xref: /openbmc/linux/include/rdma/ib_verbs.h (revision c9933d49)
1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
2 /*
3  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
4  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
5  * Copyright (c) 2004, 2020 Intel Corporation.  All rights reserved.
6  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
7  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
8  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
9  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
10  */
11 
12 #ifndef IB_VERBS_H
13 #define IB_VERBS_H
14 
15 #include <linux/ethtool.h>
16 #include <linux/types.h>
17 #include <linux/device.h>
18 #include <linux/dma-mapping.h>
19 #include <linux/kref.h>
20 #include <linux/list.h>
21 #include <linux/rwsem.h>
22 #include <linux/workqueue.h>
23 #include <linux/irq_poll.h>
24 #include <uapi/linux/if_ether.h>
25 #include <net/ipv6.h>
26 #include <net/ip.h>
27 #include <linux/string.h>
28 #include <linux/slab.h>
29 #include <linux/netdevice.h>
30 #include <linux/refcount.h>
31 #include <linux/if_link.h>
32 #include <linux/atomic.h>
33 #include <linux/mmu_notifier.h>
34 #include <linux/uaccess.h>
35 #include <linux/cgroup_rdma.h>
36 #include <linux/irqflags.h>
37 #include <linux/preempt.h>
38 #include <linux/dim.h>
39 #include <uapi/rdma/ib_user_verbs.h>
40 #include <rdma/rdma_counter.h>
41 #include <rdma/restrack.h>
42 #include <rdma/signature.h>
43 #include <uapi/rdma/rdma_user_ioctl.h>
44 #include <uapi/rdma/ib_user_ioctl_verbs.h>
45 
46 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
47 
48 struct ib_umem_odp;
49 struct ib_uqp_object;
50 struct ib_usrq_object;
51 struct ib_uwq_object;
52 struct rdma_cm_id;
53 struct ib_port;
54 struct hw_stats_device_data;
55 
56 extern struct workqueue_struct *ib_wq;
57 extern struct workqueue_struct *ib_comp_wq;
58 extern struct workqueue_struct *ib_comp_unbound_wq;
59 
60 struct ib_ucq_object;
61 
62 __printf(3, 4) __cold
63 void ibdev_printk(const char *level, const struct ib_device *ibdev,
64 		  const char *format, ...);
65 __printf(2, 3) __cold
66 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
67 __printf(2, 3) __cold
68 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
69 __printf(2, 3) __cold
70 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
71 __printf(2, 3) __cold
72 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
73 __printf(2, 3) __cold
74 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
75 __printf(2, 3) __cold
76 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
77 __printf(2, 3) __cold
78 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
79 
80 #if defined(CONFIG_DYNAMIC_DEBUG) || \
81 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
82 #define ibdev_dbg(__dev, format, args...)                       \
83 	dynamic_ibdev_dbg(__dev, format, ##args)
84 #else
85 __printf(2, 3) __cold
86 static inline
87 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
88 #endif
89 
90 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...)           \
91 do {                                                                    \
92 	static DEFINE_RATELIMIT_STATE(_rs,                              \
93 				      DEFAULT_RATELIMIT_INTERVAL,       \
94 				      DEFAULT_RATELIMIT_BURST);         \
95 	if (__ratelimit(&_rs))                                          \
96 		ibdev_level(ibdev, fmt, ##__VA_ARGS__);                 \
97 } while (0)
98 
99 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
100 	ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
101 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
102 	ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
103 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
104 	ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
105 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
106 	ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
107 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
108 	ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
109 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
110 	ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
111 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
112 	ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
113 
114 #if defined(CONFIG_DYNAMIC_DEBUG) || \
115 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
116 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
117 #define ibdev_dbg_ratelimited(ibdev, fmt, ...)                          \
118 do {                                                                    \
119 	static DEFINE_RATELIMIT_STATE(_rs,                              \
120 				      DEFAULT_RATELIMIT_INTERVAL,       \
121 				      DEFAULT_RATELIMIT_BURST);         \
122 	DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt);                 \
123 	if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs))      \
124 		__dynamic_ibdev_dbg(&descriptor, ibdev, fmt,            \
125 				    ##__VA_ARGS__);                     \
126 } while (0)
127 #else
128 __printf(2, 3) __cold
129 static inline
130 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
131 #endif
132 
133 union ib_gid {
134 	u8	raw[16];
135 	struct {
136 		__be64	subnet_prefix;
137 		__be64	interface_id;
138 	} global;
139 };
140 
141 extern union ib_gid zgid;
142 
143 enum ib_gid_type {
144 	IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
145 	IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
146 	IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
147 	IB_GID_TYPE_SIZE
148 };
149 
150 #define ROCE_V2_UDP_DPORT      4791
151 struct ib_gid_attr {
152 	struct net_device __rcu	*ndev;
153 	struct ib_device	*device;
154 	union ib_gid		gid;
155 	enum ib_gid_type	gid_type;
156 	u16			index;
157 	u32			port_num;
158 };
159 
160 enum {
161 	/* set the local administered indication */
162 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
163 };
164 
165 enum rdma_transport_type {
166 	RDMA_TRANSPORT_IB,
167 	RDMA_TRANSPORT_IWARP,
168 	RDMA_TRANSPORT_USNIC,
169 	RDMA_TRANSPORT_USNIC_UDP,
170 	RDMA_TRANSPORT_UNSPECIFIED,
171 };
172 
173 enum rdma_protocol_type {
174 	RDMA_PROTOCOL_IB,
175 	RDMA_PROTOCOL_IBOE,
176 	RDMA_PROTOCOL_IWARP,
177 	RDMA_PROTOCOL_USNIC_UDP
178 };
179 
180 __attribute_const__ enum rdma_transport_type
181 rdma_node_get_transport(unsigned int node_type);
182 
183 enum rdma_network_type {
184 	RDMA_NETWORK_IB,
185 	RDMA_NETWORK_ROCE_V1,
186 	RDMA_NETWORK_IPV4,
187 	RDMA_NETWORK_IPV6
188 };
189 
190 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
191 {
192 	if (network_type == RDMA_NETWORK_IPV4 ||
193 	    network_type == RDMA_NETWORK_IPV6)
194 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
195 	else if (network_type == RDMA_NETWORK_ROCE_V1)
196 		return IB_GID_TYPE_ROCE;
197 	else
198 		return IB_GID_TYPE_IB;
199 }
200 
201 static inline enum rdma_network_type
202 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
203 {
204 	if (attr->gid_type == IB_GID_TYPE_IB)
205 		return RDMA_NETWORK_IB;
206 
207 	if (attr->gid_type == IB_GID_TYPE_ROCE)
208 		return RDMA_NETWORK_ROCE_V1;
209 
210 	if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
211 		return RDMA_NETWORK_IPV4;
212 	else
213 		return RDMA_NETWORK_IPV6;
214 }
215 
216 enum rdma_link_layer {
217 	IB_LINK_LAYER_UNSPECIFIED,
218 	IB_LINK_LAYER_INFINIBAND,
219 	IB_LINK_LAYER_ETHERNET,
220 };
221 
222 enum ib_device_cap_flags {
223 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
224 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
225 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
226 	IB_DEVICE_RAW_MULTI			= (1 << 3),
227 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
228 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
229 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
230 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
231 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
232 	/* Not in use, former INIT_TYPE		= (1 << 9),*/
233 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
234 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
235 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
236 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
237 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
238 
239 	/*
240 	 * This device supports a per-device lkey or stag that can be
241 	 * used without performing a memory registration for the local
242 	 * memory.  Note that ULPs should never check this flag, but
243 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
244 	 * which will always contain a usable lkey.
245 	 */
246 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
247 	/* Reserved, old SEND_W_INV		= (1 << 16),*/
248 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
249 	/*
250 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
251 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
252 	 * messages and can verify the validity of checksum for
253 	 * incoming messages.  Setting this flag implies that the
254 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
255 	 */
256 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
257 	IB_DEVICE_UD_TSO			= (1 << 19),
258 	IB_DEVICE_XRC				= (1 << 20),
259 
260 	/*
261 	 * This device supports the IB "base memory management extension",
262 	 * which includes support for fast registrations (IB_WR_REG_MR,
263 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
264 	 * also be set by any iWarp device which must support FRs to comply
265 	 * to the iWarp verbs spec.  iWarp devices also support the
266 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
267 	 * stag.
268 	 */
269 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
270 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
271 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
272 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
273 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
274 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
275 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
276 	/*
277 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
278 	 * support execution of WQEs that involve synchronization
279 	 * of I/O operations with single completion queue managed
280 	 * by hardware.
281 	 */
282 	IB_DEVICE_CROSS_CHANNEL			= (1 << 27),
283 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
284 	IB_DEVICE_INTEGRITY_HANDOVER		= (1 << 30),
285 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
286 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
287 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
288 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
289 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
290 	IB_DEVICE_RDMA_NETDEV_OPA		= (1ULL << 35),
291 	/* The device supports padding incoming writes to cacheline. */
292 	IB_DEVICE_PCI_WRITE_END_PADDING		= (1ULL << 36),
293 	IB_DEVICE_ALLOW_USER_UNREG		= (1ULL << 37),
294 };
295 
296 enum ib_atomic_cap {
297 	IB_ATOMIC_NONE,
298 	IB_ATOMIC_HCA,
299 	IB_ATOMIC_GLOB
300 };
301 
302 enum ib_odp_general_cap_bits {
303 	IB_ODP_SUPPORT		= 1 << 0,
304 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
305 };
306 
307 enum ib_odp_transport_cap_bits {
308 	IB_ODP_SUPPORT_SEND	= 1 << 0,
309 	IB_ODP_SUPPORT_RECV	= 1 << 1,
310 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
311 	IB_ODP_SUPPORT_READ	= 1 << 3,
312 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
313 	IB_ODP_SUPPORT_SRQ_RECV	= 1 << 5,
314 };
315 
316 struct ib_odp_caps {
317 	uint64_t general_caps;
318 	struct {
319 		uint32_t  rc_odp_caps;
320 		uint32_t  uc_odp_caps;
321 		uint32_t  ud_odp_caps;
322 		uint32_t  xrc_odp_caps;
323 	} per_transport_caps;
324 };
325 
326 struct ib_rss_caps {
327 	/* Corresponding bit will be set if qp type from
328 	 * 'enum ib_qp_type' is supported, e.g.
329 	 * supported_qpts |= 1 << IB_QPT_UD
330 	 */
331 	u32 supported_qpts;
332 	u32 max_rwq_indirection_tables;
333 	u32 max_rwq_indirection_table_size;
334 };
335 
336 enum ib_tm_cap_flags {
337 	/*  Support tag matching with rendezvous offload for RC transport */
338 	IB_TM_CAP_RNDV_RC = 1 << 0,
339 };
340 
341 struct ib_tm_caps {
342 	/* Max size of RNDV header */
343 	u32 max_rndv_hdr_size;
344 	/* Max number of entries in tag matching list */
345 	u32 max_num_tags;
346 	/* From enum ib_tm_cap_flags */
347 	u32 flags;
348 	/* Max number of outstanding list operations */
349 	u32 max_ops;
350 	/* Max number of SGE in tag matching entry */
351 	u32 max_sge;
352 };
353 
354 struct ib_cq_init_attr {
355 	unsigned int	cqe;
356 	u32		comp_vector;
357 	u32		flags;
358 };
359 
360 enum ib_cq_attr_mask {
361 	IB_CQ_MODERATE = 1 << 0,
362 };
363 
364 struct ib_cq_caps {
365 	u16     max_cq_moderation_count;
366 	u16     max_cq_moderation_period;
367 };
368 
369 struct ib_dm_mr_attr {
370 	u64		length;
371 	u64		offset;
372 	u32		access_flags;
373 };
374 
375 struct ib_dm_alloc_attr {
376 	u64	length;
377 	u32	alignment;
378 	u32	flags;
379 };
380 
381 struct ib_device_attr {
382 	u64			fw_ver;
383 	__be64			sys_image_guid;
384 	u64			max_mr_size;
385 	u64			page_size_cap;
386 	u32			vendor_id;
387 	u32			vendor_part_id;
388 	u32			hw_ver;
389 	int			max_qp;
390 	int			max_qp_wr;
391 	u64			device_cap_flags;
392 	int			max_send_sge;
393 	int			max_recv_sge;
394 	int			max_sge_rd;
395 	int			max_cq;
396 	int			max_cqe;
397 	int			max_mr;
398 	int			max_pd;
399 	int			max_qp_rd_atom;
400 	int			max_ee_rd_atom;
401 	int			max_res_rd_atom;
402 	int			max_qp_init_rd_atom;
403 	int			max_ee_init_rd_atom;
404 	enum ib_atomic_cap	atomic_cap;
405 	enum ib_atomic_cap	masked_atomic_cap;
406 	int			max_ee;
407 	int			max_rdd;
408 	int			max_mw;
409 	int			max_raw_ipv6_qp;
410 	int			max_raw_ethy_qp;
411 	int			max_mcast_grp;
412 	int			max_mcast_qp_attach;
413 	int			max_total_mcast_qp_attach;
414 	int			max_ah;
415 	int			max_srq;
416 	int			max_srq_wr;
417 	int			max_srq_sge;
418 	unsigned int		max_fast_reg_page_list_len;
419 	unsigned int		max_pi_fast_reg_page_list_len;
420 	u16			max_pkeys;
421 	u8			local_ca_ack_delay;
422 	int			sig_prot_cap;
423 	int			sig_guard_cap;
424 	struct ib_odp_caps	odp_caps;
425 	uint64_t		timestamp_mask;
426 	uint64_t		hca_core_clock; /* in KHZ */
427 	struct ib_rss_caps	rss_caps;
428 	u32			max_wq_type_rq;
429 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
430 	struct ib_tm_caps	tm_caps;
431 	struct ib_cq_caps       cq_caps;
432 	u64			max_dm_size;
433 	/* Max entries for sgl for optimized performance per READ */
434 	u32			max_sgl_rd;
435 };
436 
437 enum ib_mtu {
438 	IB_MTU_256  = 1,
439 	IB_MTU_512  = 2,
440 	IB_MTU_1024 = 3,
441 	IB_MTU_2048 = 4,
442 	IB_MTU_4096 = 5
443 };
444 
445 enum opa_mtu {
446 	OPA_MTU_8192 = 6,
447 	OPA_MTU_10240 = 7
448 };
449 
450 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
451 {
452 	switch (mtu) {
453 	case IB_MTU_256:  return  256;
454 	case IB_MTU_512:  return  512;
455 	case IB_MTU_1024: return 1024;
456 	case IB_MTU_2048: return 2048;
457 	case IB_MTU_4096: return 4096;
458 	default: 	  return -1;
459 	}
460 }
461 
462 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
463 {
464 	if (mtu >= 4096)
465 		return IB_MTU_4096;
466 	else if (mtu >= 2048)
467 		return IB_MTU_2048;
468 	else if (mtu >= 1024)
469 		return IB_MTU_1024;
470 	else if (mtu >= 512)
471 		return IB_MTU_512;
472 	else
473 		return IB_MTU_256;
474 }
475 
476 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
477 {
478 	switch (mtu) {
479 	case OPA_MTU_8192:
480 		return 8192;
481 	case OPA_MTU_10240:
482 		return 10240;
483 	default:
484 		return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
485 	}
486 }
487 
488 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
489 {
490 	if (mtu >= 10240)
491 		return OPA_MTU_10240;
492 	else if (mtu >= 8192)
493 		return OPA_MTU_8192;
494 	else
495 		return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
496 }
497 
498 enum ib_port_state {
499 	IB_PORT_NOP		= 0,
500 	IB_PORT_DOWN		= 1,
501 	IB_PORT_INIT		= 2,
502 	IB_PORT_ARMED		= 3,
503 	IB_PORT_ACTIVE		= 4,
504 	IB_PORT_ACTIVE_DEFER	= 5
505 };
506 
507 enum ib_port_phys_state {
508 	IB_PORT_PHYS_STATE_SLEEP = 1,
509 	IB_PORT_PHYS_STATE_POLLING = 2,
510 	IB_PORT_PHYS_STATE_DISABLED = 3,
511 	IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
512 	IB_PORT_PHYS_STATE_LINK_UP = 5,
513 	IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
514 	IB_PORT_PHYS_STATE_PHY_TEST = 7,
515 };
516 
517 enum ib_port_width {
518 	IB_WIDTH_1X	= 1,
519 	IB_WIDTH_2X	= 16,
520 	IB_WIDTH_4X	= 2,
521 	IB_WIDTH_8X	= 4,
522 	IB_WIDTH_12X	= 8
523 };
524 
525 static inline int ib_width_enum_to_int(enum ib_port_width width)
526 {
527 	switch (width) {
528 	case IB_WIDTH_1X:  return  1;
529 	case IB_WIDTH_2X:  return  2;
530 	case IB_WIDTH_4X:  return  4;
531 	case IB_WIDTH_8X:  return  8;
532 	case IB_WIDTH_12X: return 12;
533 	default: 	  return -1;
534 	}
535 }
536 
537 enum ib_port_speed {
538 	IB_SPEED_SDR	= 1,
539 	IB_SPEED_DDR	= 2,
540 	IB_SPEED_QDR	= 4,
541 	IB_SPEED_FDR10	= 8,
542 	IB_SPEED_FDR	= 16,
543 	IB_SPEED_EDR	= 32,
544 	IB_SPEED_HDR	= 64,
545 	IB_SPEED_NDR	= 128,
546 };
547 
548 enum ib_stat_flag {
549 	IB_STAT_FLAG_OPTIONAL = 1 << 0,
550 };
551 
552 /**
553  * struct rdma_stat_desc
554  * @name - The name of the counter
555  * @flags - Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL
556  * @priv - Driver private information; Core code should not use
557  */
558 struct rdma_stat_desc {
559 	const char *name;
560 	unsigned int flags;
561 	const void *priv;
562 };
563 
564 /**
565  * struct rdma_hw_stats
566  * @lock - Mutex to protect parallel write access to lifespan and values
567  *    of counters, which are 64bits and not guaranteeed to be written
568  *    atomicaly on 32bits systems.
569  * @timestamp - Used by the core code to track when the last update was
570  * @lifespan - Used by the core code to determine how old the counters
571  *   should be before being updated again.  Stored in jiffies, defaults
572  *   to 10 milliseconds, drivers can override the default be specifying
573  *   their own value during their allocation routine.
574  * @descs - Array of pointers to static descriptors used for the counters
575  *   in directory.
576  * @is_disabled - A bitmap to indicate each counter is currently disabled
577  *   or not.
578  * @num_counters - How many hardware counters there are.  If name is
579  *   shorter than this number, a kernel oops will result.  Driver authors
580  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
581  *   in their code to prevent this.
582  * @value - Array of u64 counters that are accessed by the sysfs code and
583  *   filled in by the drivers get_stats routine
584  */
585 struct rdma_hw_stats {
586 	struct mutex	lock; /* Protect lifespan and values[] */
587 	unsigned long	timestamp;
588 	unsigned long	lifespan;
589 	const struct rdma_stat_desc *descs;
590 	unsigned long	*is_disabled;
591 	int		num_counters;
592 	u64		value[];
593 };
594 
595 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
596 
597 struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
598 	const struct rdma_stat_desc *descs, int num_counters,
599 	unsigned long lifespan);
600 
601 void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
602 
603 /* Define bits for the various functionality this port needs to be supported by
604  * the core.
