xref: /linux/include/rdma/ib_verbs.h (revision d5efc2e6b98fe661dbd8dd0d5d5bfb961728e57a)
1 /*
2  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
3  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
4  * Copyright (c) 2004 Intel Corporation.  All rights reserved.
5  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
6  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
7  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
9  *
10  * This software is available to you under a choice of one of two
11  * licenses.  You may choose to be licensed under the terms of the GNU
12  * General Public License (GPL) Version 2, available from the file
13  * COPYING in the main directory of this source tree, or the
14  * OpenIB.org BSD license below:
15  *
16  *     Redistribution and use in source and binary forms, with or
17  *     without modification, are permitted provided that the following
18  *     conditions are met:
19  *
20  *      - Redistributions of source code must retain the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer.
23  *
24  *      - Redistributions in binary form must reproduce the above
25  *        copyright notice, this list of conditions and the following
26  *        disclaimer in the documentation and/or other materials
27  *        provided with the distribution.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36  * SOFTWARE.
37  */
38 
39 #if !defined(IB_VERBS_H)
40 #define IB_VERBS_H
41 
42 #include <linux/types.h>
43 #include <linux/device.h>
44 #include <linux/dma-mapping.h>
45 #include <linux/kref.h>
46 #include <linux/list.h>
47 #include <linux/rwsem.h>
48 #include <linux/workqueue.h>
49 #include <linux/irq_poll.h>
50 #include <uapi/linux/if_ether.h>
51 #include <net/ipv6.h>
52 #include <net/ip.h>
53 #include <linux/string.h>
54 #include <linux/slab.h>
55 #include <linux/netdevice.h>
56 #include <linux/refcount.h>
57 #include <linux/if_link.h>
58 #include <linux/atomic.h>
59 #include <linux/mmu_notifier.h>
60 #include <linux/uaccess.h>
61 #include <linux/cgroup_rdma.h>
62 #include <linux/irqflags.h>
63 #include <linux/preempt.h>
64 #include <linux/dim.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 #include <rdma/rdma_counter.h>
67 #include <rdma/restrack.h>
68 #include <rdma/signature.h>
69 #include <uapi/rdma/rdma_user_ioctl.h>
70 #include <uapi/rdma/ib_user_ioctl_verbs.h>
71 
72 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
73 
74 struct ib_umem_odp;
75 struct ib_uqp_object;
76 struct ib_usrq_object;
77 struct ib_uwq_object;
78 
79 extern struct workqueue_struct *ib_wq;
80 extern struct workqueue_struct *ib_comp_wq;
81 extern struct workqueue_struct *ib_comp_unbound_wq;
82 
83 struct ib_ucq_object;
84 
85 __printf(3, 4) __cold
86 void ibdev_printk(const char *level, const struct ib_device *ibdev,
87 		  const char *format, ...);
88 __printf(2, 3) __cold
89 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
90 __printf(2, 3) __cold
91 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
92 __printf(2, 3) __cold
93 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
94 __printf(2, 3) __cold
95 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
96 __printf(2, 3) __cold
97 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
98 __printf(2, 3) __cold
99 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
100 __printf(2, 3) __cold
101 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
102 
103 #if defined(CONFIG_DYNAMIC_DEBUG) || \
104 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
105 #define ibdev_dbg(__dev, format, args...)                       \
106 	dynamic_ibdev_dbg(__dev, format, ##args)
107 #else
108 __printf(2, 3) __cold
109 static inline
110 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
111 #endif
112 
113 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...)           \
114 do {                                                                    \
115 	static DEFINE_RATELIMIT_STATE(_rs,                              \
116 				      DEFAULT_RATELIMIT_INTERVAL,       \
117 				      DEFAULT_RATELIMIT_BURST);         \
118 	if (__ratelimit(&_rs))                                          \
119 		ibdev_level(ibdev, fmt, ##__VA_ARGS__);                 \
120 } while (0)
121 
122 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
123 	ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
124 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
125 	ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
126 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
127 	ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
128 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
129 	ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
130 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
131 	ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
132 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
133 	ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
134 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
135 	ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
136 
137 #if defined(CONFIG_DYNAMIC_DEBUG) || \
138 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
139 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
140 #define ibdev_dbg_ratelimited(ibdev, fmt, ...)                          \
141 do {                                                                    \
142 	static DEFINE_RATELIMIT_STATE(_rs,                              \
143 				      DEFAULT_RATELIMIT_INTERVAL,       \
144 				      DEFAULT_RATELIMIT_BURST);         \
145 	DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt);                 \
146 	if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs))      \
147 		__dynamic_ibdev_dbg(&descriptor, ibdev, fmt,            \
148 				    ##__VA_ARGS__);                     \
149 } while (0)
150 #else
151 __printf(2, 3) __cold
152 static inline
153 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
154 #endif
155 
156 union ib_gid {
157 	u8	raw[16];
158 	struct {
159 		__be64	subnet_prefix;
160 		__be64	interface_id;
161 	} global;
162 };
163 
164 extern union ib_gid zgid;
165 
166 enum ib_gid_type {
167 	/* If link layer is Ethernet, this is RoCE V1 */
168 	IB_GID_TYPE_IB        = 0,
169 	IB_GID_TYPE_ROCE      = 0,
170 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
171 	IB_GID_TYPE_SIZE
172 };
173 
174 #define ROCE_V2_UDP_DPORT      4791
175 struct ib_gid_attr {
176 	struct net_device __rcu	*ndev;
177 	struct ib_device	*device;
178 	union ib_gid		gid;
179 	enum ib_gid_type	gid_type;
180 	u16			index;
181 	u8			port_num;
182 };
183 
184 enum {
185 	/* set the local administered indication */
186 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
187 };
188 
189 enum rdma_transport_type {
190 	RDMA_TRANSPORT_IB,
191 	RDMA_TRANSPORT_IWARP,
192 	RDMA_TRANSPORT_USNIC,
193 	RDMA_TRANSPORT_USNIC_UDP,
194 	RDMA_TRANSPORT_UNSPECIFIED,
195 };
196 
197 enum rdma_protocol_type {
198 	RDMA_PROTOCOL_IB,
199 	RDMA_PROTOCOL_IBOE,
200 	RDMA_PROTOCOL_IWARP,
201 	RDMA_PROTOCOL_USNIC_UDP
202 };
203 
204 __attribute_const__ enum rdma_transport_type
205 rdma_node_get_transport(unsigned int node_type);
206 
207 enum rdma_network_type {
208 	RDMA_NETWORK_IB,
209 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
210 	RDMA_NETWORK_IPV4,
211 	RDMA_NETWORK_IPV6
212 };
213 
214 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
215 {
216 	if (network_type == RDMA_NETWORK_IPV4 ||
217 	    network_type == RDMA_NETWORK_IPV6)
218 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
219 
220 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
221 	return IB_GID_TYPE_IB;
222 }
223 
224 static inline enum rdma_network_type
225 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
226 {
227 	if (attr->gid_type == IB_GID_TYPE_IB)
228 		return RDMA_NETWORK_IB;
229 
230 	if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
231 		return RDMA_NETWORK_IPV4;
232 	else
233 		return RDMA_NETWORK_IPV6;
234 }
235 
236 enum rdma_link_layer {
237 	IB_LINK_LAYER_UNSPECIFIED,
238 	IB_LINK_LAYER_INFINIBAND,
239 	IB_LINK_LAYER_ETHERNET,
240 };
241 
242 enum ib_device_cap_flags {
243 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
244 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
245 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
246 	IB_DEVICE_RAW_MULTI			= (1 << 3),
247 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
248 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
249 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
250 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
251 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
252 	/* Not in use, former INIT_TYPE		= (1 << 9),*/
253 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
254 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
255 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
256 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
257 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
258 
259 	/*
260 	 * This device supports a per-device lkey or stag that can be
261 	 * used without performing a memory registration for the local
262 	 * memory.  Note that ULPs should never check this flag, but
263 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
264 	 * which will always contain a usable lkey.
265 	 */
266 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
267 	/* Reserved, old SEND_W_INV		= (1 << 16),*/
268 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
269 	/*
270 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
271 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
272 	 * messages and can verify the validity of checksum for
273 	 * incoming messages.  Setting this flag implies that the
274 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
275 	 */
276 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
277 	IB_DEVICE_UD_TSO			= (1 << 19),
278 	IB_DEVICE_XRC				= (1 << 20),
279 
280 	/*
281 	 * This device supports the IB "base memory management extension",
282 	 * which includes support for fast registrations (IB_WR_REG_MR,
283 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
284 	 * also be set by any iWarp device which must support FRs to comply
285 	 * to the iWarp verbs spec.  iWarp devices also support the
286 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
287 	 * stag.
288 	 */
289 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
290 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
291 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
292 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
293 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
294 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
295 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
296 	/*
297 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
298 	 * support execution of WQEs that involve synchronization
299 	 * of I/O operations with single completion queue managed
300 	 * by hardware.
301 	 */
302 	IB_DEVICE_CROSS_CHANNEL			= (1 << 27),
303 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
304 	IB_DEVICE_INTEGRITY_HANDOVER		= (1 << 30),
305 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
306 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
307 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
308 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
309 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
310 	IB_DEVICE_RDMA_NETDEV_OPA		= (1ULL << 35),
311 	/* The device supports padding incoming writes to cacheline. */
312 	IB_DEVICE_PCI_WRITE_END_PADDING		= (1ULL << 36),
313 	IB_DEVICE_ALLOW_USER_UNREG		= (1ULL << 37),
314 };
315 
316 enum ib_atomic_cap {
317 	IB_ATOMIC_NONE,
318 	IB_ATOMIC_HCA,
319 	IB_ATOMIC_GLOB
320 };
321 
322 enum ib_odp_general_cap_bits {
323 	IB_ODP_SUPPORT		= 1 << 0,
324 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
325 };
326 
327 enum ib_odp_transport_cap_bits {
328 	IB_ODP_SUPPORT_SEND	= 1 << 0,
329 	IB_ODP_SUPPORT_RECV	= 1 << 1,
330 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
331 	IB_ODP_SUPPORT_READ	= 1 << 3,
332 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
333 	IB_ODP_SUPPORT_SRQ_RECV	= 1 << 5,
334 };
335 
336 struct ib_odp_caps {
337 	uint64_t general_caps;
338 	struct {
339 		uint32_t  rc_odp_caps;
340 		uint32_t  uc_odp_caps;
341 		uint32_t  ud_odp_caps;
342 		uint32_t  xrc_odp_caps;
343 	} per_transport_caps;
344 };
345 
346 struct ib_rss_caps {
347 	/* Corresponding bit will be set if qp type from
348 	 * 'enum ib_qp_type' is supported, e.g.
349 	 * supported_qpts |= 1 << IB_QPT_UD
350 	 */
351 	u32 supported_qpts;
352 	u32 max_rwq_indirection_tables;
353 	u32 max_rwq_indirection_table_size;
354 };
355 
356 enum ib_tm_cap_flags {
357 	/*  Support tag matching with rendezvous offload for RC transport */
358 	IB_TM_CAP_RNDV_RC = 1 << 0,
359 };
360 
361 struct ib_tm_caps {
362 	/* Max size of RNDV header */
363 	u32 max_rndv_hdr_size;
364 	/* Max number of entries in tag matching list */
365 	u32 max_num_tags;
366 	/* From enum ib_tm_cap_flags */
367 	u32 flags;
368 	/* Max number of outstanding list operations */
369 	u32 max_ops;
370 	/* Max number of SGE in tag matching entry */
371 	u32 max_sge;
372 };
373 
374 struct ib_cq_init_attr {
375 	unsigned int	cqe;
376 	u32		comp_vector;
377 	u32		flags;
378 };
379 
380 enum ib_cq_attr_mask {
381 	IB_CQ_MODERATE = 1 << 0,
382 };
383 
384 struct ib_cq_caps {
385 	u16     max_cq_moderation_count;
386 	u16     max_cq_moderation_period;
387 };
388 
389 struct ib_dm_mr_attr {
390 	u64		length;
391 	u64		offset;
392 	u32		access_flags;
393 };
394 
395 struct ib_dm_alloc_attr {
396 	u64	length;
397 	u32	alignment;
398 	u32	flags;
399 };
400 
401 struct ib_device_attr {
402 	u64			fw_ver;
403 	__be64			sys_image_guid;
404 	u64			max_mr_size;
405 	u64			page_size_cap;
406 	u32			vendor_id;
407 	u32			vendor_part_id;
408 	u32			hw_ver;
409 	int			max_qp;
410 	int			max_qp_wr;
411 	u64			device_cap_flags;
412 	int			max_send_sge;
413 	int			max_recv_sge;
414 	int			max_sge_rd;
415 	int			max_cq;
416 	int			max_cqe;
417 	int			max_mr;
418 	int			max_pd;
419 	int			max_qp_rd_atom;
420 	int			max_ee_rd_atom;
421 	int			max_res_rd_atom;
422 	int			max_qp_init_rd_atom;
423 	int			max_ee_init_rd_atom;
424 	enum ib_atomic_cap	atomic_cap;
425 	enum ib_atomic_cap	masked_atomic_cap;
426 	int			max_ee;
427 	int			max_rdd;
428 	int			max_mw;
429 	int			max_raw_ipv6_qp;
430 	int			max_raw_ethy_qp;
431 	int			max_mcast_grp;
432 	int			max_mcast_qp_attach;
433 	int			max_total_mcast_qp_attach;
434 	int			max_ah;
435 	int			max_srq;
436 	int			max_srq_wr;
437 	int			max_srq_sge;
438 	unsigned int		max_fast_reg_page_list_len;
439 	unsigned int		max_pi_fast_reg_page_list_len;
440 	u16			max_pkeys;
441 	u8			local_ca_ack_delay;
442 	int			sig_prot_cap;
443 	int			sig_guard_cap;
444 	struct ib_odp_caps	odp_caps;
445 	uint64_t		timestamp_mask;
446 	uint64_t		hca_core_clock; /* in KHZ */
447 	struct ib_rss_caps	rss_caps;
448 	u32			max_wq_type_rq;
449 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
450 	struct ib_tm_caps	tm_caps;
451 	struct ib_cq_caps       cq_caps;
452 	u64			max_dm_size;
453 	/* Max entries for sgl for optimized performance per READ */
454 	u32			max_sgl_rd;
455 };
456 
457 enum ib_mtu {
458 	IB_MTU_256  = 1,
459 	IB_MTU_512  = 2,
460 	IB_MTU_1024 = 3,
461 	IB_MTU_2048 = 4,
462 	IB_MTU_4096 = 5
463 };
464 
465 enum opa_mtu {
466 	OPA_MTU_8192 = 6,
467 	OPA_MTU_10240 = 7
468 };
469 
470 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
471 {
472 	switch (mtu) {
473 	case IB_MTU_256:  return  256;
474 	case IB_MTU_512:  return  512;
475 	case IB_MTU_1024: return 1024;
476 	case IB_MTU_2048: return 2048;
477 	case IB_MTU_4096: return 4096;
478 	default: 	  return -1;
479 	}
480 }
481 
482 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
483 {
484 	if (mtu >= 4096)
485 		return IB_MTU_4096;
486 	else if (mtu >= 2048)
487 		return IB_MTU_2048;
488 	else if (mtu >= 1024)
489 		return IB_MTU_1024;
490 	else if (mtu >= 512)
491 		return IB_MTU_512;
492 	else
493 		return IB_MTU_256;
494 }
495 
496 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
497 {
498 	switch (mtu) {
499 	case OPA_MTU_8192:
500 		return 8192;
501 	case OPA_MTU_10240:
502 		return 10240;
503 	default:
504 		return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
505 	}
506 }
507 
508 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
509 {
510 	if (mtu >= 10240)
511 		return OPA_MTU_10240;
512 	else if (mtu >= 8192)
513 		return OPA_MTU_8192;
514 	else
515 		return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
516 }
517 
518 enum ib_port_state {
519 	IB_PORT_NOP		= 0,
520 	IB_PORT_DOWN		= 1,
521 	IB_PORT_INIT		= 2,
522 	IB_PORT_ARMED		= 3,
523 	IB_PORT_ACTIVE		= 4,
524 	IB_PORT_ACTIVE_DEFER	= 5
525 };
526 
527 enum ib_port_phys_state {
528 	IB_PORT_PHYS_STATE_SLEEP = 1,
529 	IB_PORT_PHYS_STATE_POLLING = 2,
530 	IB_PORT_PHYS_STATE_DISABLED = 3,
531 	IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
532 	IB_PORT_PHYS_STATE_LINK_UP = 5,
533 	IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
534 	IB_PORT_PHYS_STATE_PHY_TEST = 7,
535 };
536 
537 enum ib_port_width {
538 	IB_WIDTH_1X	= 1,
539 	IB_WIDTH_2X	= 16,
540 	IB_WIDTH_4X	= 2,
541 	IB_WIDTH_8X	= 4,
542 	IB_WIDTH_12X	= 8
543 };
544 
545 static inline int ib_width_enum_to_int(enum ib_port_width width)
546 {
547 	switch (width) {
548 	case IB_WIDTH_1X:  return  1;
549 	case IB_WIDTH_2X:  return  2;
550 	case IB_WIDTH_4X:  return  4;
551 	case IB_WIDTH_8X:  return  8;
552 	case IB_WIDTH_12X: return 12;
553 	default: 	  return -1;
554 	}
555 }
556 
557 enum ib_port_speed {
558 	IB_SPEED_SDR	= 1,
559 	IB_SPEED_DDR	= 2,
560 	IB_SPEED_QDR	= 4,
561 	IB_SPEED_FDR10	= 8,
562 	IB_SPEED_FDR	= 16,
563 	IB_SPEED_EDR	= 32,
564 	IB_SPEED_HDR	= 64
565 };
566 
567 /**
568  * struct rdma_hw_stats
569  * @lock - Mutex to protect parallel write access to lifespan and values
570  *    of counters, which are 64bits and not guaranteeed to be written
571  *    atomicaly on 32bits systems.
572  * @timestamp - Used by the core code to track when the last update was
573  * @lifespan - Used by the core code to determine how old the counters
574  *   should be before being updated again.  Stored in jiffies, defaults
575  *   to 10 milliseconds, drivers can override the default be specifying
576  *   their own value during their allocation routine.
577  * @name - Array of pointers to static names used for the counters in
578  *   directory.
579  * @num_counters - How many hardware counters there are.  If name is
580  *   shorter than this number, a kernel oops will result.  Driver authors
581  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
582  *   in their code to prevent this.
583  * @value - Array of u64 counters that are accessed by the sysfs code and
584  *   filled in by the drivers get_stats routine
585  */
586 struct rdma_hw_stats {
587 	struct mutex	lock; /* Protect lifespan and values[] */
588 	unsigned long	timestamp;
589 	unsigned long	lifespan;
590 	const char * const *names;
591 	int		num_counters;
592 	u64		value[];
593 };
594 
595 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
596 /**
597  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
598  *   for drivers.
599  * @names - Array of static const char *
600  * @num_counters - How many elements in array
601  * @lifespan - How many milliseconds between updates
602  */
603 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
604 		const char * const *names, int num_counters,
605 		unsigned long lifespan)
606 {
607 	struct rdma_hw_stats *stats;
608 
609 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
610 			GFP_KERNEL);
611 	if (!stats)
612 		return NULL;
613 	stats->names = names;
614 	stats->num_counters = num_counters;
615 	stats->lifespan = msecs_to_jiffies(lifespan);
616 
617 	return stats;
618 }
619 
620 
621 /* Define bits for the various functionality this port needs to be supported by
622  * the core.
