xref: /openbmc/linux/include/rdma/ib_verbs.h (revision 367e5927)
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 	IB_DEVICE_ALLOW_USER_UNREG		= (1ULL << 37),
242 };
243 
244 enum ib_signature_prot_cap {
245 	IB_PROT_T10DIF_TYPE_1 = 1,
246 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
247 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
248 };
249 
250 enum ib_signature_guard_cap {
251 	IB_GUARD_T10DIF_CRC	= 1,
252 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
253 };
254 
255 enum ib_atomic_cap {
256 	IB_ATOMIC_NONE,
257 	IB_ATOMIC_HCA,
258 	IB_ATOMIC_GLOB
259 };
260 
261 enum ib_odp_general_cap_bits {
262 	IB_ODP_SUPPORT		= 1 << 0,
263 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
264 };
265 
266 enum ib_odp_transport_cap_bits {
267 	IB_ODP_SUPPORT_SEND	= 1 << 0,
268 	IB_ODP_SUPPORT_RECV	= 1 << 1,
269 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
270 	IB_ODP_SUPPORT_READ	= 1 << 3,
271 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
272 	IB_ODP_SUPPORT_SRQ_RECV	= 1 << 5,
273 };
274 
275 struct ib_odp_caps {
276 	uint64_t general_caps;
277 	struct {
278 		uint32_t  rc_odp_caps;
279 		uint32_t  uc_odp_caps;
280 		uint32_t  ud_odp_caps;
281 		uint32_t  xrc_odp_caps;
282 	} per_transport_caps;
283 };
284 
285 struct ib_rss_caps {
286 	/* Corresponding bit will be set if qp type from
287 	 * 'enum ib_qp_type' is supported, e.g.
288 	 * supported_qpts |= 1 << IB_QPT_UD
289 	 */
290 	u32 supported_qpts;
291 	u32 max_rwq_indirection_tables;
292 	u32 max_rwq_indirection_table_size;
293 };
294 
295 enum ib_tm_cap_flags {
296 	/*  Support tag matching on RC transport */
297 	IB_TM_CAP_RC		    = 1 << 0,
298 };
299 
300 struct ib_tm_caps {
301 	/* Max size of RNDV header */
302 	u32 max_rndv_hdr_size;
303 	/* Max number of entries in tag matching list */
304 	u32 max_num_tags;
305 	/* From enum ib_tm_cap_flags */
306 	u32 flags;
307 	/* Max number of outstanding list operations */
308 	u32 max_ops;
309 	/* Max number of SGE in tag matching entry */
310 	u32 max_sge;
311 };
312 
313 struct ib_cq_init_attr {
314 	unsigned int	cqe;
315 	int		comp_vector;
316 	u32		flags;
317 };
318 
319 enum ib_cq_attr_mask {
320 	IB_CQ_MODERATE = 1 << 0,
321 };
322 
323 struct ib_cq_caps {
324 	u16     max_cq_moderation_count;
325 	u16     max_cq_moderation_period;
326 };
327 
328 struct ib_dm_mr_attr {
329 	u64		length;
330 	u64		offset;
331 	u32		access_flags;
332 };
333 
334 struct ib_dm_alloc_attr {
335 	u64	length;
336 	u32	alignment;
337 	u32	flags;
338 };
339 
340 struct ib_device_attr {
341 	u64			fw_ver;
342 	__be64			sys_image_guid;
343 	u64			max_mr_size;
344 	u64			page_size_cap;
345 	u32			vendor_id;
346 	u32			vendor_part_id;
347 	u32			hw_ver;
348 	int			max_qp;
349 	int			max_qp_wr;
350 	u64			device_cap_flags;
351 	int			max_send_sge;
352 	int			max_recv_sge;
353 	int			max_sge_rd;
354 	int			max_cq;
355 	int			max_cqe;
356 	int			max_mr;
357 	int			max_pd;
358 	int			max_qp_rd_atom;
359 	int			max_ee_rd_atom;
360 	int			max_res_rd_atom;
361 	int			max_qp_init_rd_atom;
362 	int			max_ee_init_rd_atom;
363 	enum ib_atomic_cap	atomic_cap;
364 	enum ib_atomic_cap	masked_atomic_cap;
365 	int			max_ee;
366 	int			max_rdd;
367 	int			max_mw;
368 	int			max_raw_ipv6_qp;
369 	int			max_raw_ethy_qp;
370 	int			max_mcast_grp;
371 	int			max_mcast_qp_attach;
372 	int			max_total_mcast_qp_attach;
373 	int			max_ah;
374 	int			max_fmr;
375 	int			max_map_per_fmr;
376 	int			max_srq;
377 	int			max_srq_wr;
378 	int			max_srq_sge;
379 	unsigned int		max_fast_reg_page_list_len;
380 	u16			max_pkeys;
381 	u8			local_ca_ack_delay;
382 	int			sig_prot_cap;
383 	int			sig_guard_cap;
384 	struct ib_odp_caps	odp_caps;
385 	uint64_t		timestamp_mask;
386 	uint64_t		hca_core_clock; /* in KHZ */
387 	struct ib_rss_caps	rss_caps;
388 	u32			max_wq_type_rq;
389 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
390 	struct ib_tm_caps	tm_caps;
391 	struct ib_cq_caps       cq_caps;
392 	u64			max_dm_size;
393 };
394 
395 enum ib_mtu {
396 	IB_MTU_256  = 1,
397 	IB_MTU_512  = 2,
398 	IB_MTU_1024 = 3,
399 	IB_MTU_2048 = 4,
400 	IB_MTU_4096 = 5
401 };
402 
403 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
404 {
405 	switch (mtu) {
406 	case IB_MTU_256:  return  256;
407 	case IB_MTU_512:  return  512;
408 	case IB_MTU_1024: return 1024;
409 	case IB_MTU_2048: return 2048;
410 	case IB_MTU_4096: return 4096;
411 	default: 	  return -1;
412 	}
413 }
414 
415 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
416 {
417 	if (mtu >= 4096)
418 		return IB_MTU_4096;
419 	else if (mtu >= 2048)
420 		return IB_MTU_2048;
421 	else if (mtu >= 1024)
422 		return IB_MTU_1024;
423 	else if (mtu >= 512)
424 		return IB_MTU_512;
425 	else
426 		return IB_MTU_256;
427 }
428 
429 enum ib_port_state {
430 	IB_PORT_NOP		= 0,
431 	IB_PORT_DOWN		= 1,
432 	IB_PORT_INIT		= 2,
433 	IB_PORT_ARMED		= 3,
434 	IB_PORT_ACTIVE		= 4,
435 	IB_PORT_ACTIVE_DEFER	= 5
436 };
437 
438 enum ib_port_width {
439 	IB_WIDTH_1X	= 1,
440 	IB_WIDTH_2X	= 16,
441 	IB_WIDTH_4X	= 2,
442 	IB_WIDTH_8X	= 4,
443 	IB_WIDTH_12X	= 8
444 };
445 
446 static inline int ib_width_enum_to_int(enum ib_port_width width)
447 {
448 	switch (width) {
449 	case IB_WIDTH_1X:  return  1;
450 	case IB_WIDTH_2X:  return  2;
451 	case IB_WIDTH_4X:  return  4;
452 	case IB_WIDTH_8X:  return  8;
453 	case IB_WIDTH_12X: return 12;
454 	default: 	  return -1;
455 	}
456 }
457 
458 enum ib_port_speed {
459 	IB_SPEED_SDR	= 1,
460 	IB_SPEED_DDR	= 2,
461 	IB_SPEED_QDR	= 4,
462 	IB_SPEED_FDR10	= 8,
463 	IB_SPEED_FDR	= 16,
464 	IB_SPEED_EDR	= 32,
465 	IB_SPEED_HDR	= 64
466 };
467 
468 /**
469  * struct rdma_hw_stats
470  * @lock - Mutex to protect parallel write access to lifespan and values
471  *    of counters, which are 64bits and not guaranteeed to be written
472  *    atomicaly on 32bits systems.
473  * @timestamp - Used by the core code to track when the last update was
474  * @lifespan - Used by the core code to determine how old the counters
475  *   should be before being updated again.  Stored in jiffies, defaults
476  *   to 10 milliseconds, drivers can override the default be specifying
477  *   their own value during their allocation routine.
478  * @name - Array of pointers to static names used for the counters in
479  *   directory.
480  * @num_counters - How many hardware counters there are.  If name is
481  *   shorter than this number, a kernel oops will result.  Driver authors
482  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
483  *   in their code to prevent this.
484  * @value - Array of u64 counters that are accessed by the sysfs code and
485  *   filled in by the drivers get_stats routine
486  */
487 struct rdma_hw_stats {
488 	struct mutex	lock; /* Protect lifespan and values[] */
489 	unsigned long	timestamp;
490 	unsigned long	lifespan;
491 	const char * const *names;
492 	int		num_counters;
493 	u64		value[];
494 };
495 
496 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
497 /**
498  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
499  *   for drivers.
500  * @names - Array of static const char *
501  * @num_counters - How many elements in array
502  * @lifespan - How many milliseconds between updates
503  */
504 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
505 		const char * const *names, int num_counters,
506 		unsigned long lifespan)
507 {
508 	struct rdma_hw_stats *stats;
509 
510 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
511 			GFP_KERNEL);
512 	if (!stats)
513 		return NULL;
514 	stats->names = names;
515 	stats->num_counters = num_counters;
516 	stats->lifespan = msecs_to_jiffies(lifespan);
517 
518 	return stats;
519 }
520 
521 
522 /* Define bits for the various functionality this port needs to be supported by
523  * the core.
524  */
525 /* Management                           0x00000FFF */
526 #define RDMA_CORE_CAP_IB_MAD            0x00000001
527 #define RDMA_CORE_CAP_IB_SMI            0x00000002
528 #define RDMA_CORE_CAP_IB_CM             0x00000004
529 #define RDMA_CORE_CAP_IW_CM             0x00000008
530 #define RDMA_CORE_CAP_IB_SA             0x00000010
531 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
532 
533 /* Address format                       0x000FF000 */
534 #define RDMA_CORE_CAP_AF_IB             0x00001000
535 #define RDMA_CORE_CAP_ETH_AH            0x00002000
536 #define RDMA_CORE_CAP_OPA_AH            0x00004000
537 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
538 
539 /* Protocol                             0xFFF00000 */
540 #define RDMA_CORE_CAP_PROT_IB           0x00100000
541 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
542 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
543 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
544 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
545 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
546 
547 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
548 					| RDMA_CORE_CAP_PROT_ROCE     \
549 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
550 
551 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
552 					| RDMA_CORE_CAP_IB_MAD \
553 					| RDMA_CORE_CAP_IB_SMI \
554 					| RDMA_CORE_CAP_IB_CM  \
555 					| RDMA_CORE_CAP_IB_SA  \
556 					| RDMA_CORE_CAP_AF_IB)
557 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
558 					| RDMA_CORE_CAP_IB_MAD  \
559 					| RDMA_CORE_CAP_IB_CM   \
560 					| RDMA_CORE_CAP_AF_IB   \
561 					| RDMA_CORE_CAP_ETH_AH)
562 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
563 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
564 					| RDMA_CORE_CAP_IB_MAD  \
565 					| RDMA_CORE_CAP_IB_CM   \
566 					| RDMA_CORE_CAP_AF_IB   \
567 					| RDMA_CORE_CAP_ETH_AH)
568 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
569 					| RDMA_CORE_CAP_IW_CM)
570 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
571 					| RDMA_CORE_CAP_OPA_MAD)
572 
573 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
574 
575 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
576 
577 struct ib_port_attr {
578 	u64			subnet_prefix;
579 	enum ib_port_state	state;
580 	enum ib_mtu		max_mtu;
581 	enum ib_mtu		active_mtu;
582 	int			gid_tbl_len;
583 	unsigned int		ip_gids:1;
584 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
585 	u32			port_cap_flags;
586 	u32			max_msg_sz;
587 	u32			bad_pkey_cntr;
588 	u32			qkey_viol_cntr;
589 	u16			pkey_tbl_len;
590 	u32			sm_lid;
591 	u32			lid;
592 	u8			lmc;
593 	u8			max_vl_num;
594 	u8			sm_sl;
595 	u8			subnet_timeout;
596 	u8			init_type_reply;
597 	u8			active_width;
598 	u8			active_speed;
599 	u8                      phys_state;
600 	u16			port_cap_flags2;
601 };
602 
603 enum ib_device_modify_flags {
604 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
605 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
606 };
607 
608 #define IB_DEVICE_NODE_DESC_MAX 64
609 
610 struct ib_device_modify {
611 	u64	sys_image_guid;
612 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
613 };
614 
615 enum ib_port_modify_flags {
616 	IB_PORT_SHUTDOWN		= 1,
617 	IB_PORT_INIT_TYPE		= (1<<2),
618 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
619 	IB_PORT_OPA_MASK_CHG		= (1<<4)
620 };
621 
622 struct ib_port_modify {
623 	u32	set_port_cap_mask;
624 	u32	clr_port_cap_mask;
625 	u8	init_type;
626 };
627 
628 enum ib_event_type {
629 	IB_EVENT_CQ_ERR,
630 	IB_EVENT_QP_FATAL,
631 	IB_EVENT_QP_REQ_ERR,
632 	IB_EVENT_QP_ACCESS_ERR,
633 	IB_EVENT_COMM_EST,
634 	IB_EVENT_SQ_DRAINED,
635 	IB_EVENT_PATH_MIG,
636 	IB_EVENT_PATH_MIG_ERR,
637 	IB_EVENT_DEVICE_FATAL,
638 	IB_EVENT_PORT_ACTIVE,
639 	IB_EVENT_PORT_ERR,
640 	IB_EVENT_LID_CHANGE,
641 	IB_EVENT_PKEY_CHANGE,
642 	IB_EVENT_SM_CHANGE,
643 	IB_EVENT_SRQ_ERR,
644 	IB_EVENT_SRQ_LIMIT_REACHED,
645 	IB_EVENT_QP_LAST_WQE_REACHED,
646 	IB_EVENT_CLIENT_REREGISTER,
647 	IB_EVENT_GID_CHANGE,
648 	IB_EVENT_WQ_FATAL,
649 };
650 
651 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
652 
653 struct ib_event {
654 	struct ib_device	*device;
655 	union {
656 		struct ib_cq	*cq;
657 		struct ib_qp	*qp;
658 		struct ib_srq	*srq;
659 		struct ib_wq	*wq;
660 		u8		port_num;
661 	} element;
662 	enum ib_event_type	event;
663 };
664 
665 struct ib_event_handler {
666 	struct ib_device *device;
667 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
668 	struct list_head  list;
669 };
670 
671 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
672 	do {							\
673 		(_ptr)->device  = _device;			\
674 		(_ptr)->handler = _handler;			\
675 		INIT_LIST_HEAD(&(_ptr)->list);			\
676 	} while (0)
677 
678 struct ib_global_route {
679 	const struct ib_gid_attr *sgid_attr;
680 	union ib_gid	dgid;
681 	u32		flow_label;
682 	u8		sgid_index;
683 	u8		hop_limit;
684 	u8		traffic_class;
685 };
686 
687 struct ib_grh {
688 	__be32		version_tclass_flow;
689 	__be16		paylen;
690 	u8		next_hdr;
691 	u8		hop_limit;
692 	union ib_gid	sgid;
693 	union ib_gid	dgid;
694 };
695 
696 union rdma_network_hdr {
697 	struct ib_grh ibgrh;
698 	struct {
699 		/* The IB spec states that if it's IPv4, the header
700 		 * is located in the last 20 bytes of the header.
701 		 */
702 		u8		reserved[20];
703 		struct iphdr	roce4grh;
704 	};
705 };
706 
707 #define IB_QPN_MASK		0xFFFFFF
708 
709 enum {
710 	IB_MULTICAST_QPN = 0xffffff
711 };
712 
713 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
714 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
715 
716 enum ib_ah_flags {
717 	IB_AH_GRH	= 1
718 };
719 
720 enum ib_rate {
721 	IB_RATE_PORT_CURRENT = 0,
722 	IB_RATE_2_5_GBPS = 2,
723 	IB_RATE_5_GBPS   = 5,
724 	IB_RATE_10_GBPS  = 3,
725 	IB_RATE_20_GBPS  = 6,
726 	IB_RATE_30_GBPS  = 4,
727 	IB_RATE_40_GBPS  = 7,
728 	IB_RATE_60_GBPS  = 8,
729 	IB_RATE_80_GBPS  = 9,
730 	IB_RATE_120_GBPS = 10,
731 	IB_RATE_14_GBPS  = 11,
732 	IB_RATE_56_GBPS  = 12,
733 	IB_RATE_112_GBPS = 13,
734 	IB_RATE_168_GBPS = 14,
735 	IB_RATE_25_GBPS  = 15,
736 	IB_RATE_100_GBPS = 16,
737 	IB_RATE_200_GBPS = 17,
738 	IB_RATE_300_GBPS = 18,
739 	IB_RATE_28_GBPS  = 19,
740 	IB_RATE_50_GBPS  = 20,
741 	IB_RATE_400_GBPS = 21,
742 	IB_RATE_600_GBPS = 22,
743 };
744 
745 /**
746  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
747  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
748  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
749  * @rate: rate to convert.
