xref: /openbmc/linux/include/rdma/ib_verbs.h (revision c16c6655)
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 
76 extern struct workqueue_struct *ib_wq;
77 extern struct workqueue_struct *ib_comp_wq;
78 extern struct workqueue_struct *ib_comp_unbound_wq;
79 
80 __printf(3, 4) __cold
81 void ibdev_printk(const char *level, const struct ib_device *ibdev,
82 		  const char *format, ...);
83 __printf(2, 3) __cold
84 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
85 __printf(2, 3) __cold
86 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
87 __printf(2, 3) __cold
88 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
89 __printf(2, 3) __cold
90 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
91 __printf(2, 3) __cold
92 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
93 __printf(2, 3) __cold
94 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
95 __printf(2, 3) __cold
96 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
97 
98 #if defined(CONFIG_DYNAMIC_DEBUG)
99 #define ibdev_dbg(__dev, format, args...)                       \
100 	dynamic_ibdev_dbg(__dev, format, ##args)
101 #elif defined(DEBUG)
102 #define ibdev_dbg(__dev, format, args...)                       \
103 	ibdev_printk(KERN_DEBUG, __dev, format, ##args)
104 #else
105 __printf(2, 3) __cold
106 static inline
107 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
108 #endif
109 
110 union ib_gid {
111 	u8	raw[16];
112 	struct {
113 		__be64	subnet_prefix;
114 		__be64	interface_id;
115 	} global;
116 };
117 
118 extern union ib_gid zgid;
119 
120 enum ib_gid_type {
121 	/* If link layer is Ethernet, this is RoCE V1 */
122 	IB_GID_TYPE_IB        = 0,
123 	IB_GID_TYPE_ROCE      = 0,
124 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
125 	IB_GID_TYPE_SIZE
126 };
127 
128 #define ROCE_V2_UDP_DPORT      4791
129 struct ib_gid_attr {
130 	struct net_device __rcu	*ndev;
131 	struct ib_device	*device;
132 	union ib_gid		gid;
133 	enum ib_gid_type	gid_type;
134 	u16			index;
135 	u8			port_num;
136 };
137 
138 enum {
139 	/* set the local administered indication */
140 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
141 };
142 
143 enum rdma_transport_type {
144 	RDMA_TRANSPORT_IB,
145 	RDMA_TRANSPORT_IWARP,
146 	RDMA_TRANSPORT_USNIC,
147 	RDMA_TRANSPORT_USNIC_UDP,
148 	RDMA_TRANSPORT_UNSPECIFIED,
149 };
150 
151 enum rdma_protocol_type {
152 	RDMA_PROTOCOL_IB,
153 	RDMA_PROTOCOL_IBOE,
154 	RDMA_PROTOCOL_IWARP,
155 	RDMA_PROTOCOL_USNIC_UDP
156 };
157 
158 __attribute_const__ enum rdma_transport_type
159 rdma_node_get_transport(unsigned int node_type);
160 
161 enum rdma_network_type {
162 	RDMA_NETWORK_IB,
163 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
164 	RDMA_NETWORK_IPV4,
165 	RDMA_NETWORK_IPV6
166 };
167 
168 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
169 {
170 	if (network_type == RDMA_NETWORK_IPV4 ||
171 	    network_type == RDMA_NETWORK_IPV6)
172 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
173 
174 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
175 	return IB_GID_TYPE_IB;
176 }
177 
178 static inline enum rdma_network_type
179 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
180 {
181 	if (attr->gid_type == IB_GID_TYPE_IB)
182 		return RDMA_NETWORK_IB;
183 
184 	if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
185 		return RDMA_NETWORK_IPV4;
186 	else
187 		return RDMA_NETWORK_IPV6;
188 }
189 
190 enum rdma_link_layer {
191 	IB_LINK_LAYER_UNSPECIFIED,
192 	IB_LINK_LAYER_INFINIBAND,
193 	IB_LINK_LAYER_ETHERNET,
194 };
195 
196 enum ib_device_cap_flags {
197 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
198 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
199 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
200 	IB_DEVICE_RAW_MULTI			= (1 << 3),
201 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
202 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
203 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
204 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
205 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
206 	/* Not in use, former INIT_TYPE		= (1 << 9),*/
207 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
208 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
209 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
210 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
211 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
212 
213 	/*
214 	 * This device supports a per-device lkey or stag that can be
215 	 * used without performing a memory registration for the local
216 	 * memory.  Note that ULPs should never check this flag, but
217 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
218 	 * which will always contain a usable lkey.
219 	 */
220 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
221 	/* Reserved, old SEND_W_INV		= (1 << 16),*/
222 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
223 	/*
224 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
225 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
226 	 * messages and can verify the validity of checksum for
227 	 * incoming messages.  Setting this flag implies that the
228 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
229 	 */
230 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
231 	IB_DEVICE_UD_TSO			= (1 << 19),
232 	IB_DEVICE_XRC				= (1 << 20),
233 
234 	/*
235 	 * This device supports the IB "base memory management extension",
236 	 * which includes support for fast registrations (IB_WR_REG_MR,
237 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
238 	 * also be set by any iWarp device which must support FRs to comply
239 	 * to the iWarp verbs spec.  iWarp devices also support the
240 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
241 	 * stag.
242 	 */
243 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
244 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
245 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
246 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
247 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
248 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
249 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
250 	/*
251 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
252 	 * support execution of WQEs that involve synchronization
253 	 * of I/O operations with single completion queue managed
254 	 * by hardware.
255 	 */
256 	IB_DEVICE_CROSS_CHANNEL			= (1 << 27),
257 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
258 	IB_DEVICE_INTEGRITY_HANDOVER		= (1 << 30),
259 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
260 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
261 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
262 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
263 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
264 	IB_DEVICE_RDMA_NETDEV_OPA_VNIC		= (1ULL << 35),
265 	/* The device supports padding incoming writes to cacheline. */
266 	IB_DEVICE_PCI_WRITE_END_PADDING		= (1ULL << 36),
267 	IB_DEVICE_ALLOW_USER_UNREG		= (1ULL << 37),
268 };
269 
270 enum ib_atomic_cap {
271 	IB_ATOMIC_NONE,
272 	IB_ATOMIC_HCA,
273 	IB_ATOMIC_GLOB
274 };
275 
276 enum ib_odp_general_cap_bits {
277 	IB_ODP_SUPPORT		= 1 << 0,
278 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
279 };
280 
281 enum ib_odp_transport_cap_bits {
282 	IB_ODP_SUPPORT_SEND	= 1 << 0,
283 	IB_ODP_SUPPORT_RECV	= 1 << 1,
284 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
285 	IB_ODP_SUPPORT_READ	= 1 << 3,
286 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
287 	IB_ODP_SUPPORT_SRQ_RECV	= 1 << 5,
288 };
289 
290 struct ib_odp_caps {
291 	uint64_t general_caps;
292 	struct {
293 		uint32_t  rc_odp_caps;
294 		uint32_t  uc_odp_caps;
295 		uint32_t  ud_odp_caps;
296 		uint32_t  xrc_odp_caps;
297 	} per_transport_caps;
298 };
299 
300 struct ib_rss_caps {
301 	/* Corresponding bit will be set if qp type from
302 	 * 'enum ib_qp_type' is supported, e.g.
303 	 * supported_qpts |= 1 << IB_QPT_UD
304 	 */
305 	u32 supported_qpts;
306 	u32 max_rwq_indirection_tables;
307 	u32 max_rwq_indirection_table_size;
308 };
309 
310 enum ib_tm_cap_flags {
311 	/*  Support tag matching with rendezvous offload for RC transport */
312 	IB_TM_CAP_RNDV_RC = 1 << 0,
313 };
314 
315 struct ib_tm_caps {
316 	/* Max size of RNDV header */
317 	u32 max_rndv_hdr_size;
318 	/* Max number of entries in tag matching list */
319 	u32 max_num_tags;
320 	/* From enum ib_tm_cap_flags */
321 	u32 flags;
322 	/* Max number of outstanding list operations */
323 	u32 max_ops;
324 	/* Max number of SGE in tag matching entry */
325 	u32 max_sge;
326 };
327 
328 struct ib_cq_init_attr {
329 	unsigned int	cqe;
330 	int		comp_vector;
331 	u32		flags;
332 };
333 
334 enum ib_cq_attr_mask {
335 	IB_CQ_MODERATE = 1 << 0,
336 };
337 
338 struct ib_cq_caps {
339 	u16     max_cq_moderation_count;
340 	u16     max_cq_moderation_period;
341 };
342 
343 struct ib_dm_mr_attr {
344 	u64		length;
345 	u64		offset;
346 	u32		access_flags;
347 };
348 
349 struct ib_dm_alloc_attr {
350 	u64	length;
351 	u32	alignment;
352 	u32	flags;
353 };
354 
355 struct ib_device_attr {
356 	u64			fw_ver;
357 	__be64			sys_image_guid;
358 	u64			max_mr_size;
359 	u64			page_size_cap;
360 	u32			vendor_id;
361 	u32			vendor_part_id;
362 	u32			hw_ver;
363 	int			max_qp;
364 	int			max_qp_wr;
365 	u64			device_cap_flags;
366 	int			max_send_sge;
367 	int			max_recv_sge;
368 	int			max_sge_rd;
369 	int			max_cq;
370 	int			max_cqe;
371 	int			max_mr;
372 	int			max_pd;
373 	int			max_qp_rd_atom;
374 	int			max_ee_rd_atom;
375 	int			max_res_rd_atom;
376 	int			max_qp_init_rd_atom;
377 	int			max_ee_init_rd_atom;
378 	enum ib_atomic_cap	atomic_cap;
379 	enum ib_atomic_cap	masked_atomic_cap;
380 	int			max_ee;
381 	int			max_rdd;
382 	int			max_mw;
383 	int			max_raw_ipv6_qp;
384 	int			max_raw_ethy_qp;
385 	int			max_mcast_grp;
386 	int			max_mcast_qp_attach;
387 	int			max_total_mcast_qp_attach;
388 	int			max_ah;
389 	int			max_fmr;
390 	int			max_map_per_fmr;
391 	int			max_srq;
392 	int			max_srq_wr;
393 	int			max_srq_sge;
394 	unsigned int		max_fast_reg_page_list_len;
395 	unsigned int		max_pi_fast_reg_page_list_len;
396 	u16			max_pkeys;
397 	u8			local_ca_ack_delay;
398 	int			sig_prot_cap;
399 	int			sig_guard_cap;
400 	struct ib_odp_caps	odp_caps;
401 	uint64_t		timestamp_mask;
402 	uint64_t		hca_core_clock; /* in KHZ */
403 	struct ib_rss_caps	rss_caps;
404 	u32			max_wq_type_rq;
405 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
406 	struct ib_tm_caps	tm_caps;
407 	struct ib_cq_caps       cq_caps;
408 	u64			max_dm_size;
409 };
410 
411 enum ib_mtu {
412 	IB_MTU_256  = 1,
413 	IB_MTU_512  = 2,
414 	IB_MTU_1024 = 3,
415 	IB_MTU_2048 = 4,
416 	IB_MTU_4096 = 5
417 };
418 
419 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
420 {
421 	switch (mtu) {
422 	case IB_MTU_256:  return  256;
423 	case IB_MTU_512:  return  512;
424 	case IB_MTU_1024: return 1024;
425 	case IB_MTU_2048: return 2048;
426 	case IB_MTU_4096: return 4096;
427 	default: 	  return -1;
428 	}
429 }
430 
431 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
432 {
433 	if (mtu >= 4096)
434 		return IB_MTU_4096;
435 	else if (mtu >= 2048)
436 		return IB_MTU_2048;
437 	else if (mtu >= 1024)
438 		return IB_MTU_1024;
439 	else if (mtu >= 512)
440 		return IB_MTU_512;
441 	else
442 		return IB_MTU_256;
443 }
444 
445 enum ib_port_state {
446 	IB_PORT_NOP		= 0,
447 	IB_PORT_DOWN		= 1,
448 	IB_PORT_INIT		= 2,
449 	IB_PORT_ARMED		= 3,
450 	IB_PORT_ACTIVE		= 4,
451 	IB_PORT_ACTIVE_DEFER	= 5
452 };
453 
454 enum ib_port_width {
455 	IB_WIDTH_1X	= 1,
456 	IB_WIDTH_2X	= 16,
457 	IB_WIDTH_4X	= 2,
458 	IB_WIDTH_8X	= 4,
459 	IB_WIDTH_12X	= 8
460 };
461 
462 static inline int ib_width_enum_to_int(enum ib_port_width width)
463 {
464 	switch (width) {
465 	case IB_WIDTH_1X:  return  1;
466 	case IB_WIDTH_2X:  return  2;
467 	case IB_WIDTH_4X:  return  4;
468 	case IB_WIDTH_8X:  return  8;
469 	case IB_WIDTH_12X: return 12;
470 	default: 	  return -1;
471 	}
472 }
473 
474 enum ib_port_speed {
475 	IB_SPEED_SDR	= 1,
476 	IB_SPEED_DDR	= 2,
477 	IB_SPEED_QDR	= 4,
478 	IB_SPEED_FDR10	= 8,
479 	IB_SPEED_FDR	= 16,
480 	IB_SPEED_EDR	= 32,
481 	IB_SPEED_HDR	= 64
482 };
483 
484 /**
485  * struct rdma_hw_stats
486  * @lock - Mutex to protect parallel write access to lifespan and values
487  *    of counters, which are 64bits and not guaranteeed to be written
488  *    atomicaly on 32bits systems.
489  * @timestamp - Used by the core code to track when the last update was
490  * @lifespan - Used by the core code to determine how old the counters
491  *   should be before being updated again.  Stored in jiffies, defaults
492  *   to 10 milliseconds, drivers can override the default be specifying
493  *   their own value during their allocation routine.
494  * @name - Array of pointers to static names used for the counters in
495  *   directory.
496  * @num_counters - How many hardware counters there are.  If name is
497  *   shorter than this number, a kernel oops will result.  Driver authors
498  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
499  *   in their code to prevent this.
500  * @value - Array of u64 counters that are accessed by the sysfs code and
501  *   filled in by the drivers get_stats routine
502  */
503 struct rdma_hw_stats {
504 	struct mutex	lock; /* Protect lifespan and values[] */
505 	unsigned long	timestamp;
506 	unsigned long	lifespan;
507 	const char * const *names;
508 	int		num_counters;
509 	u64		value[];
510 };
511 
512 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
513 /**
514  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
515  *   for drivers.
516  * @names - Array of static const char *
517  * @num_counters - How many elements in array
518  * @lifespan - How many milliseconds between updates
519  */
520 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
521 		const char * const *names, int num_counters,
522 		unsigned long lifespan)
523 {
524 	struct rdma_hw_stats *stats;
525 
526 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
527 			GFP_KERNEL);
528 	if (!stats)
529 		return NULL;
530 	stats->names = names;
531 	stats->num_counters = num_counters;
532 	stats->lifespan = msecs_to_jiffies(lifespan);
533 
534 	return stats;
535 }
536 
537 
538 /* Define bits for the various functionality this port needs to be supported by
539  * the core.
540  */
541 /* Management                           0x00000FFF */
542 #define RDMA_CORE_CAP_IB_MAD            0x00000001
543 #define RDMA_CORE_CAP_IB_SMI            0x00000002
544 #define RDMA_CORE_CAP_IB_CM             0x00000004
545 #define RDMA_CORE_CAP_IW_CM             0x00000008
546 #define RDMA_CORE_CAP_IB_SA             0x00000010
547 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
548 
549 /* Address format                       0x000FF000 */
550 #define RDMA_CORE_CAP_AF_IB             0x00001000
551 #define RDMA_CORE_CAP_ETH_AH            0x00002000
552 #define RDMA_CORE_CAP_OPA_AH            0x00004000
553 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
554 
555 /* Protocol                             0xFFF00000 */
556 #define RDMA_CORE_CAP_PROT_IB           0x00100000
557 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
558 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
559 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
560 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
561 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
562 
563 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
564 					| RDMA_CORE_CAP_PROT_ROCE     \
565 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
566 
567 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
568 					| RDMA_CORE_CAP_IB_MAD \
569 					| RDMA_CORE_CAP_IB_SMI \
570 					| RDMA_CORE_CAP_IB_CM  \
571 					| RDMA_CORE_CAP_IB_SA  \
572 					| RDMA_CORE_CAP_AF_IB)
573 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
574 					| RDMA_CORE_CAP_IB_MAD  \
575 					| RDMA_CORE_CAP_IB_CM   \
576 					| RDMA_CORE_CAP_AF_IB   \
577 					| RDMA_CORE_CAP_ETH_AH)
578 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
579 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
580 					| RDMA_CORE_CAP_IB_MAD  \
581 					| RDMA_CORE_CAP_IB_CM   \
582 					| RDMA_CORE_CAP_AF_IB   \
583 					| RDMA_CORE_CAP_ETH_AH)
584 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
585 					| RDMA_CORE_CAP_IW_CM)
586 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
587 					| RDMA_CORE_CAP_OPA_MAD)
588 
589 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
590 
591 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
592 
593 struct ib_port_attr {
594 	u64			subnet_prefix;
595 	enum ib_port_state	state;
596 	enum ib_mtu		max_mtu;
597 	enum ib_mtu		active_mtu;
598 	int			gid_tbl_len;
599 	unsigned int		ip_gids:1;
600 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
601 	u32			port_cap_flags;
602 	u32			max_msg_sz;
603 	u32			bad_pkey_cntr;
604 	u32			qkey_viol_cntr;
605 	u16			pkey_tbl_len;
606 	u32			sm_lid;
607 	u32			lid;
608 	u8			lmc;
609 	u8			max_vl_num;
610 	u8			sm_sl;
611 	u8			subnet_timeout;
612 	u8			init_type_reply;
613 	u8			active_width;
614 	u8			active_speed;
615 	u8                      phys_state;
616 	u16			port_cap_flags2;
617 };
618 
619 enum ib_device_modify_flags {
620 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
621 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
622 };
623 
624 #define IB_DEVICE_NODE_DESC_MAX 64
625 
626 struct ib_device_modify {
627 	u64	sys_image_guid;
628 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
629 };
630 
631 enum ib_port_modify_flags {
632 	IB_PORT_SHUTDOWN		= 1,
633 	IB_PORT_INIT_TYPE		= (1<<2),
634 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
635 	IB_PORT_OPA_MASK_CHG		= (1<<4)
636 };
637 
638 struct ib_port_modify {
639 	u32	set_port_cap_mask;
640 	u32	clr_port_cap_mask;
641 	u8	init_type;
642 };
643 
644 enum ib_event_type {
645 	IB_EVENT_CQ_ERR,
646 	IB_EVENT_QP_FATAL,
647 	IB_EVENT_QP_REQ_ERR,
648 	IB_EVENT_QP_ACCESS_ERR,
649 	IB_EVENT_COMM_EST,
650 	IB_EVENT_SQ_DRAINED,
651 	IB_EVENT_PATH_MIG,
652 	IB_EVENT_PATH_MIG_ERR,
653 	IB_EVENT_DEVICE_FATAL,
654 	IB_EVENT_PORT_ACTIVE,
655 	IB_EVENT_PORT_ERR,
656 	IB_EVENT_LID_CHANGE,
657 	IB_EVENT_PKEY_CHANGE,
658 	IB_EVENT_SM_CHANGE,
659 	IB_EVENT_SRQ_ERR,
660 	IB_EVENT_SRQ_LIMIT_REACHED,
661 	IB_EVENT_QP_LAST_WQE_REACHED,
662 	IB_EVENT_CLIENT_REREGISTER,
663 	IB_EVENT_GID_CHANGE,
664 	IB_EVENT_WQ_FATAL,
665 };
666 
667 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
668 
669 struct ib_event {
670 	struct ib_device	*device;
671 	union {
672 		struct ib_cq	*cq;
673 		struct ib_qp	*qp;
674 		struct ib_srq	*srq;
675 		struct ib_wq	*wq;
676 		u8		port_num;
677 	} element;
678 	enum ib_event_type	event;
679 };
680 
681 struct ib_event_handler {
682 	struct ib_device *device;
683 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
684 	struct list_head  list;
685 };
686 
687 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
688 	do {							\
689 		(_ptr)->device  = _device;			\
690 		(_ptr)->handler = _handler;			\
691 		INIT_LIST_HEAD(&(_ptr)->list);			\
692 	} while (0)
693 
694 struct ib_global_route {
695 	const struct ib_gid_attr *sgid_attr;
696 	union ib_gid	dgid;
697 	u32		flow_label;
698 	u8		sgid_index;
699 	u8		hop_limit;
700 	u8		traffic_class;
701 };
702 
703 struct ib_grh {
704 	__be32		version_tclass_flow;
705 	__be16		paylen;
706 	u8		next_hdr;
707 	u8		hop_limit;
708 	union ib_gid	sgid;
709 	union ib_gid	dgid;
710 };
711 
712 union rdma_network_hdr {
713 	struct ib_grh ibgrh;
714 	struct {
715 		/* The IB spec states that if it's IPv4, the header
716 		 * is located in the last 20 bytes of the header.