623  */
624 /* Management                           0x00000FFF */
625 #define RDMA_CORE_CAP_IB_MAD            0x00000001
626 #define RDMA_CORE_CAP_IB_SMI            0x00000002
627 #define RDMA_CORE_CAP_IB_CM             0x00000004
628 #define RDMA_CORE_CAP_IW_CM             0x00000008
629 #define RDMA_CORE_CAP_IB_SA             0x00000010
630 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
631 
632 /* Address format                       0x000FF000 */
633 #define RDMA_CORE_CAP_AF_IB             0x00001000
634 #define RDMA_CORE_CAP_ETH_AH            0x00002000
635 #define RDMA_CORE_CAP_OPA_AH            0x00004000
636 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
637 
638 /* Protocol                             0xFFF00000 */
639 #define RDMA_CORE_CAP_PROT_IB           0x00100000
640 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
641 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
642 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
643 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
644 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
645 
646 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
647 					| RDMA_CORE_CAP_PROT_ROCE     \
648 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
649 
650 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
651 					| RDMA_CORE_CAP_IB_MAD \
652 					| RDMA_CORE_CAP_IB_SMI \
653 					| RDMA_CORE_CAP_IB_CM  \
654 					| RDMA_CORE_CAP_IB_SA  \
655 					| RDMA_CORE_CAP_AF_IB)
656 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
657 					| RDMA_CORE_CAP_IB_MAD  \
658 					| RDMA_CORE_CAP_IB_CM   \
659 					| RDMA_CORE_CAP_AF_IB   \
660 					| RDMA_CORE_CAP_ETH_AH)
661 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
662 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
663 					| RDMA_CORE_CAP_IB_MAD  \
664 					| RDMA_CORE_CAP_IB_CM   \
665 					| RDMA_CORE_CAP_AF_IB   \
666 					| RDMA_CORE_CAP_ETH_AH)
667 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
668 					| RDMA_CORE_CAP_IW_CM)
669 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
670 					| RDMA_CORE_CAP_OPA_MAD)
671 
672 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
673 
674 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
675 
676 struct ib_port_attr {
677 	u64			subnet_prefix;
678 	enum ib_port_state	state;
679 	enum ib_mtu		max_mtu;
680 	enum ib_mtu		active_mtu;
681 	u32                     phys_mtu;
682 	int			gid_tbl_len;
683 	unsigned int		ip_gids:1;
684 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
685 	u32			port_cap_flags;
686 	u32			max_msg_sz;
687 	u32			bad_pkey_cntr;
688 	u32			qkey_viol_cntr;
689 	u16			pkey_tbl_len;
690 	u32			sm_lid;
691 	u32			lid;
692 	u8			lmc;
693 	u8			max_vl_num;
694 	u8			sm_sl;
695 	u8			subnet_timeout;
696 	u8			init_type_reply;
697 	u8			active_width;
698 	u8			active_speed;
699 	u8                      phys_state;
700 	u16			port_cap_flags2;
701 };
702 
703 enum ib_device_modify_flags {
704 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
705 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
706 };
707 
708 #define IB_DEVICE_NODE_DESC_MAX 64
709 
710 struct ib_device_modify {
711 	u64	sys_image_guid;
712 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
713 };
714 
715 enum ib_port_modify_flags {
716 	IB_PORT_SHUTDOWN		= 1,
717 	IB_PORT_INIT_TYPE		= (1<<2),
718 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
719 	IB_PORT_OPA_MASK_CHG		= (1<<4)
720 };
721 
722 struct ib_port_modify {
723 	u32	set_port_cap_mask;
724 	u32	clr_port_cap_mask;
725 	u8	init_type;
726 };
727 
728 enum ib_event_type {
729 	IB_EVENT_CQ_ERR,
730 	IB_EVENT_QP_FATAL,
731 	IB_EVENT_QP_REQ_ERR,
732 	IB_EVENT_QP_ACCESS_ERR,
733 	IB_EVENT_COMM_EST,
734 	IB_EVENT_SQ_DRAINED,
735 	IB_EVENT_PATH_MIG,
736 	IB_EVENT_PATH_MIG_ERR,
737 	IB_EVENT_DEVICE_FATAL,
738 	IB_EVENT_PORT_ACTIVE,
739 	IB_EVENT_PORT_ERR,
740 	IB_EVENT_LID_CHANGE,
741 	IB_EVENT_PKEY_CHANGE,
742 	IB_EVENT_SM_CHANGE,
743 	IB_EVENT_SRQ_ERR,
744 	IB_EVENT_SRQ_LIMIT_REACHED,
745 	IB_EVENT_QP_LAST_WQE_REACHED,
746 	IB_EVENT_CLIENT_REREGISTER,
747 	IB_EVENT_GID_CHANGE,
748 	IB_EVENT_WQ_FATAL,
749 };
750 
751 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
752 
753 struct ib_event {
754 	struct ib_device	*device;
755 	union {
756 		struct ib_cq	*cq;
757 		struct ib_qp	*qp;
758 		struct ib_srq	*srq;
759 		struct ib_wq	*wq;
760 		u8		port_num;
761 	} element;
762 	enum ib_event_type	event;
763 };
764 
765 struct ib_event_handler {
766 	struct ib_device *device;
767 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
768 	struct list_head  list;
769 };
770 
771 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
772 	do {							\
773 		(_ptr)->device  = _device;			\
774 		(_ptr)->handler = _handler;			\
775 		INIT_LIST_HEAD(&(_ptr)->list);			\
776 	} while (0)
777 
778 struct ib_global_route {
779 	const struct ib_gid_attr *sgid_attr;
780 	union ib_gid	dgid;
781 	u32		flow_label;
782 	u8		sgid_index;
783 	u8		hop_limit;
784 	u8		traffic_class;
785 };
786 
787 struct ib_grh {
788 	__be32		version_tclass_flow;
789 	__be16		paylen;
790 	u8		next_hdr;
791 	u8		hop_limit;
792 	union ib_gid	sgid;
793 	union ib_gid	dgid;
794 };
795 
796 union rdma_network_hdr {
797 	struct ib_grh ibgrh;
798 	struct {
799 		/* The IB spec states that if it's IPv4, the header
800 		 * is located in the last 20 bytes of the header.
801 		 */
802 		u8		reserved[20];
803 		struct iphdr	roce4grh;
804 	};
805 };
806 
807 #define IB_QPN_MASK		0xFFFFFF
808 
809 enum {
810 	IB_MULTICAST_QPN = 0xffffff
811 };
812 
813 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
814 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
815 
816 enum ib_ah_flags {
817 	IB_AH_GRH	= 1
818 };
819 
820 enum ib_rate {
821 	IB_RATE_PORT_CURRENT = 0,
822 	IB_RATE_2_5_GBPS = 2,
823 	IB_RATE_5_GBPS   = 5,
824 	IB_RATE_10_GBPS  = 3,
825 	IB_RATE_20_GBPS  = 6,
826 	IB_RATE_30_GBPS  = 4,
827 	IB_RATE_40_GBPS  = 7,
828 	IB_RATE_60_GBPS  = 8,
829 	IB_RATE_80_GBPS  = 9,
830 	IB_RATE_120_GBPS = 10,
831 	IB_RATE_14_GBPS  = 11,
832 	IB_RATE_56_GBPS  = 12,
833 	IB_RATE_112_GBPS = 13,
834 	IB_RATE_168_GBPS = 14,
835 	IB_RATE_25_GBPS  = 15,
836 	IB_RATE_100_GBPS = 16,
837 	IB_RATE_200_GBPS = 17,
838 	IB_RATE_300_GBPS = 18,
839 	IB_RATE_28_GBPS  = 19,
840 	IB_RATE_50_GBPS  = 20,
841 	IB_RATE_400_GBPS = 21,
842 	IB_RATE_600_GBPS = 22,
843 };
844 
845 /**
846  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
847  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
848  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
849  * @rate: rate to convert.
850  */
851 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
852 
853 /**
854  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
855  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
856  * @rate: rate to convert.
857  */
858 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
859 
860 
861 /**
862  * enum ib_mr_type - memory region type
863  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
864  *                            normal registration
865  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
866  *                            register any arbitrary sg lists (without
867  *                            the normal mr constraints - see
868  *                            ib_map_mr_sg)
869  * @IB_MR_TYPE_DM:            memory region that is used for device
870  *                            memory registration
871  * @IB_MR_TYPE_USER:          memory region that is used for the user-space
872  *                            application
873  * @IB_MR_TYPE_DMA:           memory region that is used for DMA operations
874  *                            without address translations (VA=PA)
875  * @IB_MR_TYPE_INTEGRITY:     memory region that is used for
876  *                            data integrity operations
877  */
878 enum ib_mr_type {
879 	IB_MR_TYPE_MEM_REG,
880 	IB_MR_TYPE_SG_GAPS,
881 	IB_MR_TYPE_DM,
882 	IB_MR_TYPE_USER,
883 	IB_MR_TYPE_DMA,
884 	IB_MR_TYPE_INTEGRITY,
885 };
886 
887 enum ib_mr_status_check {
888 	IB_MR_CHECK_SIG_STATUS = 1,
889 };
890 
891 /**
892  * struct ib_mr_status - Memory region status container
893  *
894  * @fail_status: Bitmask of MR checks status. For each
895  *     failed check a corresponding status bit is set.
896  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
897  *     failure.
898  */
899 struct ib_mr_status {
900 	u32		    fail_status;
901 	struct ib_sig_err   sig_err;
902 };
903 
904 /**
905  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
906  * enum.
907  * @mult: multiple to convert.
908  */
909 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
910 
911 struct rdma_ah_init_attr {
912 	struct rdma_ah_attr *ah_attr;
913 	u32 flags;
914 	struct net_device *xmit_slave;
915 };
916 
917 enum rdma_ah_attr_type {
918 	RDMA_AH_ATTR_TYPE_UNDEFINED,
919 	RDMA_AH_ATTR_TYPE_IB,
920 	RDMA_AH_ATTR_TYPE_ROCE,
921 	RDMA_AH_ATTR_TYPE_OPA,
922 };
923 
924 struct ib_ah_attr {
925 	u16			dlid;
926 	u8			src_path_bits;
927 };
928 
929 struct roce_ah_attr {
930 	u8			dmac[ETH_ALEN];
931 };
932 
933 struct opa_ah_attr {
934 	u32			dlid;
935 	u8			src_path_bits;
936 	bool			make_grd;
937 };
938 
939 struct rdma_ah_attr {
940 	struct ib_global_route	grh;
941 	u8			sl;
942 	u8			static_rate;
943 	u8			port_num;
944 	u8			ah_flags;
945 	enum rdma_ah_attr_type type;
946 	union {
947 		struct ib_ah_attr ib;
948 		struct roce_ah_attr roce;
949 		struct opa_ah_attr opa;
950 	};
951 };
952 
953 enum ib_wc_status {
954 	IB_WC_SUCCESS,
955 	IB_WC_LOC_LEN_ERR,
956 	IB_WC_LOC_QP_OP_ERR,
957 	IB_WC_LOC_EEC_OP_ERR,
958 	IB_WC_LOC_PROT_ERR,
959 	IB_WC_WR_FLUSH_ERR,
960 	IB_WC_MW_BIND_ERR,
961 	IB_WC_BAD_RESP_ERR,
962 	IB_WC_LOC_ACCESS_ERR,
963 	IB_WC_REM_INV_REQ_ERR,
964 	IB_WC_REM_ACCESS_ERR,
965 	IB_WC_REM_OP_ERR,
966 	IB_WC_RETRY_EXC_ERR,
967 	IB_WC_RNR_RETRY_EXC_ERR,
968 	IB_WC_LOC_RDD_VIOL_ERR,
969 	IB_WC_REM_INV_RD_REQ_ERR,
970 	IB_WC_REM_ABORT_ERR,
971 	IB_WC_INV_EECN_ERR,
972 	IB_WC_INV_EEC_STATE_ERR,
973 	IB_WC_FATAL_ERR,
974 	IB_WC_RESP_TIMEOUT_ERR,
975 	IB_WC_GENERAL_ERR
976 };
977 
978 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
979 
980 enum ib_wc_opcode {
981 	IB_WC_SEND,
982 	IB_WC_RDMA_WRITE,
983 	IB_WC_RDMA_READ,
984 	IB_WC_COMP_SWAP,
985 	IB_WC_FETCH_ADD,
986 	IB_WC_LSO,
987 	IB_WC_LOCAL_INV,
988 	IB_WC_REG_MR,
989 	IB_WC_MASKED_COMP_SWAP,
990 	IB_WC_MASKED_FETCH_ADD,
991 /*
992  * Set value of IB_WC_RECV so consumers can test if a completion is a
993  * receive by testing (opcode & IB_WC_RECV).
994  */
995 	IB_WC_RECV			= 1 << 7,
996 	IB_WC_RECV_RDMA_WITH_IMM
997 };
998 
999 enum ib_wc_flags {
1000 	IB_WC_GRH		= 1,
1001 	IB_WC_WITH_IMM		= (1<<1),
1002 	IB_WC_WITH_INVALIDATE	= (1<<2),
1003 	IB_WC_IP_CSUM_OK	= (1<<3),
1004 	IB_WC_WITH_SMAC		= (1<<4),
1005 	IB_WC_WITH_VLAN		= (1<<5),
1006 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
1007 };
1008 
1009 struct ib_wc {
1010 	union {
1011 		u64		wr_id;
1012 		struct ib_cqe	*wr_cqe;
1013 	};
1014 	enum ib_wc_status	status;
1015 	enum ib_wc_opcode	opcode;
1016 	u32			vendor_err;
1017 	u32			byte_len;
1018 	struct ib_qp	       *qp;
1019 	union {
1020 		__be32		imm_data;
1021 		u32		invalidate_rkey;
1022 	} ex;
1023 	u32			src_qp;
1024 	u32			slid;
1025 	int			wc_flags;
1026 	u16			pkey_index;
1027 	u8			sl;
1028 	u8			dlid_path_bits;
1029 	u8			port_num;	/* valid only for DR SMPs on switches */
1030 	u8			smac[ETH_ALEN];
1031 	u16			vlan_id;
1032 	u8			network_hdr_type;
1033 };
1034 
1035 enum ib_cq_notify_flags {
1036 	IB_CQ_SOLICITED			= 1 << 0,
1037 	IB_CQ_NEXT_COMP			= 1 << 1,
1038 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1039 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
1040 };
1041 
1042 enum ib_srq_type {
1043 	IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1044 	IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1045 	IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1046 };
1047 
1048 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1049 {
1050 	return srq_type == IB_SRQT_XRC ||
1051 	       srq_type == IB_SRQT_TM;
1052 }
1053 
1054 enum ib_srq_attr_mask {
1055 	IB_SRQ_MAX_WR	= 1 << 0,
1056 	IB_SRQ_LIMIT	= 1 << 1,
1057 };
1058 
1059 struct ib_srq_attr {
1060 	u32	max_wr;
1061 	u32	max_sge;
1062 	u32	srq_limit;
1063 };
1064 
1065 struct ib_srq_init_attr {
1066 	void		      (*event_handler)(struct ib_event *, void *);
1067 	void		       *srq_context;
1068 	struct ib_srq_attr	attr;
1069 	enum ib_srq_type	srq_type;
1070 
1071 	struct {
1072 		struct ib_cq   *cq;
1073 		union {
1074 			struct {
1075 				struct ib_xrcd *xrcd;
1076 			} xrc;
1077 
1078 			struct {
1079 				u32		max_num_tags;
1080 			} tag_matching;
1081 		};
1082 	} ext;
1083 };
1084 
1085 struct ib_qp_cap {
1086 	u32	max_send_wr;
1087 	u32	max_recv_wr;
1088 	u32	max_send_sge;
1089 	u32	max_recv_sge;
1090 	u32	max_inline_data;
1091 
1092 	/*
1093 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1094 	 * ib_create_qp() will calculate the right amount of neededed WRs
1095 	 * and MRs based on this.
1096 	 */
1097 	u32	max_rdma_ctxs;
1098 };
1099 
1100 enum ib_sig_type {
1101 	IB_SIGNAL_ALL_WR,
1102 	IB_SIGNAL_REQ_WR
1103 };
1104 
1105 enum ib_qp_type {
1106 	/*
1107 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1108 	 * here (and in that order) since the MAD layer uses them as
1109 	 * indices into a 2-entry table.
1110 	 */
1111 	IB_QPT_SMI,
1112 	IB_QPT_GSI,
1113 
1114 	IB_QPT_RC = IB_UVERBS_QPT_RC,
1115 	IB_QPT_UC = IB_UVERBS_QPT_UC,
1116 	IB_QPT_UD = IB_UVERBS_QPT_UD,
1117 	IB_QPT_RAW_IPV6,
1118 	IB_QPT_RAW_ETHERTYPE,
1119 	IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1120 	IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1121 	IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1122 	IB_QPT_MAX,
1123 	IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1124 	/* Reserve a range for qp types internal to the low level driver.
1125 	 * These qp types will not be visible at the IB core layer, so the
1126 	 * IB_QPT_MAX usages should not be affected in the core layer
1127 	 */
1128 	IB_QPT_RESERVED1 = 0x1000,
1129 	IB_QPT_RESERVED2,
1130 	IB_QPT_RESERVED3,
1131 	IB_QPT_RESERVED4,
1132 	IB_QPT_RESERVED5,
1133 	IB_QPT_RESERVED6,
1134 	IB_QPT_RESERVED7,
1135 	IB_QPT_RESERVED8,
1136 	IB_QPT_RESERVED9,
1137 	IB_QPT_RESERVED10,
1138 };
1139 
1140 enum ib_qp_create_flags {
1141 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1142 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	=
1143 		IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1144 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1145 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1146 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1147 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1148 	IB_QP_CREATE_INTEGRITY_EN		= 1 << 6,
1149 	IB_QP_CREATE_NETDEV_USE			= 1 << 7,
1150 	IB_QP_CREATE_SCATTER_FCS		=
1151 		IB_UVERBS_QP_CREATE_SCATTER_FCS,
1152 	IB_QP_CREATE_CVLAN_STRIPPING		=
1153 		IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1154 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1155 	IB_QP_CREATE_PCI_WRITE_END_PADDING	=
1156 		IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1157 	/* reserve bits 26-31 for low level drivers' internal use */
1158 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1159 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1160 };
1161 
1162 /*
1163  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1164  * callback to destroy the passed in QP.
1165  */
1166 
1167 struct ib_qp_init_attr {
1168 	/* Consumer's event_handler callback must not block */
1169 	void                  (*event_handler)(struct ib_event *, void *);
1170 
1171 	void		       *qp_context;
1172 	struct ib_cq	       *send_cq;
1173 	struct ib_cq	       *recv_cq;
1174 	struct ib_srq	       *srq;
1175 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1176 	struct ib_qp_cap	cap;
1177 	enum ib_sig_type	sq_sig_type;
1178 	enum ib_qp_type		qp_type;
1179 	u32			create_flags;
1180 
1181 	/*
1182 	 * Only needed for special QP types, or when using the RW API.