750  */
751 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
752 
753 /**
754  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
755  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
756  * @rate: rate to convert.
757  */
758 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
759 
760 
761 /**
762  * enum ib_mr_type - memory region type
763  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
764  *                            normal registration
765  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
766  *                            signature operations (data-integrity
767  *                            capable regions)
768  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
769  *                            register any arbitrary sg lists (without
770  *                            the normal mr constraints - see
771  *                            ib_map_mr_sg)
772  */
773 enum ib_mr_type {
774 	IB_MR_TYPE_MEM_REG,
775 	IB_MR_TYPE_SIGNATURE,
776 	IB_MR_TYPE_SG_GAPS,
777 };
778 
779 /**
780  * Signature types
781  * IB_SIG_TYPE_NONE: Unprotected.
782  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
783  */
784 enum ib_signature_type {
785 	IB_SIG_TYPE_NONE,
786 	IB_SIG_TYPE_T10_DIF,
787 };
788 
789 /**
790  * Signature T10-DIF block-guard types
791  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
792  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
793  */
794 enum ib_t10_dif_bg_type {
795 	IB_T10DIF_CRC,
796 	IB_T10DIF_CSUM
797 };
798 
799 /**
800  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
801  *     domain.
802  * @bg_type: T10-DIF block guard type (CRC|CSUM)
803  * @pi_interval: protection information interval.
804  * @bg: seed of guard computation.
805  * @app_tag: application tag of guard block
806  * @ref_tag: initial guard block reference tag.
807  * @ref_remap: Indicate wethear the reftag increments each block
808  * @app_escape: Indicate to skip block check if apptag=0xffff
809  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
810  * @apptag_check_mask: check bitmask of application tag.
811  */
812 struct ib_t10_dif_domain {
813 	enum ib_t10_dif_bg_type bg_type;
814 	u16			pi_interval;
815 	u16			bg;
816 	u16			app_tag;
817 	u32			ref_tag;
818 	bool			ref_remap;
819 	bool			app_escape;
820 	bool			ref_escape;
821 	u16			apptag_check_mask;
822 };
823 
824 /**
825  * struct ib_sig_domain - Parameters for signature domain
826  * @sig_type: specific signauture type
827  * @sig: union of all signature domain attributes that may
828  *     be used to set domain layout.
829  */
830 struct ib_sig_domain {
831 	enum ib_signature_type sig_type;
832 	union {
833 		struct ib_t10_dif_domain dif;
834 	} sig;
835 };
836 
837 /**
838  * struct ib_sig_attrs - Parameters for signature handover operation
839  * @check_mask: bitmask for signature byte check (8 bytes)
840  * @mem: memory domain layout desciptor.
841  * @wire: wire domain layout desciptor.
842  */
843 struct ib_sig_attrs {
844 	u8			check_mask;
845 	struct ib_sig_domain	mem;
846 	struct ib_sig_domain	wire;
847 };
848 
849 enum ib_sig_err_type {
850 	IB_SIG_BAD_GUARD,
851 	IB_SIG_BAD_REFTAG,
852 	IB_SIG_BAD_APPTAG,
853 };
854 
855 /**
856  * Signature check masks (8 bytes in total) according to the T10-PI standard:
857  *  -------- -------- ------------
858  * | GUARD  | APPTAG |   REFTAG   |
859  * |  2B    |  2B    |    4B      |
860  *  -------- -------- ------------
861  */
862 enum {
863 	IB_SIG_CHECK_GUARD	= 0xc0,
864 	IB_SIG_CHECK_APPTAG	= 0x30,
865 	IB_SIG_CHECK_REFTAG	= 0x0f,
866 };
867 
868 /**
869  * struct ib_sig_err - signature error descriptor
870  */
871 struct ib_sig_err {
872 	enum ib_sig_err_type	err_type;
873 	u32			expected;
874 	u32			actual;
875 	u64			sig_err_offset;
876 	u32			key;
877 };
878 
879 enum ib_mr_status_check {
880 	IB_MR_CHECK_SIG_STATUS = 1,
881 };
882 
883 /**
884  * struct ib_mr_status - Memory region status container
885  *
886  * @fail_status: Bitmask of MR checks status. For each
887  *     failed check a corresponding status bit is set.
888  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
889  *     failure.
890  */
891 struct ib_mr_status {
892 	u32		    fail_status;
893 	struct ib_sig_err   sig_err;
894 };
895 
896 /**
897  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
898  * enum.
899  * @mult: multiple to convert.
900  */
901 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
902 
903 enum rdma_ah_attr_type {
904 	RDMA_AH_ATTR_TYPE_UNDEFINED,
905 	RDMA_AH_ATTR_TYPE_IB,
906 	RDMA_AH_ATTR_TYPE_ROCE,
907 	RDMA_AH_ATTR_TYPE_OPA,
908 };
909 
910 struct ib_ah_attr {
911 	u16			dlid;
912 	u8			src_path_bits;
913 };
914 
915 struct roce_ah_attr {
916 	u8			dmac[ETH_ALEN];
917 };
918 
919 struct opa_ah_attr {
920 	u32			dlid;
921 	u8			src_path_bits;
922 	bool			make_grd;
923 };
924 
925 struct rdma_ah_attr {
926 	struct ib_global_route	grh;
927 	u8			sl;
928 	u8			static_rate;
929 	u8			port_num;
930 	u8			ah_flags;
931 	enum rdma_ah_attr_type type;
932 	union {
933 		struct ib_ah_attr ib;
934 		struct roce_ah_attr roce;
935 		struct opa_ah_attr opa;
936 	};
937 };
938 
939 enum ib_wc_status {
940 	IB_WC_SUCCESS,
941 	IB_WC_LOC_LEN_ERR,
942 	IB_WC_LOC_QP_OP_ERR,
943 	IB_WC_LOC_EEC_OP_ERR,
944 	IB_WC_LOC_PROT_ERR,
945 	IB_WC_WR_FLUSH_ERR,
946 	IB_WC_MW_BIND_ERR,
947 	IB_WC_BAD_RESP_ERR,
948 	IB_WC_LOC_ACCESS_ERR,
949 	IB_WC_REM_INV_REQ_ERR,
950 	IB_WC_REM_ACCESS_ERR,
951 	IB_WC_REM_OP_ERR,
952 	IB_WC_RETRY_EXC_ERR,
953 	IB_WC_RNR_RETRY_EXC_ERR,
954 	IB_WC_LOC_RDD_VIOL_ERR,
955 	IB_WC_REM_INV_RD_REQ_ERR,
956 	IB_WC_REM_ABORT_ERR,
957 	IB_WC_INV_EECN_ERR,
958 	IB_WC_INV_EEC_STATE_ERR,
959 	IB_WC_FATAL_ERR,
960 	IB_WC_RESP_TIMEOUT_ERR,
961 	IB_WC_GENERAL_ERR
962 };
963 
964 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
965 
966 enum ib_wc_opcode {
967 	IB_WC_SEND,
968 	IB_WC_RDMA_WRITE,
969 	IB_WC_RDMA_READ,
970 	IB_WC_COMP_SWAP,
971 	IB_WC_FETCH_ADD,
972 	IB_WC_LSO,
973 	IB_WC_LOCAL_INV,
974 	IB_WC_REG_MR,
975 	IB_WC_MASKED_COMP_SWAP,
976 	IB_WC_MASKED_FETCH_ADD,
977 /*
978  * Set value of IB_WC_RECV so consumers can test if a completion is a
979  * receive by testing (opcode & IB_WC_RECV).
980  */
981 	IB_WC_RECV			= 1 << 7,
982 	IB_WC_RECV_RDMA_WITH_IMM
983 };
984 
985 enum ib_wc_flags {
986 	IB_WC_GRH		= 1,
987 	IB_WC_WITH_IMM		= (1<<1),
988 	IB_WC_WITH_INVALIDATE	= (1<<2),
989 	IB_WC_IP_CSUM_OK	= (1<<3),
990 	IB_WC_WITH_SMAC		= (1<<4),
991 	IB_WC_WITH_VLAN		= (1<<5),
992 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
993 };
994 
995 struct ib_wc {
996 	union {
997 		u64		wr_id;
998 		struct ib_cqe	*wr_cqe;
999 	};
1000 	enum ib_wc_status	status;
1001 	enum ib_wc_opcode	opcode;
1002 	u32			vendor_err;
1003 	u32			byte_len;
1004 	struct ib_qp	       *qp;
1005 	union {
1006 		__be32		imm_data;
1007 		u32		invalidate_rkey;
1008 	} ex;
1009 	u32			src_qp;
1010 	u32			slid;
1011 	int			wc_flags;
1012 	u16			pkey_index;
1013 	u8			sl;
1014 	u8			dlid_path_bits;
1015 	u8			port_num;	/* valid only for DR SMPs on switches */
1016 	u8			smac[ETH_ALEN];
1017 	u16			vlan_id;
1018 	u8			network_hdr_type;
1019 };
1020 
1021 enum ib_cq_notify_flags {
1022 	IB_CQ_SOLICITED			= 1 << 0,
1023 	IB_CQ_NEXT_COMP			= 1 << 1,
1024 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1025 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
1026 };
1027 
1028 enum ib_srq_type {
1029 	IB_SRQT_BASIC,
1030 	IB_SRQT_XRC,
1031 	IB_SRQT_TM,
1032 };
1033 
1034 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1035 {
1036 	return srq_type == IB_SRQT_XRC ||
1037 	       srq_type == IB_SRQT_TM;
1038 }
1039 
1040 enum ib_srq_attr_mask {
1041 	IB_SRQ_MAX_WR	= 1 << 0,
1042 	IB_SRQ_LIMIT	= 1 << 1,
1043 };
1044 
1045 struct ib_srq_attr {
1046 	u32	max_wr;
1047 	u32	max_sge;
1048 	u32	srq_limit;
1049 };
1050 
1051 struct ib_srq_init_attr {
1052 	void		      (*event_handler)(struct ib_event *, void *);
1053 	void		       *srq_context;
1054 	struct ib_srq_attr	attr;
1055 	enum ib_srq_type	srq_type;
1056 
1057 	struct {
1058 		struct ib_cq   *cq;
1059 		union {
1060 			struct {
1061 				struct ib_xrcd *xrcd;
1062 			} xrc;
1063 
1064 			struct {
1065 				u32		max_num_tags;
1066 			} tag_matching;
1067 		};
1068 	} ext;
1069 };
1070 
1071 struct ib_qp_cap {
1072 	u32	max_send_wr;
1073 	u32	max_recv_wr;
1074 	u32	max_send_sge;
1075 	u32	max_recv_sge;
1076 	u32	max_inline_data;
1077 
1078 	/*
1079 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1080 	 * ib_create_qp() will calculate the right amount of neededed WRs
1081 	 * and MRs based on this.
1082 	 */
1083 	u32	max_rdma_ctxs;
1084 };
1085 
1086 enum ib_sig_type {
1087 	IB_SIGNAL_ALL_WR,
1088 	IB_SIGNAL_REQ_WR
1089 };
1090 
1091 enum ib_qp_type {
1092 	/*
1093 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1094 	 * here (and in that order) since the MAD layer uses them as
1095 	 * indices into a 2-entry table.
1096 	 */
1097 	IB_QPT_SMI,
1098 	IB_QPT_GSI,
1099 
1100 	IB_QPT_RC,
1101 	IB_QPT_UC,
1102 	IB_QPT_UD,
1103 	IB_QPT_RAW_IPV6,
1104 	IB_QPT_RAW_ETHERTYPE,
1105 	IB_QPT_RAW_PACKET = 8,
1106 	IB_QPT_XRC_INI = 9,
1107 	IB_QPT_XRC_TGT,
1108 	IB_QPT_MAX,
1109 	IB_QPT_DRIVER = 0xFF,
1110 	/* Reserve a range for qp types internal to the low level driver.
1111 	 * These qp types will not be visible at the IB core layer, so the
1112 	 * IB_QPT_MAX usages should not be affected in the core layer
1113 	 */
1114 	IB_QPT_RESERVED1 = 0x1000,
1115 	IB_QPT_RESERVED2,
1116 	IB_QPT_RESERVED3,
1117 	IB_QPT_RESERVED4,
1118 	IB_QPT_RESERVED5,
1119 	IB_QPT_RESERVED6,
1120 	IB_QPT_RESERVED7,
1121 	IB_QPT_RESERVED8,
1122 	IB_QPT_RESERVED9,
1123 	IB_QPT_RESERVED10,
1124 };
1125 
1126 enum ib_qp_create_flags {
1127 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1128 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1129 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1130 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1131 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1132 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1133 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1134 	/* FREE					= 1 << 7, */
1135 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1136 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1137 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1138 	IB_QP_CREATE_PCI_WRITE_END_PADDING	= 1 << 11,
1139 	/* reserve bits 26-31 for low level drivers' internal use */
1140 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1141 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1142 };
1143 
1144 /*
1145  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1146  * callback to destroy the passed in QP.
1147  */
1148 
1149 struct ib_qp_init_attr {
1150 	/* Consumer's event_handler callback must not block */
1151 	void                  (*event_handler)(struct ib_event *, void *);
1152 
1153 	void		       *qp_context;
1154 	struct ib_cq	       *send_cq;
1155 	struct ib_cq	       *recv_cq;
1156 	struct ib_srq	       *srq;
1157 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1158 	struct ib_qp_cap	cap;
1159 	enum ib_sig_type	sq_sig_type;
1160 	enum ib_qp_type		qp_type;
1161 	u32			create_flags;
1162 
1163 	/*
1164 	 * Only needed for special QP types, or when using the RW API.
1165 	 */
1166 	u8			port_num;
1167 	struct ib_rwq_ind_table *rwq_ind_tbl;
1168 	u32			source_qpn;
1169 };
1170 
1171 struct ib_qp_open_attr {
1172 	void                  (*event_handler)(struct ib_event *, void *);
1173 	void		       *qp_context;
1174 	u32			qp_num;
1175 	enum ib_qp_type		qp_type;
1176 };
1177 
1178 enum ib_rnr_timeout {
1179 	IB_RNR_TIMER_655_36 =  0,
1180 	IB_RNR_TIMER_000_01 =  1,
1181 	IB_RNR_TIMER_000_02 =  2,
1182 	IB_RNR_TIMER_000_03 =  3,
1183 	IB_RNR_TIMER_000_04 =  4,
1184 	IB_RNR_TIMER_000_06 =  5,
1185 	IB_RNR_TIMER_000_08 =  6,
1186 	IB_RNR_TIMER_000_12 =  7,
1187 	IB_RNR_TIMER_000_16 =  8,
1188 	IB_RNR_TIMER_000_24 =  9,
1189 	IB_RNR_TIMER_000_32 = 10,
1190 	IB_RNR_TIMER_000_48 = 11,
1191 	IB_RNR_TIMER_000_64 = 12,
1192 	IB_RNR_TIMER_000_96 = 13,
1193 	IB_RNR_TIMER_001_28 = 14,
1194 	IB_RNR_TIMER_001_92 = 15,
1195 	IB_RNR_TIMER_002_56 = 16,
1196 	IB_RNR_TIMER_003_84 = 17,
1197 	IB_RNR_TIMER_005_12 = 18,
1198 	IB_RNR_TIMER_007_68 = 19,
1199 	IB_RNR_TIMER_010_24 = 20,
1200 	IB_RNR_TIMER_015_36 = 21,
1201 	IB_RNR_TIMER_020_48 = 22,
1202 	IB_RNR_TIMER_030_72 = 23,
1203 	IB_RNR_TIMER_040_96 = 24,
1204 	IB_RNR_TIMER_061_44 = 25,
1205 	IB_RNR_TIMER_081_92 = 26,
1206 	IB_RNR_TIMER_122_88 = 27,
1207 	IB_RNR_TIMER_163_84 = 28,
1208 	IB_RNR_TIMER_245_76 = 29,
1209 	IB_RNR_TIMER_327_68 = 30,
1210 	IB_RNR_TIMER_491_52 = 31
1211 };
1212 
1213 enum ib_qp_attr_mask {
1214 	IB_QP_STATE			= 1,
1215 	IB_QP_CUR_STATE			= (1<<1),
1216 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1217 	IB_QP_ACCESS_FLAGS		= (1<<3),
1218 	IB_QP_PKEY_INDEX		= (1<<4),
1219 	IB_QP_PORT			= (1<<5),
1220 	IB_QP_QKEY			= (1<<6),
1221 	IB_QP_AV			= (1<<7),
1222 	IB_QP_PATH_MTU			= (1<<8),
1223 	IB_QP_TIMEOUT			= (1<<9),
1224 	IB_QP_RETRY_CNT			= (1<<10),
1225 	IB_QP_RNR_RETRY			= (1<<11),
1226 	IB_QP_RQ_PSN			= (1<<12),
1227 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1228 	IB_QP_ALT_PATH			= (1<<14),
1229 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1230 	IB_QP_SQ_PSN			= (1<<16),
1231 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1232 	IB_QP_PATH_MIG_STATE		= (1<<18),
1233 	IB_QP_CAP			= (1<<19),
1234 	IB_QP_DEST_QPN			= (1<<20),
1235 	IB_QP_RESERVED1			= (1<<21),
1236 	IB_QP_RESERVED2			= (1<<22),
1237 	IB_QP_RESERVED3			= (1<<23),
1238 	IB_QP_RESERVED4			= (1<<24),
1239 	IB_QP_RATE_LIMIT		= (1<<25),
1240 };
1241 
1242 enum ib_qp_state {
1243 	IB_QPS_RESET,
1244 	IB_QPS_INIT,
1245 	IB_QPS_RTR,
1246 	IB_QPS_RTS,
1247 	IB_QPS_SQD,
1248 	IB_QPS_SQE,
1249 	IB_QPS_ERR
1250 };
1251 
1252 enum ib_mig_state {
1253 	IB_MIG_MIGRATED,
1254 	IB_MIG_REARM,
1255 	IB_MIG_ARMED
1256 };
1257 
1258 enum ib_mw_type {
1259 	IB_MW_TYPE_1 = 1,
1260 	IB_MW_TYPE_2 = 2
1261 };
1262 
1263 struct ib_qp_attr {
1264 	enum ib_qp_state	qp_state;
1265 	enum ib_qp_state	cur_qp_state;
1266 	enum ib_mtu		path_mtu;
1267 	enum ib_mig_state	path_mig_state;
1268 	u32			qkey;
1269 	u32			rq_psn;
1270 	u32			sq_psn;
1271 	u32			dest_qp_num;
1272 	int			qp_access_flags;
1273 	struct ib_qp_cap	cap;
1274 	struct rdma_ah_attr	ah_attr;
1275 	struct rdma_ah_attr	alt_ah_attr;
1276 	u16			pkey_index;
1277 	u16			alt_pkey_index;
1278 	u8			en_sqd_async_notify;
1279 	u8			sq_draining;
1280 	u8			max_rd_atomic;
1281 	u8			max_dest_rd_atomic;
1282 	u8			min_rnr_timer;
1283 	u8			port_num;
1284 	u8			timeout;
1285 	u8			retry_cnt;
1286 	u8			rnr_retry;
1287 	u8			alt_port_num;
1288 	u8			alt_timeout;
1289 	u32			rate_limit;
1290 };
1291 
1292 enum ib_wr_opcode {
1293 	/* These are shared with userspace */
1294 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1295 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1296 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1297 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1298 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1299 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1300 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1301 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1302 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1303 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1304 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1305 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1306 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1307 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1308 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1309 
1310 	/* These are kernel only and can not be issued by userspace */
1311 	IB_WR_REG_MR = 0x20,
1312 	IB_WR_REG_SIG_MR,
1313 
1314 	/* reserve values for low level drivers' internal use.