605  */
606 /* Management                           0x00000FFF */
607 #define RDMA_CORE_CAP_IB_MAD            0x00000001
608 #define RDMA_CORE_CAP_IB_SMI            0x00000002
609 #define RDMA_CORE_CAP_IB_CM             0x00000004
610 #define RDMA_CORE_CAP_IW_CM             0x00000008
611 #define RDMA_CORE_CAP_IB_SA             0x00000010
612 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
613 
614 /* Address format                       0x000FF000 */
615 #define RDMA_CORE_CAP_AF_IB             0x00001000
616 #define RDMA_CORE_CAP_ETH_AH            0x00002000
617 #define RDMA_CORE_CAP_OPA_AH            0x00004000
618 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
619 
620 /* Protocol                             0xFFF00000 */
621 #define RDMA_CORE_CAP_PROT_IB           0x00100000
622 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
623 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
624 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
625 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
626 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
627 
628 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
629 					| RDMA_CORE_CAP_PROT_ROCE     \
630 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
631 
632 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
633 					| RDMA_CORE_CAP_IB_MAD \
634 					| RDMA_CORE_CAP_IB_SMI \
635 					| RDMA_CORE_CAP_IB_CM  \
636 					| RDMA_CORE_CAP_IB_SA  \
637 					| RDMA_CORE_CAP_AF_IB)
638 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
639 					| RDMA_CORE_CAP_IB_MAD  \
640 					| RDMA_CORE_CAP_IB_CM   \
641 					| RDMA_CORE_CAP_AF_IB   \
642 					| RDMA_CORE_CAP_ETH_AH)
643 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
644 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
645 					| RDMA_CORE_CAP_IB_MAD  \
646 					| RDMA_CORE_CAP_IB_CM   \
647 					| RDMA_CORE_CAP_AF_IB   \
648 					| RDMA_CORE_CAP_ETH_AH)
649 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
650 					| RDMA_CORE_CAP_IW_CM)
651 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
652 					| RDMA_CORE_CAP_OPA_MAD)
653 
654 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
655 
656 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
657 
658 struct ib_port_attr {
659 	u64			subnet_prefix;
660 	enum ib_port_state	state;
661 	enum ib_mtu		max_mtu;
662 	enum ib_mtu		active_mtu;
663 	u32                     phys_mtu;
664 	int			gid_tbl_len;
665 	unsigned int		ip_gids:1;
666 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
667 	u32			port_cap_flags;
668 	u32			max_msg_sz;
669 	u32			bad_pkey_cntr;
670 	u32			qkey_viol_cntr;
671 	u16			pkey_tbl_len;
672 	u32			sm_lid;
673 	u32			lid;
674 	u8			lmc;
675 	u8			max_vl_num;
676 	u8			sm_sl;
677 	u8			subnet_timeout;
678 	u8			init_type_reply;
679 	u8			active_width;
680 	u16			active_speed;
681 	u8                      phys_state;
682 	u16			port_cap_flags2;
683 };
684 
685 enum ib_device_modify_flags {
686 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
687 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
688 };
689 
690 #define IB_DEVICE_NODE_DESC_MAX 64
691 
692 struct ib_device_modify {
693 	u64	sys_image_guid;
694 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
695 };
696 
697 enum ib_port_modify_flags {
698 	IB_PORT_SHUTDOWN		= 1,
699 	IB_PORT_INIT_TYPE		= (1<<2),
700 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
701 	IB_PORT_OPA_MASK_CHG		= (1<<4)
702 };
703 
704 struct ib_port_modify {
705 	u32	set_port_cap_mask;
706 	u32	clr_port_cap_mask;
707 	u8	init_type;
708 };
709 
710 enum ib_event_type {
711 	IB_EVENT_CQ_ERR,
712 	IB_EVENT_QP_FATAL,
713 	IB_EVENT_QP_REQ_ERR,
714 	IB_EVENT_QP_ACCESS_ERR,
715 	IB_EVENT_COMM_EST,
716 	IB_EVENT_SQ_DRAINED,
717 	IB_EVENT_PATH_MIG,
718 	IB_EVENT_PATH_MIG_ERR,
719 	IB_EVENT_DEVICE_FATAL,
720 	IB_EVENT_PORT_ACTIVE,
721 	IB_EVENT_PORT_ERR,
722 	IB_EVENT_LID_CHANGE,
723 	IB_EVENT_PKEY_CHANGE,
724 	IB_EVENT_SM_CHANGE,
725 	IB_EVENT_SRQ_ERR,
726 	IB_EVENT_SRQ_LIMIT_REACHED,
727 	IB_EVENT_QP_LAST_WQE_REACHED,
728 	IB_EVENT_CLIENT_REREGISTER,
729 	IB_EVENT_GID_CHANGE,
730 	IB_EVENT_WQ_FATAL,
731 };
732 
733 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
734 
735 struct ib_event {
736 	struct ib_device	*device;
737 	union {
738 		struct ib_cq	*cq;
739 		struct ib_qp	*qp;
740 		struct ib_srq	*srq;
741 		struct ib_wq	*wq;
742 		u32		port_num;
743 	} element;
744 	enum ib_event_type	event;
745 };
746 
747 struct ib_event_handler {
748 	struct ib_device *device;
749 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
750 	struct list_head  list;
751 };
752 
753 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
754 	do {							\
755 		(_ptr)->device  = _device;			\
756 		(_ptr)->handler = _handler;			\
757 		INIT_LIST_HEAD(&(_ptr)->list);			\
758 	} while (0)
759 
760 struct ib_global_route {
761 	const struct ib_gid_attr *sgid_attr;
762 	union ib_gid	dgid;
763 	u32		flow_label;
764 	u8		sgid_index;
765 	u8		hop_limit;
766 	u8		traffic_class;
767 };
768 
769 struct ib_grh {
770 	__be32		version_tclass_flow;
771 	__be16		paylen;
772 	u8		next_hdr;
773 	u8		hop_limit;
774 	union ib_gid	sgid;
775 	union ib_gid	dgid;
776 };
777 
778 union rdma_network_hdr {
779 	struct ib_grh ibgrh;
780 	struct {
781 		/* The IB spec states that if it's IPv4, the header
782 		 * is located in the last 20 bytes of the header.
783 		 */
784 		u8		reserved[20];
785 		struct iphdr	roce4grh;
786 	};
787 };
788 
789 #define IB_QPN_MASK		0xFFFFFF
790 
791 enum {
792 	IB_MULTICAST_QPN = 0xffffff
793 };
794 
795 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
796 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
797 
798 enum ib_ah_flags {
799 	IB_AH_GRH	= 1
800 };
801 
802 enum ib_rate {
803 	IB_RATE_PORT_CURRENT = 0,
804 	IB_RATE_2_5_GBPS = 2,
805 	IB_RATE_5_GBPS   = 5,
806 	IB_RATE_10_GBPS  = 3,
807 	IB_RATE_20_GBPS  = 6,
808 	IB_RATE_30_GBPS  = 4,
809 	IB_RATE_40_GBPS  = 7,
810 	IB_RATE_60_GBPS  = 8,
811 	IB_RATE_80_GBPS  = 9,
812 	IB_RATE_120_GBPS = 10,
813 	IB_RATE_14_GBPS  = 11,
814 	IB_RATE_56_GBPS  = 12,
815 	IB_RATE_112_GBPS = 13,
816 	IB_RATE_168_GBPS = 14,
817 	IB_RATE_25_GBPS  = 15,
818 	IB_RATE_100_GBPS = 16,
819 	IB_RATE_200_GBPS = 17,
820 	IB_RATE_300_GBPS = 18,
821 	IB_RATE_28_GBPS  = 19,
822 	IB_RATE_50_GBPS  = 20,
823 	IB_RATE_400_GBPS = 21,
824 	IB_RATE_600_GBPS = 22,
825 };
826 
827 /**
828  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
829  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
830  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
831  * @rate: rate to convert.
832  */
833 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
834 
835 /**
836  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
837  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
838  * @rate: rate to convert.
839  */
840 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
841 
842 
843 /**
844  * enum ib_mr_type - memory region type
845  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
846  *                            normal registration
847  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
848  *                            register any arbitrary sg lists (without
849  *                            the normal mr constraints - see
850  *                            ib_map_mr_sg)
851  * @IB_MR_TYPE_DM:            memory region that is used for device
852  *                            memory registration
853  * @IB_MR_TYPE_USER:          memory region that is used for the user-space
854  *                            application
855  * @IB_MR_TYPE_DMA:           memory region that is used for DMA operations
856  *                            without address translations (VA=PA)
857  * @IB_MR_TYPE_INTEGRITY:     memory region that is used for
858  *                            data integrity operations
859  */
860 enum ib_mr_type {
861 	IB_MR_TYPE_MEM_REG,
862 	IB_MR_TYPE_SG_GAPS,
863 	IB_MR_TYPE_DM,
864 	IB_MR_TYPE_USER,
865 	IB_MR_TYPE_DMA,
866 	IB_MR_TYPE_INTEGRITY,
867 };
868 
869 enum ib_mr_status_check {
870 	IB_MR_CHECK_SIG_STATUS = 1,
871 };
872 
873 /**
874  * struct ib_mr_status - Memory region status container
875  *
876  * @fail_status: Bitmask of MR checks status. For each
877  *     failed check a corresponding status bit is set.
878  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
879  *     failure.
880  */
881 struct ib_mr_status {
882 	u32		    fail_status;
883 	struct ib_sig_err   sig_err;
884 };
885 
886 /**
887  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
888  * enum.
889  * @mult: multiple to convert.
890  */
891 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
892 
893 struct rdma_ah_init_attr {
894 	struct rdma_ah_attr *ah_attr;
895 	u32 flags;
896 	struct net_device *xmit_slave;
897 };
898 
899 enum rdma_ah_attr_type {
900 	RDMA_AH_ATTR_TYPE_UNDEFINED,
901 	RDMA_AH_ATTR_TYPE_IB,
902 	RDMA_AH_ATTR_TYPE_ROCE,
903 	RDMA_AH_ATTR_TYPE_OPA,
904 };
905 
906 struct ib_ah_attr {
907 	u16			dlid;
908 	u8			src_path_bits;
909 };
910 
911 struct roce_ah_attr {
912 	u8			dmac[ETH_ALEN];
913 };
914 
915 struct opa_ah_attr {
916 	u32			dlid;
917 	u8			src_path_bits;
918 	bool			make_grd;
919 };
920 
921 struct rdma_ah_attr {
922 	struct ib_global_route	grh;
923 	u8			sl;
924 	u8			static_rate;
925 	u32			port_num;
926 	u8			ah_flags;
927 	enum rdma_ah_attr_type type;
928 	union {
929 		struct ib_ah_attr ib;
930 		struct roce_ah_attr roce;
931 		struct opa_ah_attr opa;
932 	};
933 };
934 
935 enum ib_wc_status {
936 	IB_WC_SUCCESS,
937 	IB_WC_LOC_LEN_ERR,
938 	IB_WC_LOC_QP_OP_ERR,
939 	IB_WC_LOC_EEC_OP_ERR,
940 	IB_WC_LOC_PROT_ERR,
941 	IB_WC_WR_FLUSH_ERR,
942 	IB_WC_MW_BIND_ERR,
943 	IB_WC_BAD_RESP_ERR,
944 	IB_WC_LOC_ACCESS_ERR,
945 	IB_WC_REM_INV_REQ_ERR,
946 	IB_WC_REM_ACCESS_ERR,
947 	IB_WC_REM_OP_ERR,
948 	IB_WC_RETRY_EXC_ERR,
949 	IB_WC_RNR_RETRY_EXC_ERR,
950 	IB_WC_LOC_RDD_VIOL_ERR,
951 	IB_WC_REM_INV_RD_REQ_ERR,
952 	IB_WC_REM_ABORT_ERR,
953 	IB_WC_INV_EECN_ERR,
954 	IB_WC_INV_EEC_STATE_ERR,
955 	IB_WC_FATAL_ERR,
956 	IB_WC_RESP_TIMEOUT_ERR,
957 	IB_WC_GENERAL_ERR
958 };
959 
960 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
961 
962 enum ib_wc_opcode {
963 	IB_WC_SEND = IB_UVERBS_WC_SEND,
964 	IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
965 	IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
966 	IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
967 	IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
968 	IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
969 	IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
970 	IB_WC_LSO = IB_UVERBS_WC_TSO,
971 	IB_WC_REG_MR,
972 	IB_WC_MASKED_COMP_SWAP,
973 	IB_WC_MASKED_FETCH_ADD,
974 /*
975  * Set value of IB_WC_RECV so consumers can test if a completion is a
976  * receive by testing (opcode & IB_WC_RECV).
977  */
978 	IB_WC_RECV			= 1 << 7,
979 	IB_WC_RECV_RDMA_WITH_IMM
980 };
981 
982 enum ib_wc_flags {
983 	IB_WC_GRH		= 1,
984 	IB_WC_WITH_IMM		= (1<<1),
985 	IB_WC_WITH_INVALIDATE	= (1<<2),
986 	IB_WC_IP_CSUM_OK	= (1<<3),
987 	IB_WC_WITH_SMAC		= (1<<4),
988 	IB_WC_WITH_VLAN		= (1<<5),
989 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
990 };
991 
992 struct ib_wc {
993 	union {
994 		u64		wr_id;
995 		struct ib_cqe	*wr_cqe;
996 	};
997 	enum ib_wc_status	status;
998 	enum ib_wc_opcode	opcode;
999 	u32			vendor_err;
1000 	u32			byte_len;
1001 	struct ib_qp	       *qp;
1002 	union {
1003 		__be32		imm_data;
1004 		u32		invalidate_rkey;
1005 	} ex;
1006 	u32			src_qp;
1007 	u32			slid;
1008 	int			wc_flags;
1009 	u16			pkey_index;
1010 	u8			sl;
1011 	u8			dlid_path_bits;
1012 	u32 port_num; /* valid only for DR SMPs on switches */
1013 	u8			smac[ETH_ALEN];
1014 	u16			vlan_id;
1015 	u8			network_hdr_type;
1016 };
1017 
1018 enum ib_cq_notify_flags {
1019 	IB_CQ_SOLICITED			= 1 << 0,
1020 	IB_CQ_NEXT_COMP			= 1 << 1,
1021 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1022 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
1023 };
1024 
1025 enum ib_srq_type {
1026 	IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1027 	IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1028 	IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1029 };
1030 
1031 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1032 {
1033 	return srq_type == IB_SRQT_XRC ||
1034 	       srq_type == IB_SRQT_TM;
1035 }
1036 
1037 enum ib_srq_attr_mask {
1038 	IB_SRQ_MAX_WR	= 1 << 0,
1039 	IB_SRQ_LIMIT	= 1 << 1,
1040 };
1041 
1042 struct ib_srq_attr {
1043 	u32	max_wr;
1044 	u32	max_sge;
1045 	u32	srq_limit;
1046 };
1047 
1048 struct ib_srq_init_attr {
1049 	void		      (*event_handler)(struct ib_event *, void *);
1050 	void		       *srq_context;
1051 	struct ib_srq_attr	attr;
1052 	enum ib_srq_type	srq_type;
1053 
1054 	struct {
1055 		struct ib_cq   *cq;
1056 		union {
1057 			struct {
1058 				struct ib_xrcd *xrcd;
1059 			} xrc;
1060 
1061 			struct {
1062 				u32		max_num_tags;
1063 			} tag_matching;
1064 		};
1065 	} ext;
1066 };
1067 
1068 struct ib_qp_cap {
1069 	u32	max_send_wr;
1070 	u32	max_recv_wr;
1071 	u32	max_send_sge;
1072 	u32	max_recv_sge;
1073 	u32	max_inline_data;
1074 
1075 	/*
1076 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1077 	 * ib_create_qp() will calculate the right amount of neededed WRs
1078 	 * and MRs based on this.
1079 	 */
1080 	u32	max_rdma_ctxs;
1081 };
1082 
1083 enum ib_sig_type {
1084 	IB_SIGNAL_ALL_WR,
1085 	IB_SIGNAL_REQ_WR
1086 };
1087 
1088 enum ib_qp_type {
1089 	/*
1090 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1091 	 * here (and in that order) since the MAD layer uses them as
1092 	 * indices into a 2-entry table.
1093 	 */
1094 	IB_QPT_SMI,
1095 	IB_QPT_GSI,
1096 
1097 	IB_QPT_RC = IB_UVERBS_QPT_RC,
1098 	IB_QPT_UC = IB_UVERBS_QPT_UC,
1099 	IB_QPT_UD = IB_UVERBS_QPT_UD,
1100 	IB_QPT_RAW_IPV6,
1101 	IB_QPT_RAW_ETHERTYPE,
1102 	IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1103 	IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1104 	IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1105 	IB_QPT_MAX,
1106 	IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1107 	/* Reserve a range for qp types internal to the low level driver.
1108 	 * These qp types will not be visible at the IB core layer, so the
1109 	 * IB_QPT_MAX usages should not be affected in the core layer
1110 	 */
1111 	IB_QPT_RESERVED1 = 0x1000,
1112 	IB_QPT_RESERVED2,
1113 	IB_QPT_RESERVED3,
1114 	IB_QPT_RESERVED4,
1115 	IB_QPT_RESERVED5,
1116 	IB_QPT_RESERVED6,
1117 	IB_QPT_RESERVED7,
1118 	IB_QPT_RESERVED8,
1119 	IB_QPT_RESERVED9,
1120 	IB_QPT_RESERVED10,
1121 };
1122 
1123 enum ib_qp_create_flags {
1124 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1125 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	=
1126 		IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1127 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1128 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1129 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1130 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1131 	IB_QP_CREATE_INTEGRITY_EN		= 1 << 6,
1132 	IB_QP_CREATE_NETDEV_USE			= 1 << 7,
1133 	IB_QP_CREATE_SCATTER_FCS		=
1134 		IB_UVERBS_QP_CREATE_SCATTER_FCS,
1135 	IB_QP_CREATE_CVLAN_STRIPPING		=
1136 		IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1137 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1138 	IB_QP_CREATE_PCI_WRITE_END_PADDING	=
1139 		IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1140 	/* reserve bits 26-31 for low level drivers' internal use */
1141 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1142 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1143 };
1144 
1145 /*
1146  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1147  * callback to destroy the passed in QP.
1148  */
1149 
1150 struct ib_qp_init_attr {
1151 	/* Consumer's event_handler callback must not block */
1152 	void                  (*event_handler)(struct ib_event *, void *);
1153 
1154 	void		       *qp_context;
1155 	struct ib_cq	       *send_cq;
1156 	struct ib_cq	       *recv_cq;
1157 	struct ib_srq	       *srq;
1158 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1159 	struct ib_qp_cap	cap;
1160 	enum ib_sig_type	sq_sig_type;
1161 	enum ib_qp_type		qp_type;
1162 	u32			create_flags;
1163 
1164 	/*
1165 	 * Only needed for special QP types, or when using the RW API.