717 		 */
718 		u8		reserved[20];
719 		struct iphdr	roce4grh;
720 	};
721 };
722 
723 #define IB_QPN_MASK		0xFFFFFF
724 
725 enum {
726 	IB_MULTICAST_QPN = 0xffffff
727 };
728 
729 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
730 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
731 
732 enum ib_ah_flags {
733 	IB_AH_GRH	= 1
734 };
735 
736 enum ib_rate {
737 	IB_RATE_PORT_CURRENT = 0,
738 	IB_RATE_2_5_GBPS = 2,
739 	IB_RATE_5_GBPS   = 5,
740 	IB_RATE_10_GBPS  = 3,
741 	IB_RATE_20_GBPS  = 6,
742 	IB_RATE_30_GBPS  = 4,
743 	IB_RATE_40_GBPS  = 7,
744 	IB_RATE_60_GBPS  = 8,
745 	IB_RATE_80_GBPS  = 9,
746 	IB_RATE_120_GBPS = 10,
747 	IB_RATE_14_GBPS  = 11,
748 	IB_RATE_56_GBPS  = 12,
749 	IB_RATE_112_GBPS = 13,
750 	IB_RATE_168_GBPS = 14,
751 	IB_RATE_25_GBPS  = 15,
752 	IB_RATE_100_GBPS = 16,
753 	IB_RATE_200_GBPS = 17,
754 	IB_RATE_300_GBPS = 18,
755 	IB_RATE_28_GBPS  = 19,
756 	IB_RATE_50_GBPS  = 20,
757 	IB_RATE_400_GBPS = 21,
758 	IB_RATE_600_GBPS = 22,
759 };
760 
761 /**
762  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
763  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
764  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
765  * @rate: rate to convert.
766  */
767 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
768 
769 /**
770  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
771  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
772  * @rate: rate to convert.
773  */
774 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
775 
776 
777 /**
778  * enum ib_mr_type - memory region type
779  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
780  *                            normal registration
781  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
782  *                            register any arbitrary sg lists (without
783  *                            the normal mr constraints - see
784  *                            ib_map_mr_sg)
785  * @IB_MR_TYPE_DM:            memory region that is used for device
786  *                            memory registration
787  * @IB_MR_TYPE_USER:          memory region that is used for the user-space
788  *                            application
789  * @IB_MR_TYPE_DMA:           memory region that is used for DMA operations
790  *                            without address translations (VA=PA)
791  * @IB_MR_TYPE_INTEGRITY:     memory region that is used for
792  *                            data integrity operations
793  */
794 enum ib_mr_type {
795 	IB_MR_TYPE_MEM_REG,
796 	IB_MR_TYPE_SG_GAPS,
797 	IB_MR_TYPE_DM,
798 	IB_MR_TYPE_USER,
799 	IB_MR_TYPE_DMA,
800 	IB_MR_TYPE_INTEGRITY,
801 };
802 
803 enum ib_mr_status_check {
804 	IB_MR_CHECK_SIG_STATUS = 1,
805 };
806 
807 /**
808  * struct ib_mr_status - Memory region status container
809  *
810  * @fail_status: Bitmask of MR checks status. For each
811  *     failed check a corresponding status bit is set.
812  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
813  *     failure.
814  */
815 struct ib_mr_status {
816 	u32		    fail_status;
817 	struct ib_sig_err   sig_err;
818 };
819 
820 /**
821  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
822  * enum.
823  * @mult: multiple to convert.
824  */
825 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
826 
827 enum rdma_ah_attr_type {
828 	RDMA_AH_ATTR_TYPE_UNDEFINED,
829 	RDMA_AH_ATTR_TYPE_IB,
830 	RDMA_AH_ATTR_TYPE_ROCE,
831 	RDMA_AH_ATTR_TYPE_OPA,
832 };
833 
834 struct ib_ah_attr {
835 	u16			dlid;
836 	u8			src_path_bits;
837 };
838 
839 struct roce_ah_attr {
840 	u8			dmac[ETH_ALEN];
841 };
842 
843 struct opa_ah_attr {
844 	u32			dlid;
845 	u8			src_path_bits;
846 	bool			make_grd;
847 };
848 
849 struct rdma_ah_attr {
850 	struct ib_global_route	grh;
851 	u8			sl;
852 	u8			static_rate;
853 	u8			port_num;
854 	u8			ah_flags;
855 	enum rdma_ah_attr_type type;
856 	union {
857 		struct ib_ah_attr ib;
858 		struct roce_ah_attr roce;
859 		struct opa_ah_attr opa;
860 	};
861 };
862 
863 enum ib_wc_status {
864 	IB_WC_SUCCESS,
865 	IB_WC_LOC_LEN_ERR,
866 	IB_WC_LOC_QP_OP_ERR,
867 	IB_WC_LOC_EEC_OP_ERR,
868 	IB_WC_LOC_PROT_ERR,
869 	IB_WC_WR_FLUSH_ERR,
870 	IB_WC_MW_BIND_ERR,
871 	IB_WC_BAD_RESP_ERR,
872 	IB_WC_LOC_ACCESS_ERR,
873 	IB_WC_REM_INV_REQ_ERR,
874 	IB_WC_REM_ACCESS_ERR,
875 	IB_WC_REM_OP_ERR,
876 	IB_WC_RETRY_EXC_ERR,
877 	IB_WC_RNR_RETRY_EXC_ERR,
878 	IB_WC_LOC_RDD_VIOL_ERR,
879 	IB_WC_REM_INV_RD_REQ_ERR,
880 	IB_WC_REM_ABORT_ERR,
881 	IB_WC_INV_EECN_ERR,
882 	IB_WC_INV_EEC_STATE_ERR,
883 	IB_WC_FATAL_ERR,
884 	IB_WC_RESP_TIMEOUT_ERR,
885 	IB_WC_GENERAL_ERR
886 };
887 
888 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
889 
890 enum ib_wc_opcode {
891 	IB_WC_SEND,
892 	IB_WC_RDMA_WRITE,
893 	IB_WC_RDMA_READ,
894 	IB_WC_COMP_SWAP,
895 	IB_WC_FETCH_ADD,
896 	IB_WC_LSO,
897 	IB_WC_LOCAL_INV,
898 	IB_WC_REG_MR,
899 	IB_WC_MASKED_COMP_SWAP,
900 	IB_WC_MASKED_FETCH_ADD,
901 /*
902  * Set value of IB_WC_RECV so consumers can test if a completion is a
903  * receive by testing (opcode & IB_WC_RECV).
904  */
905 	IB_WC_RECV			= 1 << 7,
906 	IB_WC_RECV_RDMA_WITH_IMM
907 };
908 
909 enum ib_wc_flags {
910 	IB_WC_GRH		= 1,
911 	IB_WC_WITH_IMM		= (1<<1),
912 	IB_WC_WITH_INVALIDATE	= (1<<2),
913 	IB_WC_IP_CSUM_OK	= (1<<3),
914 	IB_WC_WITH_SMAC		= (1<<4),
915 	IB_WC_WITH_VLAN		= (1<<5),
916 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
917 };
918 
919 struct ib_wc {
920 	union {
921 		u64		wr_id;
922 		struct ib_cqe	*wr_cqe;
923 	};
924 	enum ib_wc_status	status;
925 	enum ib_wc_opcode	opcode;
926 	u32			vendor_err;
927 	u32			byte_len;
928 	struct ib_qp	       *qp;
929 	union {
930 		__be32		imm_data;
931 		u32		invalidate_rkey;
932 	} ex;
933 	u32			src_qp;
934 	u32			slid;
935 	int			wc_flags;
936 	u16			pkey_index;
937 	u8			sl;
938 	u8			dlid_path_bits;
939 	u8			port_num;	/* valid only for DR SMPs on switches */
940 	u8			smac[ETH_ALEN];
941 	u16			vlan_id;
942 	u8			network_hdr_type;
943 };
944 
945 enum ib_cq_notify_flags {
946 	IB_CQ_SOLICITED			= 1 << 0,
947 	IB_CQ_NEXT_COMP			= 1 << 1,
948 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
949 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
950 };
951 
952 enum ib_srq_type {
953 	IB_SRQT_BASIC,
954 	IB_SRQT_XRC,
955 	IB_SRQT_TM,
956 };
957 
958 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
959 {
960 	return srq_type == IB_SRQT_XRC ||
961 	       srq_type == IB_SRQT_TM;
962 }
963 
964 enum ib_srq_attr_mask {
965 	IB_SRQ_MAX_WR	= 1 << 0,
966 	IB_SRQ_LIMIT	= 1 << 1,
967 };
968 
969 struct ib_srq_attr {
970 	u32	max_wr;
971 	u32	max_sge;
972 	u32	srq_limit;
973 };
974 
975 struct ib_srq_init_attr {
976 	void		      (*event_handler)(struct ib_event *, void *);
977 	void		       *srq_context;
978 	struct ib_srq_attr	attr;
979 	enum ib_srq_type	srq_type;
980 
981 	struct {
982 		struct ib_cq   *cq;
983 		union {
984 			struct {
985 				struct ib_xrcd *xrcd;
986 			} xrc;
987 
988 			struct {
989 				u32		max_num_tags;
990 			} tag_matching;
991 		};
992 	} ext;
993 };
994 
995 struct ib_qp_cap {
996 	u32	max_send_wr;
997 	u32	max_recv_wr;
998 	u32	max_send_sge;
999 	u32	max_recv_sge;
1000 	u32	max_inline_data;
1001 
1002 	/*
1003 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1004 	 * ib_create_qp() will calculate the right amount of neededed WRs
1005 	 * and MRs based on this.
1006 	 */
1007 	u32	max_rdma_ctxs;
1008 };
1009 
1010 enum ib_sig_type {
1011 	IB_SIGNAL_ALL_WR,
1012 	IB_SIGNAL_REQ_WR
1013 };
1014 
1015 enum ib_qp_type {
1016 	/*
1017 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1018 	 * here (and in that order) since the MAD layer uses them as
1019 	 * indices into a 2-entry table.
1020 	 */
1021 	IB_QPT_SMI,
1022 	IB_QPT_GSI,
1023 
1024 	IB_QPT_RC,
1025 	IB_QPT_UC,
1026 	IB_QPT_UD,
1027 	IB_QPT_RAW_IPV6,
1028 	IB_QPT_RAW_ETHERTYPE,
1029 	IB_QPT_RAW_PACKET = 8,
1030 	IB_QPT_XRC_INI = 9,
1031 	IB_QPT_XRC_TGT,
1032 	IB_QPT_MAX,
1033 	IB_QPT_DRIVER = 0xFF,
1034 	/* Reserve a range for qp types internal to the low level driver.
1035 	 * These qp types will not be visible at the IB core layer, so the
1036 	 * IB_QPT_MAX usages should not be affected in the core layer
1037 	 */
1038 	IB_QPT_RESERVED1 = 0x1000,
1039 	IB_QPT_RESERVED2,
1040 	IB_QPT_RESERVED3,
1041 	IB_QPT_RESERVED4,
1042 	IB_QPT_RESERVED5,
1043 	IB_QPT_RESERVED6,
1044 	IB_QPT_RESERVED7,
1045 	IB_QPT_RESERVED8,
1046 	IB_QPT_RESERVED9,
1047 	IB_QPT_RESERVED10,
1048 };
1049 
1050 enum ib_qp_create_flags {
1051 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1052 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1053 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1054 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1055 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1056 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1057 	IB_QP_CREATE_INTEGRITY_EN		= 1 << 6,
1058 	/* FREE					= 1 << 7, */
1059 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1060 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1061 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1062 	IB_QP_CREATE_PCI_WRITE_END_PADDING	= 1 << 11,
1063 	/* reserve bits 26-31 for low level drivers' internal use */
1064 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1065 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1066 };
1067 
1068 /*
1069  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1070  * callback to destroy the passed in QP.
1071  */
1072 
1073 struct ib_qp_init_attr {
1074 	/* Consumer's event_handler callback must not block */
1075 	void                  (*event_handler)(struct ib_event *, void *);
1076 
1077 	void		       *qp_context;
1078 	struct ib_cq	       *send_cq;
1079 	struct ib_cq	       *recv_cq;
1080 	struct ib_srq	       *srq;
1081 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1082 	struct ib_qp_cap	cap;
1083 	enum ib_sig_type	sq_sig_type;
1084 	enum ib_qp_type		qp_type;
1085 	u32			create_flags;
1086 
1087 	/*
1088 	 * Only needed for special QP types, or when using the RW API.
1089 	 */
1090 	u8			port_num;
1091 	struct ib_rwq_ind_table *rwq_ind_tbl;
1092 	u32			source_qpn;
1093 };
1094 
1095 struct ib_qp_open_attr {
1096 	void                  (*event_handler)(struct ib_event *, void *);
1097 	void		       *qp_context;
1098 	u32			qp_num;
1099 	enum ib_qp_type		qp_type;
1100 };
1101 
1102 enum ib_rnr_timeout {
1103 	IB_RNR_TIMER_655_36 =  0,
1104 	IB_RNR_TIMER_000_01 =  1,
1105 	IB_RNR_TIMER_000_02 =  2,
1106 	IB_RNR_TIMER_000_03 =  3,
1107 	IB_RNR_TIMER_000_04 =  4,
1108 	IB_RNR_TIMER_000_06 =  5,
1109 	IB_RNR_TIMER_000_08 =  6,
1110 	IB_RNR_TIMER_000_12 =  7,
1111 	IB_RNR_TIMER_000_16 =  8,
1112 	IB_RNR_TIMER_000_24 =  9,
1113 	IB_RNR_TIMER_000_32 = 10,
1114 	IB_RNR_TIMER_000_48 = 11,
1115 	IB_RNR_TIMER_000_64 = 12,
1116 	IB_RNR_TIMER_000_96 = 13,
1117 	IB_RNR_TIMER_001_28 = 14,
1118 	IB_RNR_TIMER_001_92 = 15,
1119 	IB_RNR_TIMER_002_56 = 16,
1120 	IB_RNR_TIMER_003_84 = 17,
1121 	IB_RNR_TIMER_005_12 = 18,
1122 	IB_RNR_TIMER_007_68 = 19,
1123 	IB_RNR_TIMER_010_24 = 20,
1124 	IB_RNR_TIMER_015_36 = 21,
1125 	IB_RNR_TIMER_020_48 = 22,
1126 	IB_RNR_TIMER_030_72 = 23,
1127 	IB_RNR_TIMER_040_96 = 24,
1128 	IB_RNR_TIMER_061_44 = 25,
1129 	IB_RNR_TIMER_081_92 = 26,
1130 	IB_RNR_TIMER_122_88 = 27,
1131 	IB_RNR_TIMER_163_84 = 28,
1132 	IB_RNR_TIMER_245_76 = 29,
1133 	IB_RNR_TIMER_327_68 = 30,
1134 	IB_RNR_TIMER_491_52 = 31
1135 };
1136 
1137 enum ib_qp_attr_mask {
1138 	IB_QP_STATE			= 1,
1139 	IB_QP_CUR_STATE			= (1<<1),
1140 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1141 	IB_QP_ACCESS_FLAGS		= (1<<3),
1142 	IB_QP_PKEY_INDEX		= (1<<4),
1143 	IB_QP_PORT			= (1<<5),
1144 	IB_QP_QKEY			= (1<<6),
1145 	IB_QP_AV			= (1<<7),
1146 	IB_QP_PATH_MTU			= (1<<8),
1147 	IB_QP_TIMEOUT			= (1<<9),
1148 	IB_QP_RETRY_CNT			= (1<<10),
1149 	IB_QP_RNR_RETRY			= (1<<11),
1150 	IB_QP_RQ_PSN			= (1<<12),
1151 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1152 	IB_QP_ALT_PATH			= (1<<14),
1153 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1154 	IB_QP_SQ_PSN			= (1<<16),
1155 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1156 	IB_QP_PATH_MIG_STATE		= (1<<18),
1157 	IB_QP_CAP			= (1<<19),
1158 	IB_QP_DEST_QPN			= (1<<20),
1159 	IB_QP_RESERVED1			= (1<<21),
1160 	IB_QP_RESERVED2			= (1<<22),
1161 	IB_QP_RESERVED3			= (1<<23),
1162 	IB_QP_RESERVED4			= (1<<24),
1163 	IB_QP_RATE_LIMIT		= (1<<25),
1164 };
1165 
1166 enum ib_qp_state {
1167 	IB_QPS_RESET,
1168 	IB_QPS_INIT,
1169 	IB_QPS_RTR,
1170 	IB_QPS_RTS,
1171 	IB_QPS_SQD,
1172 	IB_QPS_SQE,
1173 	IB_QPS_ERR
1174 };
1175 
1176 enum ib_mig_state {
1177 	IB_MIG_MIGRATED,
1178 	IB_MIG_REARM,
1179 	IB_MIG_ARMED
1180 };
1181 
1182 enum ib_mw_type {
1183 	IB_MW_TYPE_1 = 1,
1184 	IB_MW_TYPE_2 = 2
1185 };
1186 
1187 struct ib_qp_attr {
1188 	enum ib_qp_state	qp_state;
1189 	enum ib_qp_state	cur_qp_state;
1190 	enum ib_mtu		path_mtu;
1191 	enum ib_mig_state	path_mig_state;
1192 	u32			qkey;
1193 	u32			rq_psn;
1194 	u32			sq_psn;
1195 	u32			dest_qp_num;
1196 	int			qp_access_flags;
1197 	struct ib_qp_cap	cap;
1198 	struct rdma_ah_attr	ah_attr;
1199 	struct rdma_ah_attr	alt_ah_attr;
1200 	u16			pkey_index;
1201 	u16			alt_pkey_index;
1202 	u8			en_sqd_async_notify;
1203 	u8			sq_draining;
1204 	u8			max_rd_atomic;
1205 	u8			max_dest_rd_atomic;
1206 	u8			min_rnr_timer;
1207 	u8			port_num;
1208 	u8			timeout;
1209 	u8			retry_cnt;
1210 	u8			rnr_retry;
1211 	u8			alt_port_num;
1212 	u8			alt_timeout;
1213 	u32			rate_limit;
1214 };
1215 
1216 enum ib_wr_opcode {
1217 	/* These are shared with userspace */
1218 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1219 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1220 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1221 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1222 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1223 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1224 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1225 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1226 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1227 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1228 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1229 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1230 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1231 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1232 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1233 
1234 	/* These are kernel only and can not be issued by userspace */
1235 	IB_WR_REG_MR = 0x20,
1236 	IB_WR_REG_MR_INTEGRITY,
1237 
1238 	/* reserve values for low level drivers' internal use.
