cpu_addr, size_t size,
3615 				    enum dma_data_direction direction)
3616 {
3617 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3618 }
3619 
3620 /**
3621  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3622  * @dev: The device for which the DMA address was created
3623  * @addr: The DMA address
3624  * @size: The size of the region in bytes
3625  * @direction: The direction of the DMA
3626  */
3627 static inline void ib_dma_unmap_single(struct ib_device *dev,
3628 				       u64 addr, size_t size,
3629 				       enum dma_data_direction direction)
3630 {
3631 	dma_unmap_single(dev->dma_device, addr, size, direction);
3632 }
3633 
3634 /**
3635  * ib_dma_map_page - Map a physical page to DMA address
3636  * @dev: The device for which the dma_addr is to be created
3637  * @page: The page to be mapped
3638  * @offset: The offset within the page
3639  * @size: The size of the region in bytes
3640  * @direction: The direction of the DMA
3641  */
3642 static inline u64 ib_dma_map_page(struct ib_device *dev,
3643 				  struct page *page,
3644 				  unsigned long offset,
3645 				  size_t size,
3646 					 enum dma_data_direction direction)
3647 {
3648 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3649 }
3650 
3651 /**
3652  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3653  * @dev: The device for which the DMA address was created
3654  * @addr: The DMA address
3655  * @size: The size of the region in bytes
3656  * @direction: The direction of the DMA
3657  */
3658 static inline void ib_dma_unmap_page(struct ib_device *dev,
3659 				     u64 addr, size_t size,
3660 				     enum dma_data_direction direction)
3661 {
3662 	dma_unmap_page(dev->dma_device, addr, size, direction);
3663 }
3664 
3665 /**
3666  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3667  * @dev: The device for which the DMA addresses are to be created
3668  * @sg: The array of scatter/gather entries
3669  * @nents: The number of scatter/gather entries
3670  * @direction: The direction of the DMA
3671  */
3672 static inline int ib_dma_map_sg(struct ib_device *dev,
3673 				struct scatterlist *sg, int nents,
3674 				enum dma_data_direction direction)
3675 {
3676 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3677 }
3678 
3679 /**
3680  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3681  * @dev: The device for which the DMA addresses were created
3682  * @sg: The array of scatter/gather entries
3683  * @nents: The number of scatter/gather entries
3684  * @direction: The direction of the DMA
3685  */
3686 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3687 				   struct scatterlist *sg, int nents,
3688 				   enum dma_data_direction direction)
3689 {
3690 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
3691 }
3692 
3693 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3694 				      struct scatterlist *sg, int nents,
3695 				      enum dma_data_direction direction,
3696 				      unsigned long dma_attrs)
3697 {
3698 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3699 				dma_attrs);
3700 }
3701 
3702 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3703 					 struct scatterlist *sg, int nents,
3704 					 enum dma_data_direction direction,
3705 					 unsigned long dma_attrs)
3706 {
3707 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3708 }
3709 
3710 /**
3711  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
3712  * @dev: The device to query
3713  *
3714  * The returned value represents a size in bytes.