1315 	 * These values will not be used at all in the ib core layer.
1316 	 */
1317 	IB_WR_RESERVED1 = 0xf0,
1318 	IB_WR_RESERVED2,
1319 	IB_WR_RESERVED3,
1320 	IB_WR_RESERVED4,
1321 	IB_WR_RESERVED5,
1322 	IB_WR_RESERVED6,
1323 	IB_WR_RESERVED7,
1324 	IB_WR_RESERVED8,
1325 	IB_WR_RESERVED9,
1326 	IB_WR_RESERVED10,
1327 };
1328 
1329 enum ib_send_flags {
1330 	IB_SEND_FENCE		= 1,
1331 	IB_SEND_SIGNALED	= (1<<1),
1332 	IB_SEND_SOLICITED	= (1<<2),
1333 	IB_SEND_INLINE		= (1<<3),
1334 	IB_SEND_IP_CSUM		= (1<<4),
1335 
1336 	/* reserve bits 26-31 for low level drivers' internal use */
1337 	IB_SEND_RESERVED_START	= (1 << 26),
1338 	IB_SEND_RESERVED_END	= (1 << 31),
1339 };
1340 
1341 struct ib_sge {
1342 	u64	addr;
1343 	u32	length;
1344 	u32	lkey;
1345 };
1346 
1347 struct ib_cqe {
1348 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1349 };
1350 
1351 struct ib_send_wr {
1352 	struct ib_send_wr      *next;
1353 	union {
1354 		u64		wr_id;
1355 		struct ib_cqe	*wr_cqe;
1356 	};
1357 	struct ib_sge	       *sg_list;
1358 	int			num_sge;
1359 	enum ib_wr_opcode	opcode;
1360 	int			send_flags;
1361 	union {
1362 		__be32		imm_data;
1363 		u32		invalidate_rkey;
1364 	} ex;
1365 };
1366 
1367 struct ib_rdma_wr {
1368 	struct ib_send_wr	wr;
1369 	u64			remote_addr;
1370 	u32			rkey;
1371 };
1372 
1373 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1374 {
1375 	return container_of(wr, struct ib_rdma_wr, wr);
1376 }
1377 
1378 struct ib_atomic_wr {
1379 	struct ib_send_wr	wr;
1380 	u64			remote_addr;
1381 	u64			compare_add;
1382 	u64			swap;
1383 	u64			compare_add_mask;
1384 	u64			swap_mask;
1385 	u32			rkey;
1386 };
1387 
1388 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1389 {
1390 	return container_of(wr, struct ib_atomic_wr, wr);
1391 }
1392 
1393 struct ib_ud_wr {
1394 	struct ib_send_wr	wr;
1395 	struct ib_ah		*ah;
1396 	void			*header;
1397 	int			hlen;
1398 	int			mss;
1399 	u32			remote_qpn;
1400 	u32			remote_qkey;
1401 	u16			pkey_index; /* valid for GSI only */
1402 	u8			port_num;   /* valid for DR SMPs on switch only */
1403 };
1404 
1405 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1406 {
1407 	return container_of(wr, struct ib_ud_wr, wr);
1408 }
1409 
1410 struct ib_reg_wr {
1411 	struct ib_send_wr	wr;
1412 	struct ib_mr		*mr;
1413 	u32			key;
1414 	int			access;
1415 };
1416 
1417 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1418 {
1419 	return container_of(wr, struct ib_reg_wr, wr);
1420 }
1421 
1422 struct ib_sig_handover_wr {
1423 	struct ib_send_wr	wr;
1424 	struct ib_sig_attrs    *sig_attrs;
1425 	struct ib_mr	       *sig_mr;
1426 	int			access_flags;
1427 	struct ib_sge	       *prot;
1428 };
1429 
1430 static inline const struct ib_sig_handover_wr *
1431 sig_handover_wr(const struct ib_send_wr *wr)
1432 {
1433 	return container_of(wr, struct ib_sig_handover_wr, wr);
1434 }
1435 
1436 struct ib_recv_wr {
1437 	struct ib_recv_wr      *next;
1438 	union {
1439 		u64		wr_id;
1440 		struct ib_cqe	*wr_cqe;
1441 	};
1442 	struct ib_sge	       *sg_list;
1443 	int			num_sge;
1444 };
1445 
1446 enum ib_access_flags {
1447 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1448 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1449 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1450 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1451 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1452 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1453 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1454 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1455 
1456 	IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1457 };
1458 
1459 /*
1460  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1461  * are hidden here instead of a uapi header!
1462  */
1463 enum ib_mr_rereg_flags {
1464 	IB_MR_REREG_TRANS	= 1,
1465 	IB_MR_REREG_PD		= (1<<1),
1466 	IB_MR_REREG_ACCESS	= (1<<2),
1467 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1468 };
1469 
1470 struct ib_fmr_attr {
1471 	int	max_pages;
1472 	int	max_maps;
1473 	u8	page_shift;
1474 };
1475 
1476 struct ib_umem;
1477 
1478 enum rdma_remove_reason {
1479 	/*
1480 	 * Userspace requested uobject deletion or initial try
1481 	 * to remove uobject via cleanup. Call could fail
1482 	 */
1483 	RDMA_REMOVE_DESTROY,
1484 	/* Context deletion. This call should delete the actual object itself */
1485 	RDMA_REMOVE_CLOSE,
1486 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1487 	RDMA_REMOVE_DRIVER_REMOVE,
1488 	/* uobj is being cleaned-up before being committed */
1489 	RDMA_REMOVE_ABORT,
1490 };
1491 
1492 struct ib_rdmacg_object {
1493 #ifdef CONFIG_CGROUP_RDMA
1494 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1495 #endif
1496 };
1497 
1498 struct ib_ucontext {
1499 	struct ib_device       *device;
1500 	struct ib_uverbs_file  *ufile;
1501 	/*
1502 	 * 'closing' can be read by the driver only during a destroy callback,
1503 	 * it is set when we are closing the file descriptor and indicates
1504 	 * that mm_sem may be locked.
1505 	 */
1506 	bool closing;
1507 
1508 	bool cleanup_retryable;
1509 
1510 	void (*invalidate_range)(struct ib_umem_odp *umem_odp,
1511 				 unsigned long start, unsigned long end);
1512 	struct mutex per_mm_list_lock;
1513 	struct list_head per_mm_list;
1514 
1515 	struct ib_rdmacg_object	cg_obj;
1516 	/*
1517 	 * Implementation details of the RDMA core, don't use in drivers:
1518 	 */
1519 	struct rdma_restrack_entry res;
1520 };
1521 
1522 struct ib_uobject {
1523 	u64			user_handle;	/* handle given to us by userspace */
1524 	/* ufile & ucontext owning this object */
1525 	struct ib_uverbs_file  *ufile;
1526 	/* FIXME, save memory: ufile->context == context */
1527 	struct ib_ucontext     *context;	/* associated user context */
1528 	void		       *object;		/* containing object */
1529 	struct list_head	list;		/* link to context's list */
1530 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1531 	int			id;		/* index into kernel idr */
1532 	struct kref		ref;
1533 	atomic_t		usecnt;		/* protects exclusive access */
1534 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1535 
1536 	const struct uverbs_api_object *uapi_object;
1537 };
1538 
1539 struct ib_udata {
1540 	const void __user *inbuf;
1541 	void __user *outbuf;
1542 	size_t       inlen;
1543 	size_t       outlen;
1544 };
1545 
1546 struct ib_pd {
1547 	u32			local_dma_lkey;
1548 	u32			flags;
1549 	struct ib_device       *device;
1550 	struct ib_uobject      *uobject;
1551 	atomic_t          	usecnt; /* count all resources */
1552 
1553 	u32			unsafe_global_rkey;
1554 
1555 	/*
1556 	 * Implementation details of the RDMA core, don't use in drivers:
1557 	 */
1558 	struct ib_mr	       *__internal_mr;
1559 	struct rdma_restrack_entry res;
1560 };
1561 
1562 struct ib_xrcd {
1563 	struct ib_device       *device;
1564 	atomic_t		usecnt; /* count all exposed resources */
1565 	struct inode	       *inode;
1566 
1567 	struct mutex		tgt_qp_mutex;
1568 	struct list_head	tgt_qp_list;
1569 };
1570 
1571 struct ib_ah {
1572 	struct ib_device	*device;
1573 	struct ib_pd		*pd;
1574 	struct ib_uobject	*uobject;
1575 	const struct ib_gid_attr *sgid_attr;
1576 	enum rdma_ah_attr_type	type;
1577 };
1578 
1579 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1580 
1581 enum ib_poll_context {
1582 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1583 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1584 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1585 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1586 };
1587 
1588 struct ib_cq {
1589 	struct ib_device       *device;
1590 	struct ib_uobject      *uobject;
1591 	ib_comp_handler   	comp_handler;
1592 	void                  (*event_handler)(struct ib_event *, void *);
1593 	void                   *cq_context;
1594 	int               	cqe;
1595 	atomic_t          	usecnt; /* count number of work queues */
1596 	enum ib_poll_context	poll_ctx;
1597 	struct ib_wc		*wc;
1598 	union {
1599 		struct irq_poll		iop;
1600 		struct work_struct	work;
1601 	};
1602 	struct workqueue_struct *comp_wq;
1603 	/*
1604 	 * Implementation details of the RDMA core, don't use in drivers:
1605 	 */
1606 	struct rdma_restrack_entry res;
1607 };
1608 
1609 struct ib_srq {
1610 	struct ib_device       *device;
1611 	struct ib_pd	       *pd;
1612 	struct ib_uobject      *uobject;
1613 	void		      (*event_handler)(struct ib_event *, void *);
1614 	void		       *srq_context;
1615 	enum ib_srq_type	srq_type;
1616 	atomic_t		usecnt;
1617 
1618 	struct {
1619 		struct ib_cq   *cq;
1620 		union {
1621 			struct {
1622 				struct ib_xrcd *xrcd;
1623 				u32		srq_num;
1624 			} xrc;
1625 		};
1626 	} ext;
1627 };
1628 
1629 enum ib_raw_packet_caps {
1630 	/* Strip cvlan from incoming packet and report it in the matching work
1631 	 * completion is supported.
1632 	 */
1633 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1634 	/* Scatter FCS field of an incoming packet to host memory is supported.
1635 	 */
1636 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1637 	/* Checksum offloads are supported (for both send and receive). */
1638 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1639 	/* When a packet is received for an RQ with no receive WQEs, the
1640 	 * packet processing is delayed.
1641 	 */
1642 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1643 };
1644 
1645 enum ib_wq_type {
1646 	IB_WQT_RQ
1647 };
1648 
1649 enum ib_wq_state {
1650 	IB_WQS_RESET,
1651 	IB_WQS_RDY,
1652 	IB_WQS_ERR
1653 };
1654 
1655 struct ib_wq {
1656 	struct ib_device       *device;
1657 	struct ib_uobject      *uobject;
1658 	void		    *wq_context;
1659 	void		    (*event_handler)(struct ib_event *, void *);
1660 	struct ib_pd	       *pd;
1661 	struct ib_cq	       *cq;
1662 	u32		wq_num;
1663 	enum ib_wq_state       state;
1664 	enum ib_wq_type	wq_type;
1665 	atomic_t		usecnt;
1666 };
1667 
1668 enum ib_wq_flags {
1669 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1670 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1671 	IB_WQ_FLAGS_DELAY_DROP		= 1 << 2,
1672 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1673 };
1674 
1675 struct ib_wq_init_attr {
1676 	void		       *wq_context;
1677 	enum ib_wq_type	wq_type;
1678 	u32		max_wr;
1679 	u32		max_sge;
1680 	struct	ib_cq	       *cq;
1681 	void		    (*event_handler)(struct ib_event *, void *);
1682 	u32		create_flags; /* Use enum ib_wq_flags */
1683 };
1684 
1685 enum ib_wq_attr_mask {
1686 	IB_WQ_STATE		= 1 << 0,
1687 	IB_WQ_CUR_STATE		= 1 << 1,
1688 	IB_WQ_FLAGS		= 1 << 2,
1689 };
1690 
1691 struct ib_wq_attr {
1692 	enum	ib_wq_state	wq_state;
1693 	enum	ib_wq_state	curr_wq_state;
1694 	u32			flags; /* Use enum ib_wq_flags */
1695 	u32			flags_mask; /* Use enum ib_wq_flags */
1696 };
1697 
1698 struct ib_rwq_ind_table {
1699 	struct ib_device	*device;
1700 	struct ib_uobject      *uobject;
1701 	atomic_t		usecnt;
1702 	u32		ind_tbl_num;
1703 	u32		log_ind_tbl_size;
1704 	struct ib_wq	**ind_tbl;
1705 };
1706 
1707 struct ib_rwq_ind_table_init_attr {
1708 	u32		log_ind_tbl_size;
1709 	/* Each entry is a pointer to Receive Work Queue */
1710 	struct ib_wq	**ind_tbl;
1711 };
1712 
1713 enum port_pkey_state {
1714 	IB_PORT_PKEY_NOT_VALID = 0,
1715 	IB_PORT_PKEY_VALID = 1,
1716 	IB_PORT_PKEY_LISTED = 2,
1717 };
1718 
1719 struct ib_qp_security;
1720 
1721 struct ib_port_pkey {
1722 	enum port_pkey_state	state;
1723 	u16			pkey_index;
1724 	u8			port_num;
1725 	struct list_head	qp_list;
1726 	struct list_head	to_error_list;
1727 	struct ib_qp_security  *sec;
1728 };
1729 
1730 struct ib_ports_pkeys {
1731 	struct ib_port_pkey	main;
1732 	struct ib_port_pkey	alt;
1733 };
1734 
1735 struct ib_qp_security {
1736 	struct ib_qp	       *qp;
1737 	struct ib_device       *dev;
1738 	/* Hold this mutex when changing port and pkey settings. */
1739 	struct mutex		mutex;
1740 	struct ib_ports_pkeys  *ports_pkeys;
1741 	/* A list of all open shared QP handles.  Required to enforce security
1742 	 * properly for all users of a shared QP.