1239 	 * These values will not be used at all in the ib core layer.
1240 	 */
1241 	IB_WR_RESERVED1 = 0xf0,
1242 	IB_WR_RESERVED2,
1243 	IB_WR_RESERVED3,
1244 	IB_WR_RESERVED4,
1245 	IB_WR_RESERVED5,
1246 	IB_WR_RESERVED6,
1247 	IB_WR_RESERVED7,
1248 	IB_WR_RESERVED8,
1249 	IB_WR_RESERVED9,
1250 	IB_WR_RESERVED10,
1251 };
1252 
1253 enum ib_send_flags {
1254 	IB_SEND_FENCE		= 1,
1255 	IB_SEND_SIGNALED	= (1<<1),
1256 	IB_SEND_SOLICITED	= (1<<2),
1257 	IB_SEND_INLINE		= (1<<3),
1258 	IB_SEND_IP_CSUM		= (1<<4),
1259 
1260 	/* reserve bits 26-31 for low level drivers' internal use */
1261 	IB_SEND_RESERVED_START	= (1 << 26),
1262 	IB_SEND_RESERVED_END	= (1 << 31),
1263 };
1264 
1265 struct ib_sge {
1266 	u64	addr;
1267 	u32	length;
1268 	u32	lkey;
1269 };
1270 
1271 struct ib_cqe {
1272 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1273 };
1274 
1275 struct ib_send_wr {
1276 	struct ib_send_wr      *next;
1277 	union {
1278 		u64		wr_id;
1279 		struct ib_cqe	*wr_cqe;
1280 	};
1281 	struct ib_sge	       *sg_list;
1282 	int			num_sge;
1283 	enum ib_wr_opcode	opcode;
1284 	int			send_flags;
1285 	union {
1286 		__be32		imm_data;
1287 		u32		invalidate_rkey;
1288 	} ex;
1289 };
1290 
1291 struct ib_rdma_wr {
1292 	struct ib_send_wr	wr;
1293 	u64			remote_addr;
1294 	u32			rkey;
1295 };
1296 
1297 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1298 {
1299 	return container_of(wr, struct ib_rdma_wr, wr);
1300 }
1301 
1302 struct ib_atomic_wr {
1303 	struct ib_send_wr	wr;
1304 	u64			remote_addr;
1305 	u64			compare_add;
1306 	u64			swap;
1307 	u64			compare_add_mask;
1308 	u64			swap_mask;
1309 	u32			rkey;
1310 };
1311 
1312 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1313 {
1314 	return container_of(wr, struct ib_atomic_wr, wr);
1315 }
1316 
1317 struct ib_ud_wr {
1318 	struct ib_send_wr	wr;
1319 	struct ib_ah		*ah;
1320 	void			*header;
1321 	int			hlen;
1322 	int			mss;
1323 	u32			remote_qpn;
1324 	u32			remote_qkey;
1325 	u16			pkey_index; /* valid for GSI only */
1326 	u8			port_num;   /* valid for DR SMPs on switch only */
1327 };
1328 
1329 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1330 {
1331 	return container_of(wr, struct ib_ud_wr, wr);
1332 }
1333 
1334 struct ib_reg_wr {
1335 	struct ib_send_wr	wr;
1336 	struct ib_mr		*mr;
1337 	u32			key;
1338 	int			access;
1339 };
1340 
1341 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1342 {
1343 	return container_of(wr, struct ib_reg_wr, wr);
1344 }
1345 
1346 struct ib_recv_wr {
1347 	struct ib_recv_wr      *next;
1348 	union {
1349 		u64		wr_id;
1350 		struct ib_cqe	*wr_cqe;
1351 	};
1352 	struct ib_sge	       *sg_list;
1353 	int			num_sge;
1354 };
1355 
1356 enum ib_access_flags {
1357 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1358 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1359 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1360 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1361 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1362 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1363 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1364 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1365 
1366 	IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1367 };
1368 
1369 /*
1370  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1371  * are hidden here instead of a uapi header!
1372  */
1373 enum ib_mr_rereg_flags {
1374 	IB_MR_REREG_TRANS	= 1,
1375 	IB_MR_REREG_PD		= (1<<1),
1376 	IB_MR_REREG_ACCESS	= (1<<2),
1377 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1378 };
1379 
1380 struct ib_fmr_attr {
1381 	int	max_pages;
1382 	int	max_maps;
1383 	u8	page_shift;
1384 };
1385 
1386 struct ib_umem;
1387 
1388 enum rdma_remove_reason {
1389 	/*
1390 	 * Userspace requested uobject deletion or initial try
1391 	 * to remove uobject via cleanup. Call could fail
1392 	 */
1393 	RDMA_REMOVE_DESTROY,
1394 	/* Context deletion. This call should delete the actual object itself */
1395 	RDMA_REMOVE_CLOSE,
1396 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1397 	RDMA_REMOVE_DRIVER_REMOVE,
1398 	/* uobj is being cleaned-up before being committed */
1399 	RDMA_REMOVE_ABORT,
1400 };
1401 
1402 struct ib_rdmacg_object {
1403 #ifdef CONFIG_CGROUP_RDMA
1404 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1405 #endif
1406 };
1407 
1408 struct ib_ucontext {
1409 	struct ib_device       *device;
1410 	struct ib_uverbs_file  *ufile;
1411 	/*
1412 	 * 'closing' can be read by the driver only during a destroy callback,
1413 	 * it is set when we are closing the file descriptor and indicates
1414 	 * that mm_sem may be locked.
1415 	 */
1416 	bool closing;
1417 
1418 	bool cleanup_retryable;
1419 
1420 	void (*invalidate_range)(struct ib_umem_odp *umem_odp,
1421 				 unsigned long start, unsigned long end);
1422 	struct mutex per_mm_list_lock;
1423 	struct list_head per_mm_list;
1424 
1425 	struct ib_rdmacg_object	cg_obj;
1426 	/*
1427 	 * Implementation details of the RDMA core, don't use in drivers:
1428 	 */
1429 	struct rdma_restrack_entry res;
1430 };
1431 
1432 struct ib_uobject {
1433 	u64			user_handle;	/* handle given to us by userspace */
1434 	/* ufile & ucontext owning this object */
1435 	struct ib_uverbs_file  *ufile;
1436 	/* FIXME, save memory: ufile->context == context */
1437 	struct ib_ucontext     *context;	/* associated user context */
1438 	void		       *object;		/* containing object */
1439 	struct list_head	list;		/* link to context's list */
1440 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1441 	int			id;		/* index into kernel idr */
1442 	struct kref		ref;
1443 	atomic_t		usecnt;		/* protects exclusive access */
1444 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1445 
1446 	const struct uverbs_api_object *uapi_object;
1447 };
1448 
1449 struct ib_udata {
1450 	const void __user *inbuf;
1451 	void __user *outbuf;
1452 	size_t       inlen;
1453 	size_t       outlen;
1454 };
1455 
1456 struct ib_pd {
1457 	u32			local_dma_lkey;
1458 	u32			flags;
1459 	struct ib_device       *device;
1460 	struct ib_uobject      *uobject;
1461 	atomic_t          	usecnt; /* count all resources */
1462 
1463 	u32			unsafe_global_rkey;
1464 
1465 	/*
1466 	 * Implementation details of the RDMA core, don't use in drivers:
1467 	 */
1468 	struct ib_mr	       *__internal_mr;
1469 	struct rdma_restrack_entry res;
1470 };
1471 
1472 struct ib_xrcd {
1473 	struct ib_device       *device;
1474 	atomic_t		usecnt; /* count all exposed resources */
1475 	struct inode	       *inode;
1476 
1477 	struct mutex		tgt_qp_mutex;
1478 	struct list_head	tgt_qp_list;
1479 };
1480 
1481 struct ib_ah {
1482 	struct ib_device	*device;
1483 	struct ib_pd		*pd;
1484 	struct ib_uobject	*uobject;
1485 	const struct ib_gid_attr *sgid_attr;
1486 	enum rdma_ah_attr_type	type;
1487 };
1488 
1489 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1490 
1491 enum ib_poll_context {
1492 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1493 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1494 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1495 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1496 };
1497 
1498 struct ib_cq {
1499 	struct ib_device       *device;
1500 	struct ib_uobject      *uobject;
1501 	ib_comp_handler   	comp_handler;
1502 	void                  (*event_handler)(struct ib_event *, void *);
1503 	void                   *cq_context;
1504 	int               	cqe;
1505 	atomic_t          	usecnt; /* count number of work queues */
1506 	enum ib_poll_context	poll_ctx;
1507 	struct ib_wc		*wc;
1508 	union {
1509 		struct irq_poll		iop;
1510 		struct work_struct	work;
1511 	};
1512 	struct workqueue_struct *comp_wq;
1513 	struct dim *dim;
1514 	/*
1515 	 * Implementation details of the RDMA core, don't use in drivers:
1516 	 */
1517 	struct rdma_restrack_entry res;
1518 };
1519 
1520 struct ib_srq {
1521 	struct ib_device       *device;
1522 	struct ib_pd	       *pd;
1523 	struct ib_uobject      *uobject;
1524 	void		      (*event_handler)(struct ib_event *, void *);
1525 	void		       *srq_context;
1526 	enum ib_srq_type	srq_type;
1527 	atomic_t		usecnt;
1528 
1529 	struct {
1530 		struct ib_cq   *cq;
1531 		union {
1532 			struct {
1533 				struct ib_xrcd *xrcd;
1534 				u32		srq_num;
1535 			} xrc;
1536 		};
1537 	} ext;
1538 };
1539 
1540 enum ib_raw_packet_caps {
1541 	/* Strip cvlan from incoming packet and report it in the matching work
1542 	 * completion is supported.
1543 	 */
1544 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1545 	/* Scatter FCS field of an incoming packet to host memory is supported.
1546 	 */
1547 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1548 	/* Checksum offloads are supported (for both send and receive). */
1549 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1550 	/* When a packet is received for an RQ with no receive WQEs, the
1551 	 * packet processing is delayed.
1552 	 */
1553 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1554 };
1555 
1556 enum ib_wq_type {
1557 	IB_WQT_RQ
1558 };
1559 
1560 enum ib_wq_state {
1561 	IB_WQS_RESET,
1562 	IB_WQS_RDY,
1563 	IB_WQS_ERR
1564 };
1565 
1566 struct ib_wq {
1567 	struct ib_device       *device;
1568 	struct ib_uobject      *uobject;
1569 	void		    *wq_context;
1570 	void		    (*event_handler)(struct ib_event *, void *);
1571 	struct ib_pd	       *pd;
1572 	struct ib_cq	       *cq;
1573 	u32		wq_num;
1574 	enum ib_wq_state       state;
1575 	enum ib_wq_type	wq_type;
1576 	atomic_t		usecnt;
1577 };
1578 
1579 enum ib_wq_flags {
1580 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1581 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1582 	IB_WQ_FLAGS_DELAY_DROP		= 1 << 2,
1583 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1584 };
1585 
1586 struct ib_wq_init_attr {
1587 	void		       *wq_context;
1588 	enum ib_wq_type	wq_type;
1589 	u32		max_wr;
1590 	u32		max_sge;
1591 	struct	ib_cq	       *cq;
1592 	void		    (*event_handler)(struct ib_event *, void *);
1593 	u32		create_flags; /* Use enum ib_wq_flags */
1594 };
1595 
1596 enum ib_wq_attr_mask {
1597 	IB_WQ_STATE		= 1 << 0,
1598 	IB_WQ_CUR_STATE		= 1 << 1,
1599 	IB_WQ_FLAGS		= 1 << 2,
1600 };
1601 
1602 struct ib_wq_attr {
1603 	enum	ib_wq_state	wq_state;
1604 	enum	ib_wq_state	curr_wq_state;
1605 	u32			flags; /* Use enum ib_wq_flags */
1606 	u32			flags_mask; /* Use enum ib_wq_flags */
1607 };
1608 
1609 struct ib_rwq_ind_table {
1610 	struct ib_device	*device;
1611 	struct ib_uobject      *uobject;
1612 	atomic_t		usecnt;
1613 	u32		ind_tbl_num;
1614 	u32		log_ind_tbl_size;
1615 	struct ib_wq	**ind_tbl;
1616 };
1617 
1618 struct ib_rwq_ind_table_init_attr {
1619 	u32		log_ind_tbl_size;
1620 	/* Each entry is a pointer to Receive Work Queue */
1621 	struct ib_wq	**ind_tbl;
1622 };
1623 
1624 enum port_pkey_state {
1625 	IB_PORT_PKEY_NOT_VALID = 0,
1626 	IB_PORT_PKEY_VALID = 1,
1627 	IB_PORT_PKEY_LISTED = 2,
1628 };
1629 
1630 struct ib_qp_security;
1631 
1632 struct ib_port_pkey {
1633 	enum port_pkey_state	state;
1634 	u16			pkey_index;
1635 	u8			port_num;
1636 	struct list_head	qp_list;
1637 	struct list_head	to_error_list;
1638 	struct ib_qp_security  *sec;
1639 };
1640 
1641 struct ib_ports_pkeys {
1642 	struct ib_port_pkey	main;
1643 	struct ib_port_pkey	alt;
1644 };
1645 
1646 struct ib_qp_security {
1647 	struct ib_qp	       *qp;
1648 	struct ib_device       *dev;
1649 	/* Hold this mutex when changing port and pkey settings. */
1650 	struct mutex		mutex;
1651 	struct ib_ports_pkeys  *ports_pkeys;
1652 	/* A list of all open shared QP handles.  Required to enforce security
1653 	 * properly for all users of a shared QP.
1654 	 */
1655 	struct list_head        shared_qp_list;
1656 	void                   *security;
1657 	bool			destroying;
1658 	atomic_t		error_list_count;
1659 	struct completion	error_complete;
1660 	int			error_comps_pending;
1661 };
1662 
1663 /*
1664  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1665  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1666  */
1667 struct ib_qp {
1668 	struct ib_device       *device;
1669 	struct ib_pd	       *pd;
1670 	struct ib_cq	       *send_cq;
1671 	struct ib_cq	       *recv_cq;
1672 	spinlock_t		mr_lock;
1673 	int			mrs_used;
1674 	struct list_head	rdma_mrs;
1675 	struct list_head	sig_mrs;
1676 	struct ib_srq	       *srq;
1677 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1678 	struct list_head	xrcd_list;
1679 
1680 	/* count times opened, mcast attaches, flow attaches */
1681 	atomic_t		usecnt;
1682 	struct list_head	open_list;
1683 	struct ib_qp           *real_qp;
1684 	struct ib_uobject      *uobject;
1685 	void                  (*event_handler)(struct ib_event *, void *);
1686 	void		       *qp_context;
1687 	/* sgid_attrs associated with the AV's */
1688 	const struct ib_gid_attr *av_sgid_attr;
1689 	const struct ib_gid_attr *alt_path_sgid_attr;
1690 	u32			qp_num;
1691 	u32			max_write_sge;
1692 	u32			max_read_sge;
1693 	enum ib_qp_type		qp_type;
1694 	struct ib_rwq_ind_table *rwq_ind_tbl;
1695 	struct ib_qp_security  *qp_sec;
1696 	u8			port;
1697 
1698 	bool			integrity_en;
1699 	/*
1700 	 * Implementation details of the RDMA core, don't use in drivers:
1701 	 */
1702 	struct rdma_restrack_entry     res;
1703 
1704 	/* The counter the qp is bind to */
1705 	struct rdma_counter    *counter;
1706 };
1707 
1708 struct ib_dm {
1709 	struct ib_device  *device;
1710 	u32		   length;
1711 	u32		   flags;
1712 	struct ib_uobject *uobject;
1713 	atomic_t	   usecnt;
1714 };
1715 
1716 struct ib_mr {
1717 	struct ib_device  *device;
1718 	struct ib_pd	  *pd;
1719 	u32		   lkey;
1720 	u32		   rkey;
1721 	u64		   iova;
1722 	u64		   length;
1723 	unsigned int	   page_size;
1724 	enum ib_mr_type	   type;
1725 	bool		   need_inval;
1726 	union {
1727 		struct ib_uobject	*uobject;	/* user */
1728 		struct list_head	qp_entry;	/* FR */
1729 	};
1730 
1731 	struct ib_dm      *dm;
1732 	struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1733 	/*
1734 	 * Implementation details of the RDMA core, don't use in drivers:
1735 	 */
1736 	struct rdma_restrack_entry res;
1737 };
1738 
1739 struct ib_mw {
1740 	struct ib_device	*device;
1741 	struct ib_pd		*pd;
1742 	struct ib_uobject	*uobject;
1743 	u32			rkey;
1744 	enum ib_mw_type         type;
1745 };
1746 
1747 struct ib_fmr {
1748 	struct ib_device	*device;
1749 	struct ib_pd		*pd;
1750 	struct list_head	list;
1751 	u32			lkey;
1752 	u32			rkey;
1753 };
1754 
1755 /* Supported steering options */
1756 enum ib_flow_attr_type {
1757 	/* steering according to rule specifications */
1758 	IB_FLOW_ATTR_NORMAL		= 0x0,
1759 	/* default unicast and multicast rule -
1760 	 * receive all Eth traffic which isn't steered to any QP
1761 	 */
1762 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1763 	/* default multicast rule -
1764 	 * receive all Eth multicast traffic which isn't steered to any QP
1765 	 */
1766 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1767 	/* sniffer rule - receive all port traffic */
1768 	IB_FLOW_ATTR_SNIFFER		= 0x3
1769 };
1770 
1771 /* Supported steering header types */
1772 enum ib_flow_spec_type {
1773 	/* L2 headers*/
1774 	IB_FLOW_SPEC_ETH		= 0x20,
1775 	IB_FLOW_SPEC_IB			= 0x22,
1776 	/* L3 header*/
1777 	IB_FLOW_SPEC_IPV4		= 0x30,
1778 	IB_FLOW_SPEC_IPV6		= 0x31,
1779 	IB_FLOW_SPEC_ESP                = 0x34,
1780 	/* L4 headers*/
1781 	IB_FLOW_SPEC_TCP		= 0x40,
1782 	IB_FLOW_SPEC_UDP		= 0x41,
1783 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1784 	IB_FLOW_SPEC_GRE		= 0x51,
1785 	IB_FLOW_SPEC_MPLS		= 0x60,
1786 	IB_FLOW_SPEC_INNER		= 0x100,
1787 	/* Actions */
1788 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1789 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1790 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1791 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1792 };
1793 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1794 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1795 
1796 /* Flow steering rule priority is set according to it's domain.