3715  */
3716 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
3717 {
3718 	struct device_dma_parameters *p = dev->dma_device->dma_parms;
3719 
3720 	return p ? p->max_segment_size : UINT_MAX;
3721 }
3722 
3723 /**
3724  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3725  * @dev: The device for which the DMA address was created
3726  * @addr: The DMA address
3727  * @size: The size of the region in bytes
3728  * @dir: The direction of the DMA
3729  */
3730 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3731 					      u64 addr,
3732 					      size_t size,
3733 					      enum dma_data_direction dir)
3734 {
3735 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3736 }
3737 
3738 /**
3739  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3740  * @dev: The device for which the DMA address was created
3741  * @addr: The DMA address
3742  * @size: The size of the region in bytes
3743  * @dir: The direction of the DMA
3744  */
3745 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3746 						 u64 addr,
3747 						 size_t size,
3748 						 enum dma_data_direction dir)
3749 {
3750 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3751 }
3752 
3753 /**
3754  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3755  * @dev: The device for which the DMA address is requested
3756  * @size: The size of the region to allocate in bytes
3757  * @dma_handle: A pointer for returning the DMA address of the region
3758  * @flag: memory allocator flags
3759  */
3760 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3761 					   size_t size,
3762 					   dma_addr_t *dma_handle,
3763 					   gfp_t flag)
3764 {
3765 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3766 }
3767 
3768 /**
3769  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3770  * @dev: The device for which the DMA addresses were allocated
3771  * @size: The size of the region
3772  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3773  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3774  */
3775 static inline void ib_dma_free_coherent(struct ib_device *dev,
3776 					size_t size, void *cpu_addr,
3777 					dma_addr_t dma_handle)
3778 {
3779 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3780 }
3781 
3782 /**
3783  * ib_dereg_mr - Deregisters a memory region and removes it from the
3784  *   HCA translation table.
3785  * @mr: The memory region to deregister.
3786  *
3787  * This function can fail, if the memory region has memory windows bound to it.
3788  */
3789 int ib_dereg_mr(struct ib_mr *mr);
3790 
3791 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3792 			  enum ib_mr_type mr_type,
3793 			  u32 max_num_sg);
3794 
3795 /**
3796  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3797  *   R_Key and L_Key.
3798  * @mr - struct ib_mr pointer to be updated.
3799  * @newkey - new key to be used.
3800  */
3801 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3802 {
3803 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3804 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3805 }
3806 
3807 /**
3808  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3809  * for calculating a new rkey for type 2 memory windows.
3810  * @rkey - the rkey to increment.
3811  */
3812 static inline u32 ib_inc_rkey(u32 rkey)
3813 {
3814 	const u32 mask = 0x000000ff;
3815 	return ((rkey + 1) & mask) | (rkey & ~mask);
3816 }
3817 
3818 /**
3819  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3820  * @pd: The protection domain associated with the unmapped region.
3821  * @mr_access_flags: Specifies the memory access rights.
3822  * @fmr_attr: Attributes of the unmapped region.
3823  *
3824  * A fast memory region must be mapped before it can be used as part of
3825  * a work request.
3826  */
3827 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3828 			    int mr_access_flags,
3829 			    struct ib_fmr_attr *fmr_attr);
3830 
3831 /**
3832  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3833  * @fmr: The fast memory region to associate with the pages.
3834  * @page_list: An array of physical pages to map to the fast memory region.
3835  * @list_len: The number of pages in page_list.
3836  * @iova: The I/O virtual address to use with the mapped region.
3837  */
3838 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3839 				  u64 *page_list, int list_len,
3840 				  u64 iova)
3841 {
3842 	return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
3843 }
3844 
3845 /**
3846  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3847  * @fmr_list: A linked list of fast memory regions to unmap.
3848  */
3849 int ib_unmap_fmr(struct list_head *fmr_list);
3850 
3851 /**
3852  * ib_dealloc_fmr - Deallocates a fast memory region.
3853  * @fmr: The fast memory region to deallocate.
3854  */
3855 int ib_dealloc_fmr(struct ib_fmr *fmr);
3856 
3857 /**
3858  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3859  * @qp: QP to attach to the multicast group.  The QP must be type
3860  *   IB_QPT_UD.
3861  * @gid: Multicast group GID.
3862  * @lid: Multicast group LID in host byte order.
3863  *
3864  * In order to send and receive multicast packets, subnet
3865  * administration must have created the multicast group and configured
3866  * the fabric appropriately.  The port associated with the specified
3867  * QP must also be a member of the multicast group.
3868  */
3869 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3870 
3871 /**
3872  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3873  * @qp: QP to detach from the multicast group.
3874  * @gid: Multicast group GID.
3875  * @lid: Multicast group LID in host byte order.