1166 	 */
1167 	u32			port_num;
1168 	struct ib_rwq_ind_table *rwq_ind_tbl;
1169 	u32			source_qpn;
1170 };
1171 
1172 struct ib_qp_open_attr {
1173 	void                  (*event_handler)(struct ib_event *, void *);
1174 	void		       *qp_context;
1175 	u32			qp_num;
1176 	enum ib_qp_type		qp_type;
1177 };
1178 
1179 enum ib_rnr_timeout {
1180 	IB_RNR_TIMER_655_36 =  0,
1181 	IB_RNR_TIMER_000_01 =  1,
1182 	IB_RNR_TIMER_000_02 =  2,
1183 	IB_RNR_TIMER_000_03 =  3,
1184 	IB_RNR_TIMER_000_04 =  4,
1185 	IB_RNR_TIMER_000_06 =  5,
1186 	IB_RNR_TIMER_000_08 =  6,
1187 	IB_RNR_TIMER_000_12 =  7,
1188 	IB_RNR_TIMER_000_16 =  8,
1189 	IB_RNR_TIMER_000_24 =  9,
1190 	IB_RNR_TIMER_000_32 = 10,
1191 	IB_RNR_TIMER_000_48 = 11,
1192 	IB_RNR_TIMER_000_64 = 12,
1193 	IB_RNR_TIMER_000_96 = 13,
1194 	IB_RNR_TIMER_001_28 = 14,
1195 	IB_RNR_TIMER_001_92 = 15,
1196 	IB_RNR_TIMER_002_56 = 16,
1197 	IB_RNR_TIMER_003_84 = 17,
1198 	IB_RNR_TIMER_005_12 = 18,
1199 	IB_RNR_TIMER_007_68 = 19,
1200 	IB_RNR_TIMER_010_24 = 20,
1201 	IB_RNR_TIMER_015_36 = 21,
1202 	IB_RNR_TIMER_020_48 = 22,
1203 	IB_RNR_TIMER_030_72 = 23,
1204 	IB_RNR_TIMER_040_96 = 24,
1205 	IB_RNR_TIMER_061_44 = 25,
1206 	IB_RNR_TIMER_081_92 = 26,
1207 	IB_RNR_TIMER_122_88 = 27,
1208 	IB_RNR_TIMER_163_84 = 28,
1209 	IB_RNR_TIMER_245_76 = 29,
1210 	IB_RNR_TIMER_327_68 = 30,
1211 	IB_RNR_TIMER_491_52 = 31
1212 };
1213 
1214 enum ib_qp_attr_mask {
1215 	IB_QP_STATE			= 1,
1216 	IB_QP_CUR_STATE			= (1<<1),
1217 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1218 	IB_QP_ACCESS_FLAGS		= (1<<3),
1219 	IB_QP_PKEY_INDEX		= (1<<4),
1220 	IB_QP_PORT			= (1<<5),
1221 	IB_QP_QKEY			= (1<<6),
1222 	IB_QP_AV			= (1<<7),
1223 	IB_QP_PATH_MTU			= (1<<8),
1224 	IB_QP_TIMEOUT			= (1<<9),
1225 	IB_QP_RETRY_CNT			= (1<<10),
1226 	IB_QP_RNR_RETRY			= (1<<11),
1227 	IB_QP_RQ_PSN			= (1<<12),
1228 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1229 	IB_QP_ALT_PATH			= (1<<14),
1230 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1231 	IB_QP_SQ_PSN			= (1<<16),
1232 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1233 	IB_QP_PATH_MIG_STATE		= (1<<18),
1234 	IB_QP_CAP			= (1<<19),
1235 	IB_QP_DEST_QPN			= (1<<20),
1236 	IB_QP_RESERVED1			= (1<<21),
1237 	IB_QP_RESERVED2			= (1<<22),
1238 	IB_QP_RESERVED3			= (1<<23),
1239 	IB_QP_RESERVED4			= (1<<24),
1240 	IB_QP_RATE_LIMIT		= (1<<25),
1241 
1242 	IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1243 };
1244 
1245 enum ib_qp_state {
1246 	IB_QPS_RESET,
1247 	IB_QPS_INIT,
1248 	IB_QPS_RTR,
1249 	IB_QPS_RTS,
1250 	IB_QPS_SQD,
1251 	IB_QPS_SQE,
1252 	IB_QPS_ERR
1253 };
1254 
1255 enum ib_mig_state {
1256 	IB_MIG_MIGRATED,
1257 	IB_MIG_REARM,
1258 	IB_MIG_ARMED
1259 };
1260 
1261 enum ib_mw_type {
1262 	IB_MW_TYPE_1 = 1,
1263 	IB_MW_TYPE_2 = 2
1264 };
1265 
1266 struct ib_qp_attr {
1267 	enum ib_qp_state	qp_state;
1268 	enum ib_qp_state	cur_qp_state;
1269 	enum ib_mtu		path_mtu;
1270 	enum ib_mig_state	path_mig_state;
1271 	u32			qkey;
1272 	u32			rq_psn;
1273 	u32			sq_psn;
1274 	u32			dest_qp_num;
1275 	int			qp_access_flags;
1276 	struct ib_qp_cap	cap;
1277 	struct rdma_ah_attr	ah_attr;
1278 	struct rdma_ah_attr	alt_ah_attr;
1279 	u16			pkey_index;
1280 	u16			alt_pkey_index;
1281 	u8			en_sqd_async_notify;
1282 	u8			sq_draining;
1283 	u8			max_rd_atomic;
1284 	u8			max_dest_rd_atomic;
1285 	u8			min_rnr_timer;
1286 	u32			port_num;
1287 	u8			timeout;
1288 	u8			retry_cnt;
1289 	u8			rnr_retry;
1290 	u32			alt_port_num;
1291 	u8			alt_timeout;
1292 	u32			rate_limit;
1293 	struct net_device	*xmit_slave;
1294 };
1295 
1296 enum ib_wr_opcode {
1297 	/* These are shared with userspace */
1298 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1299 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1300 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1301 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1302 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1303 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1304 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1305 	IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1306 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1307 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1308 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1309 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1310 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1311 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1312 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1313 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1314 
1315 	/* These are kernel only and can not be issued by userspace */
1316 	IB_WR_REG_MR = 0x20,
1317 	IB_WR_REG_MR_INTEGRITY,
1318 
1319 	/* reserve values for low level drivers' internal use.
1320 	 * These values will not be used at all in the ib core layer.
1321 	 */
1322 	IB_WR_RESERVED1 = 0xf0,
1323 	IB_WR_RESERVED2,
1324 	IB_WR_RESERVED3,
1325 	IB_WR_RESERVED4,
1326 	IB_WR_RESERVED5,
1327 	IB_WR_RESERVED6,
1328 	IB_WR_RESERVED7,
1329 	IB_WR_RESERVED8,
1330 	IB_WR_RESERVED9,
1331 	IB_WR_RESERVED10,
1332 };
1333 
1334 enum ib_send_flags {
1335 	IB_SEND_FENCE		= 1,
1336 	IB_SEND_SIGNALED	= (1<<1),
1337 	IB_SEND_SOLICITED	= (1<<2),
1338 	IB_SEND_INLINE		= (1<<3),
1339 	IB_SEND_IP_CSUM		= (1<<4),
1340 
1341 	/* reserve bits 26-31 for low level drivers' internal use */
1342 	IB_SEND_RESERVED_START	= (1 << 26),
1343 	IB_SEND_RESERVED_END	= (1 << 31),
1344 };
1345 
1346 struct ib_sge {
1347 	u64	addr;
1348 	u32	length;
1349 	u32	lkey;
1350 };
1351 
1352 struct ib_cqe {
1353 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1354 };
1355 
1356 struct ib_send_wr {
1357 	struct ib_send_wr      *next;
1358 	union {
1359 		u64		wr_id;
1360 		struct ib_cqe	*wr_cqe;
1361 	};
1362 	struct ib_sge	       *sg_list;
1363 	int			num_sge;
1364 	enum ib_wr_opcode	opcode;
1365 	int			send_flags;
1366 	union {
1367 		__be32		imm_data;
1368 		u32		invalidate_rkey;
1369 	} ex;
1370 };
1371 
1372 struct ib_rdma_wr {
1373 	struct ib_send_wr	wr;
1374 	u64			remote_addr;
1375 	u32			rkey;
1376 };
1377 
1378 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1379 {
1380 	return container_of(wr, struct ib_rdma_wr, wr);
1381 }
1382 
1383 struct ib_atomic_wr {
1384 	struct ib_send_wr	wr;
1385 	u64			remote_addr;
1386 	u64			compare_add;
1387 	u64			swap;
1388 	u64			compare_add_mask;
1389 	u64			swap_mask;
1390 	u32			rkey;
1391 };
1392 
1393 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1394 {
1395 	return container_of(wr, struct ib_atomic_wr, wr);
1396 }
1397 
1398 struct ib_ud_wr {
1399 	struct ib_send_wr	wr;
1400 	struct ib_ah		*ah;
1401 	void			*header;
1402 	int			hlen;
1403 	int			mss;
1404 	u32			remote_qpn;
1405 	u32			remote_qkey;
1406 	u16			pkey_index; /* valid for GSI only */
1407 	u32			port_num; /* valid for DR SMPs on switch only */
1408 };
1409 
1410 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1411 {
1412 	return container_of(wr, struct ib_ud_wr, wr);
1413 }
1414 
1415 struct ib_reg_wr {
1416 	struct ib_send_wr	wr;
1417 	struct ib_mr		*mr;
1418 	u32			key;
1419 	int			access;
1420 };
1421 
1422 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1423 {
1424 	return container_of(wr, struct ib_reg_wr, wr);
1425 }
1426 
1427 struct ib_recv_wr {
1428 	struct ib_recv_wr      *next;
1429 	union {
1430 		u64		wr_id;
1431 		struct ib_cqe	*wr_cqe;
1432 	};
1433 	struct ib_sge	       *sg_list;
1434 	int			num_sge;
1435 };
1436 
1437 enum ib_access_flags {
1438 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1439 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1440 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1441 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1442 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1443 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1444 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1445 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1446 	IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1447 
1448 	IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1449 	IB_ACCESS_SUPPORTED =
1450 		((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1451 };
1452 
1453 /*
1454  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1455  * are hidden here instead of a uapi header!
1456  */
1457 enum ib_mr_rereg_flags {
1458 	IB_MR_REREG_TRANS	= 1,
1459 	IB_MR_REREG_PD		= (1<<1),
1460 	IB_MR_REREG_ACCESS	= (1<<2),
1461 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1462 };
1463 
1464 struct ib_umem;
1465 
1466 enum rdma_remove_reason {
1467 	/*
1468 	 * Userspace requested uobject deletion or initial try
1469 	 * to remove uobject via cleanup. Call could fail
1470 	 */
1471 	RDMA_REMOVE_DESTROY,
1472 	/* Context deletion. This call should delete the actual object itself */
1473 	RDMA_REMOVE_CLOSE,
1474 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1475 	RDMA_REMOVE_DRIVER_REMOVE,
1476 	/* uobj is being cleaned-up before being committed */
1477 	RDMA_REMOVE_ABORT,
1478 	/* The driver failed to destroy the uobject and is being disconnected */
1479 	RDMA_REMOVE_DRIVER_FAILURE,
1480 };
1481 
1482 struct ib_rdmacg_object {
1483 #ifdef CONFIG_CGROUP_RDMA
1484 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1485 #endif
1486 };
1487 
1488 struct ib_ucontext {
1489 	struct ib_device       *device;
1490 	struct ib_uverbs_file  *ufile;
1491 
1492 	struct ib_rdmacg_object	cg_obj;
1493 	/*
1494 	 * Implementation details of the RDMA core, don't use in drivers:
1495 	 */
1496 	struct rdma_restrack_entry res;
1497 	struct xarray mmap_xa;
1498 };
1499 
1500 struct ib_uobject {
1501 	u64			user_handle;	/* handle given to us by userspace */
1502 	/* ufile & ucontext owning this object */
1503 	struct ib_uverbs_file  *ufile;
1504 	/* FIXME, save memory: ufile->context == context */
1505 	struct ib_ucontext     *context;	/* associated user context */
1506 	void		       *object;		/* containing object */
1507 	struct list_head	list;		/* link to context's list */
1508 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1509 	int			id;		/* index into kernel idr */
1510 	struct kref		ref;
1511 	atomic_t		usecnt;		/* protects exclusive access */
1512 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1513 
1514 	const struct uverbs_api_object *uapi_object;
1515 };
1516 
1517 struct ib_udata {
1518 	const void __user *inbuf;
1519 	void __user *outbuf;
1520 	size_t       inlen;
1521 	size_t       outlen;
1522 };
1523 
1524 struct ib_pd {
1525 	u32			local_dma_lkey;
1526 	u32			flags;
1527 	struct ib_device       *device;
1528 	struct ib_uobject      *uobject;
1529 	atomic_t          	usecnt; /* count all resources */
1530 
1531 	u32			unsafe_global_rkey;
1532 
1533 	/*
1534 	 * Implementation details of the RDMA core, don't use in drivers:
1535 	 */
1536 	struct ib_mr	       *__internal_mr;
1537 	struct rdma_restrack_entry res;
1538 };
1539 
1540 struct ib_xrcd {
1541 	struct ib_device       *device;
1542 	atomic_t		usecnt; /* count all exposed resources */
1543 	struct inode	       *inode;
1544 	struct rw_semaphore	tgt_qps_rwsem;
1545 	struct xarray		tgt_qps;
1546 };
1547 
1548 struct ib_ah {
1549 	struct ib_device	*device;
1550 	struct ib_pd		*pd;
1551 	struct ib_uobject	*uobject;
1552 	const struct ib_gid_attr *sgid_attr;
1553 	enum rdma_ah_attr_type	type;
1554 };
1555 
1556 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1557 
1558 enum ib_poll_context {
1559 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1560 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1561 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1562 	IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1563 
1564 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1565 };
1566 
1567 struct ib_cq {
1568 	struct ib_device       *device;
1569 	struct ib_ucq_object   *uobject;
1570 	ib_comp_handler   	comp_handler;
1571 	void                  (*event_handler)(struct ib_event *, void *);
1572 	void                   *cq_context;
1573 	int               	cqe;
1574 	unsigned int		cqe_used;
1575 	atomic_t          	usecnt; /* count number of work queues */
1576 	enum ib_poll_context	poll_ctx;
1577 	struct ib_wc		*wc;
1578 	struct list_head        pool_entry;
1579 	union {
1580 		struct irq_poll		iop;
1581 		struct work_struct	work;
1582 	};
1583 	struct workqueue_struct *comp_wq;
1584 	struct dim *dim;
1585 
1586 	/* updated only by trace points */
1587 	ktime_t timestamp;
1588 	u8 interrupt:1;
1589 	u8 shared:1;
1590 	unsigned int comp_vector;
1591 
1592 	/*
1593 	 * Implementation details of the RDMA core, don't use in drivers:
1594 	 */
1595 	struct rdma_restrack_entry res;
1596 };
1597 
1598 struct ib_srq {
1599 	struct ib_device       *device;
1600 	struct ib_pd	       *pd;
1601 	struct ib_usrq_object  *uobject;
1602 	void		      (*event_handler)(struct ib_event *, void *);
1603 	void		       *srq_context;
1604 	enum ib_srq_type	srq_type;
1605 	atomic_t		usecnt;
1606 
1607 	struct {
1608 		struct ib_cq   *cq;
1609 		union {
1610 			struct {
1611 				struct ib_xrcd *xrcd;
1612 				u32		srq_num;
1613 			} xrc;
1614 		};
1615 	} ext;
1616 
1617 	/*
1618 	 * Implementation details of the RDMA core, don't use in drivers:
1619 	 */
1620 	struct rdma_restrack_entry res;
1621 };
1622 
1623 enum ib_raw_packet_caps {
1624 	/* Strip cvlan from incoming packet and report it in the matching work
1625 	 * completion is supported.
1626 	 */
1627 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1628 	/* Scatter FCS field of an incoming packet to host memory is supported.
1629 	 */
1630 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1631 	/* Checksum offloads are supported (for both send and receive). */
1632 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1633 	/* When a packet is received for an RQ with no receive WQEs, the
1634 	 * packet processing is delayed.
1635 	 */
1636 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1637 };
1638 
1639 enum ib_wq_type {
1640 	IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1641 };
1642 
1643 enum ib_wq_state {
1644 	IB_WQS_RESET,
1645 	IB_WQS_RDY,
1646 	IB_WQS_ERR
1647 };
1648 
1649 struct ib_wq {
1650 	struct ib_device       *device;
1651 	struct ib_uwq_object   *uobject;
1652 	void		    *wq_context;
1653 	void		    (*event_handler)(struct ib_event *, void *);
1654 	struct ib_pd	       *pd;
1655 	struct ib_cq	       *cq;
1656 	u32		wq_num;
1657 	enum ib_wq_state       state;
1658 	enum ib_wq_type	wq_type;
1659 	atomic_t		usecnt;
1660 };
1661 
1662 enum ib_wq_flags {
1663 	IB_WQ_FLAGS_CVLAN_STRIPPING	= IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1664 	IB_WQ_FLAGS_SCATTER_FCS		= IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1665 	IB_WQ_FLAGS_DELAY_DROP		= IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1666 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1667 				IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1668 };
1669 
1670 struct ib_wq_init_attr {
1671 	void		       *wq_context;
1672 	enum ib_wq_type	wq_type;
1673 	u32		max_wr;
1674 	u32		max_sge;
1675 	struct	ib_cq	       *cq;
1676 	void		    (*event_handler)(struct ib_event *, void *);
1677 	u32		create_flags; /* Use enum ib_wq_flags */
1678 };
1679 
1680 enum ib_wq_attr_mask {
1681 	IB_WQ_STATE		= 1 << 0,
1682 	IB_WQ_CUR_STATE		= 1 << 1,
1683 	IB_WQ_FLAGS		= 1 << 2,
1684 };
1685 
1686 struct ib_wq_attr {
1687 	enum	ib_wq_state	wq_state;
1688 	enum	ib_wq_state	curr_wq_state;
1689 	u32			flags; /* Use enum ib_wq_flags */
1690 	u32			flags_mask; /* Use enum ib_wq_flags */
1691 };
1692 
1693 struct ib_rwq_ind_table {
1694 	struct ib_device	*device;
1695 	struct ib_uobject      *uobject;
1696 	atomic_t		usecnt;
1697 	u32		ind_tbl_num;
1698 	u32		log_ind_tbl_size;
1699 	struct ib_wq	**ind_tbl;
1700 };
1701 
1702 struct ib_rwq_ind_table_init_attr {
1703 	u32		log_ind_tbl_size;
1704 	/* Each entry is a pointer to Receive Work Queue */
1705 	struct ib_wq	**ind_tbl;
1706 };
1707 
1708 enum port_pkey_state {
1709 	IB_PORT_PKEY_NOT_VALID = 0,
1710 	IB_PORT_PKEY_VALID = 1,
1711 	IB_PORT_PKEY_LISTED = 2,
1712 };
1713 
1714 struct ib_qp_security;
1715 
1716 struct ib_port_pkey {
1717 	enum port_pkey_state	state;
1718 	u16			pkey_index;
1719 	u32			port_num;
1720 	struct list_head	qp_list;
1721 	struct list_head	to_error_list;
1722 	struct ib_qp_security  *sec;
1723 };
1724 
1725 struct ib_ports_pkeys {
1726 	struct ib_port_pkey	main;
1727 	struct ib_port_pkey	alt;
1728 };
1729 
1730 struct ib_qp_security {
1731 	struct ib_qp	       *qp;
1732 	struct ib_device       *dev;
1733 	/* Hold this mutex when changing port and pkey settings. */
1734 	struct mutex		mutex;
1735 	struct ib_ports_pkeys  *ports_pkeys;
1736 	/* A list of all open shared QP handles.  Required to enforce security
1737 	 * properly for all users of a shared QP.