1183 	 */
1184 	u8			port_num;
1185 	struct ib_rwq_ind_table *rwq_ind_tbl;
1186 	u32			source_qpn;
1187 };
1188 
1189 struct ib_qp_open_attr {
1190 	void                  (*event_handler)(struct ib_event *, void *);
1191 	void		       *qp_context;
1192 	u32			qp_num;
1193 	enum ib_qp_type		qp_type;
1194 };
1195 
1196 enum ib_rnr_timeout {
1197 	IB_RNR_TIMER_655_36 =  0,
1198 	IB_RNR_TIMER_000_01 =  1,
1199 	IB_RNR_TIMER_000_02 =  2,
1200 	IB_RNR_TIMER_000_03 =  3,
1201 	IB_RNR_TIMER_000_04 =  4,
1202 	IB_RNR_TIMER_000_06 =  5,
1203 	IB_RNR_TIMER_000_08 =  6,
1204 	IB_RNR_TIMER_000_12 =  7,
1205 	IB_RNR_TIMER_000_16 =  8,
1206 	IB_RNR_TIMER_000_24 =  9,
1207 	IB_RNR_TIMER_000_32 = 10,
1208 	IB_RNR_TIMER_000_48 = 11,
1209 	IB_RNR_TIMER_000_64 = 12,
1210 	IB_RNR_TIMER_000_96 = 13,
1211 	IB_RNR_TIMER_001_28 = 14,
1212 	IB_RNR_TIMER_001_92 = 15,
1213 	IB_RNR_TIMER_002_56 = 16,
1214 	IB_RNR_TIMER_003_84 = 17,
1215 	IB_RNR_TIMER_005_12 = 18,
1216 	IB_RNR_TIMER_007_68 = 19,
1217 	IB_RNR_TIMER_010_24 = 20,
1218 	IB_RNR_TIMER_015_36 = 21,
1219 	IB_RNR_TIMER_020_48 = 22,
1220 	IB_RNR_TIMER_030_72 = 23,
1221 	IB_RNR_TIMER_040_96 = 24,
1222 	IB_RNR_TIMER_061_44 = 25,
1223 	IB_RNR_TIMER_081_92 = 26,
1224 	IB_RNR_TIMER_122_88 = 27,
1225 	IB_RNR_TIMER_163_84 = 28,
1226 	IB_RNR_TIMER_245_76 = 29,
1227 	IB_RNR_TIMER_327_68 = 30,
1228 	IB_RNR_TIMER_491_52 = 31
1229 };
1230 
1231 enum ib_qp_attr_mask {
1232 	IB_QP_STATE			= 1,
1233 	IB_QP_CUR_STATE			= (1<<1),
1234 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1235 	IB_QP_ACCESS_FLAGS		= (1<<3),
1236 	IB_QP_PKEY_INDEX		= (1<<4),
1237 	IB_QP_PORT			= (1<<5),
1238 	IB_QP_QKEY			= (1<<6),
1239 	IB_QP_AV			= (1<<7),
1240 	IB_QP_PATH_MTU			= (1<<8),
1241 	IB_QP_TIMEOUT			= (1<<9),
1242 	IB_QP_RETRY_CNT			= (1<<10),
1243 	IB_QP_RNR_RETRY			= (1<<11),
1244 	IB_QP_RQ_PSN			= (1<<12),
1245 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1246 	IB_QP_ALT_PATH			= (1<<14),
1247 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1248 	IB_QP_SQ_PSN			= (1<<16),
1249 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1250 	IB_QP_PATH_MIG_STATE		= (1<<18),
1251 	IB_QP_CAP			= (1<<19),
1252 	IB_QP_DEST_QPN			= (1<<20),
1253 	IB_QP_RESERVED1			= (1<<21),
1254 	IB_QP_RESERVED2			= (1<<22),
1255 	IB_QP_RESERVED3			= (1<<23),
1256 	IB_QP_RESERVED4			= (1<<24),
1257 	IB_QP_RATE_LIMIT		= (1<<25),
1258 };
1259 
1260 enum ib_qp_state {
1261 	IB_QPS_RESET,
1262 	IB_QPS_INIT,
1263 	IB_QPS_RTR,
1264 	IB_QPS_RTS,
1265 	IB_QPS_SQD,
1266 	IB_QPS_SQE,
1267 	IB_QPS_ERR
1268 };
1269 
1270 enum ib_mig_state {
1271 	IB_MIG_MIGRATED,
1272 	IB_MIG_REARM,
1273 	IB_MIG_ARMED
1274 };
1275 
1276 enum ib_mw_type {
1277 	IB_MW_TYPE_1 = 1,
1278 	IB_MW_TYPE_2 = 2
1279 };
1280 
1281 struct ib_qp_attr {
1282 	enum ib_qp_state	qp_state;
1283 	enum ib_qp_state	cur_qp_state;
1284 	enum ib_mtu		path_mtu;
1285 	enum ib_mig_state	path_mig_state;
1286 	u32			qkey;
1287 	u32			rq_psn;
1288 	u32			sq_psn;
1289 	u32			dest_qp_num;
1290 	int			qp_access_flags;
1291 	struct ib_qp_cap	cap;
1292 	struct rdma_ah_attr	ah_attr;
1293 	struct rdma_ah_attr	alt_ah_attr;
1294 	u16			pkey_index;
1295 	u16			alt_pkey_index;
1296 	u8			en_sqd_async_notify;
1297 	u8			sq_draining;
1298 	u8			max_rd_atomic;
1299 	u8			max_dest_rd_atomic;
1300 	u8			min_rnr_timer;
1301 	u8			port_num;
1302 	u8			timeout;
1303 	u8			retry_cnt;
1304 	u8			rnr_retry;
1305 	u8			alt_port_num;
1306 	u8			alt_timeout;
1307 	u32			rate_limit;
1308 	struct net_device	*xmit_slave;
1309 };
1310 
1311 enum ib_wr_opcode {
1312 	/* These are shared with userspace */
1313 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1314 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1315 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1316 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1317 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1318 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1319 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1320 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1321 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1322 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1323 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1324 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1325 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1326 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1327 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1328 
1329 	/* These are kernel only and can not be issued by userspace */
1330 	IB_WR_REG_MR = 0x20,
1331 	IB_WR_REG_MR_INTEGRITY,
1332 
1333 	/* reserve values for low level drivers' internal use.
1334 	 * These values will not be used at all in the ib core layer.
1335 	 */
1336 	IB_WR_RESERVED1 = 0xf0,
1337 	IB_WR_RESERVED2,
1338 	IB_WR_RESERVED3,
1339 	IB_WR_RESERVED4,
1340 	IB_WR_RESERVED5,
1341 	IB_WR_RESERVED6,
1342 	IB_WR_RESERVED7,
1343 	IB_WR_RESERVED8,
1344 	IB_WR_RESERVED9,
1345 	IB_WR_RESERVED10,
1346 };
1347 
1348 enum ib_send_flags {
1349 	IB_SEND_FENCE		= 1,
1350 	IB_SEND_SIGNALED	= (1<<1),
1351 	IB_SEND_SOLICITED	= (1<<2),
1352 	IB_SEND_INLINE		= (1<<3),
1353 	IB_SEND_IP_CSUM		= (1<<4),
1354 
1355 	/* reserve bits 26-31 for low level drivers' internal use */
1356 	IB_SEND_RESERVED_START	= (1 << 26),
1357 	IB_SEND_RESERVED_END	= (1 << 31),
1358 };
1359 
1360 struct ib_sge {
1361 	u64	addr;
1362 	u32	length;
1363 	u32	lkey;
1364 };
1365 
1366 struct ib_cqe {
1367 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1368 };
1369 
1370 struct ib_send_wr {
1371 	struct ib_send_wr      *next;
1372 	union {
1373 		u64		wr_id;
1374 		struct ib_cqe	*wr_cqe;
1375 	};
1376 	struct ib_sge	       *sg_list;
1377 	int			num_sge;
1378 	enum ib_wr_opcode	opcode;
1379 	int			send_flags;
1380 	union {
1381 		__be32		imm_data;
1382 		u32		invalidate_rkey;
1383 	} ex;
1384 };
1385 
1386 struct ib_rdma_wr {
1387 	struct ib_send_wr	wr;
1388 	u64			remote_addr;
1389 	u32			rkey;
1390 };
1391 
1392 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1393 {
1394 	return container_of(wr, struct ib_rdma_wr, wr);
1395 }
1396 
1397 struct ib_atomic_wr {
1398 	struct ib_send_wr	wr;
1399 	u64			remote_addr;
1400 	u64			compare_add;
1401 	u64			swap;
1402 	u64			compare_add_mask;
1403 	u64			swap_mask;
1404 	u32			rkey;
1405 };
1406 
1407 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1408 {
1409 	return container_of(wr, struct ib_atomic_wr, wr);
1410 }
1411 
1412 struct ib_ud_wr {
1413 	struct ib_send_wr	wr;
1414 	struct ib_ah		*ah;
1415 	void			*header;
1416 	int			hlen;
1417 	int			mss;
1418 	u32			remote_qpn;
1419 	u32			remote_qkey;
1420 	u16			pkey_index; /* valid for GSI only */
1421 	u8			port_num;   /* valid for DR SMPs on switch only */
1422 };
1423 
1424 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1425 {
1426 	return container_of(wr, struct ib_ud_wr, wr);
1427 }
1428 
1429 struct ib_reg_wr {
1430 	struct ib_send_wr	wr;
1431 	struct ib_mr		*mr;
1432 	u32			key;
1433 	int			access;
1434 };
1435 
1436 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1437 {
1438 	return container_of(wr, struct ib_reg_wr, wr);
1439 }
1440 
1441 struct ib_recv_wr {
1442 	struct ib_recv_wr      *next;
1443 	union {
1444 		u64		wr_id;
1445 		struct ib_cqe	*wr_cqe;
1446 	};
1447 	struct ib_sge	       *sg_list;
1448 	int			num_sge;
1449 };
1450 
1451 enum ib_access_flags {
1452 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1453 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1454 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1455 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1456 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1457 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1458 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1459 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1460 	IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1461 
1462 	IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1463 	IB_ACCESS_SUPPORTED =
1464 		((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1465 };
1466 
1467 /*
1468  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1469  * are hidden here instead of a uapi header!
1470  */
1471 enum ib_mr_rereg_flags {
1472 	IB_MR_REREG_TRANS	= 1,
1473 	IB_MR_REREG_PD		= (1<<1),
1474 	IB_MR_REREG_ACCESS	= (1<<2),
1475 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1476 };
1477 
1478 struct ib_umem;
1479 
1480 enum rdma_remove_reason {
1481 	/*
1482 	 * Userspace requested uobject deletion or initial try
1483 	 * to remove uobject via cleanup. Call could fail
1484 	 */
1485 	RDMA_REMOVE_DESTROY,
1486 	/* Context deletion. This call should delete the actual object itself */
1487 	RDMA_REMOVE_CLOSE,
1488 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1489 	RDMA_REMOVE_DRIVER_REMOVE,
1490 	/* uobj is being cleaned-up before being committed */
1491 	RDMA_REMOVE_ABORT,
1492 	/*
1493 	 * uobj has been fully created, with the uobj->object set, but is being
1494 	 * cleaned up before being comitted
1495 	 */
1496 	RDMA_REMOVE_ABORT_HWOBJ,
1497 };
1498 
1499 struct ib_rdmacg_object {
1500 #ifdef CONFIG_CGROUP_RDMA
1501 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1502 #endif
1503 };
1504 
1505 struct ib_ucontext {
1506 	struct ib_device       *device;
1507 	struct ib_uverbs_file  *ufile;
1508 	/*
1509 	 * 'closing' can be read by the driver only during a destroy callback,
1510 	 * it is set when we are closing the file descriptor and indicates
1511 	 * that mm_sem may be locked.
1512 	 */
1513 	bool closing;
1514 
1515 	bool cleanup_retryable;
1516 
1517 	struct ib_rdmacg_object	cg_obj;
1518 	/*
1519 	 * Implementation details of the RDMA core, don't use in drivers:
1520 	 */
1521 	struct rdma_restrack_entry res;
1522 	struct xarray mmap_xa;
1523 };
1524 
1525 struct ib_uobject {
1526 	u64			user_handle;	/* handle given to us by userspace */
1527 	/* ufile & ucontext owning this object */
1528 	struct ib_uverbs_file  *ufile;
1529 	/* FIXME, save memory: ufile->context == context */
1530 	struct ib_ucontext     *context;	/* associated user context */
1531 	void		       *object;		/* containing object */
1532 	struct list_head	list;		/* link to context's list */
1533 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1534 	int			id;		/* index into kernel idr */
1535 	struct kref		ref;
1536 	atomic_t		usecnt;		/* protects exclusive access */
1537 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1538 
1539 	const struct uverbs_api_object *uapi_object;
1540 };
1541 
1542 struct ib_udata {
1543 	const void __user *inbuf;
1544 	void __user *outbuf;
1545 	size_t       inlen;
1546 	size_t       outlen;
1547 };
1548 
1549 struct ib_pd {
1550 	u32			local_dma_lkey;
1551 	u32			flags;
1552 	struct ib_device       *device;
1553 	struct ib_uobject      *uobject;
1554 	atomic_t          	usecnt; /* count all resources */
1555 
1556 	u32			unsafe_global_rkey;
1557 
1558 	/*
1559 	 * Implementation details of the RDMA core, don't use in drivers:
1560 	 */
1561 	struct ib_mr	       *__internal_mr;
1562 	struct rdma_restrack_entry res;
1563 };
1564 
1565 struct ib_xrcd {
1566 	struct ib_device       *device;
1567 	atomic_t		usecnt; /* count all exposed resources */
1568 	struct inode	       *inode;
1569 
1570 	struct mutex		tgt_qp_mutex;
1571 	struct list_head	tgt_qp_list;
1572 };
1573 
1574 struct ib_ah {
1575 	struct ib_device	*device;
1576 	struct ib_pd		*pd;
1577 	struct ib_uobject	*uobject;
1578 	const struct ib_gid_attr *sgid_attr;
1579 	enum rdma_ah_attr_type	type;
1580 };
1581 
1582 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1583 
1584 enum ib_poll_context {
1585 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1586 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1587 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1588 	IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1589 
1590 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1591 };
1592 
1593 struct ib_cq {
1594 	struct ib_device       *device;
1595 	struct ib_ucq_object   *uobject;
1596 	ib_comp_handler   	comp_handler;
1597 	void                  (*event_handler)(struct ib_event *, void *);
1598 	void                   *cq_context;
1599 	int               	cqe;
1600 	unsigned int		cqe_used;
1601 	atomic_t          	usecnt; /* count number of work queues */
1602 	enum ib_poll_context	poll_ctx;
1603 	struct ib_wc		*wc;
1604 	struct list_head        pool_entry;
1605 	union {
1606 		struct irq_poll		iop;
1607 		struct work_struct	work;
1608 	};
1609 	struct workqueue_struct *comp_wq;
1610 	struct dim *dim;
1611 
1612 	/* updated only by trace points */
1613 	ktime_t timestamp;
1614 	u8 interrupt:1;
1615 	u8 shared:1;
1616 	unsigned int comp_vector;
1617 
1618 	/*
1619 	 * Implementation details of the RDMA core, don't use in drivers:
1620 	 */
1621 	struct rdma_restrack_entry res;
1622 };
1623 
1624 struct ib_srq {
1625 	struct ib_device       *device;
1626 	struct ib_pd	       *pd;
1627 	struct ib_usrq_object  *uobject;
1628 	void		      (*event_handler)(struct ib_event *, void *);
1629 	void		       *srq_context;
1630 	enum ib_srq_type	srq_type;
1631 	atomic_t		usecnt;
1632 
1633 	struct {
1634 		struct ib_cq   *cq;
1635 		union {
1636 			struct {
1637 				struct ib_xrcd *xrcd;
1638 				u32		srq_num;
1639 			} xrc;
1640 		};
1641 	} ext;
1642 };
1643 
1644 enum ib_raw_packet_caps {
1645 	/* Strip cvlan from incoming packet and report it in the matching work
1646 	 * completion is supported.
1647 	 */
1648 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1649 	/* Scatter FCS field of an incoming packet to host memory is supported.
1650 	 */
1651 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1652 	/* Checksum offloads are supported (for both send and receive). */
1653 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1654 	/* When a packet is received for an RQ with no receive WQEs, the
1655 	 * packet processing is delayed.
1656 	 */
1657 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1658 };
1659 
1660 enum ib_wq_type {
1661 	IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1662 };
1663 
1664 enum ib_wq_state {
1665 	IB_WQS_RESET,
1666 	IB_WQS_RDY,
1667 	IB_WQS_ERR
1668 };
1669 
1670 struct ib_wq {
1671 	struct ib_device       *device;
1672 	struct ib_uwq_object   *uobject;
1673 	void		    *wq_context;
1674 	void		    (*event_handler)(struct ib_event *, void *);
1675 	struct ib_pd	       *pd;
1676 	struct ib_cq	       *cq;
1677 	u32		wq_num;
1678 	enum ib_wq_state       state;
1679 	enum ib_wq_type	wq_type;
1680 	atomic_t		usecnt;
1681 };
1682 
1683 enum ib_wq_flags {
1684 	IB_WQ_FLAGS_CVLAN_STRIPPING	= IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1685 	IB_WQ_FLAGS_SCATTER_FCS		= IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1686 	IB_WQ_FLAGS_DELAY_DROP		= IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1687 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1688 				IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1689 };
1690 
1691 struct ib_wq_init_attr {
1692 	void		       *wq_context;
1693 	enum ib_wq_type	wq_type;
1694 	u32		max_wr;
1695 	u32		max_sge;
1696 	struct	ib_cq	       *cq;
1697 	void		    (*event_handler)(struct ib_event *, void *);
1698 	u32		create_flags; /* Use enum ib_wq_flags */
1699 };
1700 
1701 enum ib_wq_attr_mask {
1702 	IB_WQ_STATE		= 1 << 0,
1703 	IB_WQ_CUR_STATE		= 1 << 1,
1704 	IB_WQ_FLAGS		= 1 << 2,
1705 };
1706 
1707 struct ib_wq_attr {
1708 	enum	ib_wq_state	wq_state;
1709 	enum	ib_wq_state	curr_wq_state;
1710 	u32			flags; /* Use enum ib_wq_flags */
1711 	u32			flags_mask; /* Use enum ib_wq_flags */
1712 };
1713 
1714 struct ib_rwq_ind_table {
1715 	struct ib_device	*device;
1716 	struct ib_uobject      *uobject;
1717 	atomic_t		usecnt;
1718 	u32		ind_tbl_num;
1719 	u32		log_ind_tbl_size;
1720 	struct ib_wq	**ind_tbl;
1721 };
1722 
1723 struct ib_rwq_ind_table_init_attr {
1724 	u32		log_ind_tbl_size;
1725 	/* Each entry is a pointer to Receive Work Queue */
1726 	struct ib_wq	**ind_tbl;
1727 };
1728 
1729 enum port_pkey_state {
1730 	IB_PORT_PKEY_NOT_VALID = 0,
1731 	IB_PORT_PKEY_VALID = 1,
1732 	IB_PORT_PKEY_LISTED = 2,
1733 };
1734 
1735 struct ib_qp_security;
1736 
1737 struct ib_port_pkey {
1738 	enum port_pkey_state	state;
1739 	u16			pkey_index;
1740 	u8			port_num;
1741 	struct list_head	qp_list;
1742 	struct list_head	to_error_list;
1743 	struct ib_qp_security  *sec;
1744 };
1745 
1746 struct ib_ports_pkeys {
1747 	struct ib_port_pkey	main;
1748 	struct ib_port_pkey	alt;
1749 };
1750 
1751 struct ib_qp_security {
1752 	struct ib_qp	       *qp;
1753 	struct ib_device       *dev;
1754 	/* Hold this mutex when changing port and pkey settings. */
1755 	struct mutex		mutex;
1756 	struct ib_ports_pkeys  *ports_pkeys;
1757 	/* A list of all open shared QP handles.  Required to enforce security
1758 	 * properly for all users of a shared QP.