3876  */
3877 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3878 
3879 /**
3880  * ib_alloc_xrcd - Allocates an XRC domain.
3881  * @device: The device on which to allocate the XRC domain.
3882  * @caller: Module name for kernel consumers
3883  */
3884 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3885 #define ib_alloc_xrcd(device) \
3886 	__ib_alloc_xrcd((device), KBUILD_MODNAME)
3887 
3888 /**
3889  * ib_dealloc_xrcd - Deallocates an XRC domain.
3890  * @xrcd: The XRC domain to deallocate.
3891  */
3892 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3893 
3894 static inline int ib_check_mr_access(int flags)
3895 {
3896 	/*
3897 	 * Local write permission is required if remote write or
3898 	 * remote atomic permission is also requested.
3899 	 */
3900 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3901 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3902 		return -EINVAL;
3903 
3904 	return 0;
3905 }
3906 
3907 static inline bool ib_access_writable(int access_flags)
3908 {
3909 	/*
3910 	 * We have writable memory backing the MR if any of the following
3911 	 * access flags are set.  "Local write" and "remote write" obviously
3912 	 * require write access.  "Remote atomic" can do things like fetch and
3913 	 * add, which will modify memory, and "MW bind" can change permissions
3914 	 * by binding a window.
3915 	 */
3916 	return access_flags &
3917 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
3918 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
3919 }
3920 
3921 /**
3922  * ib_check_mr_status: lightweight check of MR status.
3923  *     This routine may provide status checks on a selected
3924  *     ib_mr. first use is for signature status check.
3925  *
3926  * @mr: A memory region.
3927  * @check_mask: Bitmask of which checks to perform from
3928  *     ib_mr_status_check enumeration.
3929  * @mr_status: The container of relevant status checks.
3930  *     failed checks will be indicated in the status bitmask
3931  *     and the relevant info shall be in the error item.
3932  */
3933 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3934 		       struct ib_mr_status *mr_status);
3935 
3936 /**
3937  * ib_device_try_get: Hold a registration lock
3938  * device: The device to lock
3939  *
3940  * A device under an active registration lock cannot become unregistered. It
3941  * is only possible to obtain a registration lock on a device that is fully
3942  * registered, otherwise this function returns false.
3943  *
3944  * The registration lock is only necessary for actions which require the
3945  * device to still be registered. Uses that only require the device pointer to
3946  * be valid should use get_device(&ibdev->dev) to hold the memory.
3947  *
3948  */
3949 static inline bool ib_device_try_get(struct ib_device *dev)
3950 {
3951 	return refcount_inc_not_zero(&dev->refcount);
3952 }
3953 
3954 void ib_device_put(struct ib_device *device);
3955 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3956 					    u16 pkey, const union ib_gid *gid,
3957 					    const struct sockaddr *addr);
3958 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3959 			   struct ib_wq_init_attr *init_attr);
3960 int ib_destroy_wq(struct ib_wq *wq);
3961 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3962 		 u32 wq_attr_mask);
3963 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3964 						 struct ib_rwq_ind_table_init_attr*
3965 						 wq_ind_table_init_attr);
3966 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3967 
3968 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3969 		 unsigned int *sg_offset, unsigned int page_size);
3970 
3971 static inline int
3972 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3973 		  unsigned int *sg_offset, unsigned int page_size)
3974 {
3975 	int n;
3976 
3977 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3978 	mr->iova = 0;
3979 
3980 	return n;
3981 }
3982 
3983 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3984 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3985 
3986 void ib_drain_rq(struct ib_qp *qp);
3987 void ib_drain_sq(struct ib_qp *qp);
3988 void ib_drain_qp(struct ib_qp *qp);
3989 
3990 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3991 
3992 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3993 {
3994 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3995 		return attr->roce.dmac;
3996 	return NULL;
3997 }
3998 
3999 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4000 {
4001 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4002 		attr->ib.dlid = (u16)dlid;
4003 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4004 		attr->opa.dlid = dlid;
4005 }
4006 
4007 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4008 {
4009 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4010 		return attr->ib.dlid;