1738 	 */
1739 	struct list_head        shared_qp_list;
1740 	void                   *security;
1741 	bool			destroying;
1742 	atomic_t		error_list_count;
1743 	struct completion	error_complete;
1744 	int			error_comps_pending;
1745 };
1746 
1747 /*
1748  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1749  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1750  */
1751 struct ib_qp {
1752 	struct ib_device       *device;
1753 	struct ib_pd	       *pd;
1754 	struct ib_cq	       *send_cq;
1755 	struct ib_cq	       *recv_cq;
1756 	spinlock_t		mr_lock;
1757 	int			mrs_used;
1758 	struct list_head	rdma_mrs;
1759 	struct list_head	sig_mrs;
1760 	struct ib_srq	       *srq;
1761 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1762 	struct list_head	xrcd_list;
1763 
1764 	/* count times opened, mcast attaches, flow attaches */
1765 	atomic_t		usecnt;
1766 	struct list_head	open_list;
1767 	struct ib_qp           *real_qp;
1768 	struct ib_uqp_object   *uobject;
1769 	void                  (*event_handler)(struct ib_event *, void *);
1770 	void		       *qp_context;
1771 	/* sgid_attrs associated with the AV's */
1772 	const struct ib_gid_attr *av_sgid_attr;
1773 	const struct ib_gid_attr *alt_path_sgid_attr;
1774 	u32			qp_num;
1775 	u32			max_write_sge;
1776 	u32			max_read_sge;
1777 	enum ib_qp_type		qp_type;
1778 	struct ib_rwq_ind_table *rwq_ind_tbl;
1779 	struct ib_qp_security  *qp_sec;
1780 	u32			port;
1781 
1782 	bool			integrity_en;
1783 	/*
1784 	 * Implementation details of the RDMA core, don't use in drivers:
1785 	 */
1786 	struct rdma_restrack_entry     res;
1787 
1788 	/* The counter the qp is bind to */
1789 	struct rdma_counter    *counter;
1790 };
1791 
1792 struct ib_dm {
1793 	struct ib_device  *device;
1794 	u32		   length;
1795 	u32		   flags;
1796 	struct ib_uobject *uobject;
1797 	atomic_t	   usecnt;
1798 };
1799 
1800 struct ib_mr {
1801 	struct ib_device  *device;
1802 	struct ib_pd	  *pd;
1803 	u32		   lkey;
1804 	u32		   rkey;
1805 	u64		   iova;
1806 	u64		   length;
1807 	unsigned int	   page_size;
1808 	enum ib_mr_type	   type;
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 	struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
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 /* Supported steering options */
1832 enum ib_flow_attr_type {
1833 	/* steering according to rule specifications */
1834 	IB_FLOW_ATTR_NORMAL		= 0x0,
1835 	/* default unicast and multicast rule -
1836 	 * receive all Eth traffic which isn't steered to any QP
1837 	 */
1838 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1839 	/* default multicast rule -
1840 	 * receive all Eth multicast traffic which isn't steered to any QP
1841 	 */
1842 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1843 	/* sniffer rule - receive all port traffic */
1844 	IB_FLOW_ATTR_SNIFFER		= 0x3
1845 };
1846 
1847 /* Supported steering header types */
1848 enum ib_flow_spec_type {
1849 	/* L2 headers*/
1850 	IB_FLOW_SPEC_ETH		= 0x20,
1851 	IB_FLOW_SPEC_IB			= 0x22,
1852 	/* L3 header*/
1853 	IB_FLOW_SPEC_IPV4		= 0x30,
1854 	IB_FLOW_SPEC_IPV6		= 0x31,
1855 	IB_FLOW_SPEC_ESP                = 0x34,
1856 	/* L4 headers*/
1857 	IB_FLOW_SPEC_TCP		= 0x40,
1858 	IB_FLOW_SPEC_UDP		= 0x41,
1859 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1860 	IB_FLOW_SPEC_GRE		= 0x51,
1861 	IB_FLOW_SPEC_MPLS		= 0x60,
1862 	IB_FLOW_SPEC_INNER		= 0x100,
1863 	/* Actions */
1864 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1865 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1866 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1867 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1868 };
1869 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1870 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1871 
1872 enum ib_flow_flags {
1873 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1874 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1875 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1876 };
1877 
1878 struct ib_flow_eth_filter {
1879 	u8	dst_mac[6];
1880 	u8	src_mac[6];
1881 	__be16	ether_type;
1882 	__be16	vlan_tag;
1883 	/* Must be last */
1884 	u8	real_sz[];
1885 };
1886 
1887 struct ib_flow_spec_eth {
1888 	u32			  type;
1889 	u16			  size;
1890 	struct ib_flow_eth_filter val;
1891 	struct ib_flow_eth_filter mask;
1892 };
1893 
1894 struct ib_flow_ib_filter {
1895 	__be16 dlid;
1896 	__u8   sl;
1897 	/* Must be last */
1898 	u8	real_sz[];
1899 };
1900 
1901 struct ib_flow_spec_ib {
1902 	u32			 type;
1903 	u16			 size;
1904 	struct ib_flow_ib_filter val;
1905 	struct ib_flow_ib_filter mask;
1906 };
1907 
1908 /* IPv4 header flags */
1909 enum ib_ipv4_flags {
1910 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1911 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1912 				    last have this flag set */
1913 };
1914 
1915 struct ib_flow_ipv4_filter {
1916 	__be32	src_ip;
1917 	__be32	dst_ip;
1918 	u8	proto;
1919 	u8	tos;
1920 	u8	ttl;
1921 	u8	flags;
1922 	/* Must be last */
1923 	u8	real_sz[];
1924 };
1925 
1926 struct ib_flow_spec_ipv4 {
1927 	u32			   type;
1928 	u16			   size;
1929 	struct ib_flow_ipv4_filter val;
1930 	struct ib_flow_ipv4_filter mask;
1931 };
1932 
1933 struct ib_flow_ipv6_filter {
1934 	u8	src_ip[16];
1935 	u8	dst_ip[16];
1936 	__be32	flow_label;
1937 	u8	next_hdr;
1938 	u8	traffic_class;
1939 	u8	hop_limit;
1940 	/* Must be last */
1941 	u8	real_sz[];
1942 };
1943 
1944 struct ib_flow_spec_ipv6 {
1945 	u32			   type;
1946 	u16			   size;
1947 	struct ib_flow_ipv6_filter val;
1948 	struct ib_flow_ipv6_filter mask;
1949 };
1950 
1951 struct ib_flow_tcp_udp_filter {
1952 	__be16	dst_port;
1953 	__be16	src_port;
1954 	/* Must be last */
1955 	u8	real_sz[];
1956 };
1957 
1958 struct ib_flow_spec_tcp_udp {
1959 	u32			      type;
1960 	u16			      size;
1961 	struct ib_flow_tcp_udp_filter val;
1962 	struct ib_flow_tcp_udp_filter mask;
1963 };
1964 
1965 struct ib_flow_tunnel_filter {
1966 	__be32	tunnel_id;
1967 	u8	real_sz[];
1968 };
1969 
1970 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1971  * the tunnel_id from val has the vni value
1972  */
1973 struct ib_flow_spec_tunnel {
1974 	u32			      type;
1975 	u16			      size;
1976 	struct ib_flow_tunnel_filter  val;
1977 	struct ib_flow_tunnel_filter  mask;
1978 };
1979 
1980 struct ib_flow_esp_filter {
1981 	__be32	spi;
1982 	__be32  seq;
1983 	/* Must be last */
1984 	u8	real_sz[];
1985 };
1986 
1987 struct ib_flow_spec_esp {
1988 	u32                           type;
1989 	u16			      size;
1990 	struct ib_flow_esp_filter     val;
1991 	struct ib_flow_esp_filter     mask;
1992 };
1993 
1994 struct ib_flow_gre_filter {
1995 	__be16 c_ks_res0_ver;
1996 	__be16 protocol;
1997 	__be32 key;
1998 	/* Must be last */
1999 	u8	real_sz[];
2000 };
2001 
2002 struct ib_flow_spec_gre {
2003 	u32                           type;
2004 	u16			      size;
2005 	struct ib_flow_gre_filter     val;
2006 	struct ib_flow_gre_filter     mask;
2007 };
2008 
2009 struct ib_flow_mpls_filter {
2010 	__be32 tag;
2011 	/* Must be last */
2012 	u8	real_sz[];
2013 };
2014 
2015 struct ib_flow_spec_mpls {
2016 	u32                           type;
2017 	u16			      size;
2018 	struct ib_flow_mpls_filter     val;
2019 	struct ib_flow_mpls_filter     mask;
2020 };
2021 
2022 struct ib_flow_spec_action_tag {
2023 	enum ib_flow_spec_type	      type;
2024 	u16			      size;
2025 	u32                           tag_id;
2026 };
2027 
2028 struct ib_flow_spec_action_drop {
2029 	enum ib_flow_spec_type	      type;
2030 	u16			      size;
2031 };
2032 
2033 struct ib_flow_spec_action_handle {
2034 	enum ib_flow_spec_type	      type;
2035 	u16			      size;
2036 	struct ib_flow_action	     *act;
2037 };
2038 
2039 enum ib_counters_description {
2040 	IB_COUNTER_PACKETS,
2041 	IB_COUNTER_BYTES,
2042 };
2043 
2044 struct ib_flow_spec_action_count {
2045 	enum ib_flow_spec_type type;
2046 	u16 size;
2047 	struct ib_counters *counters;
2048 };
2049 
2050 union ib_flow_spec {
2051 	struct {
2052 		u32			type;
2053 		u16			size;
2054 	};
2055 	struct ib_flow_spec_eth		eth;
2056 	struct ib_flow_spec_ib		ib;
2057 	struct ib_flow_spec_ipv4        ipv4;
2058 	struct ib_flow_spec_tcp_udp	tcp_udp;
2059 	struct ib_flow_spec_ipv6        ipv6;
2060 	struct ib_flow_spec_tunnel      tunnel;
2061 	struct ib_flow_spec_esp		esp;
2062 	struct ib_flow_spec_gre		gre;
2063 	struct ib_flow_spec_mpls	mpls;
2064 	struct ib_flow_spec_action_tag  flow_tag;
2065 	struct ib_flow_spec_action_drop drop;
2066 	struct ib_flow_spec_action_handle action;
2067 	struct ib_flow_spec_action_count flow_count;
2068 };
2069 
2070 struct ib_flow_attr {
2071 	enum ib_flow_attr_type type;
2072 	u16	     size;
2073 	u16	     priority;
2074 	u32	     flags;
2075 	u8	     num_of_specs;
2076 	u32	     port;
2077 	union ib_flow_spec flows[];
2078 };
2079 
2080 struct ib_flow {
2081 	struct ib_qp		*qp;
2082 	struct ib_device	*device;
2083 	struct ib_uobject	*uobject;
2084 };
2085 
2086 enum ib_flow_action_type {
2087 	IB_FLOW_ACTION_UNSPECIFIED,
2088 	IB_FLOW_ACTION_ESP = 1,
2089 };
2090 
2091 struct ib_flow_action_attrs_esp_keymats {
2092 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2093 	union {
2094 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2095 	} keymat;
2096 };
2097 
2098 struct ib_flow_action_attrs_esp_replays {
2099 	enum ib_uverbs_flow_action_esp_replay			protocol;
2100 	union {
2101 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2102 	} replay;
2103 };
2104 
2105 enum ib_flow_action_attrs_esp_flags {
2106 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2107 	 * This is done in order to share the same flags between user-space and
2108 	 * kernel and spare an unnecessary translation.
2109 	 */
2110 
2111 	/* Kernel flags */
2112 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2113 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2114 };
2115 
2116 struct ib_flow_spec_list {
2117 	struct ib_flow_spec_list	*next;
2118 	union ib_flow_spec		spec;
2119 };
2120 
2121 struct ib_flow_action_attrs_esp {
2122 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2123 	struct ib_flow_action_attrs_esp_replays		*replay;
2124 	struct ib_flow_spec_list			*encap;
2125 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2126 	 * Value of 0 is a valid value.
2127 	 */
2128 	u32						esn;
2129 	u32						spi;
2130 	u32						seq;
2131 	u32						tfc_pad;
2132 	/* Use enum ib_flow_action_attrs_esp_flags */
2133 	u64						flags;
2134 	u64						hard_limit_pkts;
2135 };
2136 
2137 struct ib_flow_action {
2138 	struct ib_device		*device;
2139 	struct ib_uobject		*uobject;
2140 	enum ib_flow_action_type	type;
2141 	atomic_t			usecnt;
2142 };
2143 
2144 struct ib_mad;
2145 
2146 enum ib_process_mad_flags {
2147 	IB_MAD_IGNORE_MKEY	= 1,
2148 	IB_MAD_IGNORE_BKEY	= 2,
2149 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2150 };
2151 
2152 enum ib_mad_result {
2153 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2154 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2155 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2156 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2157 };
2158 
2159 struct ib_port_cache {
2160 	u64		      subnet_prefix;
2161 	struct ib_pkey_cache  *pkey;
2162 	struct ib_gid_table   *gid;
2163 	u8                     lmc;
2164 	enum ib_port_state     port_state;
2165 };
2166 
2167 struct ib_port_immutable {
2168 	int                           pkey_tbl_len;
2169 	int                           gid_tbl_len;
2170 	u32                           core_cap_flags;
2171 	u32                           max_mad_size;
2172 };
2173 
2174 struct ib_port_data {
2175 	struct ib_device *ib_dev;
2176 
2177 	struct ib_port_immutable immutable;
2178 
2179 	spinlock_t pkey_list_lock;
2180 
2181 	spinlock_t netdev_lock;
2182 
2183 	struct list_head pkey_list;
2184 
2185 	struct ib_port_cache cache;
2186 
2187 	struct net_device __rcu *netdev;
2188 	struct hlist_node ndev_hash_link;
2189 	struct rdma_port_counter port_counter;
2190 	struct ib_port *sysfs;
2191 };
2192 
2193 /* rdma netdev type - specifies protocol type */
2194 enum rdma_netdev_t {
2195 	RDMA_NETDEV_OPA_VNIC,
2196 	RDMA_NETDEV_IPOIB,
2197 };
2198 
2199 /**
2200  * struct rdma_netdev - rdma netdev
2201  * For cases where netstack interfacing is required.
2202  */
2203 struct rdma_netdev {
2204 	void              *clnt_priv;
2205 	struct ib_device  *hca;
2206 	u32		   port_num;
2207 	int                mtu;
2208 
2209 	/*
2210 	 * cleanup function must be specified.
2211 	 * FIXME: This is only used for OPA_VNIC and that usage should be
2212 	 * removed too.
2213 	 */
2214 	void (*free_rdma_netdev)(struct net_device *netdev);
2215 
2216 	/* control functions */
2217 	void (*set_id)(struct net_device *netdev, int id);
2218 	/* send packet */
2219 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2220 		    struct ib_ah *address, u32 dqpn);
2221 	/* multicast */
2222 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2223 			    union ib_gid *gid, u16 mlid,
2224 			    int set_qkey, u32 qkey);
2225 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2226 			    union ib_gid *gid, u16 mlid);
2227 	/* timeout */
2228 	void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2229 };
2230 
2231 struct rdma_netdev_alloc_params {
2232 	size_t sizeof_priv;
2233 	unsigned int txqs;
2234 	unsigned int rxqs;
2235 	void *param;
2236 
2237 	int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2238 				      struct net_device *netdev, void *param);
2239 };
2240 
2241 struct ib_odp_counters {
2242 	atomic64_t faults;
2243 	atomic64_t invalidations;
2244 	atomic64_t prefetch;
2245 };
2246 
2247 struct ib_counters {
2248 	struct ib_device	*device;
2249 	struct ib_uobject	*uobject;
2250 	/* num of objects attached */
2251 	atomic_t	usecnt;
2252 };
2253 
2254 struct ib_counters_read_attr {
2255 	u64	*counters_buff;
2256 	u32	ncounters;
2257 	u32	flags; /* use enum ib_read_counters_flags */
2258 };
2259 
2260 struct uverbs_attr_bundle;
2261 struct iw_cm_id;
2262 struct iw_cm_conn_param;
2263 
2264 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member)                      \
2265 	.size_##ib_struct =                                                    \
2266 		(sizeof(struct drv_struct) +                                   \
2267 		 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) +      \
2268 		 BUILD_BUG_ON_ZERO(                                            \
2269 			 !__same_type(((struct drv_struct *)NULL)->member,     \
2270 				      struct ib_struct)))
2271 
2272 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp)                          \
2273 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2274 					   gfp, false))
2275 
2276 #define rdma_zalloc_drv_obj_numa(ib_dev, ib_type)                              \
2277 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2278 					   GFP_KERNEL, true))
2279 
2280 #define rdma_zalloc_drv_obj(ib_dev, ib_type)                                   \
2281 	rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2282 
2283 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2284 
2285 struct rdma_user_mmap_entry {
2286 	struct kref ref;
2287 	struct ib_ucontext *ucontext;
2288 	unsigned long start_pgoff;
2289 	size_t npages;
2290 	bool driver_removed;
2291 };
2292 
2293 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2294 static inline u64
2295 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2296 {
2297 	return (u64)entry->start_pgoff << PAGE_SHIFT;
2298 }
2299 
2300 /**
2301  * struct ib_device_ops - InfiniBand device operations
2302  * This structure defines all the InfiniBand device operations, providers will
2303  * need to define the supported operations, otherwise they will be set to null.
2304  */
2305 struct ib_device_ops {
2306 	struct module *owner;
2307 	enum rdma_driver_id driver_id;
2308 	u32 uverbs_abi_ver;
2309 	unsigned int uverbs_no_driver_id_binding:1;
2310 
2311 	/*
2312 	 * NOTE: New drivers should not make use of device_group; instead new
2313 	 * device parameter should be exposed via netlink command. This
2314 	 * mechanism exists only for existing drivers.
2315 	 */
2316 	const struct attribute_group *device_group;
2317 	const struct attribute_group **port_groups;
2318 
2319 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2320 			 const struct ib_send_wr **bad_send_wr);
2321 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2322 			 const struct ib_recv_wr **bad_recv_wr);
2323 	void (*drain_rq)(struct ib_qp *qp);
2324 	void (*drain_sq)(struct ib_qp *qp);
2325 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2326 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2327 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2328 	int (*post_srq_recv)(struct ib_srq *srq,
2329 			     const struct ib_recv_wr *recv_wr,
2330 			     const struct ib_recv_wr **bad_recv_wr);
2331 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2332 			   u32 port_num, const struct ib_wc *in_wc,
2333 			   const struct ib_grh *in_grh,
2334 			   const struct ib_mad *in_mad, struct ib_mad *out_mad,
2335 			   size_t *out_mad_size, u16 *out_mad_pkey_index);
2336 	int (*query_device)(struct ib_device *device,
2337 			    struct ib_device_attr *device_attr,
2338 			    struct ib_udata *udata);
2339 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2340 			     struct ib_device_modify *device_modify);
2341 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2342 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2343 						     int comp_vector);
2344 	int (*query_port)(struct ib_device *device, u32 port_num,
2345 			  struct ib_port_attr *port_attr);
2346 	int (*modify_port)(struct ib_device *device, u32 port_num,
2347 			   int port_modify_mask,
2348 			   struct ib_port_modify *port_modify);
2349 	/**
2350 	 * The following mandatory functions are used only at device
2351 	 * registration.  Keep functions such as these at the end of this
2352 	 * structure to avoid cache line misses when accessing struct ib_device
2353 	 * in fast paths.