1797  * Lower domain value means higher priority.
1798  */
1799 enum ib_flow_domain {
1800 	IB_FLOW_DOMAIN_USER,
1801 	IB_FLOW_DOMAIN_ETHTOOL,
1802 	IB_FLOW_DOMAIN_RFS,
1803 	IB_FLOW_DOMAIN_NIC,
1804 	IB_FLOW_DOMAIN_NUM /* Must be last */
1805 };
1806 
1807 enum ib_flow_flags {
1808 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1809 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1810 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1811 };
1812 
1813 struct ib_flow_eth_filter {
1814 	u8	dst_mac[6];
1815 	u8	src_mac[6];
1816 	__be16	ether_type;
1817 	__be16	vlan_tag;
1818 	/* Must be last */
1819 	u8	real_sz[0];
1820 };
1821 
1822 struct ib_flow_spec_eth {
1823 	u32			  type;
1824 	u16			  size;
1825 	struct ib_flow_eth_filter val;
1826 	struct ib_flow_eth_filter mask;
1827 };
1828 
1829 struct ib_flow_ib_filter {
1830 	__be16 dlid;
1831 	__u8   sl;
1832 	/* Must be last */
1833 	u8	real_sz[0];
1834 };
1835 
1836 struct ib_flow_spec_ib {
1837 	u32			 type;
1838 	u16			 size;
1839 	struct ib_flow_ib_filter val;
1840 	struct ib_flow_ib_filter mask;
1841 };
1842 
1843 /* IPv4 header flags */
1844 enum ib_ipv4_flags {
1845 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1846 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1847 				    last have this flag set */
1848 };
1849 
1850 struct ib_flow_ipv4_filter {
1851 	__be32	src_ip;
1852 	__be32	dst_ip;
1853 	u8	proto;
1854 	u8	tos;
1855 	u8	ttl;
1856 	u8	flags;
1857 	/* Must be last */
1858 	u8	real_sz[0];
1859 };
1860 
1861 struct ib_flow_spec_ipv4 {
1862 	u32			   type;
1863 	u16			   size;
1864 	struct ib_flow_ipv4_filter val;
1865 	struct ib_flow_ipv4_filter mask;
1866 };
1867 
1868 struct ib_flow_ipv6_filter {
1869 	u8	src_ip[16];
1870 	u8	dst_ip[16];
1871 	__be32	flow_label;
1872 	u8	next_hdr;
1873 	u8	traffic_class;
1874 	u8	hop_limit;
1875 	/* Must be last */
1876 	u8	real_sz[0];
1877 };
1878 
1879 struct ib_flow_spec_ipv6 {
1880 	u32			   type;
1881 	u16			   size;
1882 	struct ib_flow_ipv6_filter val;
1883 	struct ib_flow_ipv6_filter mask;
1884 };
1885 
1886 struct ib_flow_tcp_udp_filter {
1887 	__be16	dst_port;
1888 	__be16	src_port;
1889 	/* Must be last */
1890 	u8	real_sz[0];
1891 };
1892 
1893 struct ib_flow_spec_tcp_udp {
1894 	u32			      type;
1895 	u16			      size;
1896 	struct ib_flow_tcp_udp_filter val;
1897 	struct ib_flow_tcp_udp_filter mask;
1898 };
1899 
1900 struct ib_flow_tunnel_filter {
1901 	__be32	tunnel_id;
1902 	u8	real_sz[0];
1903 };
1904 
1905 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1906  * the tunnel_id from val has the vni value
1907  */
1908 struct ib_flow_spec_tunnel {
1909 	u32			      type;
1910 	u16			      size;
1911 	struct ib_flow_tunnel_filter  val;
1912 	struct ib_flow_tunnel_filter  mask;
1913 };
1914 
1915 struct ib_flow_esp_filter {
1916 	__be32	spi;
1917 	__be32  seq;
1918 	/* Must be last */
1919 	u8	real_sz[0];
1920 };
1921 
1922 struct ib_flow_spec_esp {
1923 	u32                           type;
1924 	u16			      size;
1925 	struct ib_flow_esp_filter     val;
1926 	struct ib_flow_esp_filter     mask;
1927 };
1928 
1929 struct ib_flow_gre_filter {
1930 	__be16 c_ks_res0_ver;
1931 	__be16 protocol;
1932 	__be32 key;
1933 	/* Must be last */
1934 	u8	real_sz[0];
1935 };
1936 
1937 struct ib_flow_spec_gre {
1938 	u32                           type;
1939 	u16			      size;
1940 	struct ib_flow_gre_filter     val;
1941 	struct ib_flow_gre_filter     mask;
1942 };
1943 
1944 struct ib_flow_mpls_filter {
1945 	__be32 tag;
1946 	/* Must be last */
1947 	u8	real_sz[0];
1948 };
1949 
1950 struct ib_flow_spec_mpls {
1951 	u32                           type;
1952 	u16			      size;
1953 	struct ib_flow_mpls_filter     val;
1954 	struct ib_flow_mpls_filter     mask;
1955 };
1956 
1957 struct ib_flow_spec_action_tag {
1958 	enum ib_flow_spec_type	      type;
1959 	u16			      size;
1960 	u32                           tag_id;
1961 };
1962 
1963 struct ib_flow_spec_action_drop {
1964 	enum ib_flow_spec_type	      type;
1965 	u16			      size;
1966 };
1967 
1968 struct ib_flow_spec_action_handle {
1969 	enum ib_flow_spec_type	      type;
1970 	u16			      size;
1971 	struct ib_flow_action	     *act;
1972 };
1973 
1974 enum ib_counters_description {
1975 	IB_COUNTER_PACKETS,
1976 	IB_COUNTER_BYTES,
1977 };
1978 
1979 struct ib_flow_spec_action_count {
1980 	enum ib_flow_spec_type type;
1981 	u16 size;
1982 	struct ib_counters *counters;
1983 };
1984 
1985 union ib_flow_spec {
1986 	struct {
1987 		u32			type;
1988 		u16			size;
1989 	};
1990 	struct ib_flow_spec_eth		eth;
1991 	struct ib_flow_spec_ib		ib;
1992 	struct ib_flow_spec_ipv4        ipv4;
1993 	struct ib_flow_spec_tcp_udp	tcp_udp;
1994 	struct ib_flow_spec_ipv6        ipv6;
1995 	struct ib_flow_spec_tunnel      tunnel;
1996 	struct ib_flow_spec_esp		esp;
1997 	struct ib_flow_spec_gre		gre;
1998 	struct ib_flow_spec_mpls	mpls;
1999 	struct ib_flow_spec_action_tag  flow_tag;
2000 	struct ib_flow_spec_action_drop drop;
2001 	struct ib_flow_spec_action_handle action;
2002 	struct ib_flow_spec_action_count flow_count;
2003 };
2004 
2005 struct ib_flow_attr {
2006 	enum ib_flow_attr_type type;
2007 	u16	     size;
2008 	u16	     priority;
2009 	u32	     flags;
2010 	u8	     num_of_specs;
2011 	u8	     port;
2012 	union ib_flow_spec flows[];
2013 };
2014 
2015 struct ib_flow {
2016 	struct ib_qp		*qp;
2017 	struct ib_device	*device;
2018 	struct ib_uobject	*uobject;
2019 };
2020 
2021 enum ib_flow_action_type {
2022 	IB_FLOW_ACTION_UNSPECIFIED,
2023 	IB_FLOW_ACTION_ESP = 1,
2024 };
2025 
2026 struct ib_flow_action_attrs_esp_keymats {
2027 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2028 	union {
2029 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2030 	} keymat;
2031 };
2032 
2033 struct ib_flow_action_attrs_esp_replays {
2034 	enum ib_uverbs_flow_action_esp_replay			protocol;
2035 	union {
2036 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2037 	} replay;
2038 };
2039 
2040 enum ib_flow_action_attrs_esp_flags {
2041 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2042 	 * This is done in order to share the same flags between user-space and
2043 	 * kernel and spare an unnecessary translation.
2044 	 */
2045 
2046 	/* Kernel flags */
2047 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2048 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2049 };
2050 
2051 struct ib_flow_spec_list {
2052 	struct ib_flow_spec_list	*next;
2053 	union ib_flow_spec		spec;
2054 };
2055 
2056 struct ib_flow_action_attrs_esp {
2057 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2058 	struct ib_flow_action_attrs_esp_replays		*replay;
2059 	struct ib_flow_spec_list			*encap;
2060 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2061 	 * Value of 0 is a valid value.
2062 	 */
2063 	u32						esn;
2064 	u32						spi;
2065 	u32						seq;
2066 	u32						tfc_pad;
2067 	/* Use enum ib_flow_action_attrs_esp_flags */
2068 	u64						flags;
2069 	u64						hard_limit_pkts;
2070 };
2071 
2072 struct ib_flow_action {
2073 	struct ib_device		*device;
2074 	struct ib_uobject		*uobject;
2075 	enum ib_flow_action_type	type;
2076 	atomic_t			usecnt;
2077 };
2078 
2079 struct ib_mad_hdr;
2080 struct ib_grh;
2081 
2082 enum ib_process_mad_flags {
2083 	IB_MAD_IGNORE_MKEY	= 1,
2084 	IB_MAD_IGNORE_BKEY	= 2,
2085 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2086 };
2087 
2088 enum ib_mad_result {
2089 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2090 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2091 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2092 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2093 };
2094 
2095 struct ib_port_cache {
2096 	u64		      subnet_prefix;
2097 	struct ib_pkey_cache  *pkey;
2098 	struct ib_gid_table   *gid;
2099 	u8                     lmc;
2100 	enum ib_port_state     port_state;
2101 };
2102 
2103 struct ib_cache {
2104 	rwlock_t                lock;
2105 	struct ib_event_handler event_handler;
2106 };
2107 
2108 struct ib_port_immutable {
2109 	int                           pkey_tbl_len;
2110 	int                           gid_tbl_len;
2111 	u32                           core_cap_flags;
2112 	u32                           max_mad_size;
2113 };
2114 
2115 struct ib_port_data {
2116 	struct ib_device *ib_dev;
2117 
2118 	struct ib_port_immutable immutable;
2119 
2120 	spinlock_t pkey_list_lock;
2121 	struct list_head pkey_list;
2122 
2123 	struct ib_port_cache cache;
2124 
2125 	spinlock_t netdev_lock;
2126 	struct net_device __rcu *netdev;
2127 	struct hlist_node ndev_hash_link;
2128 	struct rdma_port_counter port_counter;
2129 	struct rdma_hw_stats *hw_stats;
2130 };
2131 
2132 /* rdma netdev type - specifies protocol type */
2133 enum rdma_netdev_t {
2134 	RDMA_NETDEV_OPA_VNIC,
2135 	RDMA_NETDEV_IPOIB,
2136 };
2137 
2138 /**
2139  * struct rdma_netdev - rdma netdev
2140  * For cases where netstack interfacing is required.
2141  */
2142 struct rdma_netdev {
2143 	void              *clnt_priv;
2144 	struct ib_device  *hca;
2145 	u8                 port_num;
2146 
2147 	/*
2148 	 * cleanup function must be specified.
2149 	 * FIXME: This is only used for OPA_VNIC and that usage should be
2150 	 * removed too.
2151 	 */
2152 	void (*free_rdma_netdev)(struct net_device *netdev);
2153 
2154 	/* control functions */
2155 	void (*set_id)(struct net_device *netdev, int id);
2156 	/* send packet */
2157 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2158 		    struct ib_ah *address, u32 dqpn);
2159 	/* multicast */
2160 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2161 			    union ib_gid *gid, u16 mlid,
2162 			    int set_qkey, u32 qkey);
2163 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2164 			    union ib_gid *gid, u16 mlid);
2165 };
2166 
2167 struct rdma_netdev_alloc_params {
2168 	size_t sizeof_priv;
2169 	unsigned int txqs;
2170 	unsigned int rxqs;
2171 	void *param;
2172 
2173 	int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2174 				      struct net_device *netdev, void *param);
2175 };
2176 
2177 struct ib_counters {
2178 	struct ib_device	*device;
2179 	struct ib_uobject	*uobject;
2180 	/* num of objects attached */
2181 	atomic_t	usecnt;
2182 };
2183 
2184 struct ib_counters_read_attr {
2185 	u64	*counters_buff;
2186 	u32	ncounters;
2187 	u32	flags; /* use enum ib_read_counters_flags */
2188 };
2189 
2190 struct uverbs_attr_bundle;
2191 struct iw_cm_id;
2192 struct iw_cm_conn_param;
2193 
2194 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member)                      \
2195 	.size_##ib_struct =                                                    \
2196 		(sizeof(struct drv_struct) +                                   \
2197 		 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) +      \
2198 		 BUILD_BUG_ON_ZERO(                                            \
2199 			 !__same_type(((struct drv_struct *)NULL)->member,     \
2200 				      struct ib_struct)))
2201 
2202 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp)                         \
2203 	((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2204 
2205 #define rdma_zalloc_drv_obj(ib_dev, ib_type)                                   \
2206 	rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2207 
2208 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2209 
2210 /**
2211  * struct ib_device_ops - InfiniBand device operations
2212  * This structure defines all the InfiniBand device operations, providers will
2213  * need to define the supported operations, otherwise they will be set to null.
2214  */
2215 struct ib_device_ops {
2216 	struct module *owner;
2217 	enum rdma_driver_id driver_id;
2218 	u32 uverbs_abi_ver;
2219 	unsigned int uverbs_no_driver_id_binding:1;
2220 
2221 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2222 			 const struct ib_send_wr **bad_send_wr);
2223 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2224 			 const struct ib_recv_wr **bad_recv_wr);
2225 	void (*drain_rq)(struct ib_qp *qp);
2226 	void (*drain_sq)(struct ib_qp *qp);
2227 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2228 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2229 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2230 	int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2231 	int (*post_srq_recv)(struct ib_srq *srq,
2232 			     const struct ib_recv_wr *recv_wr,
2233 			     const struct ib_recv_wr **bad_recv_wr);
2234 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2235 			   u8 port_num, const struct ib_wc *in_wc,
2236 			   const struct ib_grh *in_grh,
2237 			   const struct ib_mad_hdr *in_mad, size_t in_mad_size,
2238 			   struct ib_mad_hdr *out_mad, size_t *out_mad_size,
2239 			   u16 *out_mad_pkey_index);
2240 	int (*query_device)(struct ib_device *device,
2241 			    struct ib_device_attr *device_attr,
2242 			    struct ib_udata *udata);
2243 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2244 			     struct ib_device_modify *device_modify);
2245 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2246 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2247 						     int comp_vector);
2248 	int (*query_port)(struct ib_device *device, u8 port_num,
2249 			  struct ib_port_attr *port_attr);
2250 	int (*modify_port)(struct ib_device *device, u8 port_num,
2251 			   int port_modify_mask,
2252 			   struct ib_port_modify *port_modify);
2253 	/**
2254 	 * The following mandatory functions are used only at device
2255 	 * registration.  Keep functions such as these at the end of this
2256 	 * structure to avoid cache line misses when accessing struct ib_device
2257 	 * in fast paths.
2258 	 */
2259 	int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2260 				  struct ib_port_immutable *immutable);
2261 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2262 					       u8 port_num);
2263 	/**
2264 	 * When calling get_netdev, the HW vendor's driver should return the
2265 	 * net device of device @device at port @port_num or NULL if such
2266 	 * a net device doesn't exist. The vendor driver should call dev_hold
2267 	 * on this net device. The HW vendor's device driver must guarantee
2268 	 * that this function returns NULL before the net device has finished
2269 	 * NETDEV_UNREGISTER state.
2270 	 */
2271 	struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2272 	/**
2273 	 * rdma netdev operation
2274 	 *
2275 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2276 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2277 	 */
2278 	struct net_device *(*alloc_rdma_netdev)(
2279 		struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2280 		const char *name, unsigned char name_assign_type,
2281 		void (*setup)(struct net_device *));
2282 
2283 	int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2284 				      enum rdma_netdev_t type,
2285 				      struct rdma_netdev_alloc_params *params);
2286 	/**
2287 	 * query_gid should be return GID value for @device, when @port_num
2288 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2289 	 * is RoCE link layer.