1759 	 */
1760 	struct list_head        shared_qp_list;
1761 	void                   *security;
1762 	bool			destroying;
1763 	atomic_t		error_list_count;
1764 	struct completion	error_complete;
1765 	int			error_comps_pending;
1766 };
1767 
1768 /*
1769  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1770  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1771  */
1772 struct ib_qp {
1773 	struct ib_device       *device;
1774 	struct ib_pd	       *pd;
1775 	struct ib_cq	       *send_cq;
1776 	struct ib_cq	       *recv_cq;
1777 	spinlock_t		mr_lock;
1778 	int			mrs_used;
1779 	struct list_head	rdma_mrs;
1780 	struct list_head	sig_mrs;
1781 	struct ib_srq	       *srq;
1782 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1783 	struct list_head	xrcd_list;
1784 
1785 	/* count times opened, mcast attaches, flow attaches */
1786 	atomic_t		usecnt;
1787 	struct list_head	open_list;
1788 	struct ib_qp           *real_qp;
1789 	struct ib_uqp_object   *uobject;
1790 	void                  (*event_handler)(struct ib_event *, void *);
1791 	void		       *qp_context;
1792 	/* sgid_attrs associated with the AV's */
1793 	const struct ib_gid_attr *av_sgid_attr;
1794 	const struct ib_gid_attr *alt_path_sgid_attr;
1795 	u32			qp_num;
1796 	u32			max_write_sge;
1797 	u32			max_read_sge;
1798 	enum ib_qp_type		qp_type;
1799 	struct ib_rwq_ind_table *rwq_ind_tbl;
1800 	struct ib_qp_security  *qp_sec;
1801 	u8			port;
1802 
1803 	bool			integrity_en;
1804 	/*
1805 	 * Implementation details of the RDMA core, don't use in drivers:
1806 	 */
1807 	struct rdma_restrack_entry     res;
1808 
1809 	/* The counter the qp is bind to */
1810 	struct rdma_counter    *counter;
1811 };
1812 
1813 struct ib_dm {
1814 	struct ib_device  *device;
1815 	u32		   length;
1816 	u32		   flags;
1817 	struct ib_uobject *uobject;
1818 	atomic_t	   usecnt;
1819 };
1820 
1821 struct ib_mr {
1822 	struct ib_device  *device;
1823 	struct ib_pd	  *pd;
1824 	u32		   lkey;
1825 	u32		   rkey;
1826 	u64		   iova;
1827 	u64		   length;
1828 	unsigned int	   page_size;
1829 	enum ib_mr_type	   type;
1830 	bool		   need_inval;
1831 	union {
1832 		struct ib_uobject	*uobject;	/* user */
1833 		struct list_head	qp_entry;	/* FR */
1834 	};
1835 
1836 	struct ib_dm      *dm;
1837 	struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1838 	/*
1839 	 * Implementation details of the RDMA core, don't use in drivers:
1840 	 */
1841 	struct rdma_restrack_entry res;
1842 };
1843 
1844 struct ib_mw {
1845 	struct ib_device	*device;
1846 	struct ib_pd		*pd;
1847 	struct ib_uobject	*uobject;
1848 	u32			rkey;
1849 	enum ib_mw_type         type;
1850 };
1851 
1852 /* Supported steering options */
1853 enum ib_flow_attr_type {
1854 	/* steering according to rule specifications */
1855 	IB_FLOW_ATTR_NORMAL		= 0x0,
1856 	/* default unicast and multicast rule -
1857 	 * receive all Eth traffic which isn't steered to any QP
1858 	 */
1859 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1860 	/* default multicast rule -
1861 	 * receive all Eth multicast traffic which isn't steered to any QP
1862 	 */
1863 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1864 	/* sniffer rule - receive all port traffic */
1865 	IB_FLOW_ATTR_SNIFFER		= 0x3
1866 };
1867 
1868 /* Supported steering header types */
1869 enum ib_flow_spec_type {
1870 	/* L2 headers*/
1871 	IB_FLOW_SPEC_ETH		= 0x20,
1872 	IB_FLOW_SPEC_IB			= 0x22,
1873 	/* L3 header*/
1874 	IB_FLOW_SPEC_IPV4		= 0x30,
1875 	IB_FLOW_SPEC_IPV6		= 0x31,
1876 	IB_FLOW_SPEC_ESP                = 0x34,
1877 	/* L4 headers*/
1878 	IB_FLOW_SPEC_TCP		= 0x40,
1879 	IB_FLOW_SPEC_UDP		= 0x41,
1880 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1881 	IB_FLOW_SPEC_GRE		= 0x51,
1882 	IB_FLOW_SPEC_MPLS		= 0x60,
1883 	IB_FLOW_SPEC_INNER		= 0x100,
1884 	/* Actions */
1885 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1886 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1887 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1888 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1889 };
1890 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1891 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1892 
1893 /* Flow steering rule priority is set according to it's domain.
1894  * Lower domain value means higher priority.
1895  */
1896 enum ib_flow_domain {
1897 	IB_FLOW_DOMAIN_USER,
1898 	IB_FLOW_DOMAIN_ETHTOOL,
1899 	IB_FLOW_DOMAIN_RFS,
1900 	IB_FLOW_DOMAIN_NIC,
1901 	IB_FLOW_DOMAIN_NUM /* Must be last */
1902 };
1903 
1904 enum ib_flow_flags {
1905 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1906 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1907 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1908 };
1909 
1910 struct ib_flow_eth_filter {
1911 	u8	dst_mac[6];
1912 	u8	src_mac[6];
1913 	__be16	ether_type;
1914 	__be16	vlan_tag;
1915 	/* Must be last */
1916 	u8	real_sz[];
1917 };
1918 
1919 struct ib_flow_spec_eth {
1920 	u32			  type;
1921 	u16			  size;
1922 	struct ib_flow_eth_filter val;
1923 	struct ib_flow_eth_filter mask;
1924 };
1925 
1926 struct ib_flow_ib_filter {
1927 	__be16 dlid;
1928 	__u8   sl;
1929 	/* Must be last */
1930 	u8	real_sz[];
1931 };
1932 
1933 struct ib_flow_spec_ib {
1934 	u32			 type;
1935 	u16			 size;
1936 	struct ib_flow_ib_filter val;
1937 	struct ib_flow_ib_filter mask;
1938 };
1939 
1940 /* IPv4 header flags */
1941 enum ib_ipv4_flags {
1942 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1943 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1944 				    last have this flag set */
1945 };
1946 
1947 struct ib_flow_ipv4_filter {
1948 	__be32	src_ip;
1949 	__be32	dst_ip;
1950 	u8	proto;
1951 	u8	tos;
1952 	u8	ttl;
1953 	u8	flags;
1954 	/* Must be last */
1955 	u8	real_sz[];
1956 };
1957 
1958 struct ib_flow_spec_ipv4 {
1959 	u32			   type;
1960 	u16			   size;
1961 	struct ib_flow_ipv4_filter val;
1962 	struct ib_flow_ipv4_filter mask;
1963 };
1964 
1965 struct ib_flow_ipv6_filter {
1966 	u8	src_ip[16];
1967 	u8	dst_ip[16];
1968 	__be32	flow_label;
1969 	u8	next_hdr;
1970 	u8	traffic_class;
1971 	u8	hop_limit;
1972 	/* Must be last */
1973 	u8	real_sz[];
1974 };
1975 
1976 struct ib_flow_spec_ipv6 {
1977 	u32			   type;
1978 	u16			   size;
1979 	struct ib_flow_ipv6_filter val;
1980 	struct ib_flow_ipv6_filter mask;
1981 };
1982 
1983 struct ib_flow_tcp_udp_filter {
1984 	__be16	dst_port;
1985 	__be16	src_port;
1986 	/* Must be last */
1987 	u8	real_sz[];
1988 };
1989 
1990 struct ib_flow_spec_tcp_udp {
1991 	u32			      type;
1992 	u16			      size;
1993 	struct ib_flow_tcp_udp_filter val;
1994 	struct ib_flow_tcp_udp_filter mask;
1995 };
1996 
1997 struct ib_flow_tunnel_filter {
1998 	__be32	tunnel_id;
1999 	u8	real_sz[];
2000 };
2001 
2002 /* ib_flow_spec_tunnel describes the Vxlan tunnel
2003  * the tunnel_id from val has the vni value
2004  */
2005 struct ib_flow_spec_tunnel {
2006 	u32			      type;
2007 	u16			      size;
2008 	struct ib_flow_tunnel_filter  val;
2009 	struct ib_flow_tunnel_filter  mask;
2010 };
2011 
2012 struct ib_flow_esp_filter {
2013 	__be32	spi;
2014 	__be32  seq;
2015 	/* Must be last */
2016 	u8	real_sz[];
2017 };
2018 
2019 struct ib_flow_spec_esp {
2020 	u32                           type;
2021 	u16			      size;
2022 	struct ib_flow_esp_filter     val;
2023 	struct ib_flow_esp_filter     mask;
2024 };
2025 
2026 struct ib_flow_gre_filter {
2027 	__be16 c_ks_res0_ver;
2028 	__be16 protocol;
2029 	__be32 key;
2030 	/* Must be last */
2031 	u8	real_sz[];
2032 };
2033 
2034 struct ib_flow_spec_gre {
2035 	u32                           type;
2036 	u16			      size;
2037 	struct ib_flow_gre_filter     val;
2038 	struct ib_flow_gre_filter     mask;
2039 };
2040 
2041 struct ib_flow_mpls_filter {
2042 	__be32 tag;
2043 	/* Must be last */
2044 	u8	real_sz[];
2045 };
2046 
2047 struct ib_flow_spec_mpls {
2048 	u32                           type;
2049 	u16			      size;
2050 	struct ib_flow_mpls_filter     val;
2051 	struct ib_flow_mpls_filter     mask;
2052 };
2053 
2054 struct ib_flow_spec_action_tag {
2055 	enum ib_flow_spec_type	      type;
2056 	u16			      size;
2057 	u32                           tag_id;
2058 };
2059 
2060 struct ib_flow_spec_action_drop {
2061 	enum ib_flow_spec_type	      type;
2062 	u16			      size;
2063 };
2064 
2065 struct ib_flow_spec_action_handle {
2066 	enum ib_flow_spec_type	      type;
2067 	u16			      size;
2068 	struct ib_flow_action	     *act;
2069 };
2070 
2071 enum ib_counters_description {
2072 	IB_COUNTER_PACKETS,
2073 	IB_COUNTER_BYTES,
2074 };
2075 
2076 struct ib_flow_spec_action_count {
2077 	enum ib_flow_spec_type type;
2078 	u16 size;
2079 	struct ib_counters *counters;
2080 };
2081 
2082 union ib_flow_spec {
2083 	struct {
2084 		u32			type;
2085 		u16			size;
2086 	};
2087 	struct ib_flow_spec_eth		eth;
2088 	struct ib_flow_spec_ib		ib;
2089 	struct ib_flow_spec_ipv4        ipv4;
2090 	struct ib_flow_spec_tcp_udp	tcp_udp;
2091 	struct ib_flow_spec_ipv6        ipv6;
2092 	struct ib_flow_spec_tunnel      tunnel;
2093 	struct ib_flow_spec_esp		esp;
2094 	struct ib_flow_spec_gre		gre;
2095 	struct ib_flow_spec_mpls	mpls;
2096 	struct ib_flow_spec_action_tag  flow_tag;
2097 	struct ib_flow_spec_action_drop drop;
2098 	struct ib_flow_spec_action_handle action;
2099 	struct ib_flow_spec_action_count flow_count;
2100 };
2101 
2102 struct ib_flow_attr {
2103 	enum ib_flow_attr_type type;
2104 	u16	     size;
2105 	u16	     priority;
2106 	u32	     flags;
2107 	u8	     num_of_specs;
2108 	u8	     port;
2109 	union ib_flow_spec flows[];
2110 };
2111 
2112 struct ib_flow {
2113 	struct ib_qp		*qp;
2114 	struct ib_device	*device;
2115 	struct ib_uobject	*uobject;
2116 };
2117 
2118 enum ib_flow_action_type {
2119 	IB_FLOW_ACTION_UNSPECIFIED,
2120 	IB_FLOW_ACTION_ESP = 1,
2121 };
2122 
2123 struct ib_flow_action_attrs_esp_keymats {
2124 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2125 	union {
2126 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2127 	} keymat;
2128 };
2129 
2130 struct ib_flow_action_attrs_esp_replays {
2131 	enum ib_uverbs_flow_action_esp_replay			protocol;
2132 	union {
2133 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2134 	} replay;
2135 };
2136 
2137 enum ib_flow_action_attrs_esp_flags {
2138 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2139 	 * This is done in order to share the same flags between user-space and
2140 	 * kernel and spare an unnecessary translation.
2141 	 */
2142 
2143 	/* Kernel flags */
2144 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2145 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2146 };
2147 
2148 struct ib_flow_spec_list {
2149 	struct ib_flow_spec_list	*next;
2150 	union ib_flow_spec		spec;
2151 };
2152 
2153 struct ib_flow_action_attrs_esp {
2154 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2155 	struct ib_flow_action_attrs_esp_replays		*replay;
2156 	struct ib_flow_spec_list			*encap;
2157 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2158 	 * Value of 0 is a valid value.
2159 	 */
2160 	u32						esn;
2161 	u32						spi;
2162 	u32						seq;
2163 	u32						tfc_pad;
2164 	/* Use enum ib_flow_action_attrs_esp_flags */
2165 	u64						flags;
2166 	u64						hard_limit_pkts;
2167 };
2168 
2169 struct ib_flow_action {
2170 	struct ib_device		*device;
2171 	struct ib_uobject		*uobject;
2172 	enum ib_flow_action_type	type;
2173 	atomic_t			usecnt;
2174 };
2175 
2176 struct ib_mad;
2177 struct ib_grh;
2178 
2179 enum ib_process_mad_flags {
2180 	IB_MAD_IGNORE_MKEY	= 1,
2181 	IB_MAD_IGNORE_BKEY	= 2,
2182 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2183 };
2184 
2185 enum ib_mad_result {
2186 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2187 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2188 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2189 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2190 };
2191 
2192 struct ib_port_cache {
2193 	u64		      subnet_prefix;
2194 	struct ib_pkey_cache  *pkey;
2195 	struct ib_gid_table   *gid;
2196 	u8                     lmc;
2197 	enum ib_port_state     port_state;
2198 };
2199 
2200 struct ib_port_immutable {
2201 	int                           pkey_tbl_len;
2202 	int                           gid_tbl_len;
2203 	u32                           core_cap_flags;
2204 	u32                           max_mad_size;
2205 };
2206 
2207 struct ib_port_data {
2208 	struct ib_device *ib_dev;
2209 
2210 	struct ib_port_immutable immutable;
2211 
2212 	spinlock_t pkey_list_lock;
2213 	struct list_head pkey_list;
2214 
2215 	struct ib_port_cache cache;
2216 
2217 	spinlock_t netdev_lock;
2218 	struct net_device __rcu *netdev;
2219 	struct hlist_node ndev_hash_link;
2220 	struct rdma_port_counter port_counter;
2221 	struct rdma_hw_stats *hw_stats;
2222 };
2223 
2224 /* rdma netdev type - specifies protocol type */
2225 enum rdma_netdev_t {
2226 	RDMA_NETDEV_OPA_VNIC,
2227 	RDMA_NETDEV_IPOIB,
2228 };
2229 
2230 /**
2231  * struct rdma_netdev - rdma netdev
2232  * For cases where netstack interfacing is required.
2233  */
2234 struct rdma_netdev {
2235 	void              *clnt_priv;
2236 	struct ib_device  *hca;
2237 	u8                 port_num;
2238 	int                mtu;
2239 
2240 	/*
2241 	 * cleanup function must be specified.
2242 	 * FIXME: This is only used for OPA_VNIC and that usage should be
2243 	 * removed too.
2244 	 */
2245 	void (*free_rdma_netdev)(struct net_device *netdev);
2246 
2247 	/* control functions */
2248 	void (*set_id)(struct net_device *netdev, int id);
2249 	/* send packet */
2250 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2251 		    struct ib_ah *address, u32 dqpn);
2252 	/* multicast */
2253 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2254 			    union ib_gid *gid, u16 mlid,
2255 			    int set_qkey, u32 qkey);
2256 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2257 			    union ib_gid *gid, u16 mlid);
2258 };
2259 
2260 struct rdma_netdev_alloc_params {
2261 	size_t sizeof_priv;
2262 	unsigned int txqs;
2263 	unsigned int rxqs;
2264 	void *param;
2265 
2266 	int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2267 				      struct net_device *netdev, void *param);
2268 };
2269 
2270 struct ib_odp_counters {
2271 	atomic64_t faults;
2272 	atomic64_t invalidations;
2273 };
2274 
2275 struct ib_counters {
2276 	struct ib_device	*device;
2277 	struct ib_uobject	*uobject;
2278 	/* num of objects attached */
2279 	atomic_t	usecnt;
2280 };
2281 
2282 struct ib_counters_read_attr {
2283 	u64	*counters_buff;
2284 	u32	ncounters;
2285 	u32	flags; /* use enum ib_read_counters_flags */
2286 };
2287 
2288 struct uverbs_attr_bundle;
2289 struct iw_cm_id;
2290 struct iw_cm_conn_param;
2291 
2292 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member)                      \
2293 	.size_##ib_struct =                                                    \
2294 		(sizeof(struct drv_struct) +                                   \
2295 		 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) +      \
2296 		 BUILD_BUG_ON_ZERO(                                            \
2297 			 !__same_type(((struct drv_struct *)NULL)->member,     \
2298 				      struct ib_struct)))
2299 
2300 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp)                         \
2301 	((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2302 
2303 #define rdma_zalloc_drv_obj(ib_dev, ib_type)                                   \
2304 	rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2305 
2306 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2307 
2308 struct rdma_user_mmap_entry {
2309 	struct kref ref;
2310 	struct ib_ucontext *ucontext;
2311 	unsigned long start_pgoff;
2312 	size_t npages;
2313 	bool driver_removed;
2314 };
2315 
2316 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2317 static inline u64
2318 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2319 {
2320 	return (u64)entry->start_pgoff << PAGE_SHIFT;
2321 }
2322 
2323 /**
2324  * struct ib_device_ops - InfiniBand device operations
2325  * This structure defines all the InfiniBand device operations, providers will
2326  * need to define the supported operations, otherwise they will be set to null.
2327  */
2328 struct ib_device_ops {
2329 	struct module *owner;
2330 	enum rdma_driver_id driver_id;
2331 	u32 uverbs_abi_ver;
2332 	unsigned int uverbs_no_driver_id_binding:1;
2333 
2334 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2335 			 const struct ib_send_wr **bad_send_wr);
2336 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2337 			 const struct ib_recv_wr **bad_recv_wr);
2338 	void (*drain_rq)(struct ib_qp *qp);
2339 	void (*drain_sq)(struct ib_qp *qp);
2340 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2341 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2342 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2343 	int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2344 	int (*post_srq_recv)(struct ib_srq *srq,
2345 			     const struct ib_recv_wr *recv_wr,
2346 			     const struct ib_recv_wr **bad_recv_wr);
2347 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2348 			   u8 port_num, const struct ib_wc *in_wc,
2349 			   const struct ib_grh *in_grh,
2350 			   const struct ib_mad *in_mad, struct ib_mad *out_mad,
2351 			   size_t *out_mad_size, u16 *out_mad_pkey_index);
2352 	int (*query_device)(struct ib_device *device,
2353 			    struct ib_device_attr *device_attr,
2354 			    struct ib_udata *udata);
2355 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2356 			     struct ib_device_modify *device_modify);
2357 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2358 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2359 						     int comp_vector);
2360 	int (*query_port)(struct ib_device *device, u8 port_num,
2361 			  struct ib_port_attr *port_attr);
2362 	int (*modify_port)(struct ib_device *device, u8 port_num,
2363 			   int port_modify_mask,
2364 			   struct ib_port_modify *port_modify);
2365 	/**
2366 	 * The following mandatory functions are used only at device
2367 	 * registration.  Keep functions such as these at the end of this
2368 	 * structure to avoid cache line misses when accessing struct ib_device
2369 	 * in fast paths.