2354 	 */
2355 	int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2356 				  struct ib_port_immutable *immutable);
2357 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2358 					       u32 port_num);
2359 	/**
2360 	 * When calling get_netdev, the HW vendor's driver should return the
2361 	 * net device of device @device at port @port_num or NULL if such
2362 	 * a net device doesn't exist. The vendor driver should call dev_hold
2363 	 * on this net device. The HW vendor's device driver must guarantee
2364 	 * that this function returns NULL before the net device has finished
2365 	 * NETDEV_UNREGISTER state.
2366 	 */
2367 	struct net_device *(*get_netdev)(struct ib_device *device,
2368 					 u32 port_num);
2369 	/**
2370 	 * rdma netdev operation
2371 	 *
2372 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2373 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2374 	 */
2375 	struct net_device *(*alloc_rdma_netdev)(
2376 		struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2377 		const char *name, unsigned char name_assign_type,
2378 		void (*setup)(struct net_device *));
2379 
2380 	int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2381 				      enum rdma_netdev_t type,
2382 				      struct rdma_netdev_alloc_params *params);
2383 	/**
2384 	 * query_gid should be return GID value for @device, when @port_num
2385 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2386 	 * is RoCE link layer.
2387 	 */
2388 	int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2389 			 union ib_gid *gid);
2390 	/**
2391 	 * When calling add_gid, the HW vendor's driver should add the gid
2392 	 * of device of port at gid index available at @attr. Meta-info of
2393 	 * that gid (for example, the network device related to this gid) is
2394 	 * available at @attr. @context allows the HW vendor driver to store
2395 	 * extra information together with a GID entry. The HW vendor driver may
2396 	 * allocate memory to contain this information and store it in @context
2397 	 * when a new GID entry is written to. Params are consistent until the
2398 	 * next call of add_gid or delete_gid. The function should return 0 on
2399 	 * success or error otherwise. The function could be called
2400 	 * concurrently for different ports. This function is only called when
2401 	 * roce_gid_table is used.
2402 	 */
2403 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2404 	/**
2405 	 * When calling del_gid, the HW vendor's driver should delete the
2406 	 * gid of device @device at gid index gid_index of port port_num
2407 	 * available in @attr.
2408 	 * Upon the deletion of a GID entry, the HW vendor must free any
2409 	 * allocated memory. The caller will clear @context afterwards.
2410 	 * This function is only called when roce_gid_table is used.
2411 	 */
2412 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2413 	int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2414 			  u16 *pkey);
2415 	int (*alloc_ucontext)(struct ib_ucontext *context,
2416 			      struct ib_udata *udata);
2417 	void (*dealloc_ucontext)(struct ib_ucontext *context);
2418 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2419 	/**
2420 	 * This will be called once refcount of an entry in mmap_xa reaches
2421 	 * zero. The type of the memory that was mapped may differ between
2422 	 * entries and is opaque to the rdma_user_mmap interface.
2423 	 * Therefore needs to be implemented by the driver in mmap_free.
2424 	 */
2425 	void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2426 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2427 	int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2428 	int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2429 	int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2430 			 struct ib_udata *udata);
2431 	int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2432 			      struct ib_udata *udata);
2433 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2434 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2435 	int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2436 	int (*create_srq)(struct ib_srq *srq,
2437 			  struct ib_srq_init_attr *srq_init_attr,
2438 			  struct ib_udata *udata);
2439 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2440 			  enum ib_srq_attr_mask srq_attr_mask,
2441 			  struct ib_udata *udata);
2442 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2443 	int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2444 	int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2445 			 struct ib_udata *udata);
2446 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2447 			 int qp_attr_mask, struct ib_udata *udata);
2448 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2449 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2450 	int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2451 	int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2452 			 struct ib_udata *udata);
2453 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2454 	int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2455 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2456 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2457 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2458 				     u64 virt_addr, int mr_access_flags,
2459 				     struct ib_udata *udata);
2460 	struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2461 					    u64 length, u64 virt_addr, int fd,
2462 					    int mr_access_flags,
2463 					    struct ib_udata *udata);
2464 	struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2465 				       u64 length, u64 virt_addr,
2466 				       int mr_access_flags, struct ib_pd *pd,
2467 				       struct ib_udata *udata);
2468 	int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2469 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2470 				  u32 max_num_sg);
2471 	struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2472 					    u32 max_num_data_sg,
2473 					    u32 max_num_meta_sg);
2474 	int (*advise_mr)(struct ib_pd *pd,
2475 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2476 			 struct ib_sge *sg_list, u32 num_sge,
2477 			 struct uverbs_attr_bundle *attrs);
2478 
2479 	/*
2480 	 * Kernel users should universally support relaxed ordering (RO), as
2481 	 * they are designed to read data only after observing the CQE and use
2482 	 * the DMA API correctly.
2483 	 *
2484 	 * Some drivers implicitly enable RO if platform supports it.
2485 	 */
2486 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2487 			 unsigned int *sg_offset);
2488 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2489 			       struct ib_mr_status *mr_status);
2490 	int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2491 	int (*dealloc_mw)(struct ib_mw *mw);
2492 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2493 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2494 	int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2495 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2496 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2497 				       struct ib_flow_attr *flow_attr,
2498 				       struct ib_udata *udata);
2499 	int (*destroy_flow)(struct ib_flow *flow_id);
2500 	int (*destroy_flow_action)(struct ib_flow_action *action);
2501 	int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2502 				 int state);
2503 	int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2504 			     struct ifla_vf_info *ivf);
2505 	int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2506 			    struct ifla_vf_stats *stats);
2507 	int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2508 			    struct ifla_vf_guid *node_guid,
2509 			    struct ifla_vf_guid *port_guid);
2510 	int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2511 			   int type);
2512 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2513 				   struct ib_wq_init_attr *init_attr,
2514 				   struct ib_udata *udata);
2515 	int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2516 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2517 			 u32 wq_attr_mask, struct ib_udata *udata);
2518 	int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2519 				    struct ib_rwq_ind_table_init_attr *init_attr,
2520 				    struct ib_udata *udata);
2521 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2522 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2523 				  struct ib_ucontext *context,
2524 				  struct ib_dm_alloc_attr *attr,
2525 				  struct uverbs_attr_bundle *attrs);
2526 	int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2527 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2528 				   struct ib_dm_mr_attr *attr,
2529 				   struct uverbs_attr_bundle *attrs);
2530 	int (*create_counters)(struct ib_counters *counters,
2531 			       struct uverbs_attr_bundle *attrs);
2532 	int (*destroy_counters)(struct ib_counters *counters);
2533 	int (*read_counters)(struct ib_counters *counters,
2534 			     struct ib_counters_read_attr *counters_read_attr,
2535 			     struct uverbs_attr_bundle *attrs);
2536 	int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2537 			    int data_sg_nents, unsigned int *data_sg_offset,
2538 			    struct scatterlist *meta_sg, int meta_sg_nents,
2539 			    unsigned int *meta_sg_offset);
2540 
2541 	/**
2542 	 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2543 	 *   fill in the driver initialized data.  The struct is kfree()'ed by
2544 	 *   the sysfs core when the device is removed.  A lifespan of -1 in the
2545 	 *   return struct tells the core to set a default lifespan.
2546 	 */
2547 	struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2548 	struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2549 						     u32 port_num);
2550 	/**
2551 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2552 	 * @index - The index in the value array we wish to have updated, or
2553 	 *   num_counters if we want all stats updated
2554 	 * Return codes -
2555 	 *   < 0 - Error, no counters updated
2556 	 *   index - Updated the single counter pointed to by index
2557 	 *   num_counters - Updated all counters (will reset the timestamp
2558 	 *     and prevent further calls for lifespan milliseconds)
2559 	 * Drivers are allowed to update all counters in leiu of just the
2560 	 *   one given in index at their option
2561 	 */
2562 	int (*get_hw_stats)(struct ib_device *device,
2563 			    struct rdma_hw_stats *stats, u32 port, int index);
2564 
2565 	/**
2566 	 * modify_hw_stat - Modify the counter configuration
2567 	 * @enable: true/false when enable/disable a counter
2568 	 * Return codes - 0 on success or error code otherwise.
2569 	 */
2570 	int (*modify_hw_stat)(struct ib_device *device, u32 port,
2571 			      unsigned int counter_index, bool enable);
2572 	/**
2573 	 * Allows rdma drivers to add their own restrack attributes.
2574 	 */
2575 	int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2576 	int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2577 	int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2578 	int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2579 	int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2580 	int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2581 	int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2582 
2583 	/* Device lifecycle callbacks */
2584 	/*
2585 	 * Called after the device becomes registered, before clients are
2586 	 * attached
2587 	 */
2588 	int (*enable_driver)(struct ib_device *dev);
2589 	/*
2590 	 * This is called as part of ib_dealloc_device().
2591 	 */
2592 	void (*dealloc_driver)(struct ib_device *dev);
2593 
2594 	/* iWarp CM callbacks */
2595 	void (*iw_add_ref)(struct ib_qp *qp);
2596 	void (*iw_rem_ref)(struct ib_qp *qp);
2597 	struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2598 	int (*iw_connect)(struct iw_cm_id *cm_id,
2599 			  struct iw_cm_conn_param *conn_param);
2600 	int (*iw_accept)(struct iw_cm_id *cm_id,
2601 			 struct iw_cm_conn_param *conn_param);
2602 	int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2603 			 u8 pdata_len);
2604 	int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2605 	int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2606 	/**
2607 	 * counter_bind_qp - Bind a QP to a counter.
2608 	 * @counter - The counter to be bound. If counter->id is zero then
2609 	 *   the driver needs to allocate a new counter and set counter->id
2610 	 */
2611 	int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2612 	/**
2613 	 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2614 	 *   counter and bind it onto the default one
2615 	 */
2616 	int (*counter_unbind_qp)(struct ib_qp *qp);
2617 	/**
2618 	 * counter_dealloc -De-allocate the hw counter
2619 	 */
2620 	int (*counter_dealloc)(struct rdma_counter *counter);
2621 	/**
2622 	 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2623 	 * the driver initialized data.
2624 	 */
2625 	struct rdma_hw_stats *(*counter_alloc_stats)(
2626 		struct rdma_counter *counter);
2627 	/**
2628 	 * counter_update_stats - Query the stats value of this counter
2629 	 */
2630 	int (*counter_update_stats)(struct rdma_counter *counter);
2631 
2632 	/**
2633 	 * Allows rdma drivers to add their own restrack attributes
2634 	 * dumped via 'rdma stat' iproute2 command.
2635 	 */
2636 	int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2637 
2638 	/* query driver for its ucontext properties */
2639 	int (*query_ucontext)(struct ib_ucontext *context,
2640 			      struct uverbs_attr_bundle *attrs);
2641 
2642 	/*
2643 	 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2644 	 * Everyone else relies on Linux memory management model.
2645 	 */
2646 	int (*get_numa_node)(struct ib_device *dev);
2647 
2648 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
2649 	DECLARE_RDMA_OBJ_SIZE(ib_counters);
2650 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
2651 	DECLARE_RDMA_OBJ_SIZE(ib_mw);
2652 	DECLARE_RDMA_OBJ_SIZE(ib_pd);
2653 	DECLARE_RDMA_OBJ_SIZE(ib_qp);
2654 	DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2655 	DECLARE_RDMA_OBJ_SIZE(ib_srq);
2656 	DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2657 	DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2658 };
2659 
2660 struct ib_core_device {
2661 	/* device must be the first element in structure until,
2662 	 * union of ib_core_device and device exists in ib_device.
2663 	 */
2664 	struct device dev;
2665 	possible_net_t rdma_net;
2666 	struct kobject *ports_kobj;
2667 	struct list_head port_list;
2668 	struct ib_device *owner; /* reach back to owner ib_device */
2669 };
2670 
2671 struct rdma_restrack_root;
2672 struct ib_device {
2673 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2674 	struct device                *dma_device;
2675 	struct ib_device_ops	     ops;
2676 	char                          name[IB_DEVICE_NAME_MAX];
2677 	struct rcu_head rcu_head;
2678 
2679 	struct list_head              event_handler_list;
2680 	/* Protects event_handler_list */
2681 	struct rw_semaphore event_handler_rwsem;
2682 
2683 	/* Protects QP's event_handler calls and open_qp list */
2684 	spinlock_t qp_open_list_lock;
2685 
2686 	struct rw_semaphore	      client_data_rwsem;
2687 	struct xarray                 client_data;
2688 	struct mutex                  unregistration_lock;
2689 
2690 	/* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2691 	rwlock_t cache_lock;
2692 	/**
2693 	 * port_data is indexed by port number
2694 	 */
2695 	struct ib_port_data *port_data;
2696 
2697 	int			      num_comp_vectors;
2698 
2699 	union {
2700 		struct device		dev;
2701 		struct ib_core_device	coredev;
2702 	};
2703 
2704 	/* First group is for device attributes,
2705 	 * Second group is for driver provided attributes (optional).
2706 	 * Third group is for the hw_stats
2707 	 * It is a NULL terminated array.
2708 	 */
2709 	const struct attribute_group	*groups[4];
2710 
2711 	u64			     uverbs_cmd_mask;
2712 
2713 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2714 	__be64			     node_guid;
2715 	u32			     local_dma_lkey;
2716 	u16                          is_switch:1;
2717 	/* Indicates kernel verbs support, should not be used in drivers */
2718 	u16                          kverbs_provider:1;
2719 	/* CQ adaptive moderation (RDMA DIM) */
2720 	u16                          use_cq_dim:1;
2721 	u8                           node_type;
2722 	u32			     phys_port_cnt;
2723 	struct ib_device_attr        attrs;
2724 	struct hw_stats_device_data *hw_stats_data;
2725 
2726 #ifdef CONFIG_CGROUP_RDMA
2727 	struct rdmacg_device         cg_device;
2728 #endif
2729 
2730 	u32                          index;
2731 
2732 	spinlock_t                   cq_pools_lock;
2733 	struct list_head             cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2734 
2735 	struct rdma_restrack_root *res;
2736 
2737 	const struct uapi_definition   *driver_def;
2738 
2739 	/*
2740 	 * Positive refcount indicates that the device is currently
2741 	 * registered and cannot be unregistered.
2742 	 */
2743 	refcount_t refcount;
2744 	struct completion unreg_completion;
2745 	struct work_struct unregistration_work;
2746 
2747 	const struct rdma_link_ops *link_ops;
2748 
2749 	/* Protects compat_devs xarray modifications */
2750 	struct mutex compat_devs_mutex;
2751 	/* Maintains compat devices for each net namespace */
2752 	struct xarray compat_devs;
2753 
2754 	/* Used by iWarp CM */
2755 	char iw_ifname[IFNAMSIZ];
2756 	u32 iw_driver_flags;
2757 	u32 lag_flags;
2758 };
2759 
2760 static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
2761 				    gfp_t gfp, bool is_numa_aware)
2762 {
2763 	if (is_numa_aware && dev->ops.get_numa_node)
2764 		return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
2765 
2766 	return kzalloc(size, gfp);
2767 }
2768 
2769 struct ib_client_nl_info;
2770 struct ib_client {
2771 	const char *name;
2772 	int (*add)(struct ib_device *ibdev);
2773 	void (*remove)(struct ib_device *, void *client_data);
2774 	void (*rename)(struct ib_device *dev, void *client_data);
2775 	int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2776 			   struct ib_client_nl_info *res);
2777 	int (*get_global_nl_info)(struct ib_client_nl_info *res);
2778 
2779 	/* Returns the net_dev belonging to this ib_client and matching the
2780 	 * given parameters.
2781 	 * @dev:	 An RDMA device that the net_dev use for communication.
2782 	 * @port:	 A physical port number on the RDMA device.
2783 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2784 	 * @gid:	 A GID that the net_dev uses to communicate.
2785 	 * @addr:	 An IP address the net_dev is configured with.
2786 	 * @client_data: The device's client data set by ib_set_client_data().
2787 	 *
2788 	 * An ib_client that implements a net_dev on top of RDMA devices
2789 	 * (such as IP over IB) should implement this callback, allowing the
2790 	 * rdma_cm module to find the right net_dev for a given request.
2791 	 *
2792 	 * The caller is responsible for calling dev_put on the returned
2793 	 * netdev. */
2794 	struct net_device *(*get_net_dev_by_params)(
2795 			struct ib_device *dev,
2796 			u32 port,
2797 			u16 pkey,
2798 			const union ib_gid *gid,
2799 			const struct sockaddr *addr,
2800 			void *client_data);
2801 
2802 	refcount_t uses;
2803 	struct completion uses_zero;
2804 	u32 client_id;
2805 
2806 	/* kverbs are not required by the client */
2807 	u8 no_kverbs_req:1;
2808 };
2809 
2810 /*
2811  * IB block DMA iterator
2812  *
2813  * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2814  * to a HW supported page size.
2815  */
2816 struct ib_block_iter {
2817 	/* internal states */
2818 	struct scatterlist *__sg;	/* sg holding the current aligned block */
2819 	dma_addr_t __dma_addr;		/* unaligned DMA address of this block */
2820 	unsigned int __sg_nents;	/* number of SG entries */
2821 	unsigned int __sg_advance;	/* number of bytes to advance in sg in next step */
2822 	unsigned int __pg_bit;		/* alignment of current block */
2823 };
2824 
2825 struct ib_device *_ib_alloc_device(size_t size);
2826 #define ib_alloc_device(drv_struct, member)                                    \
2827 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2828 				      BUILD_BUG_ON_ZERO(offsetof(              \
2829 					      struct drv_struct, member))),    \
2830 		     struct drv_struct, member)
2831 
2832 void ib_dealloc_device(struct ib_device *device);
2833 
2834 void ib_get_device_fw_str(struct ib_device *device, char *str);
2835 
2836 int ib_register_device(struct ib_device *device, const char *name,
2837 		       struct device *dma_device);
2838 void ib_unregister_device(struct ib_device *device);
2839 void ib_unregister_driver(enum rdma_driver_id driver_id);
2840 void ib_unregister_device_and_put(struct ib_device *device);
2841 void ib_unregister_device_queued(struct ib_device *ib_dev);
2842 
2843 int ib_register_client   (struct ib_client *client);
2844 void ib_unregister_client(struct ib_client *client);
2845 
2846 void __rdma_block_iter_start(struct ib_block_iter *biter,
2847 			     struct scatterlist *sglist,
2848 			     unsigned int nents,
2849 			     unsigned long pgsz);
2850 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2851 
2852 /**
2853  * rdma_block_iter_dma_address - get the aligned dma address of the current
2854  * block held by the block iterator.
2855  * @biter: block iterator holding the memory block
2856  */
2857 static inline dma_addr_t
2858 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2859 {
2860 	return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2861 }
2862 
2863 /**
2864  * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2865  * @sglist: sglist to iterate over
2866  * @biter: block iterator holding the memory block
2867  * @nents: maximum number of sg entries to iterate over
2868  * @pgsz: best HW supported page size to use
2869  *
2870  * Callers may use rdma_block_iter_dma_address() to get each
2871  * blocks aligned DMA address.