2290 	 */
2291 	int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2292 			 union ib_gid *gid);
2293 	/**
2294 	 * When calling add_gid, the HW vendor's driver should add the gid
2295 	 * of device of port at gid index available at @attr. Meta-info of
2296 	 * that gid (for example, the network device related to this gid) is
2297 	 * available at @attr. @context allows the HW vendor driver to store
2298 	 * extra information together with a GID entry. The HW vendor driver may
2299 	 * allocate memory to contain this information and store it in @context
2300 	 * when a new GID entry is written to. Params are consistent until the
2301 	 * next call of add_gid or delete_gid. The function should return 0 on
2302 	 * success or error otherwise. The function could be called
2303 	 * concurrently for different ports. This function is only called when
2304 	 * roce_gid_table is used.
2305 	 */
2306 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2307 	/**
2308 	 * When calling del_gid, the HW vendor's driver should delete the
2309 	 * gid of device @device at gid index gid_index of port port_num
2310 	 * available in @attr.
2311 	 * Upon the deletion of a GID entry, the HW vendor must free any
2312 	 * allocated memory. The caller will clear @context afterwards.
2313 	 * This function is only called when roce_gid_table is used.
2314 	 */
2315 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2316 	int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2317 			  u16 *pkey);
2318 	int (*alloc_ucontext)(struct ib_ucontext *context,
2319 			      struct ib_udata *udata);
2320 	void (*dealloc_ucontext)(struct ib_ucontext *context);
2321 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2322 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2323 	int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2324 	void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2325 	int (*create_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr,
2326 			 u32 flags, struct ib_udata *udata);
2327 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2328 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2329 	void (*destroy_ah)(struct ib_ah *ah, u32 flags);
2330 	int (*create_srq)(struct ib_srq *srq,
2331 			  struct ib_srq_init_attr *srq_init_attr,
2332 			  struct ib_udata *udata);
2333 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2334 			  enum ib_srq_attr_mask srq_attr_mask,
2335 			  struct ib_udata *udata);
2336 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2337 	void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2338 	struct ib_qp *(*create_qp)(struct ib_pd *pd,
2339 				   struct ib_qp_init_attr *qp_init_attr,
2340 				   struct ib_udata *udata);
2341 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2342 			 int qp_attr_mask, struct ib_udata *udata);
2343 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2344 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2345 	int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2346 	int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2347 			 struct ib_udata *udata);
2348 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2349 	void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2350 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2351 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2352 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2353 				     u64 virt_addr, int mr_access_flags,
2354 				     struct ib_udata *udata);
2355 	int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2356 			     u64 virt_addr, int mr_access_flags,
2357 			     struct ib_pd *pd, struct ib_udata *udata);
2358 	int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2359 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2360 				  u32 max_num_sg, struct ib_udata *udata);
2361 	struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2362 					    u32 max_num_data_sg,
2363 					    u32 max_num_meta_sg);
2364 	int (*advise_mr)(struct ib_pd *pd,
2365 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2366 			 struct ib_sge *sg_list, u32 num_sge,
2367 			 struct uverbs_attr_bundle *attrs);
2368 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2369 			 unsigned int *sg_offset);
2370 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2371 			       struct ib_mr_status *mr_status);
2372 	struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2373 				  struct ib_udata *udata);
2374 	int (*dealloc_mw)(struct ib_mw *mw);
2375 	struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2376 				    struct ib_fmr_attr *fmr_attr);
2377 	int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2378 			    u64 iova);
2379 	int (*unmap_fmr)(struct list_head *fmr_list);
2380 	int (*dealloc_fmr)(struct ib_fmr *fmr);
2381 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2382 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2383 	struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2384 				      struct ib_udata *udata);
2385 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2386 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2387 				       struct ib_flow_attr *flow_attr,
2388 				       int domain, struct ib_udata *udata);
2389 	int (*destroy_flow)(struct ib_flow *flow_id);
2390 	struct ib_flow_action *(*create_flow_action_esp)(
2391 		struct ib_device *device,
2392 		const struct ib_flow_action_attrs_esp *attr,
2393 		struct uverbs_attr_bundle *attrs);
2394 	int (*destroy_flow_action)(struct ib_flow_action *action);
2395 	int (*modify_flow_action_esp)(
2396 		struct ib_flow_action *action,
2397 		const struct ib_flow_action_attrs_esp *attr,
2398 		struct uverbs_attr_bundle *attrs);
2399 	int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2400 				 int state);
2401 	int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2402 			     struct ifla_vf_info *ivf);
2403 	int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2404 			    struct ifla_vf_stats *stats);
2405 	int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2406 			   int type);
2407 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2408 				   struct ib_wq_init_attr *init_attr,
2409 				   struct ib_udata *udata);
2410 	void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2411 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2412 			 u32 wq_attr_mask, struct ib_udata *udata);
2413 	struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2414 		struct ib_device *device,
2415 		struct ib_rwq_ind_table_init_attr *init_attr,
2416 		struct ib_udata *udata);
2417 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2418 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2419 				  struct ib_ucontext *context,
2420 				  struct ib_dm_alloc_attr *attr,
2421 				  struct uverbs_attr_bundle *attrs);
2422 	int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2423 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2424 				   struct ib_dm_mr_attr *attr,
2425 				   struct uverbs_attr_bundle *attrs);
2426 	struct ib_counters *(*create_counters)(
2427 		struct ib_device *device, struct uverbs_attr_bundle *attrs);
2428 	int (*destroy_counters)(struct ib_counters *counters);
2429 	int (*read_counters)(struct ib_counters *counters,
2430 			     struct ib_counters_read_attr *counters_read_attr,
2431 			     struct uverbs_attr_bundle *attrs);
2432 	int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2433 			    int data_sg_nents, unsigned int *data_sg_offset,
2434 			    struct scatterlist *meta_sg, int meta_sg_nents,
2435 			    unsigned int *meta_sg_offset);
2436 
2437 	/**
2438 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2439 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2440 	 *   core when the device is removed.  A lifespan of -1 in the return
2441 	 *   struct tells the core to set a default lifespan.
2442 	 */
2443 	struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2444 						u8 port_num);
2445 	/**
2446 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2447 	 * @index - The index in the value array we wish to have updated, or
2448 	 *   num_counters if we want all stats updated
2449 	 * Return codes -
2450 	 *   < 0 - Error, no counters updated
2451 	 *   index - Updated the single counter pointed to by index
2452 	 *   num_counters - Updated all counters (will reset the timestamp
2453 	 *     and prevent further calls for lifespan milliseconds)
2454 	 * Drivers are allowed to update all counters in leiu of just the
2455 	 *   one given in index at their option
2456 	 */
2457 	int (*get_hw_stats)(struct ib_device *device,
2458 			    struct rdma_hw_stats *stats, u8 port, int index);
2459 	/*
2460 	 * This function is called once for each port when a ib device is
2461 	 * registered.
2462 	 */
2463 	int (*init_port)(struct ib_device *device, u8 port_num,
2464 			 struct kobject *port_sysfs);
2465 	/**
2466 	 * Allows rdma drivers to add their own restrack attributes.
2467 	 */
2468 	int (*fill_res_entry)(struct sk_buff *msg,
2469 			      struct rdma_restrack_entry *entry);
2470 
2471 	/* Device lifecycle callbacks */
2472 	/*
2473 	 * Called after the device becomes registered, before clients are
2474 	 * attached
2475 	 */
2476 	int (*enable_driver)(struct ib_device *dev);
2477 	/*
2478 	 * This is called as part of ib_dealloc_device().
2479 	 */
2480 	void (*dealloc_driver)(struct ib_device *dev);
2481 
2482 	/* iWarp CM callbacks */
2483 	void (*iw_add_ref)(struct ib_qp *qp);
2484 	void (*iw_rem_ref)(struct ib_qp *qp);
2485 	struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2486 	int (*iw_connect)(struct iw_cm_id *cm_id,
2487 			  struct iw_cm_conn_param *conn_param);
2488 	int (*iw_accept)(struct iw_cm_id *cm_id,
2489 			 struct iw_cm_conn_param *conn_param);
2490 	int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2491 			 u8 pdata_len);
2492 	int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2493 	int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2494 	/**
2495 	 * counter_bind_qp - Bind a QP to a counter.
2496 	 * @counter - The counter to be bound. If counter->id is zero then
2497 	 *   the driver needs to allocate a new counter and set counter->id
2498 	 */
2499 	int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2500 	/**
2501 	 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2502 	 *   counter and bind it onto the default one
2503 	 */
2504 	int (*counter_unbind_qp)(struct ib_qp *qp);
2505 	/**
2506 	 * counter_dealloc -De-allocate the hw counter
2507 	 */
2508 	int (*counter_dealloc)(struct rdma_counter *counter);
2509 	/**
2510 	 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2511 	 * the driver initialized data.
2512 	 */
2513 	struct rdma_hw_stats *(*counter_alloc_stats)(
2514 		struct rdma_counter *counter);
2515 	/**
2516 	 * counter_update_stats - Query the stats value of this counter
2517 	 */
2518 	int (*counter_update_stats)(struct rdma_counter *counter);
2519 
2520 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
2521 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
2522 	DECLARE_RDMA_OBJ_SIZE(ib_pd);
2523 	DECLARE_RDMA_OBJ_SIZE(ib_srq);
2524 	DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2525 };
2526 
2527 struct ib_core_device {
2528 	/* device must be the first element in structure until,
2529 	 * union of ib_core_device and device exists in ib_device.
2530 	 */
2531 	struct device dev;
2532 	possible_net_t rdma_net;
2533 	struct kobject *ports_kobj;
2534 	struct list_head port_list;
2535 	struct ib_device *owner; /* reach back to owner ib_device */
2536 };
2537 
2538 struct rdma_restrack_root;
2539 struct ib_device {
2540 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2541 	struct device                *dma_device;
2542 	struct ib_device_ops	     ops;
2543 	char                          name[IB_DEVICE_NAME_MAX];
2544 	struct rcu_head rcu_head;
2545 
2546 	struct list_head              event_handler_list;
2547 	spinlock_t                    event_handler_lock;
2548 
2549 	struct rw_semaphore	      client_data_rwsem;
2550 	struct xarray                 client_data;
2551 	struct mutex                  unregistration_lock;
2552 
2553 	struct ib_cache               cache;
2554 	/**
2555 	 * port_data is indexed by port number
2556 	 */
2557 	struct ib_port_data *port_data;
2558 
2559 	int			      num_comp_vectors;
2560 
2561 	union {
2562 		struct device		dev;
2563 		struct ib_core_device	coredev;
2564 	};
2565 
2566 	/* First group for device attributes,
2567 	 * Second group for driver provided attributes (optional).
2568 	 * It is NULL terminated array.
2569 	 */
2570 	const struct attribute_group	*groups[3];
2571 
2572 	u64			     uverbs_cmd_mask;
2573 	u64			     uverbs_ex_cmd_mask;
2574 
2575 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2576 	__be64			     node_guid;
2577 	u32			     local_dma_lkey;
2578 	u16                          is_switch:1;
2579 	/* Indicates kernel verbs support, should not be used in drivers */
2580 	u16                          kverbs_provider:1;
2581 	/* CQ adaptive moderation (RDMA DIM) */
2582 	u16                          use_cq_dim:1;
2583 	u8                           node_type;
2584 	u8                           phys_port_cnt;
2585 	struct ib_device_attr        attrs;
2586 	struct attribute_group	     *hw_stats_ag;
2587 	struct rdma_hw_stats         *hw_stats;
2588 
2589 #ifdef CONFIG_CGROUP_RDMA
2590 	struct rdmacg_device         cg_device;
2591 #endif
2592 
2593 	u32                          index;
2594 	struct rdma_restrack_root *res;
2595 
2596 	const struct uapi_definition   *driver_def;
2597 
2598 	/*
2599 	 * Positive refcount indicates that the device is currently
2600 	 * registered and cannot be unregistered.
2601 	 */
2602 	refcount_t refcount;
2603 	struct completion unreg_completion;
2604 	struct work_struct unregistration_work;
2605 
2606 	const struct rdma_link_ops *link_ops;
2607 
2608 	/* Protects compat_devs xarray modifications */
2609 	struct mutex compat_devs_mutex;
2610 	/* Maintains compat devices for each net namespace */
2611 	struct xarray compat_devs;
2612 
2613 	/* Used by iWarp CM */
2614 	char iw_ifname[IFNAMSIZ];
2615 	u32 iw_driver_flags;
2616 };
2617 
2618 struct ib_client_nl_info;
2619 struct ib_client {
2620 	const char *name;
2621 	void (*add)   (struct ib_device *);
2622 	void (*remove)(struct ib_device *, void *client_data);
2623 	void (*rename)(struct ib_device *dev, void *client_data);
2624 	int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2625 			   struct ib_client_nl_info *res);
2626 	int (*get_global_nl_info)(struct ib_client_nl_info *res);
2627 
2628 	/* Returns the net_dev belonging to this ib_client and matching the
2629 	 * given parameters.
2630 	 * @dev:	 An RDMA device that the net_dev use for communication.
2631 	 * @port:	 A physical port number on the RDMA device.
2632 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2633 	 * @gid:	 A GID that the net_dev uses to communicate.
2634 	 * @addr:	 An IP address the net_dev is configured with.
2635 	 * @client_data: The device's client data set by ib_set_client_data().
2636 	 *
2637 	 * An ib_client that implements a net_dev on top of RDMA devices
2638 	 * (such as IP over IB) should implement this callback, allowing the
2639 	 * rdma_cm module to find the right net_dev for a given request.
2640 	 *
2641 	 * The caller is responsible for calling dev_put on the returned
2642 	 * netdev. */
2643 	struct net_device *(*get_net_dev_by_params)(
2644 			struct ib_device *dev,
2645 			u8 port,
2646 			u16 pkey,
2647 			const union ib_gid *gid,
2648 			const struct sockaddr *addr,
2649 			void *client_data);
2650 
2651 	refcount_t uses;
2652 	struct completion uses_zero;
2653 	u32 client_id;
2654 
2655 	/* kverbs are not required by the client */
2656 	u8 no_kverbs_req:1;
2657 };
2658 
2659 /*
2660  * IB block DMA iterator
2661  *
2662  * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2663  * to a HW supported page size.
2664  */
2665 struct ib_block_iter {
2666 	/* internal states */
2667 	struct scatterlist *__sg;	/* sg holding the current aligned block */
2668 	dma_addr_t __dma_addr;		/* unaligned DMA address of this block */
2669 	unsigned int __sg_nents;	/* number of SG entries */
2670 	unsigned int __sg_advance;	/* number of bytes to advance in sg in next step */
2671 	unsigned int __pg_bit;		/* alignment of current block */
2672 };
2673 
2674 struct ib_device *_ib_alloc_device(size_t size);
2675 #define ib_alloc_device(drv_struct, member)                                    \
2676 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2677 				      BUILD_BUG_ON_ZERO(offsetof(              \
2678 					      struct drv_struct, member))),    \
2679 		     struct drv_struct, member)
2680 
2681 void ib_dealloc_device(struct ib_device *device);
2682 
2683 void ib_get_device_fw_str(struct ib_device *device, char *str);
2684 
2685 int ib_register_device(struct ib_device *device, const char *name);
2686 void ib_unregister_device(struct ib_device *device);
2687 void ib_unregister_driver(enum rdma_driver_id driver_id);
2688 void ib_unregister_device_and_put(struct ib_device *device);
2689 void ib_unregister_device_queued(struct ib_device *ib_dev);
2690 
2691 int ib_register_client   (struct ib_client *client);
2692 void ib_unregister_client(struct ib_client *client);
2693 
2694 void __rdma_block_iter_start(struct ib_block_iter *biter,
2695 			     struct scatterlist *sglist,
2696 			     unsigned int nents,
2697 			     unsigned long pgsz);
2698 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2699 
2700 /**
2701  * rdma_block_iter_dma_address - get the aligned dma address of the current
2702  * block held by the block iterator.
2703  * @biter: block iterator holding the memory block
2704  */
2705 static inline dma_addr_t
2706 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2707 {
2708 	return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2709 }
2710 
2711 /**
2712  * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2713  * @sglist: sglist to iterate over
2714  * @biter: block iterator holding the memory block
2715  * @nents: maximum number of sg entries to iterate over
2716  * @pgsz: best HW supported page size to use
2717  *
2718  * Callers may use rdma_block_iter_dma_address() to get each
2719  * blocks aligned DMA address.
2720  */
2721 #define rdma_for_each_block(sglist, biter, nents, pgsz)		\
2722 	for (__rdma_block_iter_start(biter, sglist, nents,	\
2723 				     pgsz);			\
2724 	     __rdma_block_iter_next(biter);)
2725 
2726 /**
2727  * ib_get_client_data - Get IB client context
2728  * @device:Device to get context for
2729  * @client:Client to get context for
2730  *
2731  * ib_get_client_data() returns the client context data set with
2732  * ib_set_client_data(). This can only be called while the client is
2733  * registered to the device, once the ib_client remove() callback returns this
2734  * cannot be called.
2735  */
2736 static inline void *ib_get_client_data(struct ib_device *device,
2737 				       struct ib_client *client)
2738 {
2739 	return xa_load(&device->client_data, client->client_id);
2740 }
2741 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2742 			 void *data);
2743 void ib_set_device_ops(struct ib_device *device,
2744 		       const struct ib_device_ops *ops);
2745 
2746 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2747 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2748 		      unsigned long pfn, unsigned long size, pgprot_t prot);
2749 #else
2750 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
2751 				    struct vm_area_struct *vma,
2752 				    unsigned long pfn, unsigned long size,
2753 				    pgprot_t prot)
2754 {
2755 	return -EINVAL;
2756 }
2757 #endif
2758 
2759 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2760 {
2761 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2762 }
2763 
2764 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2765 {
2766 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2767 }
2768 
2769 static inline bool ib_is_buffer_cleared(const void __user *p,
2770 					size_t len)
2771 {
2772 	bool ret;
2773 	u8 *buf;
2774 
2775 	if (len > USHRT_MAX)
2776 		return false;
2777 
2778 	buf = memdup_user(p, len);
2779 	if (IS_ERR(buf))
2780 		return false;
2781 
2782 	ret = !memchr_inv(buf, 0, len);
2783 	kfree(buf);
2784 	return ret;
2785 }
2786 
2787 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2788 				       size_t offset,
2789 				       size_t len)
2790 {
2791 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
2792 }
2793 
2794 /**
2795  * ib_is_destroy_retryable - Check whether the uobject destruction
2796  * is retryable.
2797  * @ret: The initial destruction return code
2798  * @why: remove reason
2799  * @uobj: The uobject that is destroyed
2800  *
2801  * This function is a helper function that IB layer and low-level drivers
2802  * can use to consider whether the destruction of the given uobject is
2803  * retry-able.
2804  * It checks the original return code, if it wasn't success the destruction
2805  * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2806  * the remove reason. (i.e. why).