4011 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4012 		return attr->opa.dlid;
4013 	return 0;
4014 }
4015 
4016 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4017 {
4018 	attr->sl = sl;
4019 }
4020 
4021 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4022 {
4023 	return attr->sl;
4024 }
4025 
4026 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4027 					 u8 src_path_bits)
4028 {
4029 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4030 		attr->ib.src_path_bits = src_path_bits;
4031 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4032 		attr->opa.src_path_bits = src_path_bits;
4033 }
4034 
4035 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4036 {
4037 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4038 		return attr->ib.src_path_bits;
4039 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4040 		return attr->opa.src_path_bits;
4041 	return 0;
4042 }
4043 
4044 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4045 					bool make_grd)
4046 {
4047 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4048 		attr->opa.make_grd = make_grd;
4049 }
4050 
4051 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4052 {
4053 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4054 		return attr->opa.make_grd;
4055 	return false;
4056 }
4057 
4058 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4059 {
4060 	attr->port_num = port_num;
4061 }
4062 
4063 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4064 {
4065 	return attr->port_num;
4066 }
4067 
4068 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4069 					   u8 static_rate)
4070 {
4071 	attr->static_rate = static_rate;
4072 }
4073 
4074 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4075 {
4076 	return attr->static_rate;
4077 }
4078 
4079 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4080 					enum ib_ah_flags flag)
4081 {
4082 	attr->ah_flags = flag;
4083 }
4084 
4085 static inline enum ib_ah_flags
4086 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4087 {
4088 	return attr->ah_flags;
4089 }
4090 
4091 static inline const struct ib_global_route
4092 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4093 {
4094 	return &attr->grh;
4095 }
4096 
4097 /*To retrieve and modify the grh */
4098 static inline struct ib_global_route
4099 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4100 {
4101 	return &attr->grh;
4102 }
4103 
4104 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4105 {
4106 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4107 
4108 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4109 }
4110 
4111 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4112 					     __be64 prefix)
4113 {
4114 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4115 
4116 	grh->dgid.global.subnet_prefix = prefix;
4117 }
4118 
4119 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4120 					    __be64 if_id)
4121 {
4122 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4123 
4124 	grh->dgid.global.interface_id = if_id;
4125 }
4126 
4127 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4128 				   union ib_gid *dgid, u32 flow_label,
4129 				   u8 sgid_index, u8 hop_limit,
4130 				   u8 traffic_class)
4131 {
4132 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4133 
4134 	attr->ah_flags = IB_AH_GRH;
4135 	if (dgid)
4136 		grh->dgid = *dgid;
4137 	grh->flow_label = flow_label;
4138 	grh->sgid_index = sgid_index;
4139 	grh->hop_limit = hop_limit;
4140 	grh->traffic_class = traffic_class;
4141 	grh->sgid_attr = NULL;
4142 }
4143 
4144 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4145 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4146 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4147 			     const struct ib_gid_attr *sgid_attr);
4148 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4149 		       const struct rdma_ah_attr *src);
4150 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4151 			  const struct rdma_ah_attr *new);
4152 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4153 
4154 /**
4155  * rdma_ah_find_type - Return address handle type.
4156  *
4157  * @dev: Device to be checked
4158  * @port_num: Port number
4159  */
4160 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4161 						       u8 port_num)
4162 {
4163 	if (rdma_protocol_roce(dev, port_num))
4164 		return RDMA_AH_ATTR_TYPE_ROCE;
4165 	if (rdma_protocol_ib(dev, port_num)) {
4166 		if (rdma_cap_opa_ah(dev, port_num))
4167 			return RDMA_AH_ATTR_TYPE_OPA;
4168 		return RDMA_AH_ATTR_TYPE_IB;
4169 	}
4170 
4171 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4172 }
4173 
4174 /**
4175  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4176  *     In the current implementation the only way to get
4177  *     get the 32bit lid is from other sources for OPA.