2872  */
2873 #define rdma_for_each_block(sglist, biter, nents, pgsz)		\
2874 	for (__rdma_block_iter_start(biter, sglist, nents,	\
2875 				     pgsz);			\
2876 	     __rdma_block_iter_next(biter);)
2877 
2878 /**
2879  * ib_get_client_data - Get IB client context
2880  * @device:Device to get context for
2881  * @client:Client to get context for
2882  *
2883  * ib_get_client_data() returns the client context data set with
2884  * ib_set_client_data(). This can only be called while the client is
2885  * registered to the device, once the ib_client remove() callback returns this
2886  * cannot be called.
2887  */
2888 static inline void *ib_get_client_data(struct ib_device *device,
2889 				       struct ib_client *client)
2890 {
2891 	return xa_load(&device->client_data, client->client_id);
2892 }
2893 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2894 			 void *data);
2895 void ib_set_device_ops(struct ib_device *device,
2896 		       const struct ib_device_ops *ops);
2897 
2898 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2899 		      unsigned long pfn, unsigned long size, pgprot_t prot,
2900 		      struct rdma_user_mmap_entry *entry);
2901 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2902 				struct rdma_user_mmap_entry *entry,
2903 				size_t length);
2904 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2905 				      struct rdma_user_mmap_entry *entry,
2906 				      size_t length, u32 min_pgoff,
2907 				      u32 max_pgoff);
2908 
2909 static inline int
2910 rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
2911 				  struct rdma_user_mmap_entry *entry,
2912 				  size_t length, u32 pgoff)
2913 {
2914 	return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff,
2915 						 pgoff);
2916 }
2917 
2918 struct rdma_user_mmap_entry *
2919 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2920 			       unsigned long pgoff);
2921 struct rdma_user_mmap_entry *
2922 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2923 			 struct vm_area_struct *vma);
2924 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2925 
2926 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2927 
2928 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2929 {
2930 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2931 }
2932 
2933 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2934 {
2935 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2936 }
2937 
2938 static inline bool ib_is_buffer_cleared(const void __user *p,
2939 					size_t len)
2940 {
2941 	bool ret;
2942 	u8 *buf;
2943 
2944 	if (len > USHRT_MAX)
2945 		return false;
2946 
2947 	buf = memdup_user(p, len);
2948 	if (IS_ERR(buf))
2949 		return false;
2950 
2951 	ret = !memchr_inv(buf, 0, len);
2952 	kfree(buf);
2953 	return ret;
2954 }
2955 
2956 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2957 				       size_t offset,
2958 				       size_t len)
2959 {
2960 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
2961 }
2962 
2963 /**
2964  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2965  * contains all required attributes and no attributes not allowed for
2966  * the given QP state transition.
2967  * @cur_state: Current QP state
2968  * @next_state: Next QP state
2969  * @type: QP type
2970  * @mask: Mask of supplied QP attributes
2971  *
2972  * This function is a helper function that a low-level driver's
2973  * modify_qp method can use to validate the consumer's input.  It
2974  * checks that cur_state and next_state are valid QP states, that a
2975  * transition from cur_state to next_state is allowed by the IB spec,
2976  * and that the attribute mask supplied is allowed for the transition.
2977  */
2978 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2979 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
2980 
2981 void ib_register_event_handler(struct ib_event_handler *event_handler);
2982 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2983 void ib_dispatch_event(const struct ib_event *event);
2984 
2985 int ib_query_port(struct ib_device *device,
2986 		  u32 port_num, struct ib_port_attr *port_attr);
2987 
2988 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2989 					       u32 port_num);
2990 
2991 /**
2992  * rdma_cap_ib_switch - Check if the device is IB switch
2993  * @device: Device to check
2994  *
2995  * Device driver is responsible for setting is_switch bit on
2996  * in ib_device structure at init time.
2997  *
2998  * Return: true if the device is IB switch.
2999  */
3000 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3001 {
3002 	return device->is_switch;
3003 }
3004 
3005 /**
3006  * rdma_start_port - Return the first valid port number for the device
3007  * specified
3008  *
3009  * @device: Device to be checked
3010  *
3011  * Return start port number
3012  */
3013 static inline u32 rdma_start_port(const struct ib_device *device)
3014 {
3015 	return rdma_cap_ib_switch(device) ? 0 : 1;
3016 }
3017 
3018 /**
3019  * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3020  * @device - The struct ib_device * to iterate over
3021  * @iter - The unsigned int to store the port number
3022  */
3023 #define rdma_for_each_port(device, iter)                                       \
3024 	for (iter = rdma_start_port(device +				       \
3025 				    BUILD_BUG_ON_ZERO(!__same_type(u32,	       \
3026 								   iter)));    \
3027 	     iter <= rdma_end_port(device); iter++)
3028 
3029 /**
3030  * rdma_end_port - Return the last valid port number for the device
3031  * specified
3032  *
3033  * @device: Device to be checked
3034  *
3035  * Return last port number
3036  */
3037 static inline u32 rdma_end_port(const struct ib_device *device)
3038 {
3039 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3040 }
3041 
3042 static inline int rdma_is_port_valid(const struct ib_device *device,
3043 				     unsigned int port)
3044 {
3045 	return (port >= rdma_start_port(device) &&
3046 		port <= rdma_end_port(device));
3047 }
3048 
3049 static inline bool rdma_is_grh_required(const struct ib_device *device,
3050 					u32 port_num)
3051 {
3052 	return device->port_data[port_num].immutable.core_cap_flags &
3053 	       RDMA_CORE_PORT_IB_GRH_REQUIRED;
3054 }
3055 
3056 static inline bool rdma_protocol_ib(const struct ib_device *device,
3057 				    u32 port_num)
3058 {
3059 	return device->port_data[port_num].immutable.core_cap_flags &
3060 	       RDMA_CORE_CAP_PROT_IB;
3061 }
3062 
3063 static inline bool rdma_protocol_roce(const struct ib_device *device,
3064 				      u32 port_num)
3065 {
3066 	return device->port_data[port_num].immutable.core_cap_flags &
3067 	       (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3068 }
3069 
3070 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3071 						u32 port_num)
3072 {
3073 	return device->port_data[port_num].immutable.core_cap_flags &
3074 	       RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3075 }
3076 
3077 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3078 						u32 port_num)
3079 {
3080 	return device->port_data[port_num].immutable.core_cap_flags &
3081 	       RDMA_CORE_CAP_PROT_ROCE;
3082 }
3083 
3084 static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3085 				       u32 port_num)
3086 {
3087 	return device->port_data[port_num].immutable.core_cap_flags &
3088 	       RDMA_CORE_CAP_PROT_IWARP;
3089 }
3090 
3091 static inline bool rdma_ib_or_roce(const struct ib_device *device,
3092 				   u32 port_num)
3093 {
3094 	return rdma_protocol_ib(device, port_num) ||
3095 		rdma_protocol_roce(device, port_num);
3096 }
3097 
3098 static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3099 					    u32 port_num)
3100 {
3101 	return device->port_data[port_num].immutable.core_cap_flags &
3102 	       RDMA_CORE_CAP_PROT_RAW_PACKET;
3103 }
3104 
3105 static inline bool rdma_protocol_usnic(const struct ib_device *device,
3106 				       u32 port_num)
3107 {
3108 	return device->port_data[port_num].immutable.core_cap_flags &
3109 	       RDMA_CORE_CAP_PROT_USNIC;
3110 }
3111 
3112 /**
3113  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3114  * Management Datagrams.
3115  * @device: Device to check
3116  * @port_num: Port number to check
3117  *
3118  * Management Datagrams (MAD) are a required part of the InfiniBand
3119  * specification and are supported on all InfiniBand devices.  A slightly
3120  * extended version are also supported on OPA interfaces.
3121  *
3122  * Return: true if the port supports sending/receiving of MAD packets.
3123  */
3124 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3125 {
3126 	return device->port_data[port_num].immutable.core_cap_flags &
3127 	       RDMA_CORE_CAP_IB_MAD;
3128 }
3129 
3130 /**
3131  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3132  * Management Datagrams.
3133  * @device: Device to check
3134  * @port_num: Port number to check
3135  *
3136  * Intel OmniPath devices extend and/or replace the InfiniBand Management
3137  * datagrams with their own versions.  These OPA MADs share many but not all of
3138  * the characteristics of InfiniBand MADs.
3139  *
3140  * OPA MADs differ in the following ways:
3141  *
3142  *    1) MADs are variable size up to 2K
3143  *       IBTA defined MADs remain fixed at 256 bytes
3144  *    2) OPA SMPs must carry valid PKeys
3145  *    3) OPA SMP packets are a different format
3146  *
3147  * Return: true if the port supports OPA MAD packet formats.
3148  */
3149 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3150 {
3151 	return device->port_data[port_num].immutable.core_cap_flags &
3152 		RDMA_CORE_CAP_OPA_MAD;
3153 }
3154 
3155 /**
3156  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3157  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3158  * @device: Device to check
3159  * @port_num: Port number to check
3160  *
3161  * Each InfiniBand node is required to provide a Subnet Management Agent
3162  * that the subnet manager can access.  Prior to the fabric being fully
3163  * configured by the subnet manager, the SMA is accessed via a well known
3164  * interface called the Subnet Management Interface (SMI).  This interface
3165  * uses directed route packets to communicate with the SM to get around the
3166  * chicken and egg problem of the SM needing to know what's on the fabric
3167  * in order to configure the fabric, and needing to configure the fabric in
3168  * order to send packets to the devices on the fabric.  These directed
3169  * route packets do not need the fabric fully configured in order to reach
3170  * their destination.  The SMI is the only method allowed to send
3171  * directed route packets on an InfiniBand fabric.
3172  *
3173  * Return: true if the port provides an SMI.
3174  */
3175 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3176 {
3177 	return device->port_data[port_num].immutable.core_cap_flags &
3178 	       RDMA_CORE_CAP_IB_SMI;
3179 }
3180 
3181 /**
3182  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3183  * Communication Manager.
3184  * @device: Device to check
3185  * @port_num: Port number to check
3186  *
3187  * The InfiniBand Communication Manager is one of many pre-defined General
3188  * Service Agents (GSA) that are accessed via the General Service
3189  * Interface (GSI).  It's role is to facilitate establishment of connections
3190  * between nodes as well as other management related tasks for established
3191  * connections.
3192  *
3193  * Return: true if the port supports an IB CM (this does not guarantee that
3194  * a CM is actually running however).
3195  */
3196 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3197 {
3198 	return device->port_data[port_num].immutable.core_cap_flags &
3199 	       RDMA_CORE_CAP_IB_CM;
3200 }
3201 
3202 /**
3203  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3204  * Communication Manager.
3205  * @device: Device to check
3206  * @port_num: Port number to check
3207  *
3208  * Similar to above, but specific to iWARP connections which have a different
3209  * managment protocol than InfiniBand.
3210  *
3211  * Return: true if the port supports an iWARP CM (this does not guarantee that
3212  * a CM is actually running however).
3213  */
3214 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3215 {
3216 	return device->port_data[port_num].immutable.core_cap_flags &
3217 	       RDMA_CORE_CAP_IW_CM;
3218 }
3219 
3220 /**
3221  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3222  * Subnet Administration.
3223  * @device: Device to check
3224  * @port_num: Port number to check
3225  *
3226  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3227  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
3228  * fabrics, devices should resolve routes to other hosts by contacting the
3229  * SA to query the proper route.
3230  *
3231  * Return: true if the port should act as a client to the fabric Subnet
3232  * Administration interface.  This does not imply that the SA service is
3233  * running locally.
3234  */
3235 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3236 {
3237 	return device->port_data[port_num].immutable.core_cap_flags &
3238 	       RDMA_CORE_CAP_IB_SA;
3239 }
3240 
3241 /**
3242  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3243  * Multicast.
3244  * @device: Device to check
3245  * @port_num: Port number to check
3246  *
3247  * InfiniBand multicast registration is more complex than normal IPv4 or
3248  * IPv6 multicast registration.  Each Host Channel Adapter must register
3249  * with the Subnet Manager when it wishes to join a multicast group.  It
3250  * should do so only once regardless of how many queue pairs it subscribes
3251  * to this group.  And it should leave the group only after all queue pairs
3252  * attached to the group have been detached.
3253  *
3254  * Return: true if the port must undertake the additional adminstrative
3255  * overhead of registering/unregistering with the SM and tracking of the
3256  * total number of queue pairs attached to the multicast group.
3257  */
3258 static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3259 				     u32 port_num)
3260 {
3261 	return rdma_cap_ib_sa(device, port_num);
3262 }
3263 
3264 /**
3265  * rdma_cap_af_ib - Check if the port of device has the capability
3266  * Native Infiniband Address.
3267  * @device: Device to check
3268  * @port_num: Port number to check
3269  *
3270  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3271  * GID.  RoCE uses a different mechanism, but still generates a GID via
3272  * a prescribed mechanism and port specific data.
3273  *
3274  * Return: true if the port uses a GID address to identify devices on the
3275  * network.
3276  */
3277 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3278 {
3279 	return device->port_data[port_num].immutable.core_cap_flags &
3280 	       RDMA_CORE_CAP_AF_IB;
3281 }
3282 
3283 /**
3284  * rdma_cap_eth_ah - Check if the port of device has the capability
3285  * Ethernet Address Handle.
3286  * @device: Device to check
3287  * @port_num: Port number to check
3288  *
3289  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3290  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3291  * port.  Normally, packet headers are generated by the sending host
3292  * adapter, but when sending connectionless datagrams, we must manually
3293  * inject the proper headers for the fabric we are communicating over.
3294  *
3295  * Return: true if we are running as a RoCE port and must force the
3296  * addition of a Global Route Header built from our Ethernet Address
3297  * Handle into our header list for connectionless packets.
3298  */
3299 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3300 {
3301 	return device->port_data[port_num].immutable.core_cap_flags &
3302 	       RDMA_CORE_CAP_ETH_AH;
3303 }
3304 
3305 /**
3306  * rdma_cap_opa_ah - Check if the port of device supports
3307  * OPA Address handles
3308  * @device: Device to check
3309  * @port_num: Port number to check
3310  *
3311  * Return: true if we are running on an OPA device which supports
3312  * the extended OPA addressing.
3313  */
3314 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3315 {
3316 	return (device->port_data[port_num].immutable.core_cap_flags &
3317 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3318 }
3319 
3320 /**
3321  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3322  *
3323  * @device: Device
3324  * @port_num: Port number
3325  *
3326  * This MAD size includes the MAD headers and MAD payload.  No other headers
3327  * are included.
3328  *
3329  * Return the max MAD size required by the Port.  Will return 0 if the port
3330  * does not support MADs
3331  */
3332 static inline size_t rdma_max_mad_size(const struct ib_device *device,
3333 				       u32 port_num)
3334 {
3335 	return device->port_data[port_num].immutable.max_mad_size;
3336 }
3337 
3338 /**
3339  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3340  * @device: Device to check
3341  * @port_num: Port number to check
3342  *
3343  * RoCE GID table mechanism manages the various GIDs for a device.
3344  *
3345  * NOTE: if allocating the port's GID table has failed, this call will still
3346  * return true, but any RoCE GID table API will fail.
3347  *
3348  * Return: true if the port uses RoCE GID table mechanism in order to manage
3349  * its GIDs.
3350  */
3351 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3352 					   u32 port_num)
3353 {
3354 	return rdma_protocol_roce(device, port_num) &&
3355 		device->ops.add_gid && device->ops.del_gid;
3356 }
3357 
3358 /*
3359  * Check if the device supports READ W/ INVALIDATE.
3360  */
3361 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3362 {
3363 	/*
3364 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3365 	 * has support for it yet.
3366 	 */
3367 	return rdma_protocol_iwarp(dev, port_num);
3368 }
3369 
3370 /**
3371  * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3372  * @device: Device
3373  * @port_num: 1 based Port number
3374  *
3375  * Return true if port is an Intel OPA port , false if not
3376  */
3377 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3378 					  u32 port_num)
3379 {
3380 	return (device->port_data[port_num].immutable.core_cap_flags &
3381 		RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3382 }
3383 
3384 /**
3385  * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3386  * @device: Device
3387  * @port_num: Port number
3388  * @mtu: enum value of MTU
3389  *
3390  * Return the MTU size supported by the port as an integer value. Will return
3391  * -1 if enum value of mtu is not supported.
3392  */
3393 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3394 				       int mtu)
3395 {
3396 	if (rdma_core_cap_opa_port(device, port))
3397 		return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3398 	else
3399 		return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3400 }
3401 
3402 /**
3403  * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3404  * @device: Device
3405  * @port_num: Port number
3406  * @attr: port attribute
3407  *
3408  * Return the MTU size supported by the port as an integer value.
3409  */
3410 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3411 				     struct ib_port_attr *attr)
3412 {
3413 	if (rdma_core_cap_opa_port(device, port))
3414 		return attr->phys_mtu;
3415 	else
3416 		return ib_mtu_enum_to_int(attr->max_mtu);
3417 }
3418 
3419 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3420 			 int state);
3421 int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3422 		     struct ifla_vf_info *info);
3423 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3424 		    struct ifla_vf_stats *stats);
3425 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3426 		    struct ifla_vf_guid *node_guid,
3427 		    struct ifla_vf_guid *port_guid);
3428 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3429 		   int type);
3430 
3431 int ib_query_pkey(struct ib_device *device,
3432 		  u32 port_num, u16 index, u16 *pkey);
3433 
3434 int ib_modify_device(struct ib_device *device,
3435 		     int device_modify_mask,
3436 		     struct ib_device_modify *device_modify);
3437 
3438 int ib_modify_port(struct ib_device *device,
3439 		   u32 port_num, int port_modify_mask,
3440 		   struct ib_port_modify *port_modify);
3441 
3442 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3443 		u32 *port_num, u16 *index);
3444 
3445 int ib_find_pkey(struct ib_device *device,
3446 		 u32 port_num, u16 pkey, u16 *index);
3447 
3448 enum ib_pd_flags {
3449 	/*
3450 	 * Create a memory registration for all memory in the system and place
3451 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3452 	 * ULPs to avoid the overhead of dynamic MRs.
3453 	 *
3454 	 * This flag is generally considered unsafe and must only be used in
3455 	 * extremly trusted environments.  Every use of it will log a warning
3456 	 * in the kernel log.
3457 	 */
3458 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3459 };
3460 
3461 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3462 		const char *caller);
3463 
3464 /**
3465  * ib_alloc_pd - Allocates an unused protection domain.
3466  * @device: The device on which to allocate the protection domain.
3467  * @flags: protection domain flags
3468  *
3469  * A protection domain object provides an association between QPs, shared
3470  * receive queues, address handles, memory regions, and memory windows.
3471  *
3472  * Every PD has a local_dma_lkey which can be used as the lkey value for local
3473  * memory operations.
3474  */
3475 #define ib_alloc_pd(device, flags) \
3476 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3477 
3478 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3479 
3480 /**
3481  * ib_dealloc_pd - Deallocate kernel PD
3482  * @pd: The protection domain
3483  *
3484  * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3485  */
3486 static inline void ib_dealloc_pd(struct ib_pd *pd)
3487 {
3488 	int ret = ib_dealloc_pd_user(pd, NULL);
3489 
3490 	WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3491 }
3492 
3493 enum rdma_create_ah_flags {
3494 	/* In a sleepable context */
3495 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3496 };
3497 
3498 /**
3499  * rdma_create_ah - Creates an address handle for the given address vector.