2807  * Must be called with the object locked for destroy.
2808  */
2809 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2810 					   struct ib_uobject *uobj)
2811 {
2812 	return ret && (why == RDMA_REMOVE_DESTROY ||
2813 		       uobj->context->cleanup_retryable);
2814 }
2815 
2816 /**
2817  * ib_destroy_usecnt - Called during destruction to check the usecnt
2818  * @usecnt: The usecnt atomic
2819  * @why: remove reason
2820  * @uobj: The uobject that is destroyed
2821  *
2822  * Non-zero usecnts will block destruction unless destruction was triggered by
2823  * a ucontext cleanup.
2824  */
2825 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2826 				    enum rdma_remove_reason why,
2827 				    struct ib_uobject *uobj)
2828 {
2829 	if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2830 		return -EBUSY;
2831 	return 0;
2832 }
2833 
2834 /**
2835  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2836  * contains all required attributes and no attributes not allowed for
2837  * the given QP state transition.
2838  * @cur_state: Current QP state
2839  * @next_state: Next QP state
2840  * @type: QP type
2841  * @mask: Mask of supplied QP attributes
2842  *
2843  * This function is a helper function that a low-level driver's
2844  * modify_qp method can use to validate the consumer's input.  It
2845  * checks that cur_state and next_state are valid QP states, that a
2846  * transition from cur_state to next_state is allowed by the IB spec,
2847  * and that the attribute mask supplied is allowed for the transition.
2848  */
2849 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2850 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
2851 
2852 void ib_register_event_handler(struct ib_event_handler *event_handler);
2853 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2854 void ib_dispatch_event(struct ib_event *event);
2855 
2856 int ib_query_port(struct ib_device *device,
2857 		  u8 port_num, struct ib_port_attr *port_attr);
2858 
2859 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2860 					       u8 port_num);
2861 
2862 /**
2863  * rdma_cap_ib_switch - Check if the device is IB switch
2864  * @device: Device to check
2865  *
2866  * Device driver is responsible for setting is_switch bit on
2867  * in ib_device structure at init time.
2868  *
2869  * Return: true if the device is IB switch.
2870  */
2871 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2872 {
2873 	return device->is_switch;
2874 }
2875 
2876 /**
2877  * rdma_start_port - Return the first valid port number for the device
2878  * specified
2879  *
2880  * @device: Device to be checked
2881  *
2882  * Return start port number
2883  */
2884 static inline u8 rdma_start_port(const struct ib_device *device)
2885 {
2886 	return rdma_cap_ib_switch(device) ? 0 : 1;
2887 }
2888 
2889 /**
2890  * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2891  * @device - The struct ib_device * to iterate over
2892  * @iter - The unsigned int to store the port number
2893  */
2894 #define rdma_for_each_port(device, iter)                                       \
2895 	for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type(   \
2896 						     unsigned int, iter)));    \
2897 	     iter <= rdma_end_port(device); (iter)++)
2898 
2899 /**
2900  * rdma_end_port - Return the last valid port number for the device
2901  * specified
2902  *
2903  * @device: Device to be checked
2904  *
2905  * Return last port number
2906  */
2907 static inline u8 rdma_end_port(const struct ib_device *device)
2908 {
2909 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2910 }
2911 
2912 static inline int rdma_is_port_valid(const struct ib_device *device,
2913 				     unsigned int port)
2914 {
2915 	return (port >= rdma_start_port(device) &&
2916 		port <= rdma_end_port(device));
2917 }
2918 
2919 static inline bool rdma_is_grh_required(const struct ib_device *device,
2920 					u8 port_num)
2921 {
2922 	return device->port_data[port_num].immutable.core_cap_flags &
2923 	       RDMA_CORE_PORT_IB_GRH_REQUIRED;
2924 }
2925 
2926 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2927 {
2928 	return device->port_data[port_num].immutable.core_cap_flags &
2929 	       RDMA_CORE_CAP_PROT_IB;
2930 }
2931 
2932 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2933 {
2934 	return device->port_data[port_num].immutable.core_cap_flags &
2935 	       (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2936 }
2937 
2938 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2939 {
2940 	return device->port_data[port_num].immutable.core_cap_flags &
2941 	       RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2942 }
2943 
2944 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2945 {
2946 	return device->port_data[port_num].immutable.core_cap_flags &
2947 	       RDMA_CORE_CAP_PROT_ROCE;
2948 }
2949 
2950 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2951 {
2952 	return device->port_data[port_num].immutable.core_cap_flags &
2953 	       RDMA_CORE_CAP_PROT_IWARP;
2954 }
2955 
2956 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2957 {
2958 	return rdma_protocol_ib(device, port_num) ||
2959 		rdma_protocol_roce(device, port_num);
2960 }
2961 
2962 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2963 {
2964 	return device->port_data[port_num].immutable.core_cap_flags &
2965 	       RDMA_CORE_CAP_PROT_RAW_PACKET;
2966 }
2967 
2968 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2969 {
2970 	return device->port_data[port_num].immutable.core_cap_flags &
2971 	       RDMA_CORE_CAP_PROT_USNIC;
2972 }
2973 
2974 /**
2975  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2976  * Management Datagrams.
2977  * @device: Device to check
2978  * @port_num: Port number to check
2979  *
2980  * Management Datagrams (MAD) are a required part of the InfiniBand
2981  * specification and are supported on all InfiniBand devices.  A slightly
2982  * extended version are also supported on OPA interfaces.
2983  *
2984  * Return: true if the port supports sending/receiving of MAD packets.
2985  */
2986 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2987 {
2988 	return device->port_data[port_num].immutable.core_cap_flags &
2989 	       RDMA_CORE_CAP_IB_MAD;
2990 }
2991 
2992 /**
2993  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2994  * Management Datagrams.
2995  * @device: Device to check
2996  * @port_num: Port number to check
2997  *
2998  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2999  * datagrams with their own versions.  These OPA MADs share many but not all of
3000  * the characteristics of InfiniBand MADs.
3001  *
3002  * OPA MADs differ in the following ways:
3003  *
3004  *    1) MADs are variable size up to 2K
3005  *       IBTA defined MADs remain fixed at 256 bytes
3006  *    2) OPA SMPs must carry valid PKeys
3007  *    3) OPA SMP packets are a different format
3008  *
3009  * Return: true if the port supports OPA MAD packet formats.
3010  */
3011 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3012 {
3013 	return device->port_data[port_num].immutable.core_cap_flags &
3014 		RDMA_CORE_CAP_OPA_MAD;
3015 }
3016 
3017 /**
3018  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3019  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3020  * @device: Device to check
3021  * @port_num: Port number to check
3022  *
3023  * Each InfiniBand node is required to provide a Subnet Management Agent
3024  * that the subnet manager can access.  Prior to the fabric being fully
3025  * configured by the subnet manager, the SMA is accessed via a well known
3026  * interface called the Subnet Management Interface (SMI).  This interface
3027  * uses directed route packets to communicate with the SM to get around the
3028  * chicken and egg problem of the SM needing to know what's on the fabric
3029  * in order to configure the fabric, and needing to configure the fabric in
3030  * order to send packets to the devices on the fabric.  These directed
3031  * route packets do not need the fabric fully configured in order to reach
3032  * their destination.  The SMI is the only method allowed to send
3033  * directed route packets on an InfiniBand fabric.
3034  *
3035  * Return: true if the port provides an SMI.
3036  */
3037 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3038 {
3039 	return device->port_data[port_num].immutable.core_cap_flags &
3040 	       RDMA_CORE_CAP_IB_SMI;
3041 }
3042 
3043 /**
3044  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3045  * Communication Manager.
3046  * @device: Device to check
3047  * @port_num: Port number to check
3048  *
3049  * The InfiniBand Communication Manager is one of many pre-defined General
3050  * Service Agents (GSA) that are accessed via the General Service
3051  * Interface (GSI).  It's role is to facilitate establishment of connections
3052  * between nodes as well as other management related tasks for established
3053  * connections.
3054  *
3055  * Return: true if the port supports an IB CM (this does not guarantee that
3056  * a CM is actually running however).
3057  */
3058 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3059 {
3060 	return device->port_data[port_num].immutable.core_cap_flags &
3061 	       RDMA_CORE_CAP_IB_CM;
3062 }
3063 
3064 /**
3065  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3066  * Communication Manager.
3067  * @device: Device to check
3068  * @port_num: Port number to check
3069  *
3070  * Similar to above, but specific to iWARP connections which have a different
3071  * managment protocol than InfiniBand.
3072  *
3073  * Return: true if the port supports an iWARP CM (this does not guarantee that
3074  * a CM is actually running however).
3075  */
3076 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3077 {
3078 	return device->port_data[port_num].immutable.core_cap_flags &
3079 	       RDMA_CORE_CAP_IW_CM;
3080 }
3081 
3082 /**
3083  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3084  * Subnet Administration.
3085  * @device: Device to check
3086  * @port_num: Port number to check
3087  *
3088  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3089  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
3090  * fabrics, devices should resolve routes to other hosts by contacting the
3091  * SA to query the proper route.
3092  *
3093  * Return: true if the port should act as a client to the fabric Subnet
3094  * Administration interface.  This does not imply that the SA service is
3095  * running locally.
3096  */
3097 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3098 {
3099 	return device->port_data[port_num].immutable.core_cap_flags &
3100 	       RDMA_CORE_CAP_IB_SA;
3101 }
3102 
3103 /**
3104  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3105  * Multicast.
3106  * @device: Device to check
3107  * @port_num: Port number to check
3108  *
3109  * InfiniBand multicast registration is more complex than normal IPv4 or
3110  * IPv6 multicast registration.  Each Host Channel Adapter must register
3111  * with the Subnet Manager when it wishes to join a multicast group.  It
3112  * should do so only once regardless of how many queue pairs it subscribes
3113  * to this group.  And it should leave the group only after all queue pairs
3114  * attached to the group have been detached.
3115  *
3116  * Return: true if the port must undertake the additional adminstrative
3117  * overhead of registering/unregistering with the SM and tracking of the
3118  * total number of queue pairs attached to the multicast group.
3119  */
3120 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3121 {
3122 	return rdma_cap_ib_sa(device, port_num);
3123 }
3124 
3125 /**
3126  * rdma_cap_af_ib - Check if the port of device has the capability
3127  * Native Infiniband Address.
3128  * @device: Device to check
3129  * @port_num: Port number to check
3130  *
3131  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3132  * GID.  RoCE uses a different mechanism, but still generates a GID via
3133  * a prescribed mechanism and port specific data.
3134  *
3135  * Return: true if the port uses a GID address to identify devices on the
3136  * network.
3137  */
3138 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3139 {
3140 	return device->port_data[port_num].immutable.core_cap_flags &
3141 	       RDMA_CORE_CAP_AF_IB;
3142 }
3143 
3144 /**
3145  * rdma_cap_eth_ah - Check if the port of device has the capability
3146  * Ethernet Address Handle.
3147  * @device: Device to check
3148  * @port_num: Port number to check
3149  *
3150  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3151  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3152  * port.  Normally, packet headers are generated by the sending host
3153  * adapter, but when sending connectionless datagrams, we must manually
3154  * inject the proper headers for the fabric we are communicating over.
3155  *
3156  * Return: true if we are running as a RoCE port and must force the
3157  * addition of a Global Route Header built from our Ethernet Address
3158  * Handle into our header list for connectionless packets.
3159  */
3160 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3161 {
3162 	return device->port_data[port_num].immutable.core_cap_flags &
3163 	       RDMA_CORE_CAP_ETH_AH;
3164 }
3165 
3166 /**
3167  * rdma_cap_opa_ah - Check if the port of device supports
3168  * OPA Address handles
3169  * @device: Device to check
3170  * @port_num: Port number to check
3171  *
3172  * Return: true if we are running on an OPA device which supports
3173  * the extended OPA addressing.
3174  */
3175 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3176 {
3177 	return (device->port_data[port_num].immutable.core_cap_flags &
3178 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3179 }
3180 
3181 /**
3182  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3183  *
3184  * @device: Device
3185  * @port_num: Port number
3186  *
3187  * This MAD size includes the MAD headers and MAD payload.  No other headers
3188  * are included.
3189  *
3190  * Return the max MAD size required by the Port.  Will return 0 if the port
3191  * does not support MADs
3192  */
3193 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3194 {
3195 	return device->port_data[port_num].immutable.max_mad_size;
3196 }
3197 
3198 /**
3199  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3200  * @device: Device to check
3201  * @port_num: Port number to check
3202  *
3203  * RoCE GID table mechanism manages the various GIDs for a device.
3204  *
3205  * NOTE: if allocating the port's GID table has failed, this call will still
3206  * return true, but any RoCE GID table API will fail.
3207  *
3208  * Return: true if the port uses RoCE GID table mechanism in order to manage
3209  * its GIDs.
3210  */
3211 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3212 					   u8 port_num)
3213 {
3214 	return rdma_protocol_roce(device, port_num) &&
3215 		device->ops.add_gid && device->ops.del_gid;
3216 }
3217 
3218 /*
3219  * Check if the device supports READ W/ INVALIDATE.
3220  */
3221 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3222 {
3223 	/*
3224 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3225 	 * has support for it yet.
3226 	 */
3227 	return rdma_protocol_iwarp(dev, port_num);
3228 }
3229 
3230 /**
3231  * rdma_find_pg_bit - Find page bit given address and HW supported page sizes
3232  *
3233  * @addr: address
3234  * @pgsz_bitmap: bitmap of HW supported page sizes
3235  */
3236 static inline unsigned int rdma_find_pg_bit(unsigned long addr,
3237 					    unsigned long pgsz_bitmap)
3238 {
3239 	unsigned long align;
3240 	unsigned long pgsz;
3241 
3242 	align = addr & -addr;
3243 
3244 	/* Find page bit such that addr is aligned to the highest supported
3245 	 * HW page size
3246 	 */
3247 	pgsz = pgsz_bitmap & ~(-align << 1);
3248 	if (!pgsz)
3249 		return __ffs(pgsz_bitmap);
3250 
3251 	return __fls(pgsz);
3252 }
3253 
3254 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3255 			 int state);
3256 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3257 		     struct ifla_vf_info *info);
3258 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3259 		    struct ifla_vf_stats *stats);
3260 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3261 		   int type);
3262 
3263 int ib_query_pkey(struct ib_device *device,
3264 		  u8 port_num, u16 index, u16 *pkey);
3265 
3266 int ib_modify_device(struct ib_device *device,
3267 		     int device_modify_mask,
3268 		     struct ib_device_modify *device_modify);
3269 
3270 int ib_modify_port(struct ib_device *device,
3271 		   u8 port_num, int port_modify_mask,
3272 		   struct ib_port_modify *port_modify);
3273 
3274 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3275 		u8 *port_num, u16 *index);
3276 
3277 int ib_find_pkey(struct ib_device *device,
3278 		 u8 port_num, u16 pkey, u16 *index);
3279 
3280 enum ib_pd_flags {
3281 	/*
3282 	 * Create a memory registration for all memory in the system and place
3283 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3284 	 * ULPs to avoid the overhead of dynamic MRs.
3285 	 *
3286 	 * This flag is generally considered unsafe and must only be used in
3287 	 * extremly trusted environments.  Every use of it will log a warning
3288 	 * in the kernel log.
3289 	 */
3290 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3291 };
3292 
3293 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3294 		const char *caller);
3295 
3296 #define ib_alloc_pd(device, flags) \
3297 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3298 
3299 /**
3300  * ib_dealloc_pd_user - Deallocate kernel/user PD
3301  * @pd: The protection domain
3302  * @udata: Valid user data or NULL for kernel objects
3303  */
3304 void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3305 
3306 /**
3307  * ib_dealloc_pd - Deallocate kernel PD
3308  * @pd: The protection domain
3309  *
3310  * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3311  */
3312 static inline void ib_dealloc_pd(struct ib_pd *pd)
3313 {
3314 	ib_dealloc_pd_user(pd, NULL);
3315 }
3316 
3317 enum rdma_create_ah_flags {
3318 	/* In a sleepable context */
3319 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3320 };
3321 
3322 /**
3323  * rdma_create_ah - Creates an address handle for the given address vector.
3324  * @pd: The protection domain associated with the address handle.
3325  * @ah_attr: The attributes of the address vector.
3326  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3327  *
3328  * The address handle is used to reference a local or global destination
3329  * in all UD QP post sends.
3330  */
3331 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3332 			     u32 flags);
3333 
3334 /**
3335  * rdma_create_user_ah - Creates an address handle for the given address vector.
3336  * It resolves destination mac address for ah attribute of RoCE type.
3337  * @pd: The protection domain associated with the address handle.
3338  * @ah_attr: The attributes of the address vector.
3339  * @udata: pointer to user's input output buffer information need by
3340  *         provider driver.
3341  *
3342  * It returns 0 on success and returns appropriate error code on error.
3343  * The address handle is used to reference a local or global destination
3344  * in all UD QP post sends.
3345  */
3346 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3347 				  struct rdma_ah_attr *ah_attr,
3348 				  struct ib_udata *udata);
3349 /**
3350  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3351  *   work completion.
3352  * @hdr: the L3 header to parse
3353  * @net_type: type of header to parse
3354  * @sgid: place to store source gid
3355  * @dgid: place to store destination gid
3356  */
3357 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3358 			      enum rdma_network_type net_type,
3359 			      union ib_gid *sgid, union ib_gid *dgid);
3360 
3361 /**
3362  * ib_get_rdma_header_version - Get the header version
3363  * @hdr: the L3 header to parse
3364  */
3365 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3366 
3367 /**
3368  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3369  *   work completion.
3370  * @device: Device on which the received message arrived.
3371  * @port_num: Port on which the received message arrived.
3372  * @wc: Work completion associated with the received message.
3373  * @grh: References the received global route header.  This parameter is
3374  *   ignored unless the work completion indicates that the GRH is valid.
3375  * @ah_attr: Returned attributes that can be used when creating an address
3376  *   handle for replying to the message.
3377  * When ib_init_ah_attr_from_wc() returns success,
3378  * (a) for IB link layer it optionally contains a reference to SGID attribute
3379  * when GRH is present for IB link layer.
3380  * (b) for RoCE link layer it contains a reference to SGID attribute.
3381  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3382  * attributes which are initialized using ib_init_ah_attr_from_wc().
3383  *
3384  */
3385 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3386 			    const struct ib_wc *wc, const struct ib_grh *grh,
3387 			    struct rdma_ah_attr *ah_attr);
3388 
3389 /**
3390  * ib_create_ah_from_wc - Creates an address handle associated with the
3391  *   sender of the specified work completion.
3392  * @pd: The protection domain associated with the address handle.