1743 	 */
1744 	struct list_head        shared_qp_list;
1745 	void                   *security;
1746 	bool			destroying;
1747 	atomic_t		error_list_count;
1748 	struct completion	error_complete;
1749 	int			error_comps_pending;
1750 };
1751 
1752 /*
1753  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1754  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1755  */
1756 struct ib_qp {
1757 	struct ib_device       *device;
1758 	struct ib_pd	       *pd;
1759 	struct ib_cq	       *send_cq;
1760 	struct ib_cq	       *recv_cq;
1761 	spinlock_t		mr_lock;
1762 	int			mrs_used;
1763 	struct list_head	rdma_mrs;
1764 	struct list_head	sig_mrs;
1765 	struct ib_srq	       *srq;
1766 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1767 	struct list_head	xrcd_list;
1768 
1769 	/* count times opened, mcast attaches, flow attaches */
1770 	atomic_t		usecnt;
1771 	struct list_head	open_list;
1772 	struct ib_qp           *real_qp;
1773 	struct ib_uobject      *uobject;
1774 	void                  (*event_handler)(struct ib_event *, void *);
1775 	void		       *qp_context;
1776 	/* sgid_attrs associated with the AV's */
1777 	const struct ib_gid_attr *av_sgid_attr;
1778 	const struct ib_gid_attr *alt_path_sgid_attr;
1779 	u32			qp_num;
1780 	u32			max_write_sge;
1781 	u32			max_read_sge;
1782 	enum ib_qp_type		qp_type;
1783 	struct ib_rwq_ind_table *rwq_ind_tbl;
1784 	struct ib_qp_security  *qp_sec;
1785 	u8			port;
1786 
1787 	/*
1788 	 * Implementation details of the RDMA core, don't use in drivers:
1789 	 */
1790 	struct rdma_restrack_entry     res;
1791 };
1792 
1793 struct ib_dm {
1794 	struct ib_device  *device;
1795 	u32		   length;
1796 	u32		   flags;
1797 	struct ib_uobject *uobject;
1798 	atomic_t	   usecnt;
1799 };
1800 
1801 struct ib_mr {
1802 	struct ib_device  *device;
1803 	struct ib_pd	  *pd;
1804 	u32		   lkey;
1805 	u32		   rkey;
1806 	u64		   iova;
1807 	u64		   length;
1808 	unsigned int	   page_size;
1809 	bool		   need_inval;
1810 	union {
1811 		struct ib_uobject	*uobject;	/* user */
1812 		struct list_head	qp_entry;	/* FR */
1813 	};
1814 
1815 	struct ib_dm      *dm;
1816 
1817 	/*
1818 	 * Implementation details of the RDMA core, don't use in drivers:
1819 	 */
1820 	struct rdma_restrack_entry res;
1821 };
1822 
1823 struct ib_mw {
1824 	struct ib_device	*device;
1825 	struct ib_pd		*pd;
1826 	struct ib_uobject	*uobject;
1827 	u32			rkey;
1828 	enum ib_mw_type         type;
1829 };
1830 
1831 struct ib_fmr {
1832 	struct ib_device	*device;
1833 	struct ib_pd		*pd;
1834 	struct list_head	list;
1835 	u32			lkey;
1836 	u32			rkey;
1837 };
1838 
1839 /* Supported steering options */
1840 enum ib_flow_attr_type {
1841 	/* steering according to rule specifications */
1842 	IB_FLOW_ATTR_NORMAL		= 0x0,
1843 	/* default unicast and multicast rule -
1844 	 * receive all Eth traffic which isn't steered to any QP
1845 	 */
1846 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1847 	/* default multicast rule -
1848 	 * receive all Eth multicast traffic which isn't steered to any QP
1849 	 */
1850 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1851 	/* sniffer rule - receive all port traffic */
1852 	IB_FLOW_ATTR_SNIFFER		= 0x3
1853 };
1854 
1855 /* Supported steering header types */
1856 enum ib_flow_spec_type {
1857 	/* L2 headers*/
1858 	IB_FLOW_SPEC_ETH		= 0x20,
1859 	IB_FLOW_SPEC_IB			= 0x22,
1860 	/* L3 header*/
1861 	IB_FLOW_SPEC_IPV4		= 0x30,
1862 	IB_FLOW_SPEC_IPV6		= 0x31,
1863 	IB_FLOW_SPEC_ESP                = 0x34,
1864 	/* L4 headers*/
1865 	IB_FLOW_SPEC_TCP		= 0x40,
1866 	IB_FLOW_SPEC_UDP		= 0x41,
1867 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1868 	IB_FLOW_SPEC_GRE		= 0x51,
1869 	IB_FLOW_SPEC_MPLS		= 0x60,
1870 	IB_FLOW_SPEC_INNER		= 0x100,
1871 	/* Actions */
1872 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1873 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1874 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1875 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1876 };
1877 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1878 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1879 
1880 /* Flow steering rule priority is set according to it's domain.
1881  * Lower domain value means higher priority.
1882  */
1883 enum ib_flow_domain {
1884 	IB_FLOW_DOMAIN_USER,
1885 	IB_FLOW_DOMAIN_ETHTOOL,
1886 	IB_FLOW_DOMAIN_RFS,
1887 	IB_FLOW_DOMAIN_NIC,
1888 	IB_FLOW_DOMAIN_NUM /* Must be last */
1889 };
1890 
1891 enum ib_flow_flags {
1892 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1893 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1894 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1895 };
1896 
1897 struct ib_flow_eth_filter {
1898 	u8	dst_mac[6];
1899 	u8	src_mac[6];
1900 	__be16	ether_type;
1901 	__be16	vlan_tag;
1902 	/* Must be last */
1903 	u8	real_sz[0];
1904 };
1905 
1906 struct ib_flow_spec_eth {
1907 	u32			  type;
1908 	u16			  size;
1909 	struct ib_flow_eth_filter val;
1910 	struct ib_flow_eth_filter mask;
1911 };
1912 
1913 struct ib_flow_ib_filter {
1914 	__be16 dlid;
1915 	__u8   sl;
1916 	/* Must be last */
1917 	u8	real_sz[0];
1918 };
1919 
1920 struct ib_flow_spec_ib {
1921 	u32			 type;
1922 	u16			 size;
1923 	struct ib_flow_ib_filter val;
1924 	struct ib_flow_ib_filter mask;
1925 };
1926 
1927 /* IPv4 header flags */
1928 enum ib_ipv4_flags {
1929 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1930 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1931 				    last have this flag set */
1932 };
1933 
1934 struct ib_flow_ipv4_filter {
1935 	__be32	src_ip;
1936 	__be32	dst_ip;
1937 	u8	proto;
1938 	u8	tos;
1939 	u8	ttl;
1940 	u8	flags;
1941 	/* Must be last */
1942 	u8	real_sz[0];
1943 };
1944 
1945 struct ib_flow_spec_ipv4 {
1946 	u32			   type;
1947 	u16			   size;
1948 	struct ib_flow_ipv4_filter val;
1949 	struct ib_flow_ipv4_filter mask;
1950 };
1951 
1952 struct ib_flow_ipv6_filter {
1953 	u8	src_ip[16];
1954 	u8	dst_ip[16];
1955 	__be32	flow_label;
1956 	u8	next_hdr;
1957 	u8	traffic_class;
1958 	u8	hop_limit;
1959 	/* Must be last */
1960 	u8	real_sz[0];
1961 };
1962 
1963 struct ib_flow_spec_ipv6 {
1964 	u32			   type;
1965 	u16			   size;
1966 	struct ib_flow_ipv6_filter val;
1967 	struct ib_flow_ipv6_filter mask;
1968 };
1969 
1970 struct ib_flow_tcp_udp_filter {
1971 	__be16	dst_port;
1972 	__be16	src_port;
1973 	/* Must be last */
1974 	u8	real_sz[0];
1975 };
1976 
1977 struct ib_flow_spec_tcp_udp {
1978 	u32			      type;
1979 	u16			      size;
1980 	struct ib_flow_tcp_udp_filter val;
1981 	struct ib_flow_tcp_udp_filter mask;
1982 };
1983 
1984 struct ib_flow_tunnel_filter {
1985 	__be32	tunnel_id;
1986 	u8	real_sz[0];
1987 };
1988 
1989 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1990  * the tunnel_id from val has the vni value
1991  */
1992 struct ib_flow_spec_tunnel {
1993 	u32			      type;
1994 	u16			      size;
1995 	struct ib_flow_tunnel_filter  val;
1996 	struct ib_flow_tunnel_filter  mask;
1997 };
1998 
1999 struct ib_flow_esp_filter {
2000 	__be32	spi;
2001 	__be32  seq;
2002 	/* Must be last */
2003 	u8	real_sz[0];
2004 };
2005 
2006 struct ib_flow_spec_esp {
2007 	u32                           type;
2008 	u16			      size;
2009 	struct ib_flow_esp_filter     val;
2010 	struct ib_flow_esp_filter     mask;
2011 };
2012 
2013 struct ib_flow_gre_filter {
2014 	__be16 c_ks_res0_ver;
2015 	__be16 protocol;
2016 	__be32 key;
2017 	/* Must be last */
2018 	u8	real_sz[0];
2019 };
2020 
2021 struct ib_flow_spec_gre {
2022 	u32                           type;
2023 	u16			      size;
2024 	struct ib_flow_gre_filter     val;
2025 	struct ib_flow_gre_filter     mask;
2026 };
2027 
2028 struct ib_flow_mpls_filter {
2029 	__be32 tag;
2030 	/* Must be last */
2031 	u8	real_sz[0];
2032 };
2033 
2034 struct ib_flow_spec_mpls {
2035 	u32                           type;
2036 	u16			      size;
2037 	struct ib_flow_mpls_filter     val;
2038 	struct ib_flow_mpls_filter     mask;
2039 };
2040 
2041 struct ib_flow_spec_action_tag {
2042 	enum ib_flow_spec_type	      type;
2043 	u16			      size;
2044 	u32                           tag_id;
2045 };
2046 
2047 struct ib_flow_spec_action_drop {
2048 	enum ib_flow_spec_type	      type;
2049 	u16			      size;
2050 };
2051 
2052 struct ib_flow_spec_action_handle {
2053 	enum ib_flow_spec_type	      type;
2054 	u16			      size;
2055 	struct ib_flow_action	     *act;
2056 };
2057 
2058 enum ib_counters_description {
2059 	IB_COUNTER_PACKETS,
2060 	IB_COUNTER_BYTES,
2061 };
2062 
2063 struct ib_flow_spec_action_count {
2064 	enum ib_flow_spec_type type;
2065 	u16 size;
2066 	struct ib_counters *counters;
2067 };
2068 
2069 union ib_flow_spec {
2070 	struct {
2071 		u32			type;
2072 		u16			size;
2073 	};
2074 	struct ib_flow_spec_eth		eth;
2075 	struct ib_flow_spec_ib		ib;
2076 	struct ib_flow_spec_ipv4        ipv4;
2077 	struct ib_flow_spec_tcp_udp	tcp_udp;
2078 	struct ib_flow_spec_ipv6        ipv6;
2079 	struct ib_flow_spec_tunnel      tunnel;
2080 	struct ib_flow_spec_esp		esp;
2081 	struct ib_flow_spec_gre		gre;
2082 	struct ib_flow_spec_mpls	mpls;
2083 	struct ib_flow_spec_action_tag  flow_tag;
2084 	struct ib_flow_spec_action_drop drop;
2085 	struct ib_flow_spec_action_handle action;
2086 	struct ib_flow_spec_action_count flow_count;
2087 };
2088 
2089 struct ib_flow_attr {
2090 	enum ib_flow_attr_type type;
2091 	u16	     size;
2092 	u16	     priority;
2093 	u32	     flags;
2094 	u8	     num_of_specs;
2095 	u8	     port;
2096 	union ib_flow_spec flows[];
2097 };
2098 
2099 struct ib_flow {
2100 	struct ib_qp		*qp;
2101 	struct ib_device	*device;
2102 	struct ib_uobject	*uobject;
2103 };
2104 
2105 enum ib_flow_action_type {
2106 	IB_FLOW_ACTION_UNSPECIFIED,
2107 	IB_FLOW_ACTION_ESP = 1,
2108 };
2109 
2110 struct ib_flow_action_attrs_esp_keymats {
2111 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2112 	union {
2113 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2114 	} keymat;
2115 };
2116 
2117 struct ib_flow_action_attrs_esp_replays {
2118 	enum ib_uverbs_flow_action_esp_replay			protocol;
2119 	union {
2120 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2121 	} replay;
2122 };
2123 
2124 enum ib_flow_action_attrs_esp_flags {
2125 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2126 	 * This is done in order to share the same flags between user-space and
2127 	 * kernel and spare an unnecessary translation.
2128 	 */
2129 
2130 	/* Kernel flags */
2131 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2132 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2133 };
2134 
2135 struct ib_flow_spec_list {
2136 	struct ib_flow_spec_list	*next;
2137 	union ib_flow_spec		spec;
2138 };
2139 
2140 struct ib_flow_action_attrs_esp {
2141 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2142 	struct ib_flow_action_attrs_esp_replays		*replay;
2143 	struct ib_flow_spec_list			*encap;
2144 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2145 	 * Value of 0 is a valid value.
2146 	 */
2147 	u32						esn;
2148 	u32						spi;
2149 	u32						seq;
2150 	u32						tfc_pad;
2151 	/* Use enum ib_flow_action_attrs_esp_flags */
2152 	u64						flags;
2153 	u64						hard_limit_pkts;
2154 };
2155 
2156 struct ib_flow_action {
2157 	struct ib_device		*device;
2158 	struct ib_uobject		*uobject;
2159 	enum ib_flow_action_type	type;
2160 	atomic_t			usecnt;
2161 };
2162 
2163 struct ib_mad_hdr;
2164 struct ib_grh;
2165 
2166 enum ib_process_mad_flags {
2167 	IB_MAD_IGNORE_MKEY	= 1,
2168 	IB_MAD_IGNORE_BKEY	= 2,
2169 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2170 };
2171 
2172 enum ib_mad_result {
2173 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2174 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2175 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2176 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2177 };
2178 
2179 struct ib_port_cache {
2180 	u64		      subnet_prefix;
2181 	struct ib_pkey_cache  *pkey;
2182 	struct ib_gid_table   *gid;
2183 	u8                     lmc;
2184 	enum ib_port_state     port_state;
2185 };
2186 
2187 struct ib_cache {
2188 	rwlock_t                lock;
2189 	struct ib_event_handler event_handler;
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 struct ib_port_data {
2202 	struct ib_device *ib_dev;
2203 
2204 	struct ib_port_immutable immutable;
2205 
2206 	spinlock_t pkey_list_lock;
2207 	struct list_head pkey_list;
2208 
2209 	struct ib_port_cache cache;
2210 
2211 	spinlock_t netdev_lock;
2212 	struct net_device __rcu *netdev;
2213 	struct hlist_node ndev_hash_link;
2214 };
2215 
2216 /* rdma netdev type - specifies protocol type */
2217 enum rdma_netdev_t {
2218 	RDMA_NETDEV_OPA_VNIC,
2219 	RDMA_NETDEV_IPOIB,
2220 };
2221 
2222 /**
2223  * struct rdma_netdev - rdma netdev
2224  * For cases where netstack interfacing is required.
2225  */
2226 struct rdma_netdev {
2227 	void              *clnt_priv;
2228 	struct ib_device  *hca;
2229 	u8                 port_num;
2230 
2231 	/*
2232 	 * cleanup function must be specified.
2233 	 * FIXME: This is only used for OPA_VNIC and that usage should be
2234 	 * removed too.
2235 	 */
2236 	void (*free_rdma_netdev)(struct net_device *netdev);
2237 
2238 	/* control functions */
2239 	void (*set_id)(struct net_device *netdev, int id);
2240 	/* send packet */
2241 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2242 		    struct ib_ah *address, u32 dqpn);
2243 	/* multicast */
2244 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2245 			    union ib_gid *gid, u16 mlid,
2246 			    int set_qkey, u32 qkey);
2247 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2248 			    union ib_gid *gid, u16 mlid);
2249 };
2250 
2251 struct rdma_netdev_alloc_params {
2252 	size_t sizeof_priv;
2253 	unsigned int txqs;
2254 	unsigned int rxqs;
2255 	void *param;
2256 
2257 	int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2258 				      struct net_device *netdev, void *param);
2259 };
2260 
2261 struct ib_counters {
2262 	struct ib_device	*device;
2263 	struct ib_uobject	*uobject;
2264 	/* num of objects attached */
2265 	atomic_t	usecnt;
2266 };
2267 
2268 struct ib_counters_read_attr {
2269 	u64	*counters_buff;
2270 	u32	ncounters;
2271 	u32	flags; /* use enum ib_read_counters_flags */
2272 };
2273 
2274 struct uverbs_attr_bundle;
2275 
2276 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member)                      \
2277 	.size_##ib_struct =                                                    \
2278 		(sizeof(struct drv_struct) +                                   \
2279 		 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) +      \
2280 		 BUILD_BUG_ON_ZERO(                                            \
2281 			 !__same_type(((struct drv_struct *)NULL)->member,     \
2282 				      struct ib_struct)))
2283 
2284 #define rdma_zalloc_drv_obj(ib_dev, ib_type)                                   \
2285 	((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, GFP_KERNEL))
2286 
2287 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2288 
2289 /**
2290  * struct ib_device_ops - InfiniBand device operations
2291  * This structure defines all the InfiniBand device operations, providers will
2292  * need to define the supported operations, otherwise they will be set to null.