3500  * @pd: The protection domain associated with the address handle.
3501  * @ah_attr: The attributes of the address vector.
3502  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3503  *
3504  * The address handle is used to reference a local or global destination
3505  * in all UD QP post sends.
3506  */
3507 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3508 			     u32 flags);
3509 
3510 /**
3511  * rdma_create_user_ah - Creates an address handle for the given address vector.
3512  * It resolves destination mac address for ah attribute of RoCE type.
3513  * @pd: The protection domain associated with the address handle.
3514  * @ah_attr: The attributes of the address vector.
3515  * @udata: pointer to user's input output buffer information need by
3516  *         provider driver.
3517  *
3518  * It returns 0 on success and returns appropriate error code on error.
3519  * The address handle is used to reference a local or global destination
3520  * in all UD QP post sends.
3521  */
3522 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3523 				  struct rdma_ah_attr *ah_attr,
3524 				  struct ib_udata *udata);
3525 /**
3526  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3527  *   work completion.
3528  * @hdr: the L3 header to parse
3529  * @net_type: type of header to parse
3530  * @sgid: place to store source gid
3531  * @dgid: place to store destination gid
3532  */
3533 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3534 			      enum rdma_network_type net_type,
3535 			      union ib_gid *sgid, union ib_gid *dgid);
3536 
3537 /**
3538  * ib_get_rdma_header_version - Get the header version
3539  * @hdr: the L3 header to parse
3540  */
3541 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3542 
3543 /**
3544  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3545  *   work completion.
3546  * @device: Device on which the received message arrived.
3547  * @port_num: Port on which the received message arrived.
3548  * @wc: Work completion associated with the received message.
3549  * @grh: References the received global route header.  This parameter is
3550  *   ignored unless the work completion indicates that the GRH is valid.
3551  * @ah_attr: Returned attributes that can be used when creating an address
3552  *   handle for replying to the message.
3553  * When ib_init_ah_attr_from_wc() returns success,
3554  * (a) for IB link layer it optionally contains a reference to SGID attribute
3555  * when GRH is present for IB link layer.
3556  * (b) for RoCE link layer it contains a reference to SGID attribute.
3557  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3558  * attributes which are initialized using ib_init_ah_attr_from_wc().
3559  *
3560  */
3561 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3562 			    const struct ib_wc *wc, const struct ib_grh *grh,
3563 			    struct rdma_ah_attr *ah_attr);
3564 
3565 /**
3566  * ib_create_ah_from_wc - Creates an address handle associated with the
3567  *   sender of the specified work completion.
3568  * @pd: The protection domain associated with the address handle.
3569  * @wc: Work completion information associated with a received message.
3570  * @grh: References the received global route header.  This parameter is
3571  *   ignored unless the work completion indicates that the GRH is valid.
3572  * @port_num: The outbound port number to associate with the address.
3573  *
3574  * The address handle is used to reference a local or global destination
3575  * in all UD QP post sends.
3576  */
3577 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3578 				   const struct ib_grh *grh, u32 port_num);
3579 
3580 /**
3581  * rdma_modify_ah - Modifies the address vector associated with an address
3582  *   handle.
3583  * @ah: The address handle to modify.
3584  * @ah_attr: The new address vector attributes to associate with the
3585  *   address handle.
3586  */
3587 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3588 
3589 /**
3590  * rdma_query_ah - Queries the address vector associated with an address
3591  *   handle.
3592  * @ah: The address handle to query.
3593  * @ah_attr: The address vector attributes associated with the address
3594  *   handle.
3595  */
3596 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3597 
3598 enum rdma_destroy_ah_flags {
3599 	/* In a sleepable context */
3600 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3601 };
3602 
3603 /**
3604  * rdma_destroy_ah_user - Destroys an address handle.
3605  * @ah: The address handle to destroy.
3606  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3607  * @udata: Valid user data or NULL for kernel objects
3608  */
3609 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3610 
3611 /**
3612  * rdma_destroy_ah - Destroys an kernel address handle.
3613  * @ah: The address handle to destroy.
3614  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3615  *
3616  * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3617  */
3618 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3619 {
3620 	int ret = rdma_destroy_ah_user(ah, flags, NULL);
3621 
3622 	WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3623 }
3624 
3625 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3626 				  struct ib_srq_init_attr *srq_init_attr,
3627 				  struct ib_usrq_object *uobject,
3628 				  struct ib_udata *udata);
3629 static inline struct ib_srq *
3630 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3631 {
3632 	if (!pd->device->ops.create_srq)
3633 		return ERR_PTR(-EOPNOTSUPP);
3634 
3635 	return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3636 }
3637 
3638 /**
3639  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3640  * @srq: The SRQ to modify.
3641  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3642  *   the current values of selected SRQ attributes are returned.
3643  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3644  *   are being modified.
3645  *
3646  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3647  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3648  * the number of receives queued drops below the limit.
3649  */
3650 int ib_modify_srq(struct ib_srq *srq,
3651 		  struct ib_srq_attr *srq_attr,
3652 		  enum ib_srq_attr_mask srq_attr_mask);
3653 
3654 /**
3655  * ib_query_srq - Returns the attribute list and current values for the
3656  *   specified SRQ.
3657  * @srq: The SRQ to query.
3658  * @srq_attr: The attributes of the specified SRQ.
3659  */
3660 int ib_query_srq(struct ib_srq *srq,
3661 		 struct ib_srq_attr *srq_attr);
3662 
3663 /**
3664  * ib_destroy_srq_user - Destroys the specified SRQ.
3665  * @srq: The SRQ to destroy.
3666  * @udata: Valid user data or NULL for kernel objects
3667  */
3668 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3669 
3670 /**
3671  * ib_destroy_srq - Destroys the specified kernel SRQ.
3672  * @srq: The SRQ to destroy.
3673  *
3674  * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3675  */
3676 static inline void ib_destroy_srq(struct ib_srq *srq)
3677 {
3678 	int ret = ib_destroy_srq_user(srq, NULL);
3679 
3680 	WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3681 }
3682 
3683 /**
3684  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3685  * @srq: The SRQ to post the work request on.
3686  * @recv_wr: A list of work requests to post on the receive queue.
3687  * @bad_recv_wr: On an immediate failure, this parameter will reference
3688  *   the work request that failed to be posted on the QP.
3689  */
3690 static inline int ib_post_srq_recv(struct ib_srq *srq,
3691 				   const struct ib_recv_wr *recv_wr,
3692 				   const struct ib_recv_wr **bad_recv_wr)
3693 {
3694 	const struct ib_recv_wr *dummy;
3695 
3696 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3697 					      bad_recv_wr ? : &dummy);
3698 }
3699 
3700 struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3701 				  struct ib_qp_init_attr *qp_init_attr,
3702 				  const char *caller);
3703 /**
3704  * ib_create_qp - Creates a kernel QP associated with the specific protection
3705  * domain.
3706  * @pd: The protection domain associated with the QP.
3707  * @init_attr: A list of initial attributes required to create the
3708  *   QP.  If QP creation succeeds, then the attributes are updated to
3709  *   the actual capabilities of the created QP.
3710  */
3711 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3712 					 struct ib_qp_init_attr *init_attr)
3713 {
3714 	return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
3715 }
3716 
3717 /**
3718  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3719  * @qp: The QP to modify.
3720  * @attr: On input, specifies the QP attributes to modify.  On output,
3721  *   the current values of selected QP attributes are returned.
3722  * @attr_mask: A bit-mask used to specify which attributes of the QP
3723  *   are being modified.
3724  * @udata: pointer to user's input output buffer information
3725  *   are being modified.
3726  * It returns 0 on success and returns appropriate error code on error.
3727  */
3728 int ib_modify_qp_with_udata(struct ib_qp *qp,
3729 			    struct ib_qp_attr *attr,
3730 			    int attr_mask,
3731 			    struct ib_udata *udata);
3732 
3733 /**
3734  * ib_modify_qp - Modifies the attributes for the specified QP and then
3735  *   transitions the QP to the given state.
3736  * @qp: The QP to modify.
3737  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3738  *   the current values of selected QP attributes are returned.
3739  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3740  *   are being modified.
3741  */
3742 int ib_modify_qp(struct ib_qp *qp,
3743 		 struct ib_qp_attr *qp_attr,
3744 		 int qp_attr_mask);
3745 
3746 /**
3747  * ib_query_qp - Returns the attribute list and current values for the
3748  *   specified QP.
3749  * @qp: The QP to query.
3750  * @qp_attr: The attributes of the specified QP.
3751  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3752  * @qp_init_attr: Additional attributes of the selected QP.
3753  *
3754  * The qp_attr_mask may be used to limit the query to gathering only the
3755  * selected attributes.
3756  */
3757 int ib_query_qp(struct ib_qp *qp,
3758 		struct ib_qp_attr *qp_attr,
3759 		int qp_attr_mask,
3760 		struct ib_qp_init_attr *qp_init_attr);
3761 
3762 /**
3763  * ib_destroy_qp - Destroys the specified QP.
3764  * @qp: The QP to destroy.
3765  * @udata: Valid udata or NULL for kernel objects
3766  */
3767 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3768 
3769 /**
3770  * ib_destroy_qp - Destroys the specified kernel QP.
3771  * @qp: The QP to destroy.
3772  *
3773  * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3774  */
3775 static inline int ib_destroy_qp(struct ib_qp *qp)
3776 {
3777 	return ib_destroy_qp_user(qp, NULL);
3778 }
3779 
3780 /**
3781  * ib_open_qp - Obtain a reference to an existing sharable QP.
3782  * @xrcd - XRC domain
3783  * @qp_open_attr: Attributes identifying the QP to open.
3784  *
3785  * Returns a reference to a sharable QP.
3786  */
3787 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3788 			 struct ib_qp_open_attr *qp_open_attr);
3789 
3790 /**
3791  * ib_close_qp - Release an external reference to a QP.
3792  * @qp: The QP handle to release
3793  *
3794  * The opened QP handle is released by the caller.  The underlying
3795  * shared QP is not destroyed until all internal references are released.
3796  */
3797 int ib_close_qp(struct ib_qp *qp);
3798 
3799 /**
3800  * ib_post_send - Posts a list of work requests to the send queue of
3801  *   the specified QP.
3802  * @qp: The QP to post the work request on.
3803  * @send_wr: A list of work requests to post on the send queue.
3804  * @bad_send_wr: On an immediate failure, this parameter will reference
3805  *   the work request that failed to be posted on the QP.
3806  *
3807  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3808  * error is returned, the QP state shall not be affected,
3809  * ib_post_send() will return an immediate error after queueing any
3810  * earlier work requests in the list.
3811  */
3812 static inline int ib_post_send(struct ib_qp *qp,
3813 			       const struct ib_send_wr *send_wr,
3814 			       const struct ib_send_wr **bad_send_wr)
3815 {
3816 	const struct ib_send_wr *dummy;
3817 
3818 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3819 }
3820 
3821 /**
3822  * ib_post_recv - Posts a list of work requests to the receive queue of
3823  *   the specified QP.
3824  * @qp: The QP to post the work request on.
3825  * @recv_wr: A list of work requests to post on the receive queue.
3826  * @bad_recv_wr: On an immediate failure, this parameter will reference
3827  *   the work request that failed to be posted on the QP.
3828  */
3829 static inline int ib_post_recv(struct ib_qp *qp,
3830 			       const struct ib_recv_wr *recv_wr,
3831 			       const struct ib_recv_wr **bad_recv_wr)
3832 {
3833 	const struct ib_recv_wr *dummy;
3834 
3835 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3836 }
3837 
3838 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3839 			    int comp_vector, enum ib_poll_context poll_ctx,
3840 			    const char *caller);
3841 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3842 					int nr_cqe, int comp_vector,
3843 					enum ib_poll_context poll_ctx)
3844 {
3845 	return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3846 			     KBUILD_MODNAME);
3847 }
3848 
3849 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3850 				int nr_cqe, enum ib_poll_context poll_ctx,
3851 				const char *caller);
3852 
3853 /**
3854  * ib_alloc_cq_any: Allocate kernel CQ
3855  * @dev: The IB device
3856  * @private: Private data attached to the CQE
3857  * @nr_cqe: Number of CQEs in the CQ
3858  * @poll_ctx: Context used for polling the CQ
3859  */
3860 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3861 					    void *private, int nr_cqe,
3862 					    enum ib_poll_context poll_ctx)
3863 {
3864 	return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3865 				 KBUILD_MODNAME);
3866 }
3867 
3868 void ib_free_cq(struct ib_cq *cq);
3869 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3870 
3871 /**
3872  * ib_create_cq - Creates a CQ on the specified device.
3873  * @device: The device on which to create the CQ.
3874  * @comp_handler: A user-specified callback that is invoked when a
3875  *   completion event occurs on the CQ.
3876  * @event_handler: A user-specified callback that is invoked when an
3877  *   asynchronous event not associated with a completion occurs on the CQ.
3878  * @cq_context: Context associated with the CQ returned to the user via
3879  *   the associated completion and event handlers.
3880  * @cq_attr: The attributes the CQ should be created upon.
3881  *
3882  * Users can examine the cq structure to determine the actual CQ size.
3883  */
3884 struct ib_cq *__ib_create_cq(struct ib_device *device,
3885 			     ib_comp_handler comp_handler,
3886 			     void (*event_handler)(struct ib_event *, void *),
3887 			     void *cq_context,
3888 			     const struct ib_cq_init_attr *cq_attr,
3889 			     const char *caller);
3890 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3891 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3892 
3893 /**
3894  * ib_resize_cq - Modifies the capacity of the CQ.
3895  * @cq: The CQ to resize.
3896  * @cqe: The minimum size of the CQ.
3897  *
3898  * Users can examine the cq structure to determine the actual CQ size.
3899  */
3900 int ib_resize_cq(struct ib_cq *cq, int cqe);
3901 
3902 /**
3903  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3904  * @cq: The CQ to modify.
3905  * @cq_count: number of CQEs that will trigger an event
3906  * @cq_period: max period of time in usec before triggering an event
3907  *
3908  */
3909 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3910 
3911 /**
3912  * ib_destroy_cq_user - Destroys the specified CQ.
3913  * @cq: The CQ to destroy.
3914  * @udata: Valid user data or NULL for kernel objects
3915  */
3916 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3917 
3918 /**
3919  * ib_destroy_cq - Destroys the specified kernel CQ.
3920  * @cq: The CQ to destroy.
3921  *
3922  * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3923  */
3924 static inline void ib_destroy_cq(struct ib_cq *cq)
3925 {
3926 	int ret = ib_destroy_cq_user(cq, NULL);
3927 
3928 	WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3929 }
3930 
3931 /**
3932  * ib_poll_cq - poll a CQ for completion(s)
3933  * @cq:the CQ being polled
3934  * @num_entries:maximum number of completions to return
3935  * @wc:array of at least @num_entries &struct ib_wc where completions
3936  *   will be returned
3937  *
3938  * Poll a CQ for (possibly multiple) completions.  If the return value
3939  * is < 0, an error occurred.  If the return value is >= 0, it is the
3940  * number of completions returned.  If the return value is
3941  * non-negative and < num_entries, then the CQ was emptied.
3942  */
3943 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3944 			     struct ib_wc *wc)
3945 {
3946 	return cq->device->ops.poll_cq(cq, num_entries, wc);
3947 }
3948 
3949 /**
3950  * ib_req_notify_cq - Request completion notification on a CQ.
3951  * @cq: The CQ to generate an event for.
3952  * @flags:
3953  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3954  *   to request an event on the next solicited event or next work
3955  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3956  *   may also be |ed in to request a hint about missed events, as
3957  *   described below.
3958  *
3959  * Return Value:
3960  *    < 0 means an error occurred while requesting notification
3961  *   == 0 means notification was requested successfully, and if
3962  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3963  *        were missed and it is safe to wait for another event.  In
3964  *        this case is it guaranteed that any work completions added
3965  *        to the CQ since the last CQ poll will trigger a completion
3966  *        notification event.
3967  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3968  *        in.  It means that the consumer must poll the CQ again to
3969  *        make sure it is empty to avoid missing an event because of a
3970  *        race between requesting notification and an entry being
3971  *        added to the CQ.  This return value means it is possible
3972  *        (but not guaranteed) that a work completion has been added
3973  *        to the CQ since the last poll without triggering a
3974  *        completion notification event.
3975  */
3976 static inline int ib_req_notify_cq(struct ib_cq *cq,
3977 				   enum ib_cq_notify_flags flags)
3978 {
3979 	return cq->device->ops.req_notify_cq(cq, flags);
3980 }
3981 
3982 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
3983 			     int comp_vector_hint,
3984 			     enum ib_poll_context poll_ctx);
3985 
3986 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
3987 
3988 /*
3989  * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
3990  * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
3991  * address into the dma address.