4178  *     For IB, lids will always be 16bits so cast the
4179  *     value accordingly.
4180  *
4181  * @lid: A 32bit LID
4182  */
4183 static inline u16 ib_lid_cpu16(u32 lid)
4184 {
4185 	WARN_ON_ONCE(lid & 0xFFFF0000);
4186 	return (u16)lid;
4187 }
4188 
4189 /**
4190  * ib_lid_be16 - Return lid in 16bit BE encoding.
4191  *
4192  * @lid: A 32bit LID
4193  */
4194 static inline __be16 ib_lid_be16(u32 lid)
4195 {
4196 	WARN_ON_ONCE(lid & 0xFFFF0000);
4197 	return cpu_to_be16((u16)lid);
4198 }
4199 
4200 /**
4201  * ib_get_vector_affinity - Get the affinity mappings of a given completion
4202  *   vector
4203  * @device:         the rdma device
4204  * @comp_vector:    index of completion vector
4205  *
4206  * Returns NULL on failure, otherwise a corresponding cpu map of the
4207  * completion vector (returns all-cpus map if the device driver doesn't
4208  * implement get_vector_affinity).
4209  */
4210 static inline const struct cpumask *
4211 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4212 {
4213 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4214 	    !device->ops.get_vector_affinity)
4215 		return NULL;
4216 
4217 	return device->ops.get_vector_affinity(device, comp_vector);
4218 
4219 }
4220 
4221 /**
4222  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4223  * and add their gids, as needed, to the relevant RoCE devices.
4224  *
4225  * @device:         the rdma device
4226  */
4227 void rdma_roce_rescan_device(struct ib_device *ibdev);
4228 
4229 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4230 
4231 struct ib_ucontext *rdma_get_ucontext(struct ib_udata *udata);
4232 
4233 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4234 
4235 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4236 				     enum rdma_netdev_t type, const char *name,
4237 				     unsigned char name_assign_type,
4238 				     void (*setup)(struct net_device *));
4239 
4240 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4241 		     enum rdma_netdev_t type, const char *name,
4242 		     unsigned char name_assign_type,
4243 		     void (*setup)(struct net_device *),
4244 		     struct net_device *netdev);
4245 
4246 /**
4247  * rdma_set_device_sysfs_group - Set device attributes group to have
4248  *				 driver specific sysfs entries at
4249  *				 for infiniband class.
4250  *
4251  * @device:	device pointer for which attributes to be created
4252  * @group:	Pointer to group which should be added when device
4253  *		is registered with sysfs.
4254  * rdma_set_device_sysfs_group() allows existing drivers to expose one
4255  * group per device to have sysfs attributes.
4256  *
4257  * NOTE: New drivers should not make use of this API; instead new device
4258  * parameter should be exposed via netlink command. This API and mechanism
4259  * exist only for existing drivers.
4260  */
4261 static inline void
4262 rdma_set_device_sysfs_group(struct ib_device *dev,
4263 			    const struct attribute_group *group)
4264 {
4265 	dev->groups[1] = group;
4266 }
4267 
4268 /**
4269  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4270  *
4271  * @device:	device pointer for which ib_device pointer to retrieve
4272  *
4273  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4274  *
4275  */
4276 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4277 {
4278 	return container_of(device, struct ib_device, dev);
4279 }
4280 
4281 /**
4282  * rdma_device_to_drv_device - Helper macro to reach back to driver's
4283  *			       ib_device holder structure from device pointer.
4284  *
4285  * NOTE: New drivers should not make use of this API; This API is only for
4286  * existing drivers who have exposed sysfs entries using
4287  * rdma_set_device_sysfs_group().
4288  */
4289 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
4290 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4291 #endif /* IB_VERBS_H */
4292