2293  */
2294 struct ib_device_ops {
2295 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2296 			 const struct ib_send_wr **bad_send_wr);
2297 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2298 			 const struct ib_recv_wr **bad_recv_wr);
2299 	void (*drain_rq)(struct ib_qp *qp);
2300 	void (*drain_sq)(struct ib_qp *qp);
2301 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2302 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2303 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2304 	int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2305 	int (*post_srq_recv)(struct ib_srq *srq,
2306 			     const struct ib_recv_wr *recv_wr,
2307 			     const struct ib_recv_wr **bad_recv_wr);
2308 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2309 			   u8 port_num, const struct ib_wc *in_wc,
2310 			   const struct ib_grh *in_grh,
2311 			   const struct ib_mad_hdr *in_mad, size_t in_mad_size,
2312 			   struct ib_mad_hdr *out_mad, size_t *out_mad_size,
2313 			   u16 *out_mad_pkey_index);
2314 	int (*query_device)(struct ib_device *device,
2315 			    struct ib_device_attr *device_attr,
2316 			    struct ib_udata *udata);
2317 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2318 			     struct ib_device_modify *device_modify);
2319 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2320 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2321 						     int comp_vector);
2322 	int (*query_port)(struct ib_device *device, u8 port_num,
2323 			  struct ib_port_attr *port_attr);
2324 	int (*modify_port)(struct ib_device *device, u8 port_num,
2325 			   int port_modify_mask,
2326 			   struct ib_port_modify *port_modify);
2327 	/**
2328 	 * The following mandatory functions are used only at device
2329 	 * registration.  Keep functions such as these at the end of this
2330 	 * structure to avoid cache line misses when accessing struct ib_device
2331 	 * in fast paths.
2332 	 */
2333 	int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2334 				  struct ib_port_immutable *immutable);
2335 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2336 					       u8 port_num);
2337 	/**
2338 	 * When calling get_netdev, the HW vendor's driver should return the
2339 	 * net device of device @device at port @port_num or NULL if such
2340 	 * a net device doesn't exist. The vendor driver should call dev_hold
2341 	 * on this net device. The HW vendor's device driver must guarantee
2342 	 * that this function returns NULL before the net device has finished
2343 	 * NETDEV_UNREGISTER state.
2344 	 */
2345 	struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2346 	/**
2347 	 * rdma netdev operation
2348 	 *
2349 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2350 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2351 	 */
2352 	struct net_device *(*alloc_rdma_netdev)(
2353 		struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2354 		const char *name, unsigned char name_assign_type,
2355 		void (*setup)(struct net_device *));
2356 
2357 	int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2358 				      enum rdma_netdev_t type,
2359 				      struct rdma_netdev_alloc_params *params);
2360 	/**
2361 	 * query_gid should be return GID value for @device, when @port_num
2362 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2363 	 * is RoCE link layer.
2364 	 */
2365 	int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2366 			 union ib_gid *gid);
2367 	/**
2368 	 * When calling add_gid, the HW vendor's driver should add the gid
2369 	 * of device of port at gid index available at @attr. Meta-info of
2370 	 * that gid (for example, the network device related to this gid) is
2371 	 * available at @attr. @context allows the HW vendor driver to store
2372 	 * extra information together with a GID entry. The HW vendor driver may
2373 	 * allocate memory to contain this information and store it in @context
2374 	 * when a new GID entry is written to. Params are consistent until the
2375 	 * next call of add_gid or delete_gid. The function should return 0 on
2376 	 * success or error otherwise. The function could be called
2377 	 * concurrently for different ports. This function is only called when
2378 	 * roce_gid_table is used.
2379 	 */
2380 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2381 	/**
2382 	 * When calling del_gid, the HW vendor's driver should delete the
2383 	 * gid of device @device at gid index gid_index of port port_num
2384 	 * available in @attr.
2385 	 * Upon the deletion of a GID entry, the HW vendor must free any
2386 	 * allocated memory. The caller will clear @context afterwards.
2387 	 * This function is only called when roce_gid_table is used.
2388 	 */
2389 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2390 	int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2391 			  u16 *pkey);
2392 	int (*alloc_ucontext)(struct ib_ucontext *context,
2393 			      struct ib_udata *udata);
2394 	void (*dealloc_ucontext)(struct ib_ucontext *context);
2395 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2396 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2397 	int (*alloc_pd)(struct ib_pd *pd, struct ib_ucontext *context,
2398 			struct ib_udata *udata);
2399 	void (*dealloc_pd)(struct ib_pd *pd);
2400 	struct ib_ah *(*create_ah)(struct ib_pd *pd,
2401 				   struct rdma_ah_attr *ah_attr, u32 flags,
2402 				   struct ib_udata *udata);
2403 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2404 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2405 	int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2406 	struct ib_srq *(*create_srq)(struct ib_pd *pd,
2407 				     struct ib_srq_init_attr *srq_init_attr,
2408 				     struct ib_udata *udata);
2409 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2410 			  enum ib_srq_attr_mask srq_attr_mask,
2411 			  struct ib_udata *udata);
2412 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2413 	int (*destroy_srq)(struct ib_srq *srq);
2414 	struct ib_qp *(*create_qp)(struct ib_pd *pd,
2415 				   struct ib_qp_init_attr *qp_init_attr,
2416 				   struct ib_udata *udata);
2417 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2418 			 int qp_attr_mask, struct ib_udata *udata);
2419 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2420 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2421 	int (*destroy_qp)(struct ib_qp *qp);
2422 	struct ib_cq *(*create_cq)(struct ib_device *device,
2423 				   const struct ib_cq_init_attr *attr,
2424 				   struct ib_ucontext *context,
2425 				   struct ib_udata *udata);
2426 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2427 	int (*destroy_cq)(struct ib_cq *cq);
2428 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2429 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2430 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2431 				     u64 virt_addr, int mr_access_flags,
2432 				     struct ib_udata *udata);
2433 	int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2434 			     u64 virt_addr, int mr_access_flags,
2435 			     struct ib_pd *pd, struct ib_udata *udata);
2436 	int (*dereg_mr)(struct ib_mr *mr);
2437 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2438 				  u32 max_num_sg);
2439 	int (*advise_mr)(struct ib_pd *pd,
2440 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2441 			 struct ib_sge *sg_list, u32 num_sge,
2442 			 struct uverbs_attr_bundle *attrs);
2443 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2444 			 unsigned int *sg_offset);
2445 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2446 			       struct ib_mr_status *mr_status);
2447 	struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2448 				  struct ib_udata *udata);
2449 	int (*dealloc_mw)(struct ib_mw *mw);
2450 	struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2451 				    struct ib_fmr_attr *fmr_attr);
2452 	int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2453 			    u64 iova);
2454 	int (*unmap_fmr)(struct list_head *fmr_list);
2455 	int (*dealloc_fmr)(struct ib_fmr *fmr);
2456 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2457 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2458 	struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2459 				      struct ib_ucontext *ucontext,
2460 				      struct ib_udata *udata);
2461 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2462 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2463 				       struct ib_flow_attr *flow_attr,
2464 				       int domain, struct ib_udata *udata);
2465 	int (*destroy_flow)(struct ib_flow *flow_id);
2466 	struct ib_flow_action *(*create_flow_action_esp)(
2467 		struct ib_device *device,
2468 		const struct ib_flow_action_attrs_esp *attr,
2469 		struct uverbs_attr_bundle *attrs);
2470 	int (*destroy_flow_action)(struct ib_flow_action *action);
2471 	int (*modify_flow_action_esp)(
2472 		struct ib_flow_action *action,
2473 		const struct ib_flow_action_attrs_esp *attr,
2474 		struct uverbs_attr_bundle *attrs);
2475 	int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2476 				 int state);
2477 	int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2478 			     struct ifla_vf_info *ivf);
2479 	int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2480 			    struct ifla_vf_stats *stats);
2481 	int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2482 			   int type);
2483 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2484 				   struct ib_wq_init_attr *init_attr,
2485 				   struct ib_udata *udata);
2486 	int (*destroy_wq)(struct ib_wq *wq);
2487 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2488 			 u32 wq_attr_mask, struct ib_udata *udata);
2489 	struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2490 		struct ib_device *device,
2491 		struct ib_rwq_ind_table_init_attr *init_attr,
2492 		struct ib_udata *udata);
2493 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2494 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2495 				  struct ib_ucontext *context,
2496 				  struct ib_dm_alloc_attr *attr,
2497 				  struct uverbs_attr_bundle *attrs);
2498 	int (*dealloc_dm)(struct ib_dm *dm);
2499 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2500 				   struct ib_dm_mr_attr *attr,
2501 				   struct uverbs_attr_bundle *attrs);
2502 	struct ib_counters *(*create_counters)(
2503 		struct ib_device *device, struct uverbs_attr_bundle *attrs);
2504 	int (*destroy_counters)(struct ib_counters *counters);
2505 	int (*read_counters)(struct ib_counters *counters,
2506 			     struct ib_counters_read_attr *counters_read_attr,
2507 			     struct uverbs_attr_bundle *attrs);
2508 	/**
2509 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2510 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2511 	 *   core when the device is removed.  A lifespan of -1 in the return
2512 	 *   struct tells the core to set a default lifespan.
2513 	 */
2514 	struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2515 						u8 port_num);
2516 	/**
2517 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2518 	 * @index - The index in the value array we wish to have updated, or
2519 	 *   num_counters if we want all stats updated
2520 	 * Return codes -
2521 	 *   < 0 - Error, no counters updated
2522 	 *   index - Updated the single counter pointed to by index
2523 	 *   num_counters - Updated all counters (will reset the timestamp
2524 	 *     and prevent further calls for lifespan milliseconds)
2525 	 * Drivers are allowed to update all counters in leiu of just the
2526 	 *   one given in index at their option
2527 	 */
2528 	int (*get_hw_stats)(struct ib_device *device,
2529 			    struct rdma_hw_stats *stats, u8 port, int index);
2530 	/*
2531 	 * This function is called once for each port when a ib device is
2532 	 * registered.
2533 	 */
2534 	int (*init_port)(struct ib_device *device, u8 port_num,
2535 			 struct kobject *port_sysfs);
2536 	/**
2537 	 * Allows rdma drivers to add their own restrack attributes.
2538 	 */
2539 	int (*fill_res_entry)(struct sk_buff *msg,
2540 			      struct rdma_restrack_entry *entry);
2541 
2542 	/* Device lifecycle callbacks */
2543 	/*
2544 	 * Called after the device becomes registered, before clients are
2545 	 * attached
2546 	 */
2547 	int (*enable_driver)(struct ib_device *dev);
2548 	/*
2549 	 * This is called as part of ib_dealloc_device().
2550 	 */
2551 	void (*dealloc_driver)(struct ib_device *dev);
2552 
2553 	DECLARE_RDMA_OBJ_SIZE(ib_pd);
2554 	DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2555 };
2556 
2557 struct rdma_restrack_root;
2558 
2559 struct ib_device {
2560 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2561 	struct device                *dma_device;
2562 	struct ib_device_ops	     ops;
2563 	char                          name[IB_DEVICE_NAME_MAX];
2564 	struct rcu_head rcu_head;
2565 
2566 	struct list_head              event_handler_list;
2567 	spinlock_t                    event_handler_lock;
2568 
2569 	struct rw_semaphore	      client_data_rwsem;
2570 	struct xarray                 client_data;
2571 	struct mutex                  unregistration_lock;
2572 
2573 	struct ib_cache               cache;
2574 	/**
2575 	 * port_data is indexed by port number
2576 	 */
2577 	struct ib_port_data *port_data;
2578 
2579 	int			      num_comp_vectors;
2580 
2581 	struct iw_cm_verbs	     *iwcm;
2582 
2583 	struct module               *owner;
2584 	struct device                dev;
2585 	/* First group for device attributes,
2586 	 * Second group for driver provided attributes (optional).
2587 	 * It is NULL terminated array.
2588 	 */
2589 	const struct attribute_group	*groups[3];
2590 
2591 	struct kobject			*ports_kobj;
2592 	struct list_head             port_list;
2593 
2594 	int			     uverbs_abi_ver;
2595 	u64			     uverbs_cmd_mask;
2596 	u64			     uverbs_ex_cmd_mask;
2597 
2598 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2599 	__be64			     node_guid;
2600 	u32			     local_dma_lkey;
2601 	u16                          is_switch:1;
2602 	/* Indicates kernel verbs support, should not be used in drivers */
2603 	u16                          kverbs_provider:1;
2604 	u8                           node_type;
2605 	u8                           phys_port_cnt;
2606 	struct ib_device_attr        attrs;
2607 	struct attribute_group	     *hw_stats_ag;
2608 	struct rdma_hw_stats         *hw_stats;
2609 
2610 #ifdef CONFIG_CGROUP_RDMA
2611 	struct rdmacg_device         cg_device;
2612 #endif
2613 
2614 	u32                          index;
2615 	struct rdma_restrack_root *res;
2616 
2617 	const struct uapi_definition   *driver_def;
2618 	enum rdma_driver_id		driver_id;
2619 
2620 	/*
2621 	 * Positive refcount indicates that the device is currently
2622 	 * registered and cannot be unregistered.
2623 	 */
2624 	refcount_t refcount;
2625 	struct completion unreg_completion;
2626 	struct work_struct unregistration_work;
2627 
2628 	const struct rdma_link_ops *link_ops;
2629 };
2630 
2631 struct ib_client {
2632 	const char *name;
2633 	void (*add)   (struct ib_device *);
2634 	void (*remove)(struct ib_device *, void *client_data);
2635 
2636 	/* Returns the net_dev belonging to this ib_client and matching the
2637 	 * given parameters.
2638 	 * @dev:	 An RDMA device that the net_dev use for communication.
2639 	 * @port:	 A physical port number on the RDMA device.
2640 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2641 	 * @gid:	 A GID that the net_dev uses to communicate.
2642 	 * @addr:	 An IP address the net_dev is configured with.
2643 	 * @client_data: The device's client data set by ib_set_client_data().
2644 	 *
2645 	 * An ib_client that implements a net_dev on top of RDMA devices
2646 	 * (such as IP over IB) should implement this callback, allowing the
2647 	 * rdma_cm module to find the right net_dev for a given request.
2648 	 *
2649 	 * The caller is responsible for calling dev_put on the returned
2650 	 * netdev. */
2651 	struct net_device *(*get_net_dev_by_params)(
2652 			struct ib_device *dev,
2653 			u8 port,
2654 			u16 pkey,
2655 			const union ib_gid *gid,
2656 			const struct sockaddr *addr,
2657 			void *client_data);
2658 	struct list_head list;
2659 	u32 client_id;
2660 
2661 	/* kverbs are not required by the client */
2662 	u8 no_kverbs_req:1;
2663 };
2664 
2665 struct ib_device *_ib_alloc_device(size_t size);
2666 #define ib_alloc_device(drv_struct, member)                                    \
2667 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2668 				      BUILD_BUG_ON_ZERO(offsetof(              \
2669 					      struct drv_struct, member))),    \
2670 		     struct drv_struct, member)
2671 
2672 void ib_dealloc_device(struct ib_device *device);
2673 
2674 void ib_get_device_fw_str(struct ib_device *device, char *str);
2675 
2676 int ib_register_device(struct ib_device *device, const char *name);
2677 void ib_unregister_device(struct ib_device *device);
2678 void ib_unregister_driver(enum rdma_driver_id driver_id);
2679 void ib_unregister_device_and_put(struct ib_device *device);
2680 void ib_unregister_device_queued(struct ib_device *ib_dev);
2681 
2682 int ib_register_client   (struct ib_client *client);
2683 void ib_unregister_client(struct ib_client *client);
2684 
2685 /**
2686  * ib_get_client_data - Get IB client context
2687  * @device:Device to get context for
2688  * @client:Client to get context for
2689  *
2690  * ib_get_client_data() returns the client context data set with
2691  * ib_set_client_data(). This can only be called while the client is
2692  * registered to the device, once the ib_client remove() callback returns this
2693  * cannot be called.