2370 	 */
2371 	int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2372 				  struct ib_port_immutable *immutable);
2373 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2374 					       u8 port_num);
2375 	/**
2376 	 * When calling get_netdev, the HW vendor's driver should return the
2377 	 * net device of device @device at port @port_num or NULL if such
2378 	 * a net device doesn't exist. The vendor driver should call dev_hold
2379 	 * on this net device. The HW vendor's device driver must guarantee
2380 	 * that this function returns NULL before the net device has finished
2381 	 * NETDEV_UNREGISTER state.
2382 	 */
2383 	struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2384 	/**
2385 	 * rdma netdev operation
2386 	 *
2387 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2388 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2389 	 */
2390 	struct net_device *(*alloc_rdma_netdev)(
2391 		struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2392 		const char *name, unsigned char name_assign_type,
2393 		void (*setup)(struct net_device *));
2394 
2395 	int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2396 				      enum rdma_netdev_t type,
2397 				      struct rdma_netdev_alloc_params *params);
2398 	/**
2399 	 * query_gid should be return GID value for @device, when @port_num
2400 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2401 	 * is RoCE link layer.
2402 	 */
2403 	int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2404 			 union ib_gid *gid);
2405 	/**
2406 	 * When calling add_gid, the HW vendor's driver should add the gid
2407 	 * of device of port at gid index available at @attr. Meta-info of
2408 	 * that gid (for example, the network device related to this gid) is
2409 	 * available at @attr. @context allows the HW vendor driver to store
2410 	 * extra information together with a GID entry. The HW vendor driver may
2411 	 * allocate memory to contain this information and store it in @context
2412 	 * when a new GID entry is written to. Params are consistent until the
2413 	 * next call of add_gid or delete_gid. The function should return 0 on
2414 	 * success or error otherwise. The function could be called
2415 	 * concurrently for different ports. This function is only called when
2416 	 * roce_gid_table is used.
2417 	 */
2418 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2419 	/**
2420 	 * When calling del_gid, the HW vendor's driver should delete the
2421 	 * gid of device @device at gid index gid_index of port port_num
2422 	 * available in @attr.
2423 	 * Upon the deletion of a GID entry, the HW vendor must free any
2424 	 * allocated memory. The caller will clear @context afterwards.
2425 	 * This function is only called when roce_gid_table is used.
2426 	 */
2427 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2428 	int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2429 			  u16 *pkey);
2430 	int (*alloc_ucontext)(struct ib_ucontext *context,
2431 			      struct ib_udata *udata);
2432 	void (*dealloc_ucontext)(struct ib_ucontext *context);
2433 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2434 	/**
2435 	 * This will be called once refcount of an entry in mmap_xa reaches
2436 	 * zero. The type of the memory that was mapped may differ between
2437 	 * entries and is opaque to the rdma_user_mmap interface.
2438 	 * Therefore needs to be implemented by the driver in mmap_free.
2439 	 */
2440 	void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2441 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2442 	int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2443 	void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2444 	int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2445 			 struct ib_udata *udata);
2446 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2447 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2448 	void (*destroy_ah)(struct ib_ah *ah, u32 flags);
2449 	int (*create_srq)(struct ib_srq *srq,
2450 			  struct ib_srq_init_attr *srq_init_attr,
2451 			  struct ib_udata *udata);
2452 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2453 			  enum ib_srq_attr_mask srq_attr_mask,
2454 			  struct ib_udata *udata);
2455 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2456 	void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2457 	struct ib_qp *(*create_qp)(struct ib_pd *pd,
2458 				   struct ib_qp_init_attr *qp_init_attr,
2459 				   struct ib_udata *udata);
2460 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2461 			 int qp_attr_mask, struct ib_udata *udata);
2462 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2463 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2464 	int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2465 	int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2466 			 struct ib_udata *udata);
2467 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2468 	void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2469 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2470 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2471 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2472 				     u64 virt_addr, int mr_access_flags,
2473 				     struct ib_udata *udata);
2474 	int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2475 			     u64 virt_addr, int mr_access_flags,
2476 			     struct ib_pd *pd, struct ib_udata *udata);
2477 	int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2478 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2479 				  u32 max_num_sg, struct ib_udata *udata);
2480 	struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2481 					    u32 max_num_data_sg,
2482 					    u32 max_num_meta_sg);
2483 	int (*advise_mr)(struct ib_pd *pd,
2484 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2485 			 struct ib_sge *sg_list, u32 num_sge,
2486 			 struct uverbs_attr_bundle *attrs);
2487 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2488 			 unsigned int *sg_offset);
2489 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2490 			       struct ib_mr_status *mr_status);
2491 	struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2492 				  struct ib_udata *udata);
2493 	int (*dealloc_mw)(struct ib_mw *mw);
2494 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2495 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2496 	struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2497 				      struct ib_udata *udata);
2498 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2499 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2500 				       struct ib_flow_attr *flow_attr,
2501 				       int domain, struct ib_udata *udata);
2502 	int (*destroy_flow)(struct ib_flow *flow_id);
2503 	struct ib_flow_action *(*create_flow_action_esp)(
2504 		struct ib_device *device,
2505 		const struct ib_flow_action_attrs_esp *attr,
2506 		struct uverbs_attr_bundle *attrs);
2507 	int (*destroy_flow_action)(struct ib_flow_action *action);
2508 	int (*modify_flow_action_esp)(
2509 		struct ib_flow_action *action,
2510 		const struct ib_flow_action_attrs_esp *attr,
2511 		struct uverbs_attr_bundle *attrs);
2512 	int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2513 				 int state);
2514 	int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2515 			     struct ifla_vf_info *ivf);
2516 	int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2517 			    struct ifla_vf_stats *stats);
2518 	int (*get_vf_guid)(struct ib_device *device, int vf, u8 port,
2519 			    struct ifla_vf_guid *node_guid,
2520 			    struct ifla_vf_guid *port_guid);
2521 	int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2522 			   int type);
2523 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2524 				   struct ib_wq_init_attr *init_attr,
2525 				   struct ib_udata *udata);
2526 	void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2527 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2528 			 u32 wq_attr_mask, struct ib_udata *udata);
2529 	struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2530 		struct ib_device *device,
2531 		struct ib_rwq_ind_table_init_attr *init_attr,
2532 		struct ib_udata *udata);
2533 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2534 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2535 				  struct ib_ucontext *context,
2536 				  struct ib_dm_alloc_attr *attr,
2537 				  struct uverbs_attr_bundle *attrs);
2538 	int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2539 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2540 				   struct ib_dm_mr_attr *attr,
2541 				   struct uverbs_attr_bundle *attrs);
2542 	struct ib_counters *(*create_counters)(
2543 		struct ib_device *device, struct uverbs_attr_bundle *attrs);
2544 	int (*destroy_counters)(struct ib_counters *counters);
2545 	int (*read_counters)(struct ib_counters *counters,
2546 			     struct ib_counters_read_attr *counters_read_attr,
2547 			     struct uverbs_attr_bundle *attrs);
2548 	int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2549 			    int data_sg_nents, unsigned int *data_sg_offset,
2550 			    struct scatterlist *meta_sg, int meta_sg_nents,
2551 			    unsigned int *meta_sg_offset);
2552 
2553 	/**
2554 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2555 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2556 	 *   core when the device is removed.  A lifespan of -1 in the return
2557 	 *   struct tells the core to set a default lifespan.
2558 	 */
2559 	struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2560 						u8 port_num);
2561 	/**
2562 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2563 	 * @index - The index in the value array we wish to have updated, or
2564 	 *   num_counters if we want all stats updated
2565 	 * Return codes -
2566 	 *   < 0 - Error, no counters updated
2567 	 *   index - Updated the single counter pointed to by index
2568 	 *   num_counters - Updated all counters (will reset the timestamp
2569 	 *     and prevent further calls for lifespan milliseconds)
2570 	 * Drivers are allowed to update all counters in leiu of just the
2571 	 *   one given in index at their option
2572 	 */
2573 	int (*get_hw_stats)(struct ib_device *device,
2574 			    struct rdma_hw_stats *stats, u8 port, int index);
2575 	/*
2576 	 * This function is called once for each port when a ib device is
2577 	 * registered.
2578 	 */
2579 	int (*init_port)(struct ib_device *device, u8 port_num,
2580 			 struct kobject *port_sysfs);
2581 	/**
2582 	 * Allows rdma drivers to add their own restrack attributes.
2583 	 */
2584 	int (*fill_res_entry)(struct sk_buff *msg,
2585 			      struct rdma_restrack_entry *entry);
2586 
2587 	/* Device lifecycle callbacks */
2588 	/*
2589 	 * Called after the device becomes registered, before clients are
2590 	 * attached
2591 	 */
2592 	int (*enable_driver)(struct ib_device *dev);
2593 	/*
2594 	 * This is called as part of ib_dealloc_device().
2595 	 */
2596 	void (*dealloc_driver)(struct ib_device *dev);
2597 
2598 	/* iWarp CM callbacks */
2599 	void (*iw_add_ref)(struct ib_qp *qp);
2600 	void (*iw_rem_ref)(struct ib_qp *qp);
2601 	struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2602 	int (*iw_connect)(struct iw_cm_id *cm_id,
2603 			  struct iw_cm_conn_param *conn_param);
2604 	int (*iw_accept)(struct iw_cm_id *cm_id,
2605 			 struct iw_cm_conn_param *conn_param);
2606 	int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2607 			 u8 pdata_len);
2608 	int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2609 	int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2610 	/**
2611 	 * counter_bind_qp - Bind a QP to a counter.
2612 	 * @counter - The counter to be bound. If counter->id is zero then
2613 	 *   the driver needs to allocate a new counter and set counter->id
2614 	 */
2615 	int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2616 	/**
2617 	 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2618 	 *   counter and bind it onto the default one
2619 	 */
2620 	int (*counter_unbind_qp)(struct ib_qp *qp);
2621 	/**
2622 	 * counter_dealloc -De-allocate the hw counter
2623 	 */
2624 	int (*counter_dealloc)(struct rdma_counter *counter);
2625 	/**
2626 	 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2627 	 * the driver initialized data.
2628 	 */
2629 	struct rdma_hw_stats *(*counter_alloc_stats)(
2630 		struct rdma_counter *counter);
2631 	/**
2632 	 * counter_update_stats - Query the stats value of this counter
2633 	 */
2634 	int (*counter_update_stats)(struct rdma_counter *counter);
2635 
2636 	/**
2637 	 * Allows rdma drivers to add their own restrack attributes
2638 	 * dumped via 'rdma stat' iproute2 command.
2639 	 */
2640 	int (*fill_stat_entry)(struct sk_buff *msg,
2641 			       struct rdma_restrack_entry *entry);
2642 
2643 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
2644 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
2645 	DECLARE_RDMA_OBJ_SIZE(ib_pd);
2646 	DECLARE_RDMA_OBJ_SIZE(ib_srq);
2647 	DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2648 };
2649 
2650 struct ib_core_device {
2651 	/* device must be the first element in structure until,
2652 	 * union of ib_core_device and device exists in ib_device.
2653 	 */
2654 	struct device dev;
2655 	possible_net_t rdma_net;
2656 	struct kobject *ports_kobj;
2657 	struct list_head port_list;
2658 	struct ib_device *owner; /* reach back to owner ib_device */
2659 };
2660 
2661 struct rdma_restrack_root;
2662 struct ib_device {
2663 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2664 	struct device                *dma_device;
2665 	struct ib_device_ops	     ops;
2666 	char                          name[IB_DEVICE_NAME_MAX];
2667 	struct rcu_head rcu_head;
2668 
2669 	struct list_head              event_handler_list;
2670 	/* Protects event_handler_list */
2671 	struct rw_semaphore event_handler_rwsem;
2672 
2673 	/* Protects QP's event_handler calls and open_qp list */
2674 	spinlock_t qp_open_list_lock;
2675 
2676 	struct rw_semaphore	      client_data_rwsem;
2677 	struct xarray                 client_data;
2678 	struct mutex                  unregistration_lock;
2679 
2680 	/* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2681 	rwlock_t cache_lock;
2682 	/**
2683 	 * port_data is indexed by port number
2684 	 */
2685 	struct ib_port_data *port_data;
2686 
2687 	int			      num_comp_vectors;
2688 
2689 	union {
2690 		struct device		dev;
2691 		struct ib_core_device	coredev;
2692 	};
2693 
2694 	/* First group for device attributes,
2695 	 * Second group for driver provided attributes (optional).
2696 	 * It is NULL terminated array.
2697 	 */
2698 	const struct attribute_group	*groups[3];
2699 
2700 	u64			     uverbs_cmd_mask;
2701 	u64			     uverbs_ex_cmd_mask;
2702 
2703 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2704 	__be64			     node_guid;
2705 	u32			     local_dma_lkey;
2706 	u16                          is_switch:1;
2707 	/* Indicates kernel verbs support, should not be used in drivers */
2708 	u16                          kverbs_provider:1;
2709 	/* CQ adaptive moderation (RDMA DIM) */
2710 	u16                          use_cq_dim:1;
2711 	u8                           node_type;
2712 	u8                           phys_port_cnt;
2713 	struct ib_device_attr        attrs;
2714 	struct attribute_group	     *hw_stats_ag;
2715 	struct rdma_hw_stats         *hw_stats;
2716 
2717 #ifdef CONFIG_CGROUP_RDMA
2718 	struct rdmacg_device         cg_device;
2719 #endif
2720 
2721 	u32                          index;
2722 
2723 	spinlock_t                   cq_pools_lock;
2724 	struct list_head             cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2725 
2726 	struct rdma_restrack_root *res;
2727 
2728 	const struct uapi_definition   *driver_def;
2729 
2730 	/*
2731 	 * Positive refcount indicates that the device is currently
2732 	 * registered and cannot be unregistered.
2733 	 */
2734 	refcount_t refcount;
2735 	struct completion unreg_completion;
2736 	struct work_struct unregistration_work;
2737 
2738 	const struct rdma_link_ops *link_ops;
2739 
2740 	/* Protects compat_devs xarray modifications */
2741 	struct mutex compat_devs_mutex;
2742 	/* Maintains compat devices for each net namespace */
2743 	struct xarray compat_devs;
2744 
2745 	/* Used by iWarp CM */
2746 	char iw_ifname[IFNAMSIZ];
2747 	u32 iw_driver_flags;
2748 	u32 lag_flags;
2749 };
2750 
2751 struct ib_client_nl_info;
2752 struct ib_client {
2753 	const char *name;
2754 	int (*add)(struct ib_device *ibdev);
2755 	void (*remove)(struct ib_device *, void *client_data);
2756 	void (*rename)(struct ib_device *dev, void *client_data);
2757 	int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2758 			   struct ib_client_nl_info *res);
2759 	int (*get_global_nl_info)(struct ib_client_nl_info *res);
2760 
2761 	/* Returns the net_dev belonging to this ib_client and matching the
2762 	 * given parameters.
2763 	 * @dev:	 An RDMA device that the net_dev use for communication.
2764 	 * @port:	 A physical port number on the RDMA device.
2765 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2766 	 * @gid:	 A GID that the net_dev uses to communicate.
2767 	 * @addr:	 An IP address the net_dev is configured with.
2768 	 * @client_data: The device's client data set by ib_set_client_data().
2769 	 *
2770 	 * An ib_client that implements a net_dev on top of RDMA devices
2771 	 * (such as IP over IB) should implement this callback, allowing the
2772 	 * rdma_cm module to find the right net_dev for a given request.
2773 	 *
2774 	 * The caller is responsible for calling dev_put on the returned
2775 	 * netdev. */
2776 	struct net_device *(*get_net_dev_by_params)(
2777 			struct ib_device *dev,
2778 			u8 port,
2779 			u16 pkey,
2780 			const union ib_gid *gid,
2781 			const struct sockaddr *addr,
2782 			void *client_data);
2783 
2784 	refcount_t uses;
2785 	struct completion uses_zero;
2786 	u32 client_id;
2787 
2788 	/* kverbs are not required by the client */
2789 	u8 no_kverbs_req:1;
2790 };
2791 
2792 /*
2793  * IB block DMA iterator
2794  *
2795  * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2796  * to a HW supported page size.
2797  */
2798 struct ib_block_iter {
2799 	/* internal states */
2800 	struct scatterlist *__sg;	/* sg holding the current aligned block */
2801 	dma_addr_t __dma_addr;		/* unaligned DMA address of this block */
2802 	unsigned int __sg_nents;	/* number of SG entries */
2803 	unsigned int __sg_advance;	/* number of bytes to advance in sg in next step */
2804 	unsigned int __pg_bit;		/* alignment of current block */
2805 };
2806 
2807 struct ib_device *_ib_alloc_device(size_t size);
2808 #define ib_alloc_device(drv_struct, member)                                    \
2809 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2810 				      BUILD_BUG_ON_ZERO(offsetof(              \
2811 					      struct drv_struct, member))),    \
2812 		     struct drv_struct, member)
2813 
2814 void ib_dealloc_device(struct ib_device *device);
2815 
2816 void ib_get_device_fw_str(struct ib_device *device, char *str);
2817 
2818 int ib_register_device(struct ib_device *device, const char *name);
2819 void ib_unregister_device(struct ib_device *device);
2820 void ib_unregister_driver(enum rdma_driver_id driver_id);
2821 void ib_unregister_device_and_put(struct ib_device *device);
2822 void ib_unregister_device_queued(struct ib_device *ib_dev);
2823 
2824 int ib_register_client   (struct ib_client *client);
2825 void ib_unregister_client(struct ib_client *client);
2826 
2827 void __rdma_block_iter_start(struct ib_block_iter *biter,
2828 			     struct scatterlist *sglist,
2829 			     unsigned int nents,
2830 			     unsigned long pgsz);
2831 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2832 
2833 /**
2834  * rdma_block_iter_dma_address - get the aligned dma address of the current
2835  * block held by the block iterator.
2836  * @biter: block iterator holding the memory block
2837  */
2838 static inline dma_addr_t
2839 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2840 {
2841 	return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2842 }
2843 
2844 /**
2845  * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2846  * @sglist: sglist to iterate over
2847  * @biter: block iterator holding the memory block
2848  * @nents: maximum number of sg entries to iterate over
2849  * @pgsz: best HW supported page size to use
2850  *
2851  * Callers may use rdma_block_iter_dma_address() to get each
2852  * blocks aligned DMA address.
2853  */
2854 #define rdma_for_each_block(sglist, biter, nents, pgsz)		\
2855 	for (__rdma_block_iter_start(biter, sglist, nents,	\
2856 				     pgsz);			\
2857 	     __rdma_block_iter_next(biter);)
2858 
2859 /**
2860  * ib_get_client_data - Get IB client context
2861  * @device:Device to get context for
2862  * @client:Client to get context for
2863  *
2864  * ib_get_client_data() returns the client context data set with
2865  * ib_set_client_data(). This can only be called while the client is
2866  * registered to the device, once the ib_client remove() callback returns this
2867  * cannot be called.