3393  * @wc: Work completion information associated with a received message.
3394  * @grh: References the received global route header.  This parameter is
3395  *   ignored unless the work completion indicates that the GRH is valid.
3396  * @port_num: The outbound port number to associate with the address.
3397  *
3398  * The address handle is used to reference a local or global destination
3399  * in all UD QP post sends.
3400  */
3401 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3402 				   const struct ib_grh *grh, u8 port_num);
3403 
3404 /**
3405  * rdma_modify_ah - Modifies the address vector associated with an address
3406  *   handle.
3407  * @ah: The address handle to modify.
3408  * @ah_attr: The new address vector attributes to associate with the
3409  *   address handle.
3410  */
3411 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3412 
3413 /**
3414  * rdma_query_ah - Queries the address vector associated with an address
3415  *   handle.
3416  * @ah: The address handle to query.
3417  * @ah_attr: The address vector attributes associated with the address
3418  *   handle.
3419  */
3420 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3421 
3422 enum rdma_destroy_ah_flags {
3423 	/* In a sleepable context */
3424 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3425 };
3426 
3427 /**
3428  * rdma_destroy_ah_user - Destroys an address handle.
3429  * @ah: The address handle to destroy.
3430  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3431  * @udata: Valid user data or NULL for kernel objects
3432  */
3433 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3434 
3435 /**
3436  * rdma_destroy_ah - Destroys an kernel address handle.
3437  * @ah: The address handle to destroy.
3438  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3439  *
3440  * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3441  */
3442 static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3443 {
3444 	return rdma_destroy_ah_user(ah, flags, NULL);
3445 }
3446 
3447 /**
3448  * ib_create_srq - Creates a SRQ associated with the specified protection
3449  *   domain.
3450  * @pd: The protection domain associated with the SRQ.
3451  * @srq_init_attr: A list of initial attributes required to create the
3452  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
3453  *   the actual capabilities of the created SRQ.
3454  *
3455  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3456  * requested size of the SRQ, and set to the actual values allocated
3457  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
3458  * will always be at least as large as the requested values.
3459  */
3460 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3461 			     struct ib_srq_init_attr *srq_init_attr);
3462 
3463 /**
3464  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3465  * @srq: The SRQ to modify.
3466  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3467  *   the current values of selected SRQ attributes are returned.
3468  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3469  *   are being modified.
3470  *
3471  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3472  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3473  * the number of receives queued drops below the limit.
3474  */
3475 int ib_modify_srq(struct ib_srq *srq,
3476 		  struct ib_srq_attr *srq_attr,
3477 		  enum ib_srq_attr_mask srq_attr_mask);
3478 
3479 /**
3480  * ib_query_srq - Returns the attribute list and current values for the
3481  *   specified SRQ.
3482  * @srq: The SRQ to query.
3483  * @srq_attr: The attributes of the specified SRQ.
3484  */
3485 int ib_query_srq(struct ib_srq *srq,
3486 		 struct ib_srq_attr *srq_attr);
3487 
3488 /**
3489  * ib_destroy_srq_user - Destroys the specified SRQ.
3490  * @srq: The SRQ to destroy.
3491  * @udata: Valid user data or NULL for kernel objects
3492  */
3493 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3494 
3495 /**
3496  * ib_destroy_srq - Destroys the specified kernel SRQ.
3497  * @srq: The SRQ to destroy.
3498  *
3499  * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3500  */
3501 static inline int ib_destroy_srq(struct ib_srq *srq)
3502 {
3503 	return ib_destroy_srq_user(srq, NULL);
3504 }
3505 
3506 /**
3507  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3508  * @srq: The SRQ to post the work request on.
3509  * @recv_wr: A list of work requests to post on the receive queue.
3510  * @bad_recv_wr: On an immediate failure, this parameter will reference
3511  *   the work request that failed to be posted on the QP.
3512  */
3513 static inline int ib_post_srq_recv(struct ib_srq *srq,
3514 				   const struct ib_recv_wr *recv_wr,
3515 				   const struct ib_recv_wr **bad_recv_wr)
3516 {
3517 	const struct ib_recv_wr *dummy;
3518 
3519 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3520 					      bad_recv_wr ? : &dummy);
3521 }
3522 
3523 /**
3524  * ib_create_qp_user - Creates a QP associated with the specified protection
3525  *   domain.
3526  * @pd: The protection domain associated with the QP.
3527  * @qp_init_attr: A list of initial attributes required to create the
3528  *   QP.  If QP creation succeeds, then the attributes are updated to
3529  *   the actual capabilities of the created QP.
3530  * @udata: Valid user data or NULL for kernel objects
3531  */
3532 struct ib_qp *ib_create_qp_user(struct ib_pd *pd,
3533 				struct ib_qp_init_attr *qp_init_attr,
3534 				struct ib_udata *udata);
3535 
3536 /**
3537  * ib_create_qp - Creates a kernel QP associated with the specified protection
3538  *   domain.
3539  * @pd: The protection domain associated with the QP.
3540  * @qp_init_attr: A list of initial attributes required to create the
3541  *   QP.  If QP creation succeeds, then the attributes are updated to
3542  *   the actual capabilities of the created QP.
3543  * @udata: Valid user data or NULL for kernel objects
3544  *
3545  * NOTE: for user qp use ib_create_qp_user with valid udata!
3546  */
3547 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3548 					 struct ib_qp_init_attr *qp_init_attr)
3549 {
3550 	return ib_create_qp_user(pd, qp_init_attr, NULL);
3551 }
3552 
3553 /**
3554  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3555  * @qp: The QP to modify.
3556  * @attr: On input, specifies the QP attributes to modify.  On output,
3557  *   the current values of selected QP attributes are returned.
3558  * @attr_mask: A bit-mask used to specify which attributes of the QP
3559  *   are being modified.
3560  * @udata: pointer to user's input output buffer information
3561  *   are being modified.
3562  * It returns 0 on success and returns appropriate error code on error.
3563  */
3564 int ib_modify_qp_with_udata(struct ib_qp *qp,
3565 			    struct ib_qp_attr *attr,
3566 			    int attr_mask,
3567 			    struct ib_udata *udata);
3568 
3569 /**
3570  * ib_modify_qp - Modifies the attributes for the specified QP and then
3571  *   transitions the QP to the given state.
3572  * @qp: The QP to modify.
3573  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3574  *   the current values of selected QP attributes are returned.
3575  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3576  *   are being modified.
3577  */
3578 int ib_modify_qp(struct ib_qp *qp,
3579 		 struct ib_qp_attr *qp_attr,
3580 		 int qp_attr_mask);
3581 
3582 /**
3583  * ib_query_qp - Returns the attribute list and current values for the
3584  *   specified QP.
3585  * @qp: The QP to query.
3586  * @qp_attr: The attributes of the specified QP.
3587  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3588  * @qp_init_attr: Additional attributes of the selected QP.
3589  *
3590  * The qp_attr_mask may be used to limit the query to gathering only the
3591  * selected attributes.
3592  */
3593 int ib_query_qp(struct ib_qp *qp,
3594 		struct ib_qp_attr *qp_attr,
3595 		int qp_attr_mask,
3596 		struct ib_qp_init_attr *qp_init_attr);
3597 
3598 /**
3599  * ib_destroy_qp - Destroys the specified QP.
3600  * @qp: The QP to destroy.
3601  * @udata: Valid udata or NULL for kernel objects
3602  */
3603 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3604 
3605 /**
3606  * ib_destroy_qp - Destroys the specified kernel QP.
3607  * @qp: The QP to destroy.
3608  *
3609  * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3610  */
3611 static inline int ib_destroy_qp(struct ib_qp *qp)
3612 {
3613 	return ib_destroy_qp_user(qp, NULL);
3614 }
3615 
3616 /**
3617  * ib_open_qp - Obtain a reference to an existing sharable QP.
3618  * @xrcd - XRC domain
3619  * @qp_open_attr: Attributes identifying the QP to open.
3620  *
3621  * Returns a reference to a sharable QP.
3622  */
3623 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3624 			 struct ib_qp_open_attr *qp_open_attr);
3625 
3626 /**
3627  * ib_close_qp - Release an external reference to a QP.
3628  * @qp: The QP handle to release
3629  *
3630  * The opened QP handle is released by the caller.  The underlying
3631  * shared QP is not destroyed until all internal references are released.
3632  */
3633 int ib_close_qp(struct ib_qp *qp);
3634 
3635 /**
3636  * ib_post_send - Posts a list of work requests to the send queue of
3637  *   the specified QP.
3638  * @qp: The QP to post the work request on.
3639  * @send_wr: A list of work requests to post on the send queue.
3640  * @bad_send_wr: On an immediate failure, this parameter will reference
3641  *   the work request that failed to be posted on the QP.
3642  *
3643  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3644  * error is returned, the QP state shall not be affected,
3645  * ib_post_send() will return an immediate error after queueing any
3646  * earlier work requests in the list.
3647  */
3648 static inline int ib_post_send(struct ib_qp *qp,
3649 			       const struct ib_send_wr *send_wr,
3650 			       const struct ib_send_wr **bad_send_wr)
3651 {
3652 	const struct ib_send_wr *dummy;
3653 
3654 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3655 }
3656 
3657 /**
3658  * ib_post_recv - Posts a list of work requests to the receive queue of
3659  *   the specified QP.
3660  * @qp: The QP to post the work request on.
3661  * @recv_wr: A list of work requests to post on the receive queue.
3662  * @bad_recv_wr: On an immediate failure, this parameter will reference
3663  *   the work request that failed to be posted on the QP.
3664  */
3665 static inline int ib_post_recv(struct ib_qp *qp,
3666 			       const struct ib_recv_wr *recv_wr,
3667 			       const struct ib_recv_wr **bad_recv_wr)
3668 {
3669 	const struct ib_recv_wr *dummy;
3670 
3671 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3672 }
3673 
3674 struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
3675 				 int nr_cqe, int comp_vector,
3676 				 enum ib_poll_context poll_ctx,
3677 				 const char *caller, struct ib_udata *udata);
3678 
3679 /**
3680  * ib_alloc_cq_user: Allocate kernel/user CQ
3681  * @dev: The IB device
3682  * @private: Private data attached to the CQE
3683  * @nr_cqe: Number of CQEs in the CQ
3684  * @comp_vector: Completion vector used for the IRQs
3685  * @poll_ctx: Context used for polling the CQ
3686  * @udata: Valid user data or NULL for kernel objects
3687  */
3688 static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
3689 					     void *private, int nr_cqe,
3690 					     int comp_vector,
3691 					     enum ib_poll_context poll_ctx,
3692 					     struct ib_udata *udata)
3693 {
3694 	return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3695 				  KBUILD_MODNAME, udata);
3696 }
3697 
3698 /**
3699  * ib_alloc_cq: Allocate kernel CQ
3700  * @dev: The IB device
3701  * @private: Private data attached to the CQE
3702  * @nr_cqe: Number of CQEs in the CQ
3703  * @comp_vector: Completion vector used for the IRQs
3704  * @poll_ctx: Context used for polling the CQ
3705  *
3706  * NOTE: for user cq use ib_alloc_cq_user with valid udata!
3707  */
3708 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3709 					int nr_cqe, int comp_vector,
3710 					enum ib_poll_context poll_ctx)
3711 {
3712 	return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3713 				NULL);
3714 }
3715 
3716 /**
3717  * ib_free_cq_user - Free kernel/user CQ
3718  * @cq: The CQ to free
3719  * @udata: Valid user data or NULL for kernel objects
3720  */
3721 void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3722 
3723 /**
3724  * ib_free_cq - Free kernel CQ
3725  * @cq: The CQ to free
3726  *
3727  * NOTE: for user cq use ib_free_cq_user with valid udata!
3728  */
3729 static inline void ib_free_cq(struct ib_cq *cq)
3730 {
3731 	ib_free_cq_user(cq, NULL);
3732 }
3733 
3734 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3735 
3736 /**
3737  * ib_create_cq - Creates a CQ on the specified device.
3738  * @device: The device on which to create the CQ.
3739  * @comp_handler: A user-specified callback that is invoked when a
3740  *   completion event occurs on the CQ.
3741  * @event_handler: A user-specified callback that is invoked when an
3742  *   asynchronous event not associated with a completion occurs on the CQ.
3743  * @cq_context: Context associated with the CQ returned to the user via
3744  *   the associated completion and event handlers.
3745  * @cq_attr: The attributes the CQ should be created upon.
3746  *
3747  * Users can examine the cq structure to determine the actual CQ size.
3748  */
3749 struct ib_cq *__ib_create_cq(struct ib_device *device,
3750 			     ib_comp_handler comp_handler,
3751 			     void (*event_handler)(struct ib_event *, void *),
3752 			     void *cq_context,
3753 			     const struct ib_cq_init_attr *cq_attr,
3754 			     const char *caller);
3755 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3756 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3757 
3758 /**
3759  * ib_resize_cq - Modifies the capacity of the CQ.
3760  * @cq: The CQ to resize.
3761  * @cqe: The minimum size of the CQ.
3762  *
3763  * Users can examine the cq structure to determine the actual CQ size.
3764  */
3765 int ib_resize_cq(struct ib_cq *cq, int cqe);
3766 
3767 /**
3768  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3769  * @cq: The CQ to modify.
3770  * @cq_count: number of CQEs that will trigger an event
3771  * @cq_period: max period of time in usec before triggering an event
3772  *
3773  */
3774 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3775 
3776 /**
3777  * ib_destroy_cq_user - Destroys the specified CQ.
3778  * @cq: The CQ to destroy.
3779  * @udata: Valid user data or NULL for kernel objects
3780  */
3781 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3782 
3783 /**
3784  * ib_destroy_cq - Destroys the specified kernel CQ.
3785  * @cq: The CQ to destroy.
3786  *
3787  * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3788  */
3789 static inline void ib_destroy_cq(struct ib_cq *cq)
3790 {
3791 	ib_destroy_cq_user(cq, NULL);
3792 }
3793 
3794 /**
3795  * ib_poll_cq - poll a CQ for completion(s)
3796  * @cq:the CQ being polled
3797  * @num_entries:maximum number of completions to return
3798  * @wc:array of at least @num_entries &struct ib_wc where completions
3799  *   will be returned
3800  *
3801  * Poll a CQ for (possibly multiple) completions.  If the return value
3802  * is < 0, an error occurred.  If the return value is >= 0, it is the
3803  * number of completions returned.  If the return value is
3804  * non-negative and < num_entries, then the CQ was emptied.
3805  */
3806 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3807 			     struct ib_wc *wc)
3808 {
3809 	return cq->device->ops.poll_cq(cq, num_entries, wc);
3810 }
3811 
3812 /**
3813  * ib_req_notify_cq - Request completion notification on a CQ.
3814  * @cq: The CQ to generate an event for.
3815  * @flags:
3816  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3817  *   to request an event on the next solicited event or next work
3818  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3819  *   may also be |ed in to request a hint about missed events, as
3820  *   described below.
3821  *
3822  * Return Value:
3823  *    < 0 means an error occurred while requesting notification
3824  *   == 0 means notification was requested successfully, and if
3825  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3826  *        were missed and it is safe to wait for another event.  In
3827  *        this case is it guaranteed that any work completions added
3828  *        to the CQ since the last CQ poll will trigger a completion
3829  *        notification event.
3830  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3831  *        in.  It means that the consumer must poll the CQ again to
3832  *        make sure it is empty to avoid missing an event because of a
3833  *        race between requesting notification and an entry being
3834  *        added to the CQ.  This return value means it is possible
3835  *        (but not guaranteed) that a work completion has been added
3836  *        to the CQ since the last poll without triggering a
3837  *        completion notification event.
3838  */
3839 static inline int ib_req_notify_cq(struct ib_cq *cq,
3840 				   enum ib_cq_notify_flags flags)
3841 {
3842 	return cq->device->ops.req_notify_cq(cq, flags);
3843 }
3844 
3845 /**
3846  * ib_req_ncomp_notif - Request completion notification when there are
3847  *   at least the specified number of unreaped completions on the CQ.
3848  * @cq: The CQ to generate an event for.
3849  * @wc_cnt: The number of unreaped completions that should be on the
3850  *   CQ before an event is generated.
3851  */
3852 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3853 {
3854 	return cq->device->ops.req_ncomp_notif ?
3855 		cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3856 		-ENOSYS;
3857 }
3858 
3859 /**
3860  * ib_dma_mapping_error - check a DMA addr for error
3861  * @dev: The device for which the dma_addr was created
3862  * @dma_addr: The DMA address to check
3863  */
3864 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3865 {
3866 	return dma_mapping_error(dev->dma_device, dma_addr);
3867 }
3868 
3869 /**
3870  * ib_dma_map_single - Map a kernel virtual address to DMA address
3871  * @dev: The device for which the dma_addr is to be created
3872  * @cpu_addr: The kernel virtual address
3873  * @size: The size of the region in bytes
3874  * @direction: The direction of the DMA
3875  */
3876 static inline u64 ib_dma_map_single(struct ib_device *dev,
3877 				    void *cpu_addr, size_t size,
3878 				    enum dma_data_direction direction)
3879 {
3880 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3881 }
3882 
3883 /**
3884  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3885  * @dev: The device for which the DMA address was created
3886  * @addr: The DMA address
3887  * @size: The size of the region in bytes
3888  * @direction: The direction of the DMA
3889  */
3890 static inline void ib_dma_unmap_single(struct ib_device *dev,
3891 				       u64 addr, size_t size,
3892 				       enum dma_data_direction direction)
3893 {
3894 	dma_unmap_single(dev->dma_device, addr, size, direction);
3895 }
3896 
3897 /**
3898  * ib_dma_map_page - Map a physical page to DMA address
3899  * @dev: The device for which the dma_addr is to be created
3900  * @page: The page to be mapped
3901  * @offset: The offset within the page
3902  * @size: The size of the region in bytes
3903  * @direction: The direction of the DMA
3904  */
3905 static inline u64 ib_dma_map_page(struct ib_device *dev,
3906 				  struct page *page,
3907 				  unsigned long offset,
3908 				  size_t size,
3909 					 enum dma_data_direction direction)
3910 {
3911 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3912 }
3913 
3914 /**
3915  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3916  * @dev: The device for which the DMA address was created
3917  * @addr: The DMA address
3918  * @size: The size of the region in bytes
3919  * @direction: The direction of the DMA
3920  */
3921 static inline void ib_dma_unmap_page(struct ib_device *dev,
3922 				     u64 addr, size_t size,
3923 				     enum dma_data_direction direction)
3924 {
3925 	dma_unmap_page(dev->dma_device, addr, size, direction);
3926 }
3927 
3928 /**
3929  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3930  * @dev: The device for which the DMA addresses are to be created
3931  * @sg: The array of scatter/gather entries
3932  * @nents: The number of scatter/gather entries
3933  * @direction: The direction of the DMA
3934  */
3935 static inline int ib_dma_map_sg(struct ib_device *dev,
3936 				struct scatterlist *sg, int nents,
3937 				enum dma_data_direction direction)
3938 {
3939 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3940 }
3941 
3942 /**
3943  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3944  * @dev: The device for which the DMA addresses were created
3945  * @sg: The array of scatter/gather entries
3946  * @nents: The number of scatter/gather entries
3947  * @direction: The direction of the DMA
3948  */
3949 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3950 				   struct scatterlist *sg, int nents,
3951 				   enum dma_data_direction direction)
3952 {
3953 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
3954 }
3955 
3956 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3957 				      struct scatterlist *sg, int nents,
3958 				      enum dma_data_direction direction,
3959 				      unsigned long dma_attrs)
3960 {
3961 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3962 				dma_attrs);
3963 }
3964 
3965 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3966 					 struct scatterlist *sg, int nents,
3967 					 enum dma_data_direction direction,
3968 					 unsigned long dma_attrs)
3969 {
3970 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3971 }
3972 
3973 /**
3974  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
3975  * @dev: The device to query
3976  *
3977  * The returned value represents a size in bytes.