2694  */
2695 static inline void *ib_get_client_data(struct ib_device *device,
2696 				       struct ib_client *client)
2697 {
2698 	return xa_load(&device->client_data, client->client_id);
2699 }
2700 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2701 			 void *data);
2702 void ib_set_device_ops(struct ib_device *device,
2703 		       const struct ib_device_ops *ops);
2704 
2705 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2706 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2707 		      unsigned long pfn, unsigned long size, pgprot_t prot);
2708 int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2709 			struct vm_area_struct *vma, struct page *page,
2710 			unsigned long size);
2711 #else
2712 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
2713 				    struct vm_area_struct *vma,
2714 				    unsigned long pfn, unsigned long size,
2715 				    pgprot_t prot)
2716 {
2717 	return -EINVAL;
2718 }
2719 static inline int rdma_user_mmap_page(struct ib_ucontext *ucontext,
2720 				struct vm_area_struct *vma, struct page *page,
2721 				unsigned long size)
2722 {
2723 	return -EINVAL;
2724 }
2725 #endif
2726 
2727 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2728 {
2729 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2730 }
2731 
2732 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2733 {
2734 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2735 }
2736 
2737 static inline bool ib_is_buffer_cleared(const void __user *p,
2738 					size_t len)
2739 {
2740 	bool ret;
2741 	u8 *buf;
2742 
2743 	if (len > USHRT_MAX)
2744 		return false;
2745 
2746 	buf = memdup_user(p, len);
2747 	if (IS_ERR(buf))
2748 		return false;
2749 
2750 	ret = !memchr_inv(buf, 0, len);
2751 	kfree(buf);
2752 	return ret;
2753 }
2754 
2755 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2756 				       size_t offset,
2757 				       size_t len)
2758 {
2759 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
2760 }
2761 
2762 /**
2763  * ib_is_destroy_retryable - Check whether the uobject destruction
2764  * is retryable.
2765  * @ret: The initial destruction return code
2766  * @why: remove reason
2767  * @uobj: The uobject that is destroyed
2768  *
2769  * This function is a helper function that IB layer and low-level drivers
2770  * can use to consider whether the destruction of the given uobject is
2771  * retry-able.
2772  * It checks the original return code, if it wasn't success the destruction
2773  * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2774  * the remove reason. (i.e. why).
2775  * Must be called with the object locked for destroy.
2776  */
2777 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2778 					   struct ib_uobject *uobj)
2779 {
2780 	return ret && (why == RDMA_REMOVE_DESTROY ||
2781 		       uobj->context->cleanup_retryable);
2782 }
2783 
2784 /**
2785  * ib_destroy_usecnt - Called during destruction to check the usecnt
2786  * @usecnt: The usecnt atomic
2787  * @why: remove reason
2788  * @uobj: The uobject that is destroyed
2789  *
2790  * Non-zero usecnts will block destruction unless destruction was triggered by
2791  * a ucontext cleanup.
2792  */
2793 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2794 				    enum rdma_remove_reason why,
2795 				    struct ib_uobject *uobj)
2796 {
2797 	if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2798 		return -EBUSY;
2799 	return 0;
2800 }
2801 
2802 /**
2803  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2804  * contains all required attributes and no attributes not allowed for
2805  * the given QP state transition.
2806  * @cur_state: Current QP state
2807  * @next_state: Next QP state
2808  * @type: QP type
2809  * @mask: Mask of supplied QP attributes
2810  *
2811  * This function is a helper function that a low-level driver's
2812  * modify_qp method can use to validate the consumer's input.  It
2813  * checks that cur_state and next_state are valid QP states, that a
2814  * transition from cur_state to next_state is allowed by the IB spec,
2815  * and that the attribute mask supplied is allowed for the transition.
2816  */
2817 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2818 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
2819 
2820 void ib_register_event_handler(struct ib_event_handler *event_handler);
2821 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2822 void ib_dispatch_event(struct ib_event *event);
2823 
2824 int ib_query_port(struct ib_device *device,
2825 		  u8 port_num, struct ib_port_attr *port_attr);
2826 
2827 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2828 					       u8 port_num);
2829 
2830 /**
2831  * rdma_cap_ib_switch - Check if the device is IB switch
2832  * @device: Device to check
2833  *
2834  * Device driver is responsible for setting is_switch bit on
2835  * in ib_device structure at init time.
2836  *
2837  * Return: true if the device is IB switch.
2838  */
2839 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2840 {
2841 	return device->is_switch;
2842 }
2843 
2844 /**
2845  * rdma_start_port - Return the first valid port number for the device
2846  * specified
2847  *
2848  * @device: Device to be checked
2849  *
2850  * Return start port number
2851  */
2852 static inline u8 rdma_start_port(const struct ib_device *device)
2853 {
2854 	return rdma_cap_ib_switch(device) ? 0 : 1;
2855 }
2856 
2857 /**
2858  * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2859  * @device - The struct ib_device * to iterate over
2860  * @iter - The unsigned int to store the port number
2861  */
2862 #define rdma_for_each_port(device, iter)                                       \
2863 	for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type(   \
2864 						     unsigned int, iter)));    \
2865 	     iter <= rdma_end_port(device); (iter)++)
2866 
2867 /**
2868  * rdma_end_port - Return the last valid port number for the device
2869  * specified
2870  *
2871  * @device: Device to be checked
2872  *
2873  * Return last port number
2874  */
2875 static inline u8 rdma_end_port(const struct ib_device *device)
2876 {
2877 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2878 }
2879 
2880 static inline int rdma_is_port_valid(const struct ib_device *device,
2881 				     unsigned int port)
2882 {
2883 	return (port >= rdma_start_port(device) &&
2884 		port <= rdma_end_port(device));
2885 }
2886 
2887 static inline bool rdma_is_grh_required(const struct ib_device *device,
2888 					u8 port_num)
2889 {
2890 	return device->port_data[port_num].immutable.core_cap_flags &
2891 	       RDMA_CORE_PORT_IB_GRH_REQUIRED;
2892 }
2893 
2894 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2895 {
2896 	return device->port_data[port_num].immutable.core_cap_flags &
2897 	       RDMA_CORE_CAP_PROT_IB;
2898 }
2899 
2900 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2901 {
2902 	return device->port_data[port_num].immutable.core_cap_flags &
2903 	       (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2904 }
2905 
2906 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2907 {
2908 	return device->port_data[port_num].immutable.core_cap_flags &
2909 	       RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2910 }
2911 
2912 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2913 {
2914 	return device->port_data[port_num].immutable.core_cap_flags &
2915 	       RDMA_CORE_CAP_PROT_ROCE;
2916 }
2917 
2918 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2919 {
2920 	return device->port_data[port_num].immutable.core_cap_flags &
2921 	       RDMA_CORE_CAP_PROT_IWARP;
2922 }
2923 
2924 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2925 {
2926 	return rdma_protocol_ib(device, port_num) ||
2927 		rdma_protocol_roce(device, port_num);
2928 }
2929 
2930 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2931 {
2932 	return device->port_data[port_num].immutable.core_cap_flags &
2933 	       RDMA_CORE_CAP_PROT_RAW_PACKET;
2934 }
2935 
2936 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2937 {
2938 	return device->port_data[port_num].immutable.core_cap_flags &
2939 	       RDMA_CORE_CAP_PROT_USNIC;
2940 }
2941 
2942 /**
2943  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2944  * Management Datagrams.
2945  * @device: Device to check
2946  * @port_num: Port number to check
2947  *
2948  * Management Datagrams (MAD) are a required part of the InfiniBand
2949  * specification and are supported on all InfiniBand devices.  A slightly
2950  * extended version are also supported on OPA interfaces.
2951  *
2952  * Return: true if the port supports sending/receiving of MAD packets.
2953  */
2954 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2955 {
2956 	return device->port_data[port_num].immutable.core_cap_flags &
2957 	       RDMA_CORE_CAP_IB_MAD;
2958 }
2959 
2960 /**
2961  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2962  * Management Datagrams.
2963  * @device: Device to check
2964  * @port_num: Port number to check
2965  *
2966  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2967  * datagrams with their own versions.  These OPA MADs share many but not all of
2968  * the characteristics of InfiniBand MADs.
2969  *
2970  * OPA MADs differ in the following ways:
2971  *
2972  *    1) MADs are variable size up to 2K
2973  *       IBTA defined MADs remain fixed at 256 bytes
2974  *    2) OPA SMPs must carry valid PKeys
2975  *    3) OPA SMP packets are a different format
2976  *
2977  * Return: true if the port supports OPA MAD packet formats.
2978  */
2979 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2980 {
2981 	return (device->port_data[port_num].immutable.core_cap_flags &
2982 		RDMA_CORE_CAP_OPA_MAD) == RDMA_CORE_CAP_OPA_MAD;
2983 }
2984 
2985 /**
2986  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2987  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2988  * @device: Device to check
2989  * @port_num: Port number to check
2990  *
2991  * Each InfiniBand node is required to provide a Subnet Management Agent
2992  * that the subnet manager can access.  Prior to the fabric being fully
2993  * configured by the subnet manager, the SMA is accessed via a well known
2994  * interface called the Subnet Management Interface (SMI).  This interface
2995  * uses directed route packets to communicate with the SM to get around the
2996  * chicken and egg problem of the SM needing to know what's on the fabric
2997  * in order to configure the fabric, and needing to configure the fabric in
2998  * order to send packets to the devices on the fabric.  These directed
2999  * route packets do not need the fabric fully configured in order to reach
3000  * their destination.  The SMI is the only method allowed to send
3001  * directed route packets on an InfiniBand fabric.
3002  *
3003  * Return: true if the port provides an SMI.
3004  */
3005 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3006 {
3007 	return device->port_data[port_num].immutable.core_cap_flags &
3008 	       RDMA_CORE_CAP_IB_SMI;
3009 }
3010 
3011 /**
3012  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3013  * Communication Manager.
3014  * @device: Device to check
3015  * @port_num: Port number to check
3016  *
3017  * The InfiniBand Communication Manager is one of many pre-defined General
3018  * Service Agents (GSA) that are accessed via the General Service
3019  * Interface (GSI).  It's role is to facilitate establishment of connections
3020  * between nodes as well as other management related tasks for established
3021  * connections.
3022  *
3023  * Return: true if the port supports an IB CM (this does not guarantee that
3024  * a CM is actually running however).
3025  */
3026 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3027 {
3028 	return device->port_data[port_num].immutable.core_cap_flags &
3029 	       RDMA_CORE_CAP_IB_CM;
3030 }
3031 
3032 /**
3033  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3034  * Communication Manager.
3035  * @device: Device to check
3036  * @port_num: Port number to check
3037  *
3038  * Similar to above, but specific to iWARP connections which have a different
3039  * managment protocol than InfiniBand.
3040  *
3041  * Return: true if the port supports an iWARP CM (this does not guarantee that
3042  * a CM is actually running however).
3043  */
3044 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3045 {
3046 	return device->port_data[port_num].immutable.core_cap_flags &
3047 	       RDMA_CORE_CAP_IW_CM;
3048 }
3049 
3050 /**
3051  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3052  * Subnet Administration.
3053  * @device: Device to check
3054  * @port_num: Port number to check
3055  *
3056  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3057  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
3058  * fabrics, devices should resolve routes to other hosts by contacting the
3059  * SA to query the proper route.
3060  *
3061  * Return: true if the port should act as a client to the fabric Subnet
3062  * Administration interface.  This does not imply that the SA service is
3063  * running locally.
3064  */
3065 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3066 {
3067 	return device->port_data[port_num].immutable.core_cap_flags &
3068 	       RDMA_CORE_CAP_IB_SA;
3069 }
3070 
3071 /**
3072  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3073  * Multicast.
3074  * @device: Device to check
3075  * @port_num: Port number to check
3076  *
3077  * InfiniBand multicast registration is more complex than normal IPv4 or
3078  * IPv6 multicast registration.  Each Host Channel Adapter must register
3079  * with the Subnet Manager when it wishes to join a multicast group.  It
3080  * should do so only once regardless of how many queue pairs it subscribes
3081  * to this group.  And it should leave the group only after all queue pairs
3082  * attached to the group have been detached.
3083  *
3084  * Return: true if the port must undertake the additional adminstrative
3085  * overhead of registering/unregistering with the SM and tracking of the
3086  * total number of queue pairs attached to the multicast group.
3087  */
3088 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3089 {
3090 	return rdma_cap_ib_sa(device, port_num);
3091 }
3092 
3093 /**
3094  * rdma_cap_af_ib - Check if the port of device has the capability
3095  * Native Infiniband Address.
3096  * @device: Device to check
3097  * @port_num: Port number to check
3098  *
3099  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3100  * GID.  RoCE uses a different mechanism, but still generates a GID via
3101  * a prescribed mechanism and port specific data.
3102  *
3103  * Return: true if the port uses a GID address to identify devices on the
3104  * network.
3105  */
3106 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3107 {
3108 	return device->port_data[port_num].immutable.core_cap_flags &
3109 	       RDMA_CORE_CAP_AF_IB;
3110 }
3111 
3112 /**
3113  * rdma_cap_eth_ah - Check if the port of device has the capability
3114  * Ethernet Address Handle.
3115  * @device: Device to check
3116  * @port_num: Port number to check
3117  *
3118  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3119  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3120  * port.  Normally, packet headers are generated by the sending host
3121  * adapter, but when sending connectionless datagrams, we must manually
3122  * inject the proper headers for the fabric we are communicating over.
3123  *
3124  * Return: true if we are running as a RoCE port and must force the
3125  * addition of a Global Route Header built from our Ethernet Address
3126  * Handle into our header list for connectionless packets.
3127  */
3128 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3129 {
3130 	return device->port_data[port_num].immutable.core_cap_flags &
3131 	       RDMA_CORE_CAP_ETH_AH;
3132 }
3133 
3134 /**
3135  * rdma_cap_opa_ah - Check if the port of device supports
3136  * OPA Address handles
3137  * @device: Device to check
3138  * @port_num: Port number to check
3139  *
3140  * Return: true if we are running on an OPA device which supports
3141  * the extended OPA addressing.
3142  */
3143 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3144 {
3145 	return (device->port_data[port_num].immutable.core_cap_flags &
3146 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3147 }
3148 
3149 /**
3150  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3151  *
3152  * @device: Device
3153  * @port_num: Port number
3154  *
3155  * This MAD size includes the MAD headers and MAD payload.  No other headers
3156  * are included.
3157  *
3158  * Return the max MAD size required by the Port.  Will return 0 if the port
3159  * does not support MADs
3160  */
3161 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3162 {
3163 	return device->port_data[port_num].immutable.max_mad_size;
3164 }
3165 
3166 /**
3167  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3168  * @device: Device to check
3169  * @port_num: Port number to check
3170  *
3171  * RoCE GID table mechanism manages the various GIDs for a device.
3172  *
3173  * NOTE: if allocating the port's GID table has failed, this call will still
3174  * return true, but any RoCE GID table API will fail.
3175  *
3176  * Return: true if the port uses RoCE GID table mechanism in order to manage
3177  * its GIDs.
3178  */
3179 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3180 					   u8 port_num)
3181 {
3182 	return rdma_protocol_roce(device, port_num) &&
3183 		device->ops.add_gid && device->ops.del_gid;
3184 }
3185 
3186 /*
3187  * Check if the device supports READ W/ INVALIDATE.
3188  */
3189 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3190 {
3191 	/*
3192 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3193 	 * has support for it yet.
3194 	 */
3195 	return rdma_protocol_iwarp(dev, port_num);
3196 }
3197 
3198 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3199 			 int state);
3200 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3201 		     struct ifla_vf_info *info);
3202 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3203 		    struct ifla_vf_stats *stats);
3204 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3205 		   int type);
3206 
3207 int ib_query_pkey(struct ib_device *device,
3208 		  u8 port_num, u16 index, u16 *pkey);
3209 
3210 int ib_modify_device(struct ib_device *device,
3211 		     int device_modify_mask,
3212 		     struct ib_device_modify *device_modify);
3213 
3214 int ib_modify_port(struct ib_device *device,
3215 		   u8 port_num, int port_modify_mask,
3216 		   struct ib_port_modify *port_modify);
3217 
3218 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3219 		u8 *port_num, u16 *index);
3220 
3221 int ib_find_pkey(struct ib_device *device,
3222 		 u8 port_num, u16 pkey, u16 *index);
3223 
3224 enum ib_pd_flags {
3225 	/*
3226 	 * Create a memory registration for all memory in the system and place
3227 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3228 	 * ULPs to avoid the overhead of dynamic MRs.
3229 	 *
3230 	 * This flag is generally considered unsafe and must only be used in
3231 	 * extremly trusted environments.  Every use of it will log a warning
3232 	 * in the kernel log.
3233 	 */
3234 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3235 };
3236 
3237 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3238 		const char *caller);
3239 #define ib_alloc_pd(device, flags) \
3240 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3241 void ib_dealloc_pd(struct ib_pd *pd);
3242 
3243 enum rdma_create_ah_flags {
3244 	/* In a sleepable context */
3245 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3246 };
3247 
3248 /**
3249  * rdma_create_ah - Creates an address handle for the given address vector.
3250  * @pd: The protection domain associated with the address handle.
3251  * @ah_attr: The attributes of the address vector.
3252  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3253  *
3254  * The address handle is used to reference a local or global destination
3255  * in all UD QP post sends.
3256  */
3257 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3258 			     u32 flags);
3259 
3260 /**
3261  * rdma_create_user_ah - Creates an address handle for the given address vector.
3262  * It resolves destination mac address for ah attribute of RoCE type.
3263  * @pd: The protection domain associated with the address handle.
3264  * @ah_attr: The attributes of the address vector.