2868  */
2869 static inline void *ib_get_client_data(struct ib_device *device,
2870 				       struct ib_client *client)
2871 {
2872 	return xa_load(&device->client_data, client->client_id);
2873 }
2874 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2875 			 void *data);
2876 void ib_set_device_ops(struct ib_device *device,
2877 		       const struct ib_device_ops *ops);
2878 
2879 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2880 		      unsigned long pfn, unsigned long size, pgprot_t prot,
2881 		      struct rdma_user_mmap_entry *entry);
2882 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2883 				struct rdma_user_mmap_entry *entry,
2884 				size_t length);
2885 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2886 				      struct rdma_user_mmap_entry *entry,
2887 				      size_t length, u32 min_pgoff,
2888 				      u32 max_pgoff);
2889 
2890 struct rdma_user_mmap_entry *
2891 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2892 			       unsigned long pgoff);
2893 struct rdma_user_mmap_entry *
2894 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2895 			 struct vm_area_struct *vma);
2896 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2897 
2898 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2899 
2900 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2901 {
2902 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2903 }
2904 
2905 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2906 {
2907 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2908 }
2909 
2910 static inline bool ib_is_buffer_cleared(const void __user *p,
2911 					size_t len)
2912 {
2913 	bool ret;
2914 	u8 *buf;
2915 
2916 	if (len > USHRT_MAX)
2917 		return false;
2918 
2919 	buf = memdup_user(p, len);
2920 	if (IS_ERR(buf))
2921 		return false;
2922 
2923 	ret = !memchr_inv(buf, 0, len);
2924 	kfree(buf);
2925 	return ret;
2926 }
2927 
2928 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2929 				       size_t offset,
2930 				       size_t len)
2931 {
2932 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
2933 }
2934 
2935 /**
2936  * ib_is_destroy_retryable - Check whether the uobject destruction
2937  * is retryable.
2938  * @ret: The initial destruction return code
2939  * @why: remove reason
2940  * @uobj: The uobject that is destroyed
2941  *
2942  * This function is a helper function that IB layer and low-level drivers
2943  * can use to consider whether the destruction of the given uobject is
2944  * retry-able.
2945  * It checks the original return code, if it wasn't success the destruction
2946  * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2947  * the remove reason. (i.e. why).
2948  * Must be called with the object locked for destroy.
2949  */
2950 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2951 					   struct ib_uobject *uobj)
2952 {
2953 	return ret && (why == RDMA_REMOVE_DESTROY ||
2954 		       uobj->context->cleanup_retryable);
2955 }
2956 
2957 /**
2958  * ib_destroy_usecnt - Called during destruction to check the usecnt
2959  * @usecnt: The usecnt atomic
2960  * @why: remove reason
2961  * @uobj: The uobject that is destroyed
2962  *
2963  * Non-zero usecnts will block destruction unless destruction was triggered by
2964  * a ucontext cleanup.
2965  */
2966 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2967 				    enum rdma_remove_reason why,
2968 				    struct ib_uobject *uobj)
2969 {
2970 	if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2971 		return -EBUSY;
2972 	return 0;
2973 }
2974 
2975 /**
2976  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2977  * contains all required attributes and no attributes not allowed for
2978  * the given QP state transition.
2979  * @cur_state: Current QP state
2980  * @next_state: Next QP state
2981  * @type: QP type
2982  * @mask: Mask of supplied QP attributes
2983  *
2984  * This function is a helper function that a low-level driver's
2985  * modify_qp method can use to validate the consumer's input.  It
2986  * checks that cur_state and next_state are valid QP states, that a
2987  * transition from cur_state to next_state is allowed by the IB spec,
2988  * and that the attribute mask supplied is allowed for the transition.
2989  */
2990 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2991 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
2992 
2993 void ib_register_event_handler(struct ib_event_handler *event_handler);
2994 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2995 void ib_dispatch_event(const struct ib_event *event);
2996 
2997 int ib_query_port(struct ib_device *device,
2998 		  u8 port_num, struct ib_port_attr *port_attr);
2999 
3000 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3001 					       u8 port_num);
3002 
3003 /**
3004  * rdma_cap_ib_switch - Check if the device is IB switch
3005  * @device: Device to check
3006  *
3007  * Device driver is responsible for setting is_switch bit on
3008  * in ib_device structure at init time.
3009  *
3010  * Return: true if the device is IB switch.
3011  */
3012 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3013 {
3014 	return device->is_switch;
3015 }
3016 
3017 /**
3018  * rdma_start_port - Return the first valid port number for the device
3019  * specified
3020  *
3021  * @device: Device to be checked
3022  *
3023  * Return start port number
3024  */
3025 static inline u8 rdma_start_port(const struct ib_device *device)
3026 {
3027 	return rdma_cap_ib_switch(device) ? 0 : 1;
3028 }
3029 
3030 /**
3031  * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3032  * @device - The struct ib_device * to iterate over
3033  * @iter - The unsigned int to store the port number
3034  */
3035 #define rdma_for_each_port(device, iter)                                       \
3036 	for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type(   \
3037 						     unsigned int, iter)));    \
3038 	     iter <= rdma_end_port(device); (iter)++)
3039 
3040 /**
3041  * rdma_end_port - Return the last valid port number for the device
3042  * specified
3043  *
3044  * @device: Device to be checked
3045  *
3046  * Return last port number
3047  */
3048 static inline u8 rdma_end_port(const struct ib_device *device)
3049 {
3050 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3051 }
3052 
3053 static inline int rdma_is_port_valid(const struct ib_device *device,
3054 				     unsigned int port)
3055 {
3056 	return (port >= rdma_start_port(device) &&
3057 		port <= rdma_end_port(device));
3058 }
3059 
3060 static inline bool rdma_is_grh_required(const struct ib_device *device,
3061 					u8 port_num)
3062 {
3063 	return device->port_data[port_num].immutable.core_cap_flags &
3064 	       RDMA_CORE_PORT_IB_GRH_REQUIRED;
3065 }
3066 
3067 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
3068 {
3069 	return device->port_data[port_num].immutable.core_cap_flags &
3070 	       RDMA_CORE_CAP_PROT_IB;
3071 }
3072 
3073 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
3074 {
3075 	return device->port_data[port_num].immutable.core_cap_flags &
3076 	       (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3077 }
3078 
3079 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
3080 {
3081 	return device->port_data[port_num].immutable.core_cap_flags &
3082 	       RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3083 }
3084 
3085 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
3086 {
3087 	return device->port_data[port_num].immutable.core_cap_flags &
3088 	       RDMA_CORE_CAP_PROT_ROCE;
3089 }
3090 
3091 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
3092 {
3093 	return device->port_data[port_num].immutable.core_cap_flags &
3094 	       RDMA_CORE_CAP_PROT_IWARP;
3095 }
3096 
3097 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
3098 {
3099 	return rdma_protocol_ib(device, port_num) ||
3100 		rdma_protocol_roce(device, port_num);
3101 }
3102 
3103 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
3104 {
3105 	return device->port_data[port_num].immutable.core_cap_flags &
3106 	       RDMA_CORE_CAP_PROT_RAW_PACKET;
3107 }
3108 
3109 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
3110 {
3111 	return device->port_data[port_num].immutable.core_cap_flags &
3112 	       RDMA_CORE_CAP_PROT_USNIC;
3113 }
3114 
3115 /**
3116  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3117  * Management Datagrams.
3118  * @device: Device to check
3119  * @port_num: Port number to check
3120  *
3121  * Management Datagrams (MAD) are a required part of the InfiniBand
3122  * specification and are supported on all InfiniBand devices.  A slightly
3123  * extended version are also supported on OPA interfaces.
3124  *
3125  * Return: true if the port supports sending/receiving of MAD packets.
3126  */
3127 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
3128 {
3129 	return device->port_data[port_num].immutable.core_cap_flags &
3130 	       RDMA_CORE_CAP_IB_MAD;
3131 }
3132 
3133 /**
3134  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3135  * Management Datagrams.
3136  * @device: Device to check
3137  * @port_num: Port number to check
3138  *
3139  * Intel OmniPath devices extend and/or replace the InfiniBand Management
3140  * datagrams with their own versions.  These OPA MADs share many but not all of
3141  * the characteristics of InfiniBand MADs.
3142  *
3143  * OPA MADs differ in the following ways:
3144  *
3145  *    1) MADs are variable size up to 2K
3146  *       IBTA defined MADs remain fixed at 256 bytes
3147  *    2) OPA SMPs must carry valid PKeys
3148  *    3) OPA SMP packets are a different format
3149  *
3150  * Return: true if the port supports OPA MAD packet formats.
3151  */
3152 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3153 {
3154 	return device->port_data[port_num].immutable.core_cap_flags &
3155 		RDMA_CORE_CAP_OPA_MAD;
3156 }
3157 
3158 /**
3159  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3160  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3161  * @device: Device to check
3162  * @port_num: Port number to check
3163  *
3164  * Each InfiniBand node is required to provide a Subnet Management Agent
3165  * that the subnet manager can access.  Prior to the fabric being fully
3166  * configured by the subnet manager, the SMA is accessed via a well known
3167  * interface called the Subnet Management Interface (SMI).  This interface
3168  * uses directed route packets to communicate with the SM to get around the
3169  * chicken and egg problem of the SM needing to know what's on the fabric
3170  * in order to configure the fabric, and needing to configure the fabric in
3171  * order to send packets to the devices on the fabric.  These directed
3172  * route packets do not need the fabric fully configured in order to reach
3173  * their destination.  The SMI is the only method allowed to send
3174  * directed route packets on an InfiniBand fabric.
3175  *
3176  * Return: true if the port provides an SMI.
3177  */
3178 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3179 {
3180 	return device->port_data[port_num].immutable.core_cap_flags &
3181 	       RDMA_CORE_CAP_IB_SMI;
3182 }
3183 
3184 /**
3185  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3186  * Communication Manager.
3187  * @device: Device to check
3188  * @port_num: Port number to check
3189  *
3190  * The InfiniBand Communication Manager is one of many pre-defined General
3191  * Service Agents (GSA) that are accessed via the General Service
3192  * Interface (GSI).  It's role is to facilitate establishment of connections
3193  * between nodes as well as other management related tasks for established
3194  * connections.
3195  *
3196  * Return: true if the port supports an IB CM (this does not guarantee that
3197  * a CM is actually running however).
3198  */
3199 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3200 {
3201 	return device->port_data[port_num].immutable.core_cap_flags &
3202 	       RDMA_CORE_CAP_IB_CM;
3203 }
3204 
3205 /**
3206  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3207  * Communication Manager.
3208  * @device: Device to check
3209  * @port_num: Port number to check
3210  *
3211  * Similar to above, but specific to iWARP connections which have a different
3212  * managment protocol than InfiniBand.
3213  *
3214  * Return: true if the port supports an iWARP CM (this does not guarantee that
3215  * a CM is actually running however).
3216  */
3217 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3218 {
3219 	return device->port_data[port_num].immutable.core_cap_flags &
3220 	       RDMA_CORE_CAP_IW_CM;
3221 }
3222 
3223 /**
3224  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3225  * Subnet Administration.
3226  * @device: Device to check
3227  * @port_num: Port number to check
3228  *
3229  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3230  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
3231  * fabrics, devices should resolve routes to other hosts by contacting the
3232  * SA to query the proper route.
3233  *
3234  * Return: true if the port should act as a client to the fabric Subnet
3235  * Administration interface.  This does not imply that the SA service is
3236  * running locally.
3237  */
3238 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3239 {
3240 	return device->port_data[port_num].immutable.core_cap_flags &
3241 	       RDMA_CORE_CAP_IB_SA;
3242 }
3243 
3244 /**
3245  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3246  * Multicast.
3247  * @device: Device to check
3248  * @port_num: Port number to check
3249  *
3250  * InfiniBand multicast registration is more complex than normal IPv4 or
3251  * IPv6 multicast registration.  Each Host Channel Adapter must register
3252  * with the Subnet Manager when it wishes to join a multicast group.  It
3253  * should do so only once regardless of how many queue pairs it subscribes
3254  * to this group.  And it should leave the group only after all queue pairs
3255  * attached to the group have been detached.
3256  *
3257  * Return: true if the port must undertake the additional adminstrative
3258  * overhead of registering/unregistering with the SM and tracking of the
3259  * total number of queue pairs attached to the multicast group.
3260  */
3261 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3262 {
3263 	return rdma_cap_ib_sa(device, port_num);
3264 }
3265 
3266 /**
3267  * rdma_cap_af_ib - Check if the port of device has the capability
3268  * Native Infiniband Address.
3269  * @device: Device to check
3270  * @port_num: Port number to check
3271  *
3272  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3273  * GID.  RoCE uses a different mechanism, but still generates a GID via
3274  * a prescribed mechanism and port specific data.
3275  *
3276  * Return: true if the port uses a GID address to identify devices on the
3277  * network.
3278  */
3279 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3280 {
3281 	return device->port_data[port_num].immutable.core_cap_flags &
3282 	       RDMA_CORE_CAP_AF_IB;
3283 }
3284 
3285 /**
3286  * rdma_cap_eth_ah - Check if the port of device has the capability
3287  * Ethernet Address Handle.
3288  * @device: Device to check
3289  * @port_num: Port number to check
3290  *
3291  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3292  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3293  * port.  Normally, packet headers are generated by the sending host
3294  * adapter, but when sending connectionless datagrams, we must manually
3295  * inject the proper headers for the fabric we are communicating over.
3296  *
3297  * Return: true if we are running as a RoCE port and must force the
3298  * addition of a Global Route Header built from our Ethernet Address
3299  * Handle into our header list for connectionless packets.
3300  */
3301 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3302 {
3303 	return device->port_data[port_num].immutable.core_cap_flags &
3304 	       RDMA_CORE_CAP_ETH_AH;
3305 }
3306 
3307 /**
3308  * rdma_cap_opa_ah - Check if the port of device supports
3309  * OPA Address handles
3310  * @device: Device to check
3311  * @port_num: Port number to check
3312  *
3313  * Return: true if we are running on an OPA device which supports
3314  * the extended OPA addressing.
3315  */
3316 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3317 {
3318 	return (device->port_data[port_num].immutable.core_cap_flags &
3319 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3320 }
3321 
3322 /**
3323  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3324  *
3325  * @device: Device
3326  * @port_num: Port number
3327  *
3328  * This MAD size includes the MAD headers and MAD payload.  No other headers
3329  * are included.
3330  *
3331  * Return the max MAD size required by the Port.  Will return 0 if the port
3332  * does not support MADs
3333  */
3334 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3335 {
3336 	return device->port_data[port_num].immutable.max_mad_size;
3337 }
3338 
3339 /**
3340  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3341  * @device: Device to check
3342  * @port_num: Port number to check
3343  *
3344  * RoCE GID table mechanism manages the various GIDs for a device.
3345  *
3346  * NOTE: if allocating the port's GID table has failed, this call will still
3347  * return true, but any RoCE GID table API will fail.
3348  *
3349  * Return: true if the port uses RoCE GID table mechanism in order to manage
3350  * its GIDs.
3351  */
3352 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3353 					   u8 port_num)
3354 {
3355 	return rdma_protocol_roce(device, port_num) &&
3356 		device->ops.add_gid && device->ops.del_gid;
3357 }
3358 
3359 /*
3360  * Check if the device supports READ W/ INVALIDATE.
3361  */
3362 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3363 {
3364 	/*
3365 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3366 	 * has support for it yet.
3367 	 */
3368 	return rdma_protocol_iwarp(dev, port_num);
3369 }
3370 
3371 /**
3372  * rdma_find_pg_bit - Find page bit given address and HW supported page sizes
3373  *
3374  * @addr: address
3375  * @pgsz_bitmap: bitmap of HW supported page sizes
3376  */
3377 static inline unsigned int rdma_find_pg_bit(unsigned long addr,
3378 					    unsigned long pgsz_bitmap)
3379 {
3380 	unsigned long align;
3381 	unsigned long pgsz;
3382 
3383 	align = addr & -addr;
3384 
3385 	/* Find page bit such that addr is aligned to the highest supported
3386 	 * HW page size
3387 	 */
3388 	pgsz = pgsz_bitmap & ~(-align << 1);
3389 	if (!pgsz)
3390 		return __ffs(pgsz_bitmap);
3391 
3392 	return __fls(pgsz);
3393 }
3394 
3395 /**
3396  * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3397  * @device: Device
3398  * @port_num: 1 based Port number
3399  *
3400  * Return true if port is an Intel OPA port , false if not
3401  */
3402 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3403 					  u32 port_num)
3404 {
3405 	return (device->port_data[port_num].immutable.core_cap_flags &
3406 		RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3407 }
3408 
3409 /**
3410  * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3411  * @device: Device
3412  * @port_num: Port number
3413  * @mtu: enum value of MTU
3414  *
3415  * Return the MTU size supported by the port as an integer value. Will return
3416  * -1 if enum value of mtu is not supported.
3417  */
3418 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u8 port,
3419 				       int mtu)
3420 {
3421 	if (rdma_core_cap_opa_port(device, port))
3422 		return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3423 	else
3424 		return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3425 }
3426 
3427 /**
3428  * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3429  * @device: Device
3430  * @port_num: Port number
3431  * @attr: port attribute
3432  *
3433  * Return the MTU size supported by the port as an integer value.
3434  */
3435 static inline int rdma_mtu_from_attr(struct ib_device *device, u8 port,
3436 				     struct ib_port_attr *attr)
3437 {
3438 	if (rdma_core_cap_opa_port(device, port))
3439 		return attr->phys_mtu;
3440 	else
3441 		return ib_mtu_enum_to_int(attr->max_mtu);
3442 }
3443 
3444 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3445 			 int state);
3446 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3447 		     struct ifla_vf_info *info);
3448 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3449 		    struct ifla_vf_stats *stats);
3450 int ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
3451 		    struct ifla_vf_guid *node_guid,
3452 		    struct ifla_vf_guid *port_guid);
3453 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3454 		   int type);
3455 
3456 int ib_query_pkey(struct ib_device *device,
3457 		  u8 port_num, u16 index, u16 *pkey);
3458 
3459 int ib_modify_device(struct ib_device *device,
3460 		     int device_modify_mask,
3461 		     struct ib_device_modify *device_modify);
3462 
3463 int ib_modify_port(struct ib_device *device,
3464 		   u8 port_num, int port_modify_mask,
3465 		   struct ib_port_modify *port_modify);
3466 
3467 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3468 		u8 *port_num, u16 *index);
3469 
3470 int ib_find_pkey(struct ib_device *device,
3471 		 u8 port_num, u16 pkey, u16 *index);
3472 
3473 enum ib_pd_flags {
3474 	/*
3475 	 * Create a memory registration for all memory in the system and place
3476 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3477 	 * ULPs to avoid the overhead of dynamic MRs.
3478 	 *
3479 	 * This flag is generally considered unsafe and must only be used in
3480 	 * extremly trusted environments.  Every use of it will log a warning
3481 	 * in the kernel log.
3482 	 */
3483 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3484 };
3485 
3486 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3487 		const char *caller);
3488 
3489 #define ib_alloc_pd(device, flags) \
3490 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3491 
3492 /**
3493  * ib_dealloc_pd_user - Deallocate kernel/user PD
3494  * @pd: The protection domain
3495  * @udata: Valid user data or NULL for kernel objects
3496  */
3497 void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3498 
3499 /**
3500  * ib_dealloc_pd - Deallocate kernel PD
3501  * @pd: The protection domain
3502  *
3503  * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3504  */
3505 static inline void ib_dealloc_pd(struct ib_pd *pd)
3506 {
3507 	ib_dealloc_pd_user(pd, NULL);
3508 }
3509 
3510 enum rdma_create_ah_flags {
3511 	/* In a sleepable context */
3512 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3513 };
3514 
3515 /**
3516  * rdma_create_ah - Creates an address handle for the given address vector.
3517  * @pd: The protection domain associated with the address handle.
3518  * @ah_attr: The attributes of the address vector.
3519  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3520  *
3521  * The address handle is used to reference a local or global destination
3522  * in all UD QP post sends.