3978  */
3979 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
3980 {
3981 	struct device_dma_parameters *p = dev->dma_device->dma_parms;
3982 
3983 	return p ? p->max_segment_size : UINT_MAX;
3984 }
3985 
3986 /**
3987  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3988  * @dev: The device for which the DMA address was created
3989  * @addr: The DMA address
3990  * @size: The size of the region in bytes
3991  * @dir: The direction of the DMA
3992  */
3993 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3994 					      u64 addr,
3995 					      size_t size,
3996 					      enum dma_data_direction dir)
3997 {
3998 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3999 }
4000 
4001 /**
4002  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4003  * @dev: The device for which the DMA address was created
4004  * @addr: The DMA address
4005  * @size: The size of the region in bytes
4006  * @dir: The direction of the DMA
4007  */
4008 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4009 						 u64 addr,
4010 						 size_t size,
4011 						 enum dma_data_direction dir)
4012 {
4013 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4014 }
4015 
4016 /**
4017  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4018  * @dev: The device for which the DMA address is requested
4019  * @size: The size of the region to allocate in bytes
4020  * @dma_handle: A pointer for returning the DMA address of the region
4021  * @flag: memory allocator flags
4022  */
4023 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4024 					   size_t size,
4025 					   dma_addr_t *dma_handle,
4026 					   gfp_t flag)
4027 {
4028 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4029 }
4030 
4031 /**
4032  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4033  * @dev: The device for which the DMA addresses were allocated
4034  * @size: The size of the region
4035  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4036  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4037  */
4038 static inline void ib_dma_free_coherent(struct ib_device *dev,
4039 					size_t size, void *cpu_addr,
4040 					dma_addr_t dma_handle)
4041 {
4042 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4043 }
4044 
4045 /**
4046  * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4047  *   HCA translation table.
4048  * @mr: The memory region to deregister.
4049  * @udata: Valid user data or NULL for kernel object
4050  *
4051  * This function can fail, if the memory region has memory windows bound to it.
4052  */
4053 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4054 
4055 /**
4056  * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4057  *   HCA translation table.
4058  * @mr: The memory region to deregister.
4059  *
4060  * This function can fail, if the memory region has memory windows bound to it.
4061  *
4062  * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4063  */
4064 static inline int ib_dereg_mr(struct ib_mr *mr)
4065 {
4066 	return ib_dereg_mr_user(mr, NULL);
4067 }
4068 
4069 struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type,
4070 			       u32 max_num_sg, struct ib_udata *udata);
4071 
4072 static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
4073 					enum ib_mr_type mr_type, u32 max_num_sg)
4074 {
4075 	return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL);
4076 }
4077 
4078 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4079 				    u32 max_num_data_sg,
4080 				    u32 max_num_meta_sg);
4081 
4082 /**
4083  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4084  *   R_Key and L_Key.
4085  * @mr - struct ib_mr pointer to be updated.
4086  * @newkey - new key to be used.
4087  */
4088 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4089 {
4090 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4091 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4092 }
4093 
4094 /**
4095  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4096  * for calculating a new rkey for type 2 memory windows.
4097  * @rkey - the rkey to increment.
4098  */
4099 static inline u32 ib_inc_rkey(u32 rkey)
4100 {
4101 	const u32 mask = 0x000000ff;
4102 	return ((rkey + 1) & mask) | (rkey & ~mask);
4103 }
4104 
4105 /**
4106  * ib_alloc_fmr - Allocates a unmapped fast memory region.
4107  * @pd: The protection domain associated with the unmapped region.
4108  * @mr_access_flags: Specifies the memory access rights.
4109  * @fmr_attr: Attributes of the unmapped region.
4110  *
4111  * A fast memory region must be mapped before it can be used as part of
4112  * a work request.
4113  */
4114 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
4115 			    int mr_access_flags,
4116 			    struct ib_fmr_attr *fmr_attr);
4117 
4118 /**
4119  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
4120  * @fmr: The fast memory region to associate with the pages.
4121  * @page_list: An array of physical pages to map to the fast memory region.
4122  * @list_len: The number of pages in page_list.
4123  * @iova: The I/O virtual address to use with the mapped region.
4124  */
4125 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
4126 				  u64 *page_list, int list_len,
4127 				  u64 iova)
4128 {
4129 	return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
4130 }
4131 
4132 /**
4133  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
4134  * @fmr_list: A linked list of fast memory regions to unmap.
4135  */
4136 int ib_unmap_fmr(struct list_head *fmr_list);
4137 
4138 /**
4139  * ib_dealloc_fmr - Deallocates a fast memory region.
4140  * @fmr: The fast memory region to deallocate.
4141  */
4142 int ib_dealloc_fmr(struct ib_fmr *fmr);
4143 
4144 /**
4145  * ib_attach_mcast - Attaches the specified QP to a multicast group.
4146  * @qp: QP to attach to the multicast group.  The QP must be type
4147  *   IB_QPT_UD.
4148  * @gid: Multicast group GID.
4149  * @lid: Multicast group LID in host byte order.
4150  *
4151  * In order to send and receive multicast packets, subnet
4152  * administration must have created the multicast group and configured
4153  * the fabric appropriately.  The port associated with the specified
4154  * QP must also be a member of the multicast group.
4155  */
4156 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4157 
4158 /**
4159  * ib_detach_mcast - Detaches the specified QP from a multicast group.
4160  * @qp: QP to detach from the multicast group.
4161  * @gid: Multicast group GID.
4162  * @lid: Multicast group LID in host byte order.
4163  */
4164 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4165 
4166 /**
4167  * ib_alloc_xrcd - Allocates an XRC domain.
4168  * @device: The device on which to allocate the XRC domain.
4169  * @caller: Module name for kernel consumers
4170  */
4171 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
4172 #define ib_alloc_xrcd(device) \
4173 	__ib_alloc_xrcd((device), KBUILD_MODNAME)
4174 
4175 /**
4176  * ib_dealloc_xrcd - Deallocates an XRC domain.
4177  * @xrcd: The XRC domain to deallocate.
4178  * @udata: Valid user data or NULL for kernel object
4179  */
4180 int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
4181 
4182 static inline int ib_check_mr_access(int flags)
4183 {
4184 	/*
4185 	 * Local write permission is required if remote write or
4186 	 * remote atomic permission is also requested.
4187 	 */
4188 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4189 	    !(flags & IB_ACCESS_LOCAL_WRITE))
4190 		return -EINVAL;
4191 
4192 	return 0;
4193 }
4194 
4195 static inline bool ib_access_writable(int access_flags)
4196 {
4197 	/*
4198 	 * We have writable memory backing the MR if any of the following
4199 	 * access flags are set.  "Local write" and "remote write" obviously
4200 	 * require write access.  "Remote atomic" can do things like fetch and
4201 	 * add, which will modify memory, and "MW bind" can change permissions
4202 	 * by binding a window.
4203 	 */
4204 	return access_flags &
4205 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
4206 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4207 }
4208 
4209 /**
4210  * ib_check_mr_status: lightweight check of MR status.
4211  *     This routine may provide status checks on a selected
4212  *     ib_mr. first use is for signature status check.
4213  *
4214  * @mr: A memory region.
4215  * @check_mask: Bitmask of which checks to perform from
4216  *     ib_mr_status_check enumeration.
4217  * @mr_status: The container of relevant status checks.
4218  *     failed checks will be indicated in the status bitmask
4219  *     and the relevant info shall be in the error item.
4220  */
4221 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4222 		       struct ib_mr_status *mr_status);
4223 
4224 /**
4225  * ib_device_try_get: Hold a registration lock
4226  * device: The device to lock
4227  *
4228  * A device under an active registration lock cannot become unregistered. It
4229  * is only possible to obtain a registration lock on a device that is fully
4230  * registered, otherwise this function returns false.
4231  *
4232  * The registration lock is only necessary for actions which require the
4233  * device to still be registered. Uses that only require the device pointer to
4234  * be valid should use get_device(&ibdev->dev) to hold the memory.
4235  *
4236  */
4237 static inline bool ib_device_try_get(struct ib_device *dev)
4238 {
4239 	return refcount_inc_not_zero(&dev->refcount);
4240 }
4241 
4242 void ib_device_put(struct ib_device *device);
4243 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4244 					  enum rdma_driver_id driver_id);
4245 struct ib_device *ib_device_get_by_name(const char *name,
4246 					enum rdma_driver_id driver_id);
4247 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4248 					    u16 pkey, const union ib_gid *gid,
4249 					    const struct sockaddr *addr);
4250 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4251 			 unsigned int port);
4252 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4253 
4254 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4255 			   struct ib_wq_init_attr *init_attr);
4256 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
4257 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4258 		 u32 wq_attr_mask);
4259 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4260 						 struct ib_rwq_ind_table_init_attr*
4261 						 wq_ind_table_init_attr);
4262 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4263 
4264 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4265 		 unsigned int *sg_offset, unsigned int page_size);
4266 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4267 		    int data_sg_nents, unsigned int *data_sg_offset,
4268 		    struct scatterlist *meta_sg, int meta_sg_nents,
4269 		    unsigned int *meta_sg_offset, unsigned int page_size);
4270 
4271 static inline int
4272 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4273 		  unsigned int *sg_offset, unsigned int page_size)
4274 {
4275 	int n;
4276 
4277 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4278 	mr->iova = 0;
4279 
4280 	return n;
4281 }
4282 
4283 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4284 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4285 
4286 void ib_drain_rq(struct ib_qp *qp);
4287 void ib_drain_sq(struct ib_qp *qp);
4288 void ib_drain_qp(struct ib_qp *qp);
4289 
4290 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4291 
4292 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4293 {
4294 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4295 		return attr->roce.dmac;
4296 	return NULL;
4297 }
4298 
4299 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4300 {
4301 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4302 		attr->ib.dlid = (u16)dlid;
4303 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4304 		attr->opa.dlid = dlid;
4305 }
4306 
4307 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4308 {
4309 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4310 		return attr->ib.dlid;
4311 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4312 		return attr->opa.dlid;
4313 	return 0;
4314 }
4315 
4316 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4317 {
4318 	attr->sl = sl;
4319 }
4320 
4321 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4322 {
4323 	return attr->sl;
4324 }
4325 
4326 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4327 					 u8 src_path_bits)
4328 {
4329 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4330 		attr->ib.src_path_bits = src_path_bits;
4331 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4332 		attr->opa.src_path_bits = src_path_bits;
4333 }
4334 
4335 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4336 {
4337 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4338 		return attr->ib.src_path_bits;
4339 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4340 		return attr->opa.src_path_bits;
4341 	return 0;
4342 }
4343 
4344 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4345 					bool make_grd)
4346 {
4347 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4348 		attr->opa.make_grd = make_grd;
4349 }
4350 
4351 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4352 {
4353 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4354 		return attr->opa.make_grd;
4355 	return false;
4356 }
4357 
4358 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4359 {
4360 	attr->port_num = port_num;
4361 }
4362 
4363 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4364 {
4365 	return attr->port_num;
4366 }
4367 
4368 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4369 					   u8 static_rate)
4370 {
4371 	attr->static_rate = static_rate;
4372 }
4373 
4374 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4375 {
4376 	return attr->static_rate;
4377 }
4378 
4379 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4380 					enum ib_ah_flags flag)
4381 {
4382 	attr->ah_flags = flag;
4383 }
4384 
4385 static inline enum ib_ah_flags
4386 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4387 {
4388 	return attr->ah_flags;
4389 }
4390 
4391 static inline const struct ib_global_route
4392 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4393 {
4394 	return &attr->grh;
4395 }
4396 
4397 /*To retrieve and modify the grh */
4398 static inline struct ib_global_route
4399 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4400 {
4401 	return &attr->grh;
4402 }
4403 
4404 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4405 {
4406 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4407 
4408 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4409 }
4410 
4411 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4412 					     __be64 prefix)
4413 {
4414 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4415 
4416 	grh->dgid.global.subnet_prefix = prefix;
4417 }
4418 
4419 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4420 					    __be64 if_id)
4421 {
4422 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4423 
4424 	grh->dgid.global.interface_id = if_id;
4425 }
4426 
4427 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4428 				   union ib_gid *dgid, u32 flow_label,
4429 				   u8 sgid_index, u8 hop_limit,
4430 				   u8 traffic_class)
4431 {
4432 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4433 
4434 	attr->ah_flags = IB_AH_GRH;
4435 	if (dgid)
4436 		grh->dgid = *dgid;
4437 	grh->flow_label = flow_label;
4438 	grh->sgid_index = sgid_index;
4439 	grh->hop_limit = hop_limit;
4440 	grh->traffic_class = traffic_class;
4441 	grh->sgid_attr = NULL;
4442 }
4443 
4444 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4445 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4446 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4447 			     const struct ib_gid_attr *sgid_attr);
4448 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4449 		       const struct rdma_ah_attr *src);
4450 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4451 			  const struct rdma_ah_attr *new);
4452 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4453 
4454 /**
4455  * rdma_ah_find_type - Return address handle type.
4456  *
4457  * @dev: Device to be checked
4458  * @port_num: Port number
4459  */
4460 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4461 						       u8 port_num)
4462 {
4463 	if (rdma_protocol_roce(dev, port_num))
4464 		return RDMA_AH_ATTR_TYPE_ROCE;
4465 	if (rdma_protocol_ib(dev, port_num)) {
4466 		if (rdma_cap_opa_ah(dev, port_num))
4467 			return RDMA_AH_ATTR_TYPE_OPA;
4468 		return RDMA_AH_ATTR_TYPE_IB;
4469 	}
4470 
4471 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4472 }
4473 
4474 /**
4475  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4476  *     In the current implementation the only way to get
4477  *     get the 32bit lid is from other sources for OPA.
4478  *     For IB, lids will always be 16bits so cast the
4479  *     value accordingly.
4480  *
4481  * @lid: A 32bit LID
4482  */
4483 static inline u16 ib_lid_cpu16(u32 lid)
4484 {
4485 	WARN_ON_ONCE(lid & 0xFFFF0000);
4486 	return (u16)lid;
4487 }
4488 
4489 /**
4490  * ib_lid_be16 - Return lid in 16bit BE encoding.
4491  *
4492  * @lid: A 32bit LID
4493  */
4494 static inline __be16 ib_lid_be16(u32 lid)
4495 {
4496 	WARN_ON_ONCE(lid & 0xFFFF0000);
4497 	return cpu_to_be16((u16)lid);
4498 }
4499 
4500 /**
4501  * ib_get_vector_affinity - Get the affinity mappings of a given completion
4502  *   vector
4503  * @device:         the rdma device
4504  * @comp_vector:    index of completion vector
4505  *
4506  * Returns NULL on failure, otherwise a corresponding cpu map of the
4507  * completion vector (returns all-cpus map if the device driver doesn't
4508  * implement get_vector_affinity).
4509  */
4510 static inline const struct cpumask *
4511 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4512 {
4513 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4514 	    !device->ops.get_vector_affinity)
4515 		return NULL;
4516 
4517 	return device->ops.get_vector_affinity(device, comp_vector);
4518 
4519 }
4520 
4521 /**
4522  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4523  * and add their gids, as needed, to the relevant RoCE devices.
4524  *
4525  * @device:         the rdma device
4526  */
4527 void rdma_roce_rescan_device(struct ib_device *ibdev);
4528 
4529 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4530 
4531 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4532 
4533 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4534 				     enum rdma_netdev_t type, const char *name,
4535 				     unsigned char name_assign_type,
4536 				     void (*setup)(struct net_device *));
4537 
4538 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4539 		     enum rdma_netdev_t type, const char *name,
4540 		     unsigned char name_assign_type,
4541 		     void (*setup)(struct net_device *),
4542 		     struct net_device *netdev);
4543 
4544 /**
4545  * rdma_set_device_sysfs_group - Set device attributes group to have
4546  *				 driver specific sysfs entries at
4547  *				 for infiniband class.
4548  *
4549  * @device:	device pointer for which attributes to be created
4550  * @group:	Pointer to group which should be added when device
4551  *		is registered with sysfs.
4552  * rdma_set_device_sysfs_group() allows existing drivers to expose one
4553  * group per device to have sysfs attributes.
4554  *
4555  * NOTE: New drivers should not make use of this API; instead new device
4556  * parameter should be exposed via netlink command. This API and mechanism
4557  * exist only for existing drivers.
4558  */
4559 static inline void
4560 rdma_set_device_sysfs_group(struct ib_device *dev,
4561 			    const struct attribute_group *group)
4562 {
4563 	dev->groups[1] = group;
4564 }
4565 
4566 /**
4567  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4568  *
4569  * @device:	device pointer for which ib_device pointer to retrieve
4570  *
4571  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4572  *
4573  */
4574 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4575 {
4576 	struct ib_core_device *coredev =
4577 		container_of(device, struct ib_core_device, dev);
4578 
4579 	return coredev->owner;
4580 }
4581 
4582 /**
4583  * rdma_device_to_drv_device - Helper macro to reach back to driver's
4584  *			       ib_device holder structure from device pointer.
4585  *
4586  * NOTE: New drivers should not make use of this API; This API is only for
4587  * existing drivers who have exposed sysfs entries using
4588  * rdma_set_device_sysfs_group().
4589  */
4590 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
4591 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4592 
4593 bool rdma_dev_access_netns(const struct ib_device *device,
4594 			   const struct net *net);
4595 #endif /* IB_VERBS_H */
4596