3992  */
3993 static inline bool ib_uses_virt_dma(struct ib_device *dev)
3994 {
3995 	return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
3996 }
3997 
3998 /**
3999  * ib_dma_mapping_error - check a DMA addr for error
4000  * @dev: The device for which the dma_addr was created
4001  * @dma_addr: The DMA address to check
4002  */
4003 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4004 {
4005 	if (ib_uses_virt_dma(dev))
4006 		return 0;
4007 	return dma_mapping_error(dev->dma_device, dma_addr);
4008 }
4009 
4010 /**
4011  * ib_dma_map_single - Map a kernel virtual address to DMA address
4012  * @dev: The device for which the dma_addr is to be created
4013  * @cpu_addr: The kernel virtual address
4014  * @size: The size of the region in bytes
4015  * @direction: The direction of the DMA
4016  */
4017 static inline u64 ib_dma_map_single(struct ib_device *dev,
4018 				    void *cpu_addr, size_t size,
4019 				    enum dma_data_direction direction)
4020 {
4021 	if (ib_uses_virt_dma(dev))
4022 		return (uintptr_t)cpu_addr;
4023 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4024 }
4025 
4026 /**
4027  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4028  * @dev: The device for which the DMA address was created
4029  * @addr: The DMA address
4030  * @size: The size of the region in bytes
4031  * @direction: The direction of the DMA
4032  */
4033 static inline void ib_dma_unmap_single(struct ib_device *dev,
4034 				       u64 addr, size_t size,
4035 				       enum dma_data_direction direction)
4036 {
4037 	if (!ib_uses_virt_dma(dev))
4038 		dma_unmap_single(dev->dma_device, addr, size, direction);
4039 }
4040 
4041 /**
4042  * ib_dma_map_page - Map a physical page to DMA address
4043  * @dev: The device for which the dma_addr is to be created
4044  * @page: The page to be mapped
4045  * @offset: The offset within the page
4046  * @size: The size of the region in bytes
4047  * @direction: The direction of the DMA
4048  */
4049 static inline u64 ib_dma_map_page(struct ib_device *dev,
4050 				  struct page *page,
4051 				  unsigned long offset,
4052 				  size_t size,
4053 					 enum dma_data_direction direction)
4054 {
4055 	if (ib_uses_virt_dma(dev))
4056 		return (uintptr_t)(page_address(page) + offset);
4057 	return dma_map_page(dev->dma_device, page, offset, size, direction);
4058 }
4059 
4060 /**
4061  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4062  * @dev: The device for which the DMA address was created
4063  * @addr: The DMA address
4064  * @size: The size of the region in bytes
4065  * @direction: The direction of the DMA
4066  */
4067 static inline void ib_dma_unmap_page(struct ib_device *dev,
4068 				     u64 addr, size_t size,
4069 				     enum dma_data_direction direction)
4070 {
4071 	if (!ib_uses_virt_dma(dev))
4072 		dma_unmap_page(dev->dma_device, addr, size, direction);
4073 }
4074 
4075 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4076 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4077 				      struct scatterlist *sg, int nents,
4078 				      enum dma_data_direction direction,
4079 				      unsigned long dma_attrs)
4080 {
4081 	if (ib_uses_virt_dma(dev))
4082 		return ib_dma_virt_map_sg(dev, sg, nents);
4083 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4084 				dma_attrs);
4085 }
4086 
4087 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4088 					 struct scatterlist *sg, int nents,
4089 					 enum dma_data_direction direction,
4090 					 unsigned long dma_attrs)
4091 {
4092 	if (!ib_uses_virt_dma(dev))
4093 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4094 				   dma_attrs);
4095 }
4096 
4097 /**
4098  * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4099  * @dev: The device for which the DMA addresses are to be created
4100  * @sg: The sg_table object describing the buffer
4101  * @direction: The direction of the DMA
4102  * @attrs: Optional DMA attributes for the map operation
4103  */
4104 static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4105 					   struct sg_table *sgt,
4106 					   enum dma_data_direction direction,
4107 					   unsigned long dma_attrs)
4108 {
4109 	int nents;
4110 
4111 	if (ib_uses_virt_dma(dev)) {
4112 		nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4113 		if (!nents)
4114 			return -EIO;
4115 		sgt->nents = nents;
4116 		return 0;
4117 	}
4118 	return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4119 }
4120 
4121 static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4122 					      struct sg_table *sgt,
4123 					      enum dma_data_direction direction,
4124 					      unsigned long dma_attrs)
4125 {
4126 	if (!ib_uses_virt_dma(dev))
4127 		dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4128 }
4129 
4130 /**
4131  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4132  * @dev: The device for which the DMA addresses are to be created
4133  * @sg: The array of scatter/gather entries
4134  * @nents: The number of scatter/gather entries
4135  * @direction: The direction of the DMA
4136  */
4137 static inline int ib_dma_map_sg(struct ib_device *dev,
4138 				struct scatterlist *sg, int nents,
4139 				enum dma_data_direction direction)
4140 {
4141 	return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4142 }
4143 
4144 /**
4145  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4146  * @dev: The device for which the DMA addresses were created
4147  * @sg: The array of scatter/gather entries
4148  * @nents: The number of scatter/gather entries
4149  * @direction: The direction of the DMA
4150  */
4151 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4152 				   struct scatterlist *sg, int nents,
4153 				   enum dma_data_direction direction)
4154 {
4155 	ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4156 }
4157 
4158 /**
4159  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4160  * @dev: The device to query
4161  *
4162  * The returned value represents a size in bytes.
4163  */
4164 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4165 {
4166 	if (ib_uses_virt_dma(dev))
4167 		return UINT_MAX;
4168 	return dma_get_max_seg_size(dev->dma_device);
4169 }
4170 
4171 /**
4172  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4173  * @dev: The device for which the DMA address was created
4174  * @addr: The DMA address
4175  * @size: The size of the region in bytes
4176  * @dir: The direction of the DMA
4177  */
4178 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4179 					      u64 addr,
4180 					      size_t size,
4181 					      enum dma_data_direction dir)
4182 {
4183 	if (!ib_uses_virt_dma(dev))
4184 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4185 }
4186 
4187 /**
4188  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4189  * @dev: The device for which the DMA address was created
4190  * @addr: The DMA address
4191  * @size: The size of the region in bytes
4192  * @dir: The direction of the DMA
4193  */
4194 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4195 						 u64 addr,
4196 						 size_t size,
4197 						 enum dma_data_direction dir)
4198 {
4199 	if (!ib_uses_virt_dma(dev))
4200 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4201 }
4202 
4203 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4204  * space. This function should be called when 'current' is the owning MM.
4205  */
4206 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4207 			     u64 virt_addr, int mr_access_flags);
4208 
4209 /* ib_advise_mr -  give an advice about an address range in a memory region */
4210 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4211 		 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4212 /**
4213  * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4214  *   HCA translation table.
4215  * @mr: The memory region to deregister.
4216  * @udata: Valid user data or NULL for kernel object
4217  *
4218  * This function can fail, if the memory region has memory windows bound to it.
4219  */
4220 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4221 
4222 /**
4223  * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4224  *   HCA translation table.
4225  * @mr: The memory region to deregister.
4226  *
4227  * This function can fail, if the memory region has memory windows bound to it.
4228  *
4229  * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4230  */
4231 static inline int ib_dereg_mr(struct ib_mr *mr)
4232 {
4233 	return ib_dereg_mr_user(mr, NULL);
4234 }
4235 
4236 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4237 			  u32 max_num_sg);
4238 
4239 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4240 				    u32 max_num_data_sg,
4241 				    u32 max_num_meta_sg);
4242 
4243 /**
4244  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4245  *   R_Key and L_Key.
4246  * @mr - struct ib_mr pointer to be updated.
4247  * @newkey - new key to be used.
4248  */
4249 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4250 {
4251 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4252 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4253 }
4254 
4255 /**
4256  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4257  * for calculating a new rkey for type 2 memory windows.
4258  * @rkey - the rkey to increment.
4259  */
4260 static inline u32 ib_inc_rkey(u32 rkey)
4261 {
4262 	const u32 mask = 0x000000ff;
4263 	return ((rkey + 1) & mask) | (rkey & ~mask);
4264 }
4265 
4266 /**
4267  * ib_attach_mcast - Attaches the specified QP to a multicast group.
4268  * @qp: QP to attach to the multicast group.  The QP must be type
4269  *   IB_QPT_UD.
4270  * @gid: Multicast group GID.
4271  * @lid: Multicast group LID in host byte order.
4272  *
4273  * In order to send and receive multicast packets, subnet
4274  * administration must have created the multicast group and configured
4275  * the fabric appropriately.  The port associated with the specified
4276  * QP must also be a member of the multicast group.
4277  */
4278 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4279 
4280 /**
4281  * ib_detach_mcast - Detaches the specified QP from a multicast group.
4282  * @qp: QP to detach from the multicast group.
4283  * @gid: Multicast group GID.
4284  * @lid: Multicast group LID in host byte order.
4285  */
4286 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4287 
4288 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4289 				   struct inode *inode, struct ib_udata *udata);
4290 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4291 
4292 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4293 				     unsigned int flags)
4294 {
4295 	/*
4296 	 * Local write permission is required if remote write or
4297 	 * remote atomic permission is also requested.
4298 	 */
4299 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4300 	    !(flags & IB_ACCESS_LOCAL_WRITE))
4301 		return -EINVAL;
4302 
4303 	if (flags & ~IB_ACCESS_SUPPORTED)
4304 		return -EINVAL;
4305 
4306 	if (flags & IB_ACCESS_ON_DEMAND &&
4307 	    !(ib_dev->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING))
4308 		return -EINVAL;
4309 	return 0;
4310 }
4311 
4312 static inline bool ib_access_writable(int access_flags)
4313 {
4314 	/*
4315 	 * We have writable memory backing the MR if any of the following
4316 	 * access flags are set.  "Local write" and "remote write" obviously
4317 	 * require write access.  "Remote atomic" can do things like fetch and
4318 	 * add, which will modify memory, and "MW bind" can change permissions
4319 	 * by binding a window.
4320 	 */
4321 	return access_flags &
4322 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
4323 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4324 }
4325 
4326 /**
4327  * ib_check_mr_status: lightweight check of MR status.
4328  *     This routine may provide status checks on a selected
4329  *     ib_mr. first use is for signature status check.
4330  *
4331  * @mr: A memory region.
4332  * @check_mask: Bitmask of which checks to perform from
4333  *     ib_mr_status_check enumeration.
4334  * @mr_status: The container of relevant status checks.
4335  *     failed checks will be indicated in the status bitmask
4336  *     and the relevant info shall be in the error item.
4337  */
4338 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4339 		       struct ib_mr_status *mr_status);
4340 
4341 /**
4342  * ib_device_try_get: Hold a registration lock
4343  * device: The device to lock
4344  *
4345  * A device under an active registration lock cannot become unregistered. It
4346  * is only possible to obtain a registration lock on a device that is fully
4347  * registered, otherwise this function returns false.
4348  *
4349  * The registration lock is only necessary for actions which require the
4350  * device to still be registered. Uses that only require the device pointer to
4351  * be valid should use get_device(&ibdev->dev) to hold the memory.
4352  *
4353  */
4354 static inline bool ib_device_try_get(struct ib_device *dev)
4355 {
4356 	return refcount_inc_not_zero(&dev->refcount);
4357 }
4358 
4359 void ib_device_put(struct ib_device *device);
4360 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4361 					  enum rdma_driver_id driver_id);
4362 struct ib_device *ib_device_get_by_name(const char *name,
4363 					enum rdma_driver_id driver_id);
4364 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4365 					    u16 pkey, const union ib_gid *gid,
4366 					    const struct sockaddr *addr);
4367 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4368 			 unsigned int port);
4369 struct net_device *ib_device_netdev(struct ib_device *dev, u32 port);
4370 
4371 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4372 			   struct ib_wq_init_attr *init_attr);
4373 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4374 
4375 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4376 		 unsigned int *sg_offset, unsigned int page_size);
4377 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4378 		    int data_sg_nents, unsigned int *data_sg_offset,
4379 		    struct scatterlist *meta_sg, int meta_sg_nents,
4380 		    unsigned int *meta_sg_offset, unsigned int page_size);
4381 
4382 static inline int
4383 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4384 		  unsigned int *sg_offset, unsigned int page_size)
4385 {
4386 	int n;
4387 
4388 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4389 	mr->iova = 0;
4390 
4391 	return n;
4392 }
4393 
4394 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4395 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4396 
4397 void ib_drain_rq(struct ib_qp *qp);
4398 void ib_drain_sq(struct ib_qp *qp);
4399 void ib_drain_qp(struct ib_qp *qp);
4400 
4401 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4402 		     u8 *width);
4403 
4404 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4405 {
4406 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4407 		return attr->roce.dmac;
4408 	return NULL;
4409 }
4410 
4411 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4412 {
4413 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4414 		attr->ib.dlid = (u16)dlid;
4415 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4416 		attr->opa.dlid = dlid;
4417 }
4418 
4419 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4420 {
4421 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4422 		return attr->ib.dlid;
4423 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4424 		return attr->opa.dlid;
4425 	return 0;
4426 }
4427 
4428 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4429 {
4430 	attr->sl = sl;
4431 }
4432 
4433 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4434 {
4435 	return attr->sl;
4436 }
4437 
4438 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4439 					 u8 src_path_bits)
4440 {
4441 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4442 		attr->ib.src_path_bits = src_path_bits;
4443 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4444 		attr->opa.src_path_bits = src_path_bits;
4445 }
4446 
4447 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4448 {
4449 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4450 		return attr->ib.src_path_bits;
4451 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4452 		return attr->opa.src_path_bits;
4453 	return 0;
4454 }
4455 
4456 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4457 					bool make_grd)
4458 {
4459 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4460 		attr->opa.make_grd = make_grd;
4461 }
4462 
4463 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4464 {
4465 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4466 		return attr->opa.make_grd;
4467 	return false;
4468 }
4469 
4470 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4471 {
4472 	attr->port_num = port_num;
4473 }
4474 
4475 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4476 {
4477 	return attr->port_num;
4478 }
4479 
4480 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4481 					   u8 static_rate)
4482 {
4483 	attr->static_rate = static_rate;
4484 }
4485 
4486 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4487 {
4488 	return attr->static_rate;
4489 }
4490 
4491 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4492 					enum ib_ah_flags flag)
4493 {
4494 	attr->ah_flags = flag;
4495 }
4496 
4497 static inline enum ib_ah_flags
4498 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4499 {
4500 	return attr->ah_flags;
4501 }
4502 
4503 static inline const struct ib_global_route
4504 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4505 {
4506 	return &attr->grh;
4507 }
4508 
4509 /*To retrieve and modify the grh */
4510 static inline struct ib_global_route
4511 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4512 {
4513 	return &attr->grh;
4514 }
4515 
4516 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4517 {
4518 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4519 
4520 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4521 }
4522 
4523 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4524 					     __be64 prefix)
4525 {
4526 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4527 
4528 	grh->dgid.global.subnet_prefix = prefix;
4529 }
4530 
4531 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4532 					    __be64 if_id)
4533 {
4534 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4535 
4536 	grh->dgid.global.interface_id = if_id;
4537 }
4538 
4539 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4540 				   union ib_gid *dgid, u32 flow_label,
4541 				   u8 sgid_index, u8 hop_limit,
4542 				   u8 traffic_class)
4543 {
4544 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4545 
4546 	attr->ah_flags = IB_AH_GRH;
4547 	if (dgid)
4548 		grh->dgid = *dgid;
4549 	grh->flow_label = flow_label;
4550 	grh->sgid_index = sgid_index;
4551 	grh->hop_limit = hop_limit;
4552 	grh->traffic_class = traffic_class;
4553 	grh->sgid_attr = NULL;
4554 }
4555 
4556 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4557 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4558 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4559 			     const struct ib_gid_attr *sgid_attr);
4560 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4561 		       const struct rdma_ah_attr *src);
4562 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4563 			  const struct rdma_ah_attr *new);
4564 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4565 
4566 /**
4567  * rdma_ah_find_type - Return address handle type.
4568  *
4569  * @dev: Device to be checked
4570  * @port_num: Port number
4571  */
4572 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4573 						       u32 port_num)
4574 {
4575 	if (rdma_protocol_roce(dev, port_num))
4576 		return RDMA_AH_ATTR_TYPE_ROCE;
4577 	if (rdma_protocol_ib(dev, port_num)) {
4578 		if (rdma_cap_opa_ah(dev, port_num))
4579 			return RDMA_AH_ATTR_TYPE_OPA;
4580 		return RDMA_AH_ATTR_TYPE_IB;
4581 	}
4582 
4583 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4584 }
4585 
4586 /**
4587  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4588  *     In the current implementation the only way to get
4589  *     get the 32bit lid is from other sources for OPA.
4590  *     For IB, lids will always be 16bits so cast the
4591  *     value accordingly.
4592  *
4593  * @lid: A 32bit LID
4594  */
4595 static inline u16 ib_lid_cpu16(u32 lid)
4596 {
4597 	WARN_ON_ONCE(lid & 0xFFFF0000);
4598 	return (u16)lid;
4599 }
4600 
4601 /**
4602  * ib_lid_be16 - Return lid in 16bit BE encoding.
4603  *
4604  * @lid: A 32bit LID
4605  */
4606 static inline __be16 ib_lid_be16(u32 lid)
4607 {
4608 	WARN_ON_ONCE(lid & 0xFFFF0000);
4609 	return cpu_to_be16((u16)lid);
4610 }
4611 
4612 /**
4613  * ib_get_vector_affinity - Get the affinity mappings of a given completion
4614  *   vector
4615  * @device:         the rdma device
4616  * @comp_vector:    index of completion vector
4617  *
4618  * Returns NULL on failure, otherwise a corresponding cpu map of the
4619  * completion vector (returns all-cpus map if the device driver doesn't
4620  * implement get_vector_affinity).
4621  */
4622 static inline const struct cpumask *
4623 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4624 {
4625 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4626 	    !device->ops.get_vector_affinity)
4627 		return NULL;
4628 
4629 	return device->ops.get_vector_affinity(device, comp_vector);
4630 
4631 }
4632 
4633 /**
4634  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4635  * and add their gids, as needed, to the relevant RoCE devices.
4636  *
4637  * @device:         the rdma device
4638  */
4639 void rdma_roce_rescan_device(struct ib_device *ibdev);
4640 
4641 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4642 
4643 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4644 
4645 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4646 				     enum rdma_netdev_t type, const char *name,
4647 				     unsigned char name_assign_type,
4648 				     void (*setup)(struct net_device *));
4649 
4650 int rdma_init_netdev(struct ib_device *device, u32 port_num,
4651 		     enum rdma_netdev_t type, const char *name,
4652 		     unsigned char name_assign_type,
4653 		     void (*setup)(struct net_device *),
4654 		     struct net_device *netdev);
4655 
4656 /**
4657  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4658  *
4659  * @device:	device pointer for which ib_device pointer to retrieve
4660  *
4661  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4662  *
4663  */
4664 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4665 {
4666 	struct ib_core_device *coredev =
4667 		container_of(device, struct ib_core_device, dev);
4668 
4669 	return coredev->owner;
4670 }
4671 
4672 /**
4673  * ibdev_to_node - return the NUMA node for a given ib_device
4674  * @dev:	device to get the NUMA node for.
4675  */
4676 static inline int ibdev_to_node(struct ib_device *ibdev)
4677 {
4678 	struct device *parent = ibdev->dev.parent;
4679 
4680 	if (!parent)
4681 		return NUMA_NO_NODE;
4682 	return dev_to_node(parent);
4683 }
4684 
4685 /**
4686  * rdma_device_to_drv_device - Helper macro to reach back to driver's
4687  *			       ib_device holder structure from device pointer.
4688  *
4689  * NOTE: New drivers should not make use of this API; This API is only for
4690  * existing drivers who have exposed sysfs entries using
4691  * ops->device_group.
4692  */
4693 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
4694 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4695 
4696 bool rdma_dev_access_netns(const struct ib_device *device,
4697 			   const struct net *net);
4698 
4699 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4700 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4701 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4702 
4703 /**
4704  * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4705  *                               on the flow_label
4706  *
4707  * This function will convert the 20 bit flow_label input to a valid RoCE v2
4708  * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4709  * convention.
4710  */
4711 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4712 {
4713 	u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4714 
4715 	fl_low ^= fl_high >> 14;
4716 	return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4717 }
4718 
4719 /**
4720  * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4721  *                        local and remote qpn values
4722  *
4723  * This function folded the multiplication results of two qpns, 24 bit each,
4724  * fields, and converts it to a 20 bit results.
4725  *
4726  * This function will create symmetric flow_label value based on the local
4727  * and remote qpn values. this will allow both the requester and responder
4728  * to calculate the same flow_label for a given connection.
4729  *
4730  * This helper function should be used by driver in case the upper layer
4731  * provide a zero flow_label value. This is to improve entropy of RDMA
4732  * traffic in the network.
4733  */
4734 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4735 {
4736 	u64 v = (u64)lqpn * rqpn;
4737 
4738 	v ^= v >> 20;
4739 	v ^= v >> 40;
4740 
4741 	return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4742 }
4743 
4744 /**
4745  * rdma_get_udp_sport - Calculate and set UDP source port based on the flow
4746  *                      label. If flow label is not defined in GRH then
4747  *                      calculate it based on lqpn/rqpn.
4748  *
4749  * @fl:                 flow label from GRH
4750  * @lqpn:               local qp number
4751  * @rqpn:               remote qp number
4752  */
4753 static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
4754 {
4755 	if (!fl)
4756 		fl = rdma_calc_flow_label(lqpn, rqpn);
4757 
4758 	return rdma_flow_label_to_udp_sport(fl);
4759 }
4760 
4761 const struct ib_port_immutable*
4762 ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4763 #endif /* IB_VERBS_H */
4764