3523  */
3524 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3525 			     u32 flags);
3526 
3527 /**
3528  * rdma_create_user_ah - Creates an address handle for the given address vector.
3529  * It resolves destination mac address for ah attribute of RoCE type.
3530  * @pd: The protection domain associated with the address handle.
3531  * @ah_attr: The attributes of the address vector.
3532  * @udata: pointer to user's input output buffer information need by
3533  *         provider driver.
3534  *
3535  * It returns 0 on success and returns appropriate error code on error.
3536  * The address handle is used to reference a local or global destination
3537  * in all UD QP post sends.
3538  */
3539 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3540 				  struct rdma_ah_attr *ah_attr,
3541 				  struct ib_udata *udata);
3542 /**
3543  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3544  *   work completion.
3545  * @hdr: the L3 header to parse
3546  * @net_type: type of header to parse
3547  * @sgid: place to store source gid
3548  * @dgid: place to store destination gid
3549  */
3550 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3551 			      enum rdma_network_type net_type,
3552 			      union ib_gid *sgid, union ib_gid *dgid);
3553 
3554 /**
3555  * ib_get_rdma_header_version - Get the header version
3556  * @hdr: the L3 header to parse
3557  */
3558 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3559 
3560 /**
3561  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3562  *   work completion.
3563  * @device: Device on which the received message arrived.
3564  * @port_num: Port on which the received message arrived.
3565  * @wc: Work completion associated with the received message.
3566  * @grh: References the received global route header.  This parameter is
3567  *   ignored unless the work completion indicates that the GRH is valid.
3568  * @ah_attr: Returned attributes that can be used when creating an address
3569  *   handle for replying to the message.
3570  * When ib_init_ah_attr_from_wc() returns success,
3571  * (a) for IB link layer it optionally contains a reference to SGID attribute
3572  * when GRH is present for IB link layer.
3573  * (b) for RoCE link layer it contains a reference to SGID attribute.
3574  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3575  * attributes which are initialized using ib_init_ah_attr_from_wc().
3576  *
3577  */
3578 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3579 			    const struct ib_wc *wc, const struct ib_grh *grh,
3580 			    struct rdma_ah_attr *ah_attr);
3581 
3582 /**
3583  * ib_create_ah_from_wc - Creates an address handle associated with the
3584  *   sender of the specified work completion.
3585  * @pd: The protection domain associated with the address handle.
3586  * @wc: Work completion information associated with a received message.
3587  * @grh: References the received global route header.  This parameter is
3588  *   ignored unless the work completion indicates that the GRH is valid.
3589  * @port_num: The outbound port number to associate with the address.
3590  *
3591  * The address handle is used to reference a local or global destination
3592  * in all UD QP post sends.
3593  */
3594 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3595 				   const struct ib_grh *grh, u8 port_num);
3596 
3597 /**
3598  * rdma_modify_ah - Modifies the address vector associated with an address
3599  *   handle.
3600  * @ah: The address handle to modify.
3601  * @ah_attr: The new address vector attributes to associate with the
3602  *   address handle.
3603  */
3604 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3605 
3606 /**
3607  * rdma_query_ah - Queries the address vector associated with an address
3608  *   handle.
3609  * @ah: The address handle to query.
3610  * @ah_attr: The address vector attributes associated with the address
3611  *   handle.
3612  */
3613 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3614 
3615 enum rdma_destroy_ah_flags {
3616 	/* In a sleepable context */
3617 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3618 };
3619 
3620 /**
3621  * rdma_destroy_ah_user - Destroys an address handle.
3622  * @ah: The address handle to destroy.
3623  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3624  * @udata: Valid user data or NULL for kernel objects
3625  */
3626 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3627 
3628 /**
3629  * rdma_destroy_ah - Destroys an kernel address handle.
3630  * @ah: The address handle to destroy.
3631  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3632  *
3633  * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3634  */
3635 static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3636 {
3637 	return rdma_destroy_ah_user(ah, flags, NULL);
3638 }
3639 
3640 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3641 				  struct ib_srq_init_attr *srq_init_attr,
3642 				  struct ib_usrq_object *uobject,
3643 				  struct ib_udata *udata);
3644 static inline struct ib_srq *
3645 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3646 {
3647 	if (!pd->device->ops.create_srq)
3648 		return ERR_PTR(-EOPNOTSUPP);
3649 
3650 	return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3651 }
3652 
3653 /**
3654  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3655  * @srq: The SRQ to modify.
3656  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3657  *   the current values of selected SRQ attributes are returned.
3658  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3659  *   are being modified.
3660  *
3661  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3662  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3663  * the number of receives queued drops below the limit.
3664  */
3665 int ib_modify_srq(struct ib_srq *srq,
3666 		  struct ib_srq_attr *srq_attr,
3667 		  enum ib_srq_attr_mask srq_attr_mask);
3668 
3669 /**
3670  * ib_query_srq - Returns the attribute list and current values for the
3671  *   specified SRQ.
3672  * @srq: The SRQ to query.
3673  * @srq_attr: The attributes of the specified SRQ.
3674  */
3675 int ib_query_srq(struct ib_srq *srq,
3676 		 struct ib_srq_attr *srq_attr);
3677 
3678 /**
3679  * ib_destroy_srq_user - Destroys the specified SRQ.
3680  * @srq: The SRQ to destroy.
3681  * @udata: Valid user data or NULL for kernel objects
3682  */
3683 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3684 
3685 /**
3686  * ib_destroy_srq - Destroys the specified kernel SRQ.
3687  * @srq: The SRQ to destroy.
3688  *
3689  * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3690  */
3691 static inline int ib_destroy_srq(struct ib_srq *srq)
3692 {
3693 	return ib_destroy_srq_user(srq, NULL);
3694 }
3695 
3696 /**
3697  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3698  * @srq: The SRQ to post the work request on.
3699  * @recv_wr: A list of work requests to post on the receive queue.
3700  * @bad_recv_wr: On an immediate failure, this parameter will reference
3701  *   the work request that failed to be posted on the QP.
3702  */
3703 static inline int ib_post_srq_recv(struct ib_srq *srq,
3704 				   const struct ib_recv_wr *recv_wr,
3705 				   const struct ib_recv_wr **bad_recv_wr)
3706 {
3707 	const struct ib_recv_wr *dummy;
3708 
3709 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3710 					      bad_recv_wr ? : &dummy);
3711 }
3712 
3713 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3714 			   struct ib_qp_init_attr *qp_init_attr);
3715 
3716 /**
3717  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3718  * @qp: The QP to modify.
3719  * @attr: On input, specifies the QP attributes to modify.  On output,
3720  *   the current values of selected QP attributes are returned.
3721  * @attr_mask: A bit-mask used to specify which attributes of the QP
3722  *   are being modified.
3723  * @udata: pointer to user's input output buffer information
3724  *   are being modified.
3725  * It returns 0 on success and returns appropriate error code on error.
3726  */
3727 int ib_modify_qp_with_udata(struct ib_qp *qp,
3728 			    struct ib_qp_attr *attr,
3729 			    int attr_mask,
3730 			    struct ib_udata *udata);
3731 
3732 /**
3733  * ib_modify_qp - Modifies the attributes for the specified QP and then
3734  *   transitions the QP to the given state.
3735  * @qp: The QP to modify.
3736  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3737  *   the current values of selected QP attributes are returned.
3738  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3739  *   are being modified.
3740  */
3741 int ib_modify_qp(struct ib_qp *qp,
3742 		 struct ib_qp_attr *qp_attr,
3743 		 int qp_attr_mask);
3744 
3745 /**
3746  * ib_query_qp - Returns the attribute list and current values for the
3747  *   specified QP.
3748  * @qp: The QP to query.
3749  * @qp_attr: The attributes of the specified QP.
3750  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3751  * @qp_init_attr: Additional attributes of the selected QP.
3752  *
3753  * The qp_attr_mask may be used to limit the query to gathering only the
3754  * selected attributes.
3755  */
3756 int ib_query_qp(struct ib_qp *qp,
3757 		struct ib_qp_attr *qp_attr,
3758 		int qp_attr_mask,
3759 		struct ib_qp_init_attr *qp_init_attr);
3760 
3761 /**
3762  * ib_destroy_qp - Destroys the specified QP.
3763  * @qp: The QP to destroy.
3764  * @udata: Valid udata or NULL for kernel objects
3765  */
3766 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3767 
3768 /**
3769  * ib_destroy_qp - Destroys the specified kernel QP.
3770  * @qp: The QP to destroy.
3771  *
3772  * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3773  */
3774 static inline int ib_destroy_qp(struct ib_qp *qp)
3775 {
3776 	return ib_destroy_qp_user(qp, NULL);
3777 }
3778 
3779 /**
3780  * ib_open_qp - Obtain a reference to an existing sharable QP.
3781  * @xrcd - XRC domain
3782  * @qp_open_attr: Attributes identifying the QP to open.
3783  *
3784  * Returns a reference to a sharable QP.
3785  */
3786 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3787 			 struct ib_qp_open_attr *qp_open_attr);
3788 
3789 /**
3790  * ib_close_qp - Release an external reference to a QP.
3791  * @qp: The QP handle to release
3792  *
3793  * The opened QP handle is released by the caller.  The underlying
3794  * shared QP is not destroyed until all internal references are released.
3795  */
3796 int ib_close_qp(struct ib_qp *qp);
3797 
3798 /**
3799  * ib_post_send - Posts a list of work requests to the send queue of
3800  *   the specified QP.
3801  * @qp: The QP to post the work request on.
3802  * @send_wr: A list of work requests to post on the send queue.
3803  * @bad_send_wr: On an immediate failure, this parameter will reference
3804  *   the work request that failed to be posted on the QP.
3805  *
3806  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3807  * error is returned, the QP state shall not be affected,
3808  * ib_post_send() will return an immediate error after queueing any
3809  * earlier work requests in the list.
3810  */
3811 static inline int ib_post_send(struct ib_qp *qp,
3812 			       const struct ib_send_wr *send_wr,
3813 			       const struct ib_send_wr **bad_send_wr)
3814 {
3815 	const struct ib_send_wr *dummy;
3816 
3817 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3818 }
3819 
3820 /**
3821  * ib_post_recv - Posts a list of work requests to the receive queue of
3822  *   the specified QP.
3823  * @qp: The QP to post the work request on.
3824  * @recv_wr: A list of work requests to post on the receive queue.
3825  * @bad_recv_wr: On an immediate failure, this parameter will reference
3826  *   the work request that failed to be posted on the QP.
3827  */
3828 static inline int ib_post_recv(struct ib_qp *qp,
3829 			       const struct ib_recv_wr *recv_wr,
3830 			       const struct ib_recv_wr **bad_recv_wr)
3831 {
3832 	const struct ib_recv_wr *dummy;
3833 
3834 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3835 }
3836 
3837 struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
3838 				 int nr_cqe, int comp_vector,
3839 				 enum ib_poll_context poll_ctx,
3840 				 const char *caller, struct ib_udata *udata);
3841 
3842 /**
3843  * ib_alloc_cq_user: Allocate kernel/user CQ
3844  * @dev: The IB device
3845  * @private: Private data attached to the CQE
3846  * @nr_cqe: Number of CQEs in the CQ
3847  * @comp_vector: Completion vector used for the IRQs
3848  * @poll_ctx: Context used for polling the CQ
3849  * @udata: Valid user data or NULL for kernel objects
3850  */
3851 static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
3852 					     void *private, int nr_cqe,
3853 					     int comp_vector,
3854 					     enum ib_poll_context poll_ctx,
3855 					     struct ib_udata *udata)
3856 {
3857 	return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3858 				  KBUILD_MODNAME, udata);
3859 }
3860 
3861 /**
3862  * ib_alloc_cq: Allocate kernel CQ
3863  * @dev: The IB device
3864  * @private: Private data attached to the CQE
3865  * @nr_cqe: Number of CQEs in the CQ
3866  * @comp_vector: Completion vector used for the IRQs
3867  * @poll_ctx: Context used for polling the CQ
3868  *
3869  * NOTE: for user cq use ib_alloc_cq_user with valid udata!
3870  */
3871 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3872 					int nr_cqe, int comp_vector,
3873 					enum ib_poll_context poll_ctx)
3874 {
3875 	return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3876 				NULL);
3877 }
3878 
3879 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3880 				int nr_cqe, enum ib_poll_context poll_ctx,
3881 				const char *caller);
3882 
3883 /**
3884  * ib_alloc_cq_any: Allocate kernel CQ
3885  * @dev: The IB device
3886  * @private: Private data attached to the CQE
3887  * @nr_cqe: Number of CQEs in the CQ
3888  * @poll_ctx: Context used for polling the CQ
3889  */
3890 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3891 					    void *private, int nr_cqe,
3892 					    enum ib_poll_context poll_ctx)
3893 {
3894 	return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3895 				 KBUILD_MODNAME);
3896 }
3897 
3898 /**
3899  * ib_free_cq_user - Free kernel/user CQ
3900  * @cq: The CQ to free
3901  * @udata: Valid user data or NULL for kernel objects
3902  *
3903  * NOTE: This function shouldn't be called on shared CQs.
3904  */
3905 void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3906 
3907 /**
3908  * ib_free_cq - Free kernel CQ
3909  * @cq: The CQ to free
3910  *
3911  * NOTE: for user cq use ib_free_cq_user with valid udata!
3912  */
3913 static inline void ib_free_cq(struct ib_cq *cq)
3914 {
3915 	ib_free_cq_user(cq, NULL);
3916 }
3917 
3918 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3919 
3920 /**
3921  * ib_create_cq - Creates a CQ on the specified device.
3922  * @device: The device on which to create the CQ.
3923  * @comp_handler: A user-specified callback that is invoked when a
3924  *   completion event occurs on the CQ.
3925  * @event_handler: A user-specified callback that is invoked when an
3926  *   asynchronous event not associated with a completion occurs on the CQ.
3927  * @cq_context: Context associated with the CQ returned to the user via
3928  *   the associated completion and event handlers.
3929  * @cq_attr: The attributes the CQ should be created upon.
3930  *
3931  * Users can examine the cq structure to determine the actual CQ size.
3932  */
3933 struct ib_cq *__ib_create_cq(struct ib_device *device,
3934 			     ib_comp_handler comp_handler,
3935 			     void (*event_handler)(struct ib_event *, void *),
3936 			     void *cq_context,
3937 			     const struct ib_cq_init_attr *cq_attr,
3938 			     const char *caller);
3939 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3940 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3941 
3942 /**
3943  * ib_resize_cq - Modifies the capacity of the CQ.
3944  * @cq: The CQ to resize.
3945  * @cqe: The minimum size of the CQ.
3946  *
3947  * Users can examine the cq structure to determine the actual CQ size.
3948  */
3949 int ib_resize_cq(struct ib_cq *cq, int cqe);
3950 
3951 /**
3952  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3953  * @cq: The CQ to modify.
3954  * @cq_count: number of CQEs that will trigger an event
3955  * @cq_period: max period of time in usec before triggering an event
3956  *
3957  */
3958 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3959 
3960 /**
3961  * ib_destroy_cq_user - Destroys the specified CQ.
3962  * @cq: The CQ to destroy.
3963  * @udata: Valid user data or NULL for kernel objects
3964  */
3965 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3966 
3967 /**
3968  * ib_destroy_cq - Destroys the specified kernel CQ.
3969  * @cq: The CQ to destroy.
3970  *
3971  * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3972  */
3973 static inline void ib_destroy_cq(struct ib_cq *cq)
3974 {
3975 	ib_destroy_cq_user(cq, NULL);
3976 }
3977 
3978 /**
3979  * ib_poll_cq - poll a CQ for completion(s)
3980  * @cq:the CQ being polled
3981  * @num_entries:maximum number of completions to return
3982  * @wc:array of at least @num_entries &struct ib_wc where completions
3983  *   will be returned
3984  *
3985  * Poll a CQ for (possibly multiple) completions.  If the return value
3986  * is < 0, an error occurred.  If the return value is >= 0, it is the
3987  * number of completions returned.  If the return value is
3988  * non-negative and < num_entries, then the CQ was emptied.
3989  */
3990 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3991 			     struct ib_wc *wc)
3992 {
3993 	return cq->device->ops.poll_cq(cq, num_entries, wc);
3994 }
3995 
3996 /**
3997  * ib_req_notify_cq - Request completion notification on a CQ.
3998  * @cq: The CQ to generate an event for.
3999  * @flags:
4000  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
4001  *   to request an event on the next solicited event or next work
4002  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
4003  *   may also be |ed in to request a hint about missed events, as
4004  *   described below.
4005  *
4006  * Return Value:
4007  *    < 0 means an error occurred while requesting notification
4008  *   == 0 means notification was requested successfully, and if
4009  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
4010  *        were missed and it is safe to wait for another event.  In
4011  *        this case is it guaranteed that any work completions added
4012  *        to the CQ since the last CQ poll will trigger a completion
4013  *        notification event.
4014  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
4015  *        in.  It means that the consumer must poll the CQ again to
4016  *        make sure it is empty to avoid missing an event because of a
4017  *        race between requesting notification and an entry being
4018  *        added to the CQ.  This return value means it is possible
4019  *        (but not guaranteed) that a work completion has been added
4020  *        to the CQ since the last poll without triggering a
4021  *        completion notification event.
4022  */
4023 static inline int ib_req_notify_cq(struct ib_cq *cq,
4024 				   enum ib_cq_notify_flags flags)
4025 {
4026 	return cq->device->ops.req_notify_cq(cq, flags);
4027 }
4028 
4029 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4030 			     int comp_vector_hint,
4031 			     enum ib_poll_context poll_ctx);
4032 
4033 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4034 
4035 /**
4036  * ib_req_ncomp_notif - Request completion notification when there are
4037  *   at least the specified number of unreaped completions on the CQ.
4038  * @cq: The CQ to generate an event for.
4039  * @wc_cnt: The number of unreaped completions that should be on the
4040  *   CQ before an event is generated.
4041  */
4042 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
4043 {
4044 	return cq->device->ops.req_ncomp_notif ?