3265  * @udata: pointer to user's input output buffer information need by
3266  *         provider driver.
3267  *
3268  * It returns 0 on success and returns appropriate error code on error.
3269  * The address handle is used to reference a local or global destination
3270  * in all UD QP post sends.
3271  */
3272 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3273 				  struct rdma_ah_attr *ah_attr,
3274 				  struct ib_udata *udata);
3275 /**
3276  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3277  *   work completion.
3278  * @hdr: the L3 header to parse
3279  * @net_type: type of header to parse
3280  * @sgid: place to store source gid
3281  * @dgid: place to store destination gid
3282  */
3283 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3284 			      enum rdma_network_type net_type,
3285 			      union ib_gid *sgid, union ib_gid *dgid);
3286 
3287 /**
3288  * ib_get_rdma_header_version - Get the header version
3289  * @hdr: the L3 header to parse
3290  */
3291 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3292 
3293 /**
3294  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3295  *   work completion.
3296  * @device: Device on which the received message arrived.
3297  * @port_num: Port on which the received message arrived.
3298  * @wc: Work completion associated with the received message.
3299  * @grh: References the received global route header.  This parameter is
3300  *   ignored unless the work completion indicates that the GRH is valid.
3301  * @ah_attr: Returned attributes that can be used when creating an address
3302  *   handle for replying to the message.
3303  * When ib_init_ah_attr_from_wc() returns success,
3304  * (a) for IB link layer it optionally contains a reference to SGID attribute
3305  * when GRH is present for IB link layer.
3306  * (b) for RoCE link layer it contains a reference to SGID attribute.
3307  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3308  * attributes which are initialized using ib_init_ah_attr_from_wc().
3309  *
3310  */
3311 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3312 			    const struct ib_wc *wc, const struct ib_grh *grh,
3313 			    struct rdma_ah_attr *ah_attr);
3314 
3315 /**
3316  * ib_create_ah_from_wc - Creates an address handle associated with the
3317  *   sender of the specified work completion.
3318  * @pd: The protection domain associated with the address handle.
3319  * @wc: Work completion information associated with a received message.
3320  * @grh: References the received global route header.  This parameter is
3321  *   ignored unless the work completion indicates that the GRH is valid.
3322  * @port_num: The outbound port number to associate with the address.
3323  *
3324  * The address handle is used to reference a local or global destination
3325  * in all UD QP post sends.
3326  */
3327 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3328 				   const struct ib_grh *grh, u8 port_num);
3329 
3330 /**
3331  * rdma_modify_ah - Modifies the address vector associated with an address
3332  *   handle.
3333  * @ah: The address handle to modify.
3334  * @ah_attr: The new address vector attributes to associate with the
3335  *   address handle.
3336  */
3337 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3338 
3339 /**
3340  * rdma_query_ah - Queries the address vector associated with an address
3341  *   handle.
3342  * @ah: The address handle to query.
3343  * @ah_attr: The address vector attributes associated with the address
3344  *   handle.
3345  */
3346 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3347 
3348 enum rdma_destroy_ah_flags {
3349 	/* In a sleepable context */
3350 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3351 };
3352 
3353 /**
3354  * rdma_destroy_ah - Destroys an address handle.
3355  * @ah: The address handle to destroy.
3356  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3357  */
3358 int rdma_destroy_ah(struct ib_ah *ah, u32 flags);
3359 
3360 /**
3361  * ib_create_srq - Creates a SRQ associated with the specified protection
3362  *   domain.
3363  * @pd: The protection domain associated with the SRQ.
3364  * @srq_init_attr: A list of initial attributes required to create the
3365  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
3366  *   the actual capabilities of the created SRQ.
3367  *
3368  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3369  * requested size of the SRQ, and set to the actual values allocated
3370  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
3371  * will always be at least as large as the requested values.
3372  */
3373 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3374 			     struct ib_srq_init_attr *srq_init_attr);
3375 
3376 /**
3377  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3378  * @srq: The SRQ to modify.
3379  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3380  *   the current values of selected SRQ attributes are returned.
3381  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3382  *   are being modified.
3383  *
3384  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3385  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3386  * the number of receives queued drops below the limit.
3387  */
3388 int ib_modify_srq(struct ib_srq *srq,
3389 		  struct ib_srq_attr *srq_attr,
3390 		  enum ib_srq_attr_mask srq_attr_mask);
3391 
3392 /**
3393  * ib_query_srq - Returns the attribute list and current values for the
3394  *   specified SRQ.
3395  * @srq: The SRQ to query.
3396  * @srq_attr: The attributes of the specified SRQ.
3397  */
3398 int ib_query_srq(struct ib_srq *srq,
3399 		 struct ib_srq_attr *srq_attr);
3400 
3401 /**
3402  * ib_destroy_srq - Destroys the specified SRQ.
3403  * @srq: The SRQ to destroy.
3404  */
3405 int ib_destroy_srq(struct ib_srq *srq);
3406 
3407 /**
3408  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3409  * @srq: The SRQ to post the work request on.
3410  * @recv_wr: A list of work requests to post on the receive queue.
3411  * @bad_recv_wr: On an immediate failure, this parameter will reference
3412  *   the work request that failed to be posted on the QP.
3413  */
3414 static inline int ib_post_srq_recv(struct ib_srq *srq,
3415 				   const struct ib_recv_wr *recv_wr,
3416 				   const struct ib_recv_wr **bad_recv_wr)
3417 {
3418 	const struct ib_recv_wr *dummy;
3419 
3420 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3421 					      bad_recv_wr ? : &dummy);
3422 }
3423 
3424 /**
3425  * ib_create_qp - Creates a QP associated with the specified protection
3426  *   domain.
3427  * @pd: The protection domain associated with the QP.
3428  * @qp_init_attr: A list of initial attributes required to create the
3429  *   QP.  If QP creation succeeds, then the attributes are updated to
3430  *   the actual capabilities of the created QP.
3431  */
3432 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3433 			   struct ib_qp_init_attr *qp_init_attr);
3434 
3435 /**
3436  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3437  * @qp: The QP to modify.
3438  * @attr: On input, specifies the QP attributes to modify.  On output,
3439  *   the current values of selected QP attributes are returned.
3440  * @attr_mask: A bit-mask used to specify which attributes of the QP
3441  *   are being modified.
3442  * @udata: pointer to user's input output buffer information
3443  *   are being modified.
3444  * It returns 0 on success and returns appropriate error code on error.
3445  */
3446 int ib_modify_qp_with_udata(struct ib_qp *qp,
3447 			    struct ib_qp_attr *attr,
3448 			    int attr_mask,
3449 			    struct ib_udata *udata);
3450 
3451 /**
3452  * ib_modify_qp - Modifies the attributes for the specified QP and then
3453  *   transitions the QP to the given state.
3454  * @qp: The QP to modify.
3455  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3456  *   the current values of selected QP attributes are returned.
3457  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3458  *   are being modified.
3459  */
3460 int ib_modify_qp(struct ib_qp *qp,
3461 		 struct ib_qp_attr *qp_attr,
3462 		 int qp_attr_mask);
3463 
3464 /**
3465  * ib_query_qp - Returns the attribute list and current values for the
3466  *   specified QP.
3467  * @qp: The QP to query.
3468  * @qp_attr: The attributes of the specified QP.
3469  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3470  * @qp_init_attr: Additional attributes of the selected QP.
3471  *
3472  * The qp_attr_mask may be used to limit the query to gathering only the
3473  * selected attributes.
3474  */
3475 int ib_query_qp(struct ib_qp *qp,
3476 		struct ib_qp_attr *qp_attr,
3477 		int qp_attr_mask,
3478 		struct ib_qp_init_attr *qp_init_attr);
3479 
3480 /**
3481  * ib_destroy_qp - Destroys the specified QP.
3482  * @qp: The QP to destroy.
3483  */
3484 int ib_destroy_qp(struct ib_qp *qp);
3485 
3486 /**
3487  * ib_open_qp - Obtain a reference to an existing sharable QP.
3488  * @xrcd - XRC domain
3489  * @qp_open_attr: Attributes identifying the QP to open.
3490  *
3491  * Returns a reference to a sharable QP.
3492  */
3493 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3494 			 struct ib_qp_open_attr *qp_open_attr);
3495 
3496 /**
3497  * ib_close_qp - Release an external reference to a QP.
3498  * @qp: The QP handle to release
3499  *
3500  * The opened QP handle is released by the caller.  The underlying
3501  * shared QP is not destroyed until all internal references are released.
3502  */
3503 int ib_close_qp(struct ib_qp *qp);
3504 
3505 /**
3506  * ib_post_send - Posts a list of work requests to the send queue of
3507  *   the specified QP.
3508  * @qp: The QP to post the work request on.
3509  * @send_wr: A list of work requests to post on the send queue.
3510  * @bad_send_wr: On an immediate failure, this parameter will reference
3511  *   the work request that failed to be posted on the QP.
3512  *
3513  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3514  * error is returned, the QP state shall not be affected,
3515  * ib_post_send() will return an immediate error after queueing any
3516  * earlier work requests in the list.
3517  */
3518 static inline int ib_post_send(struct ib_qp *qp,
3519 			       const struct ib_send_wr *send_wr,
3520 			       const struct ib_send_wr **bad_send_wr)
3521 {
3522 	const struct ib_send_wr *dummy;
3523 
3524 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3525 }
3526 
3527 /**
3528  * ib_post_recv - Posts a list of work requests to the receive queue of
3529  *   the specified QP.
3530  * @qp: The QP to post the work request on.
3531  * @recv_wr: A list of work requests to post on the receive queue.
3532  * @bad_recv_wr: On an immediate failure, this parameter will reference
3533  *   the work request that failed to be posted on the QP.
3534  */
3535 static inline int ib_post_recv(struct ib_qp *qp,
3536 			       const struct ib_recv_wr *recv_wr,
3537 			       const struct ib_recv_wr **bad_recv_wr)
3538 {
3539 	const struct ib_recv_wr *dummy;
3540 
3541 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3542 }
3543 
3544 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3545 			    int nr_cqe, int comp_vector,
3546 			    enum ib_poll_context poll_ctx, const char *caller);
3547 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3548 	__ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3549 
3550 void ib_free_cq(struct ib_cq *cq);
3551 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3552 
3553 /**
3554  * ib_create_cq - Creates a CQ on the specified device.
3555  * @device: The device on which to create the CQ.
3556  * @comp_handler: A user-specified callback that is invoked when a
3557  *   completion event occurs on the CQ.
3558  * @event_handler: A user-specified callback that is invoked when an
3559  *   asynchronous event not associated with a completion occurs on the CQ.
3560  * @cq_context: Context associated with the CQ returned to the user via
3561  *   the associated completion and event handlers.
3562  * @cq_attr: The attributes the CQ should be created upon.
3563  *
3564  * Users can examine the cq structure to determine the actual CQ size.
3565  */
3566 struct ib_cq *__ib_create_cq(struct ib_device *device,
3567 			     ib_comp_handler comp_handler,
3568 			     void (*event_handler)(struct ib_event *, void *),
3569 			     void *cq_context,
3570 			     const struct ib_cq_init_attr *cq_attr,
3571 			     const char *caller);
3572 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3573 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3574 
3575 /**
3576  * ib_resize_cq - Modifies the capacity of the CQ.
3577  * @cq: The CQ to resize.
3578  * @cqe: The minimum size of the CQ.
3579  *
3580  * Users can examine the cq structure to determine the actual CQ size.
3581  */
3582 int ib_resize_cq(struct ib_cq *cq, int cqe);
3583 
3584 /**
3585  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3586  * @cq: The CQ to modify.
3587  * @cq_count: number of CQEs that will trigger an event
3588  * @cq_period: max period of time in usec before triggering an event
3589  *
3590  */
3591 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3592 
3593 /**
3594  * ib_destroy_cq - Destroys the specified CQ.
3595  * @cq: The CQ to destroy.
3596  */
3597 int ib_destroy_cq(struct ib_cq *cq);
3598 
3599 /**
3600  * ib_poll_cq - poll a CQ for completion(s)
3601  * @cq:the CQ being polled
3602  * @num_entries:maximum number of completions to return
3603  * @wc:array of at least @num_entries &struct ib_wc where completions
3604  *   will be returned
3605  *
3606  * Poll a CQ for (possibly multiple) completions.  If the return value
3607  * is < 0, an error occurred.  If the return value is >= 0, it is the
3608  * number of completions returned.  If the return value is
3609  * non-negative and < num_entries, then the CQ was emptied.
3610  */
3611 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3612 			     struct ib_wc *wc)
3613 {
3614 	return cq->device->ops.poll_cq(cq, num_entries, wc);
3615 }
3616 
3617 /**
3618  * ib_req_notify_cq - Request completion notification on a CQ.
3619  * @cq: The CQ to generate an event for.
3620  * @flags:
3621  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3622  *   to request an event on the next solicited event or next work
3623  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3624  *   may also be |ed in to request a hint about missed events, as
3625  *   described below.
3626  *
3627  * Return Value:
3628  *    < 0 means an error occurred while requesting notification
3629  *   == 0 means notification was requested successfully, and if
3630  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3631  *        were missed and it is safe to wait for another event.  In
3632  *        this case is it guaranteed that any work completions added
3633  *        to the CQ since the last CQ poll will trigger a completion
3634  *        notification event.
3635  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3636  *        in.  It means that the consumer must poll the CQ again to
3637  *        make sure it is empty to avoid missing an event because of a
3638  *        race between requesting notification and an entry being
3639  *        added to the CQ.  This return value means it is possible
3640  *        (but not guaranteed) that a work completion has been added
3641  *        to the CQ since the last poll without triggering a
3642  *        completion notification event.
3643  */
3644 static inline int ib_req_notify_cq(struct ib_cq *cq,
3645 				   enum ib_cq_notify_flags flags)
3646 {
3647 	return cq->device->ops.req_notify_cq(cq, flags);
3648 }
3649 
3650 /**
3651  * ib_req_ncomp_notif - Request completion notification when there are
3652  *   at least the specified number of unreaped completions on the CQ.
3653  * @cq: The CQ to generate an event for.
3654  * @wc_cnt: The number of unreaped completions that should be on the
3655  *   CQ before an event is generated.
3656  */
3657 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3658 {
3659 	return cq->device->ops.req_ncomp_notif ?
3660 		cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3661 		-ENOSYS;
3662 }
3663 
3664 /**
3665  * ib_dma_mapping_error - check a DMA addr for error
3666  * @dev: The device for which the dma_addr was created
3667  * @dma_addr: The DMA address to check
3668  */
3669 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3670 {
3671 	return dma_mapping_error(dev->dma_device, dma_addr);
3672 }
3673 
3674 /**
3675  * ib_dma_map_single - Map a kernel virtual address to DMA address
3676  * @dev: The device for which the dma_addr is to be created
3677  * @cpu_addr: The kernel virtual address
3678  * @size: The size of the region in bytes
3679  * @direction: The direction of the DMA
3680  */
3681 static inline u64 ib_dma_map_single(struct ib_device *dev,
3682 				    void *cpu_addr, size_t size,
3683 				    enum dma_data_direction direction)
3684 {
3685 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3686 }
3687 
3688 /**
3689  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3690  * @dev: The device for which the DMA address was created
3691  * @addr: The DMA address
3692  * @size: The size of the region in bytes
3693  * @direction: The direction of the DMA
3694  */
3695 static inline void ib_dma_unmap_single(struct ib_device *dev,
3696 				       u64 addr, size_t size,
3697 				       enum dma_data_direction direction)
3698 {
3699 	dma_unmap_single(dev->dma_device, addr, size, direction);
3700 }
3701 
3702 /**
3703  * ib_dma_map_page - Map a physical page to DMA address
3704  * @dev: The device for which the dma_addr is to be created
3705  * @page: The page to be mapped
3706  * @offset: The offset within the page
3707  * @size: The size of the region in bytes
3708  * @direction: The direction of the DMA
3709  */
3710 static inline u64 ib_dma_map_page(struct ib_device *dev,
3711 				  struct page *page,
3712 				  unsigned long offset,
3713 				  size_t size,
3714 					 enum dma_data_direction direction)
3715 {
3716 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3717 }
3718 
3719 /**
3720  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3721  * @dev: The device for which the DMA address was created
3722  * @addr: The DMA address
3723  * @size: The size of the region in bytes
3724  * @direction: The direction of the DMA
3725  */
3726 static inline void ib_dma_unmap_page(struct ib_device *dev,
3727 				     u64 addr, size_t size,
3728 				     enum dma_data_direction direction)
3729 {
3730 	dma_unmap_page(dev->dma_device, addr, size, direction);
3731 }
3732 
3733 /**
3734  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3735  * @dev: The device for which the DMA addresses are to be created
3736  * @sg: The array of scatter/gather entries
3737  * @nents: The number of scatter/gather entries
3738  * @direction: The direction of the DMA
3739  */
3740 static inline int ib_dma_map_sg(struct ib_device *dev,
3741 				struct scatterlist *sg, int nents,
3742 				enum dma_data_direction direction)
3743 {
3744 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3745 }
3746 
3747 /**
3748  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3749  * @dev: The device for which the DMA addresses were created
3750  * @sg: The array of scatter/gather entries
3751  * @nents: The number of scatter/gather entries
3752  * @direction: The direction of the DMA
3753  */
3754 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3755 				   struct scatterlist *sg, int nents,
3756 				   enum dma_data_direction direction)
3757 {
3758 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
3759 }
3760 
3761 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3762 				      struct scatterlist *sg, int nents,
3763 				      enum dma_data_direction direction,
3764 				      unsigned long dma_attrs)
3765 {
3766 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3767 				dma_attrs);
3768 }
3769 
3770 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3771 					 struct scatterlist *sg, int nents,
3772 					 enum dma_data_direction direction,
3773 					 unsigned long dma_attrs)
3774 {
3775 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3776 }
3777 
3778 /**
3779  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
3780  * @dev: The device to query
3781  *
3782  * The returned value represents a size in bytes.
3783  */
3784 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
3785 {
3786 	struct device_dma_parameters *p = dev->dma_device->dma_parms;
3787 
3788 	return p ? p->max_segment_size : UINT_MAX;
3789 }
3790 
3791 /**
3792  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3793  * @dev: The device for which the DMA address was created
3794  * @addr: The DMA address
3795  * @size: The size of the region in bytes
3796  * @dir: The direction of the DMA
3797  */
3798 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3799 					      u64 addr,
3800 					      size_t size,
3801 					      enum dma_data_direction dir)
3802 {
3803 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3804 }
3805 
3806 /**
3807  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3808  * @dev: The device for which the DMA address was created
3809  * @addr: The DMA address
3810  * @size: The size of the region in bytes
3811  * @dir: The direction of the DMA
3812  */
3813 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3814 						 u64 addr,
3815 						 size_t size,
3816 						 enum dma_data_direction dir)
3817 {
3818 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3819 }
3820 
3821 /**
3822  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3823  * @dev: The device for which the DMA address is requested
3824  * @size: The size of the region to allocate in bytes
3825  * @dma_handle: A pointer for returning the DMA address of the region
3826  * @flag: memory allocator flags
3827  */
3828 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3829 					   size_t size,
3830 					   dma_addr_t *dma_handle,
3831 					   gfp_t flag)
3832 {
3833 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3834 }
3835 
3836 /**
3837  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3838  * @dev: The device for which the DMA addresses were allocated
3839  * @size: The size of the region
3840  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3841  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3842  */
3843 static inline void ib_dma_free_coherent(struct ib_device *dev,
3844 					size_t size, void *cpu_addr,
3845 					dma_addr_t dma_handle)
3846 {
3847 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3848 }
3849 
3850 /**
3851  * ib_dereg_mr - Deregisters a memory region and removes it from the
3852  *   HCA translation table.
3853  * @mr: The memory region to deregister.
3854  *
3855  * This function can fail, if the memory region has memory windows bound to it.
3856  */
3857 int ib_dereg_mr(struct ib_mr *mr);
3858 
3859 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3860 			  enum ib_mr_type mr_type,
3861 			  u32 max_num_sg);
3862 
3863 /**
3864  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3865  *   R_Key and L_Key.
3866  * @mr - struct ib_mr pointer to be updated.
3867  * @newkey - new key to be used.
3868  */
3869 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3870 {
3871 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3872 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3873 }
3874 
3875 /**
3876  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3877  * for calculating a new rkey for type 2 memory windows.
3878  * @rkey - the rkey to increment.
3879  */
3880 static inline u32 ib_inc_rkey(u32 rkey)
3881 {
3882 	const u32 mask = 0x000000ff;
3883 	return ((rkey + 1) & mask) | (rkey & ~mask);
3884 }
3885 
3886 /**
3887  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3888  * @pd: The protection domain associated with the unmapped region.
3889  * @mr_access_flags: Specifies the memory access rights.
3890  * @fmr_attr: Attributes of the unmapped region.
3891  *
3892  * A fast memory region must be mapped before it can be used as part of
3893  * a work request.
3894  */
3895 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3896 			    int mr_access_flags,
3897 			    struct ib_fmr_attr *fmr_attr);
3898 
3899 /**
3900  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3901  * @fmr: The fast memory region to associate with the pages.
3902  * @page_list: An array of physical pages to map to the fast memory region.
3903  * @list_len: The number of pages in page_list.
3904  * @iova: The I/O virtual address to use with the mapped region.
3905  */
3906 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3907 				  u64 *page_list, int list_len,
3908 				  u64 iova)
3909 {
3910 	return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
3911 }
3912 
3913 /**
3914  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3915  * @fmr_list: A linked list of fast memory regions to unmap.
3916  */
3917 int ib_unmap_fmr(struct list_head *fmr_list);
3918 
3919 /**
3920  * ib_dealloc_fmr - Deallocates a fast memory region.
3921  * @fmr: The fast memory region to deallocate.
3922  */
3923 int ib_dealloc_fmr(struct ib_fmr *fmr);
3924 
3925 /**
3926  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3927  * @qp: QP to attach to the multicast group.  The QP must be type
3928  *   IB_QPT_UD.
3929  * @gid: Multicast group GID.
3930  * @lid: Multicast group LID in host byte order.
3931  *
3932  * In order to send and receive multicast packets, subnet
3933  * administration must have created the multicast group and configured
3934  * the fabric appropriately.  The port associated with the specified
3935  * QP must also be a member of the multicast group.
3936  */
3937 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3938 
3939 /**
3940  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3941  * @qp: QP to detach from the multicast group.
3942  * @gid: Multicast group GID.
3943  * @lid: Multicast group LID in host byte order.
3944  */
3945 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3946 
3947 /**
3948  * ib_alloc_xrcd - Allocates an XRC domain.
3949  * @device: The device on which to allocate the XRC domain.
3950  * @caller: Module name for kernel consumers
3951  */
3952 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3953 #define ib_alloc_xrcd(device) \
3954 	__ib_alloc_xrcd((device), KBUILD_MODNAME)
3955 
3956 /**
3957  * ib_dealloc_xrcd - Deallocates an XRC domain.
3958  * @xrcd: The XRC domain to deallocate.
3959  */
3960 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3961 
3962 static inline int ib_check_mr_access(int flags)
3963 {
3964 	/*
3965 	 * Local write permission is required if remote write or
3966 	 * remote atomic permission is also requested.
3967 	 */
3968 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3969 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3970 		return -EINVAL;
3971 
3972 	return 0;
3973 }
3974 
3975 static inline bool ib_access_writable(int access_flags)
3976 {
3977 	/*
3978 	 * We have writable memory backing the MR if any of the following
3979 	 * access flags are set.  "Local write" and "remote write" obviously
3980 	 * require write access.  "Remote atomic" can do things like fetch and
3981 	 * add, which will modify memory, and "MW bind" can change permissions
3982 	 * by binding a window.
3983 	 */
3984 	return access_flags &
3985 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
3986 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
3987 }
3988 
3989 /**
3990  * ib_check_mr_status: lightweight check of MR status.
3991  *     This routine may provide status checks on a selected
3992  *     ib_mr. first use is for signature status check.
3993  *
3994  * @mr: A memory region.
3995  * @check_mask: Bitmask of which checks to perform from
3996  *     ib_mr_status_check enumeration.
3997  * @mr_status: The container of relevant status checks.
3998  *     failed checks will be indicated in the status bitmask
3999  *     and the relevant info shall be in the error item.
4000  */
4001 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4002 		       struct ib_mr_status *mr_status);
4003 
4004 /**
4005  * ib_device_try_get: Hold a registration lock
4006  * device: The device to lock
4007  *
4008  * A device under an active registration lock cannot become unregistered. It
4009  * is only possible to obtain a registration lock on a device that is fully
4010  * registered, otherwise this function returns false.
4011  *
4012  * The registration lock is only necessary for actions which require the
4013  * device to still be registered. Uses that only require the device pointer to
4014  * be valid should use get_device(&ibdev->dev) to hold the memory.
4015  *
4016  */
4017 static inline bool ib_device_try_get(struct ib_device *dev)
4018 {
4019 	return refcount_inc_not_zero(&dev->refcount);
4020 }
4021 
4022 void ib_device_put(struct ib_device *device);
4023 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4024 					  enum rdma_driver_id driver_id);
4025 struct ib_device *ib_device_get_by_name(const char *name,
4026 					enum rdma_driver_id driver_id);
4027 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4028 					    u16 pkey, const union ib_gid *gid,
4029 					    const struct sockaddr *addr);
4030 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4031 			 unsigned int port);
4032 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4033 
4034 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4035 			   struct ib_wq_init_attr *init_attr);
4036 int ib_destroy_wq(struct ib_wq *wq);
4037 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4038 		 u32 wq_attr_mask);
4039 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4040 						 struct ib_rwq_ind_table_init_attr*
4041 						 wq_ind_table_init_attr);
4042 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4043 
4044 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4045 		 unsigned int *sg_offset, unsigned int page_size);
4046 
4047 static inline int
4048 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4049 		  unsigned int *sg_offset, unsigned int page_size)
4050 {
4051 	int n;
4052 
4053 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4054 	mr->iova = 0;
4055 
4056 	return n;
4057 }
4058 
4059 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4060 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4061 
4062 void ib_drain_rq(struct ib_qp *qp);
4063 void ib_drain_sq(struct ib_qp *qp);
4064 void ib_drain_qp(struct ib_qp *qp);
4065 
4066 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4067 
4068 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4069 {
4070 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4071 		return attr->roce.dmac;
4072 	return NULL;
4073 }
4074 
4075 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4076 {
4077 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4078 		attr->ib.dlid = (u16)dlid;
4079 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4080 		attr->opa.dlid = dlid;
4081 }
4082 
4083 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4084 {
4085 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4086 		return attr->ib.dlid;
4087 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4088 		return attr->opa.dlid;
4089 	return 0;
4090 }
4091 
4092 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4093 {
4094 	attr->sl = sl;
4095 }
4096 
4097 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4098 {
4099 	return attr->sl;
4100 }
4101 
4102 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4103 					 u8 src_path_bits)
4104 {
4105 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4106 		attr->ib.src_path_bits = src_path_bits;
4107 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4108 		attr->opa.src_path_bits = src_path_bits;
4109 }
4110 
4111 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4112 {
4113 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4114 		return attr->ib.src_path_bits;
4115 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4116 		return attr->opa.src_path_bits;
4117 	return 0;
4118 }
4119 
4120 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4121 					bool make_grd)
4122 {
4123 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4124 		attr->opa.make_grd = make_grd;
4125 }
4126 
4127 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4128 {
4129 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4130 		return attr->opa.make_grd;
4131 	return false;
4132 }
4133 
4134 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4135 {
4136 	attr->port_num = port_num;
4137 }
4138 
4139 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4140 {
4141 	return attr->port_num;
4142 }
4143 
4144 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4145 					   u8 static_rate)
4146 {
4147 	attr->static_rate = static_rate;
4148 }
4149 
4150 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4151 {
4152 	return attr->static_rate;
4153 }
4154 
4155 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4156 					enum ib_ah_flags flag)
4157 {
4158 	attr->ah_flags = flag;
4159 }
4160 
4161 static inline enum ib_ah_flags
4162 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4163 {
4164 	return attr->ah_flags;
4165 }
4166 
4167 static inline const struct ib_global_route
4168 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4169 {
4170 	return &attr->grh;
4171 }
4172 
4173 /*To retrieve and modify the grh */
4174 static inline struct ib_global_route
4175 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4176 {
4177 	return &attr->grh;
4178 }
4179 
4180 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4181 {
4182 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4183 
4184 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4185 }
4186 
4187 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4188 					     __be64 prefix)
4189 {
4190 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4191 
4192 	grh->dgid.global.subnet_prefix = prefix;
4193 }
4194 
4195 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4196 					    __be64 if_id)
4197 {
4198 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4199 
4200 	grh->dgid.global.interface_id = if_id;
4201 }
4202 
4203 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4204 				   union ib_gid *dgid, u32 flow_label,
4205 				   u8 sgid_index, u8 hop_limit,
4206 				   u8 traffic_class)
4207 {
4208 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4209 
4210 	attr->ah_flags = IB_AH_GRH;
4211 	if (dgid)
4212 		grh->dgid = *dgid;
4213 	grh->flow_label = flow_label;
4214 	grh->sgid_index = sgid_index;
4215 	grh->hop_limit = hop_limit;
4216 	grh->traffic_class = traffic_class;
4217 	grh->sgid_attr = NULL;
4218 }
4219 
4220 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4221 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4222 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4223 			     const struct ib_gid_attr *sgid_attr);
4224 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4225 		       const struct rdma_ah_attr *src);
4226 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4227 			  const struct rdma_ah_attr *new);
4228 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4229 
4230 /**
4231  * rdma_ah_find_type - Return address handle type.
4232  *
4233  * @dev: Device to be checked
4234  * @port_num: Port number
4235  */
4236 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4237 						       u8 port_num)
4238 {
4239 	if (rdma_protocol_roce(dev, port_num))
4240 		return RDMA_AH_ATTR_TYPE_ROCE;
4241 	if (rdma_protocol_ib(dev, port_num)) {
4242 		if (rdma_cap_opa_ah(dev, port_num))
4243 			return RDMA_AH_ATTR_TYPE_OPA;
4244 		return RDMA_AH_ATTR_TYPE_IB;
4245 	}
4246 
4247 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4248 }
4249 
4250 /**
4251  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4252  *     In the current implementation the only way to get
4253  *     get the 32bit lid is from other sources for OPA.
4254  *     For IB, lids will always be 16bits so cast the
4255  *     value accordingly.
4256  *
4257  * @lid: A 32bit LID
4258  */
4259 static inline u16 ib_lid_cpu16(u32 lid)
4260 {
4261 	WARN_ON_ONCE(lid & 0xFFFF0000);
4262 	return (u16)lid;
4263 }
4264 
4265 /**
4266  * ib_lid_be16 - Return lid in 16bit BE encoding.
4267  *
4268  * @lid: A 32bit LID
4269  */
4270 static inline __be16 ib_lid_be16(u32 lid)
4271 {
4272 	WARN_ON_ONCE(lid & 0xFFFF0000);
4273 	return cpu_to_be16((u16)lid);
4274 }
4275 
4276 /**
4277  * ib_get_vector_affinity - Get the affinity mappings of a given completion
4278  *   vector
4279  * @device:         the rdma device
4280  * @comp_vector:    index of completion vector
4281  *
4282  * Returns NULL on failure, otherwise a corresponding cpu map of the
4283  * completion vector (returns all-cpus map if the device driver doesn't
4284  * implement get_vector_affinity).
4285  */
4286 static inline const struct cpumask *
4287 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4288 {
4289 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4290 	    !device->ops.get_vector_affinity)
4291 		return NULL;
4292 
4293 	return device->ops.get_vector_affinity(device, comp_vector);
4294 
4295 }
4296 
4297 /**
4298  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4299  * and add their gids, as needed, to the relevant RoCE devices.
4300  *
4301  * @device:         the rdma device
4302  */
4303 void rdma_roce_rescan_device(struct ib_device *ibdev);
4304 
4305 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4306 
4307 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4308 
4309 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4310 				     enum rdma_netdev_t type, const char *name,
4311 				     unsigned char name_assign_type,
4312 				     void (*setup)(struct net_device *));
4313 
4314 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4315 		     enum rdma_netdev_t type, const char *name,
4316 		     unsigned char name_assign_type,
4317 		     void (*setup)(struct net_device *),
4318 		     struct net_device *netdev);
4319 
4320 /**
4321  * rdma_set_device_sysfs_group - Set device attributes group to have
4322  *				 driver specific sysfs entries at
4323  *				 for infiniband class.
4324  *
4325  * @device:	device pointer for which attributes to be created
4326  * @group:	Pointer to group which should be added when device
4327  *		is registered with sysfs.
4328  * rdma_set_device_sysfs_group() allows existing drivers to expose one
4329  * group per device to have sysfs attributes.
4330  *
4331  * NOTE: New drivers should not make use of this API; instead new device
4332  * parameter should be exposed via netlink command. This API and mechanism
4333  * exist only for existing drivers.
4334  */
4335 static inline void
4336 rdma_set_device_sysfs_group(struct ib_device *dev,
4337 			    const struct attribute_group *group)
4338 {
4339 	dev->groups[1] = group;
4340 }
4341 
4342 /**
4343  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4344  *
4345  * @device:	device pointer for which ib_device pointer to retrieve
4346  *
4347  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4348  *
4349  */
4350 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4351 {
4352 	return container_of(device, struct ib_device, dev);
4353 }
4354 
4355 /**
4356  * rdma_device_to_drv_device - Helper macro to reach back to driver's
4357  *			       ib_device holder structure from device pointer.
4358  *
4359  * NOTE: New drivers should not make use of this API; This API is only for
4360  * existing drivers who have exposed sysfs entries using
4361  * rdma_set_device_sysfs_group().
4362  */
4363 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
4364 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4365 #endif /* IB_VERBS_H */
4366