4045 		cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
4046 		-ENOSYS;
4047 }
4048 
4049 /**
4050  * ib_dma_mapping_error - check a DMA addr for error
4051  * @dev: The device for which the dma_addr was created
4052  * @dma_addr: The DMA address to check
4053  */
4054 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4055 {
4056 	return dma_mapping_error(dev->dma_device, dma_addr);
4057 }
4058 
4059 /**
4060  * ib_dma_map_single - Map a kernel virtual address to DMA address
4061  * @dev: The device for which the dma_addr is to be created
4062  * @cpu_addr: The kernel virtual address
4063  * @size: The size of the region in bytes
4064  * @direction: The direction of the DMA
4065  */
4066 static inline u64 ib_dma_map_single(struct ib_device *dev,
4067 				    void *cpu_addr, size_t size,
4068 				    enum dma_data_direction direction)
4069 {
4070 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4071 }
4072 
4073 /**
4074  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4075  * @dev: The device for which the DMA address was created
4076  * @addr: The DMA address
4077  * @size: The size of the region in bytes
4078  * @direction: The direction of the DMA
4079  */
4080 static inline void ib_dma_unmap_single(struct ib_device *dev,
4081 				       u64 addr, size_t size,
4082 				       enum dma_data_direction direction)
4083 {
4084 	dma_unmap_single(dev->dma_device, addr, size, direction);
4085 }
4086 
4087 /**
4088  * ib_dma_map_page - Map a physical page to DMA address
4089  * @dev: The device for which the dma_addr is to be created
4090  * @page: The page to be mapped
4091  * @offset: The offset within the page
4092  * @size: The size of the region in bytes
4093  * @direction: The direction of the DMA
4094  */
4095 static inline u64 ib_dma_map_page(struct ib_device *dev,
4096 				  struct page *page,
4097 				  unsigned long offset,
4098 				  size_t size,
4099 					 enum dma_data_direction direction)
4100 {
4101 	return dma_map_page(dev->dma_device, page, offset, size, direction);
4102 }
4103 
4104 /**
4105  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4106  * @dev: The device for which the DMA address was created
4107  * @addr: The DMA address
4108  * @size: The size of the region in bytes
4109  * @direction: The direction of the DMA
4110  */
4111 static inline void ib_dma_unmap_page(struct ib_device *dev,
4112 				     u64 addr, size_t size,
4113 				     enum dma_data_direction direction)
4114 {
4115 	dma_unmap_page(dev->dma_device, addr, size, direction);
4116 }
4117 
4118 /**
4119  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4120  * @dev: The device for which the DMA addresses are to be created
4121  * @sg: The array of scatter/gather entries
4122  * @nents: The number of scatter/gather entries
4123  * @direction: The direction of the DMA
4124  */
4125 static inline int ib_dma_map_sg(struct ib_device *dev,
4126 				struct scatterlist *sg, int nents,
4127 				enum dma_data_direction direction)
4128 {
4129 	return dma_map_sg(dev->dma_device, sg, nents, direction);
4130 }
4131 
4132 /**
4133  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4134  * @dev: The device for which the DMA addresses were created
4135  * @sg: The array of scatter/gather entries
4136  * @nents: The number of scatter/gather entries
4137  * @direction: The direction of the DMA
4138  */
4139 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4140 				   struct scatterlist *sg, int nents,
4141 				   enum dma_data_direction direction)
4142 {
4143 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
4144 }
4145 
4146 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4147 				      struct scatterlist *sg, int nents,
4148 				      enum dma_data_direction direction,
4149 				      unsigned long dma_attrs)
4150 {
4151 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4152 				dma_attrs);
4153 }
4154 
4155 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4156 					 struct scatterlist *sg, int nents,
4157 					 enum dma_data_direction direction,
4158 					 unsigned long dma_attrs)
4159 {
4160 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
4161 }
4162 
4163 /**
4164  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4165  * @dev: The device to query
4166  *
4167  * The returned value represents a size in bytes.
4168  */
4169 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4170 {
4171 	return dma_get_max_seg_size(dev->dma_device);
4172 }
4173 
4174 /**
4175  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4176  * @dev: The device for which the DMA address was created
4177  * @addr: The DMA address
4178  * @size: The size of the region in bytes
4179  * @dir: The direction of the DMA
4180  */
4181 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4182 					      u64 addr,
4183 					      size_t size,
4184 					      enum dma_data_direction dir)
4185 {
4186 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4187 }
4188 
4189 /**
4190  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4191  * @dev: The device for which the DMA address was created
4192  * @addr: The DMA address
4193  * @size: The size of the region in bytes
4194  * @dir: The direction of the DMA
4195  */
4196 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4197 						 u64 addr,
4198 						 size_t size,
4199 						 enum dma_data_direction dir)
4200 {
4201 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4202 }
4203 
4204 /**
4205  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4206  * @dev: The device for which the DMA address is requested
4207  * @size: The size of the region to allocate in bytes
4208  * @dma_handle: A pointer for returning the DMA address of the region
4209  * @flag: memory allocator flags
4210  */
4211 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4212 					   size_t size,
4213 					   dma_addr_t *dma_handle,
4214 					   gfp_t flag)
4215 {
4216 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4217 }
4218 
4219 /**
4220  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4221  * @dev: The device for which the DMA addresses were allocated
4222  * @size: The size of the region
4223  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4224  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4225  */
4226 static inline void ib_dma_free_coherent(struct ib_device *dev,
4227 					size_t size, void *cpu_addr,
4228 					dma_addr_t dma_handle)
4229 {
4230 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4231 }
4232 
4233 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4234  * space. This function should be called when 'current' is the owning MM.
4235  */
4236 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4237 			     u64 virt_addr, int mr_access_flags);
4238 
4239 /* ib_advise_mr -  give an advice about an address range in a memory region */
4240 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4241 		 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4242 /**
4243  * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4244  *   HCA translation table.
4245  * @mr: The memory region to deregister.
4246  * @udata: Valid user data or NULL for kernel object
4247  *
4248  * This function can fail, if the memory region has memory windows bound to it.
4249  */
4250 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4251 
4252 /**
4253  * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4254  *   HCA translation table.
4255  * @mr: The memory region to deregister.
4256  *
4257  * This function can fail, if the memory region has memory windows bound to it.
4258  *
4259  * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4260  */
4261 static inline int ib_dereg_mr(struct ib_mr *mr)
4262 {
4263 	return ib_dereg_mr_user(mr, NULL);
4264 }
4265 
4266 struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type,
4267 			       u32 max_num_sg, struct ib_udata *udata);
4268 
4269 static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
4270 					enum ib_mr_type mr_type, u32 max_num_sg)
4271 {
4272 	return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL);
4273 }
4274 
4275 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4276 				    u32 max_num_data_sg,
4277 				    u32 max_num_meta_sg);
4278 
4279 /**
4280  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4281  *   R_Key and L_Key.
4282  * @mr - struct ib_mr pointer to be updated.
4283  * @newkey - new key to be used.
4284  */
4285 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4286 {
4287 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4288 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4289 }
4290 
4291 /**
4292  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4293  * for calculating a new rkey for type 2 memory windows.
4294  * @rkey - the rkey to increment.
4295  */
4296 static inline u32 ib_inc_rkey(u32 rkey)
4297 {
4298 	const u32 mask = 0x000000ff;
4299 	return ((rkey + 1) & mask) | (rkey & ~mask);
4300 }
4301 
4302 /**
4303  * ib_attach_mcast - Attaches the specified QP to a multicast group.
4304  * @qp: QP to attach to the multicast group.  The QP must be type
4305  *   IB_QPT_UD.
4306  * @gid: Multicast group GID.
4307  * @lid: Multicast group LID in host byte order.
4308  *
4309  * In order to send and receive multicast packets, subnet
4310  * administration must have created the multicast group and configured
4311  * the fabric appropriately.  The port associated with the specified
4312  * QP must also be a member of the multicast group.
4313  */
4314 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4315 
4316 /**
4317  * ib_detach_mcast - Detaches the specified QP from a multicast group.
4318  * @qp: QP to detach from the multicast group.
4319  * @gid: Multicast group GID.
4320  * @lid: Multicast group LID in host byte order.
4321  */
4322 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4323 
4324 /**
4325  * ib_alloc_xrcd - Allocates an XRC domain.
4326  * @device: The device on which to allocate the XRC domain.
4327  * @caller: Module name for kernel consumers
4328  */
4329 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
4330 #define ib_alloc_xrcd(device) \
4331 	__ib_alloc_xrcd((device), KBUILD_MODNAME)
4332 
4333 /**
4334  * ib_dealloc_xrcd - Deallocates an XRC domain.
4335  * @xrcd: The XRC domain to deallocate.
4336  * @udata: Valid user data or NULL for kernel object
4337  */
4338 int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
4339 
4340 static inline int ib_check_mr_access(int flags)
4341 {
4342 	/*
4343 	 * Local write permission is required if remote write or
4344 	 * remote atomic permission is also requested.
4345 	 */
4346 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4347 	    !(flags & IB_ACCESS_LOCAL_WRITE))
4348 		return -EINVAL;
4349 
4350 	if (flags & ~IB_ACCESS_SUPPORTED)
4351 		return -EINVAL;
4352 
4353 	return 0;
4354 }
4355 
4356 static inline bool ib_access_writable(int access_flags)
4357 {
4358 	/*
4359 	 * We have writable memory backing the MR if any of the following
4360 	 * access flags are set.  "Local write" and "remote write" obviously
4361 	 * require write access.  "Remote atomic" can do things like fetch and
4362 	 * add, which will modify memory, and "MW bind" can change permissions
4363 	 * by binding a window.
4364 	 */
4365 	return access_flags &
4366 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
4367 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4368 }
4369 
4370 /**
4371  * ib_check_mr_status: lightweight check of MR status.
4372  *     This routine may provide status checks on a selected
4373  *     ib_mr. first use is for signature status check.
4374  *
4375  * @mr: A memory region.
4376  * @check_mask: Bitmask of which checks to perform from
4377  *     ib_mr_status_check enumeration.
4378  * @mr_status: The container of relevant status checks.
4379  *     failed checks will be indicated in the status bitmask
4380  *     and the relevant info shall be in the error item.
4381  */
4382 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4383 		       struct ib_mr_status *mr_status);
4384 
4385 /**
4386  * ib_device_try_get: Hold a registration lock
4387  * device: The device to lock
4388  *
4389  * A device under an active registration lock cannot become unregistered. It
4390  * is only possible to obtain a registration lock on a device that is fully
4391  * registered, otherwise this function returns false.
4392  *
4393  * The registration lock is only necessary for actions which require the
4394  * device to still be registered. Uses that only require the device pointer to
4395  * be valid should use get_device(&ibdev->dev) to hold the memory.
4396  *
4397  */
4398 static inline bool ib_device_try_get(struct ib_device *dev)
4399 {
4400 	return refcount_inc_not_zero(&dev->refcount);
4401 }
4402 
4403 void ib_device_put(struct ib_device *device);
4404 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4405 					  enum rdma_driver_id driver_id);
4406 struct ib_device *ib_device_get_by_name(const char *name,
4407 					enum rdma_driver_id driver_id);
4408 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4409 					    u16 pkey, const union ib_gid *gid,
4410 					    const struct sockaddr *addr);
4411 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4412 			 unsigned int port);
4413 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4414 
4415 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4416 			   struct ib_wq_init_attr *init_attr);
4417 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
4418 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4419 		 u32 wq_attr_mask);
4420 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4421 						 struct ib_rwq_ind_table_init_attr*
4422 						 wq_ind_table_init_attr);
4423 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4424 
4425 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4426 		 unsigned int *sg_offset, unsigned int page_size);
4427 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4428 		    int data_sg_nents, unsigned int *data_sg_offset,
4429 		    struct scatterlist *meta_sg, int meta_sg_nents,
4430 		    unsigned int *meta_sg_offset, unsigned int page_size);
4431 
4432 static inline int
4433 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4434 		  unsigned int *sg_offset, unsigned int page_size)
4435 {
4436 	int n;
4437 
4438 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4439 	mr->iova = 0;
4440 
4441 	return n;
4442 }
4443 
4444 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4445 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4446 
4447 void ib_drain_rq(struct ib_qp *qp);
4448 void ib_drain_sq(struct ib_qp *qp);
4449 void ib_drain_qp(struct ib_qp *qp);
4450 
4451 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4452 
4453 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4454 {
4455 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4456 		return attr->roce.dmac;
4457 	return NULL;
4458 }
4459 
4460 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4461 {
4462 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4463 		attr->ib.dlid = (u16)dlid;
4464 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4465 		attr->opa.dlid = dlid;
4466 }
4467 
4468 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4469 {
4470 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4471 		return attr->ib.dlid;
4472 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4473 		return attr->opa.dlid;
4474 	return 0;
4475 }
4476 
4477 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4478 {
4479 	attr->sl = sl;
4480 }
4481 
4482 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4483 {
4484 	return attr->sl;
4485 }
4486 
4487 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4488 					 u8 src_path_bits)
4489 {
4490 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4491 		attr->ib.src_path_bits = src_path_bits;
4492 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4493 		attr->opa.src_path_bits = src_path_bits;
4494 }
4495 
4496 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4497 {
4498 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4499 		return attr->ib.src_path_bits;
4500 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4501 		return attr->opa.src_path_bits;
4502 	return 0;
4503 }
4504 
4505 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4506 					bool make_grd)
4507 {
4508 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4509 		attr->opa.make_grd = make_grd;
4510 }
4511 
4512 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4513 {
4514 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4515 		return attr->opa.make_grd;
4516 	return false;
4517 }
4518 
4519 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4520 {
4521 	attr->port_num = port_num;
4522 }
4523 
4524 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4525 {
4526 	return attr->port_num;
4527 }
4528 
4529 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4530 					   u8 static_rate)
4531 {
4532 	attr->static_rate = static_rate;
4533 }
4534 
4535 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4536 {
4537 	return attr->static_rate;
4538 }
4539 
4540 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4541 					enum ib_ah_flags flag)
4542 {
4543 	attr->ah_flags = flag;
4544 }
4545 
4546 static inline enum ib_ah_flags
4547 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4548 {
4549 	return attr->ah_flags;
4550 }
4551 
4552 static inline const struct ib_global_route
4553 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4554 {
4555 	return &attr->grh;
4556 }
4557 
4558 /*To retrieve and modify the grh */
4559 static inline struct ib_global_route
4560 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4561 {
4562 	return &attr->grh;
4563 }
4564 
4565 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4566 {
4567 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4568 
4569 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4570 }
4571 
4572 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4573 					     __be64 prefix)
4574 {
4575 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4576 
4577 	grh->dgid.global.subnet_prefix = prefix;
4578 }
4579 
4580 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4581 					    __be64 if_id)
4582 {
4583 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4584 
4585 	grh->dgid.global.interface_id = if_id;
4586 }
4587 
4588 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4589 				   union ib_gid *dgid, u32 flow_label,
4590 				   u8 sgid_index, u8 hop_limit,
4591 				   u8 traffic_class)
4592 {
4593 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4594 
4595 	attr->ah_flags = IB_AH_GRH;
4596 	if (dgid)
4597 		grh->dgid = *dgid;
4598 	grh->flow_label = flow_label;
4599 	grh->sgid_index = sgid_index;
4600 	grh->hop_limit = hop_limit;
4601 	grh->traffic_class = traffic_class;
4602 	grh->sgid_attr = NULL;
4603 }
4604 
4605 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4606 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4607 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4608 			     const struct ib_gid_attr *sgid_attr);
4609 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4610 		       const struct rdma_ah_attr *src);
4611 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4612 			  const struct rdma_ah_attr *new);
4613 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4614 
4615 /**
4616  * rdma_ah_find_type - Return address handle type.
4617  *
4618  * @dev: Device to be checked
4619  * @port_num: Port number
4620  */
4621 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4622 						       u8 port_num)
4623 {
4624 	if (rdma_protocol_roce(dev, port_num))
4625 		return RDMA_AH_ATTR_TYPE_ROCE;
4626 	if (rdma_protocol_ib(dev, port_num)) {
4627 		if (rdma_cap_opa_ah(dev, port_num))
4628 			return RDMA_AH_ATTR_TYPE_OPA;
4629 		return RDMA_AH_ATTR_TYPE_IB;
4630 	}
4631 
4632 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4633 }
4634 
4635 /**
4636  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4637  *     In the current implementation the only way to get
4638  *     get the 32bit lid is from other sources for OPA.
4639  *     For IB, lids will always be 16bits so cast the
4640  *     value accordingly.
4641  *
4642  * @lid: A 32bit LID
4643  */
4644 static inline u16 ib_lid_cpu16(u32 lid)
4645 {
4646 	WARN_ON_ONCE(lid & 0xFFFF0000);
4647 	return (u16)lid;
4648 }
4649 
4650 /**
4651  * ib_lid_be16 - Return lid in 16bit BE encoding.
4652  *
4653  * @lid: A 32bit LID
4654  */
4655 static inline __be16 ib_lid_be16(u32 lid)
4656 {
4657 	WARN_ON_ONCE(lid & 0xFFFF0000);
4658 	return cpu_to_be16((u16)lid);
4659 }
4660 
4661 /**
4662  * ib_get_vector_affinity - Get the affinity mappings of a given completion
4663  *   vector
4664  * @device:         the rdma device
4665  * @comp_vector:    index of completion vector
4666  *
4667  * Returns NULL on failure, otherwise a corresponding cpu map of the
4668  * completion vector (returns all-cpus map if the device driver doesn't
4669  * implement get_vector_affinity).
4670  */
4671 static inline const struct cpumask *
4672 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4673 {
4674 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4675 	    !device->ops.get_vector_affinity)
4676 		return NULL;
4677 
4678 	return device->ops.get_vector_affinity(device, comp_vector);
4679 
4680 }
4681 
4682 /**
4683  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4684  * and add their gids, as needed, to the relevant RoCE devices.
4685  *
4686  * @device:         the rdma device
4687  */
4688 void rdma_roce_rescan_device(struct ib_device *ibdev);
4689 
4690 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4691 
4692 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4693 
4694 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4695 				     enum rdma_netdev_t type, const char *name,
4696 				     unsigned char name_assign_type,
4697 				     void (*setup)(struct net_device *));
4698 
4699 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4700 		     enum rdma_netdev_t type, const char *name,
4701 		     unsigned char name_assign_type,
4702 		     void (*setup)(struct net_device *),
4703 		     struct net_device *netdev);
4704 
4705 /**
4706  * rdma_set_device_sysfs_group - Set device attributes group to have
4707  *				 driver specific sysfs entries at
4708  *				 for infiniband class.
4709  *
4710  * @device:	device pointer for which attributes to be created
4711  * @group:	Pointer to group which should be added when device
4712  *		is registered with sysfs.
4713  * rdma_set_device_sysfs_group() allows existing drivers to expose one
4714  * group per device to have sysfs attributes.
4715  *
4716  * NOTE: New drivers should not make use of this API; instead new device
4717  * parameter should be exposed via netlink command. This API and mechanism
4718  * exist only for existing drivers.
4719  */
4720 static inline void
4721 rdma_set_device_sysfs_group(struct ib_device *dev,
4722 			    const struct attribute_group *group)
4723 {
4724 	dev->groups[1] = group;
4725 }
4726 
4727 /**
4728  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4729  *
4730  * @device:	device pointer for which ib_device pointer to retrieve
4731  *
4732  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4733  *
4734  */
4735 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4736 {
4737 	struct ib_core_device *coredev =
4738 		container_of(device, struct ib_core_device, dev);
4739 
4740 	return coredev->owner;
4741 }
4742 
4743 /**
4744  * rdma_device_to_drv_device - Helper macro to reach back to driver's
4745  *			       ib_device holder structure from device pointer.
4746  *
4747  * NOTE: New drivers should not make use of this API; This API is only for
4748  * existing drivers who have exposed sysfs entries using
4749  * rdma_set_device_sysfs_group().
4750  */
4751 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
4752 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4753 
4754 bool rdma_dev_access_netns(const struct ib_device *device,
4755 			   const struct net *net);
4756 
4757 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4758 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4759 
4760 /**
4761  * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4762  *                               on the flow_label
4763  *
4764  * This function will convert the 20 bit flow_label input to a valid RoCE v2
4765  * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4766  * convention.
4767  */
4768 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4769 {
4770 	u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4771 
4772 	fl_low ^= fl_high >> 14;
4773 	return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4774 }
4775 
4776 /**
4777  * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4778  *                        local and remote qpn values
4779  *
4780  * This function folded the multiplication results of two qpns, 24 bit each,
4781  * fields, and converts it to a 20 bit results.
4782  *
4783  * This function will create symmetric flow_label value based on the local
4784  * and remote qpn values. this will allow both the requester and responder
4785  * to calculate the same flow_label for a given connection.
4786  *
4787  * This helper function should be used by driver in case the upper layer
4788  * provide a zero flow_label value. This is to improve entropy of RDMA
4789  * traffic in the network.
4790  */
4791 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4792 {
4793 	u64 v = (u64)lqpn * rqpn;
4794 
4795 	v ^= v >> 20;
4796 	v ^= v >> 40;
4797 
4798 	return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4799 }
4800 #endif /* IB_VERBS_H */
4801