xref: /openbmc/linux/include/rdma/ib_verbs.h (revision e2c75e76)
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/mm.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
54 #include <net/ipv6.h>
55 #include <net/ip.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/netdevice.h>
59 
60 #include <linux/if_link.h>
61 #include <linux/atomic.h>
62 #include <linux/mmu_notifier.h>
63 #include <linux/uaccess.h>
64 #include <linux/cgroup_rdma.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 #include <rdma/restrack.h>
67 
68 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
69 
70 extern struct workqueue_struct *ib_wq;
71 extern struct workqueue_struct *ib_comp_wq;
72 
73 union ib_gid {
74 	u8	raw[16];
75 	struct {
76 		__be64	subnet_prefix;
77 		__be64	interface_id;
78 	} global;
79 };
80 
81 extern union ib_gid zgid;
82 
83 enum ib_gid_type {
84 	/* If link layer is Ethernet, this is RoCE V1 */
85 	IB_GID_TYPE_IB        = 0,
86 	IB_GID_TYPE_ROCE      = 0,
87 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
88 	IB_GID_TYPE_SIZE
89 };
90 
91 #define ROCE_V2_UDP_DPORT      4791
92 struct ib_gid_attr {
93 	enum ib_gid_type	gid_type;
94 	struct net_device	*ndev;
95 };
96 
97 enum rdma_node_type {
98 	/* IB values map to NodeInfo:NodeType. */
99 	RDMA_NODE_IB_CA 	= 1,
100 	RDMA_NODE_IB_SWITCH,
101 	RDMA_NODE_IB_ROUTER,
102 	RDMA_NODE_RNIC,
103 	RDMA_NODE_USNIC,
104 	RDMA_NODE_USNIC_UDP,
105 };
106 
107 enum {
108 	/* set the local administered indication */
109 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
110 };
111 
112 enum rdma_transport_type {
113 	RDMA_TRANSPORT_IB,
114 	RDMA_TRANSPORT_IWARP,
115 	RDMA_TRANSPORT_USNIC,
116 	RDMA_TRANSPORT_USNIC_UDP
117 };
118 
119 enum rdma_protocol_type {
120 	RDMA_PROTOCOL_IB,
121 	RDMA_PROTOCOL_IBOE,
122 	RDMA_PROTOCOL_IWARP,
123 	RDMA_PROTOCOL_USNIC_UDP
124 };
125 
126 __attribute_const__ enum rdma_transport_type
127 rdma_node_get_transport(enum rdma_node_type node_type);
128 
129 enum rdma_network_type {
130 	RDMA_NETWORK_IB,
131 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
132 	RDMA_NETWORK_IPV4,
133 	RDMA_NETWORK_IPV6
134 };
135 
136 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
137 {
138 	if (network_type == RDMA_NETWORK_IPV4 ||
139 	    network_type == RDMA_NETWORK_IPV6)
140 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
141 
142 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
143 	return IB_GID_TYPE_IB;
144 }
145 
146 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
147 							    union ib_gid *gid)
148 {
149 	if (gid_type == IB_GID_TYPE_IB)
150 		return RDMA_NETWORK_IB;
151 
152 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
153 		return RDMA_NETWORK_IPV4;
154 	else
155 		return RDMA_NETWORK_IPV6;
156 }
157 
158 enum rdma_link_layer {
159 	IB_LINK_LAYER_UNSPECIFIED,
160 	IB_LINK_LAYER_INFINIBAND,
161 	IB_LINK_LAYER_ETHERNET,
162 };
163 
164 enum ib_device_cap_flags {
165 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
166 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
167 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
168 	IB_DEVICE_RAW_MULTI			= (1 << 3),
169 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
170 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
171 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
172 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
173 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
174 	/* Not in use, former INIT_TYPE		= (1 << 9),*/
175 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
176 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
177 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
178 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
179 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
180 
181 	/*
182 	 * This device supports a per-device lkey or stag that can be
183 	 * used without performing a memory registration for the local
184 	 * memory.  Note that ULPs should never check this flag, but
185 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
186 	 * which will always contain a usable lkey.
187 	 */
188 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
189 	/* Reserved, old SEND_W_INV		= (1 << 16),*/
190 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
191 	/*
192 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
193 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
194 	 * messages and can verify the validity of checksum for
195 	 * incoming messages.  Setting this flag implies that the
196 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
197 	 */
198 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
199 	IB_DEVICE_UD_TSO			= (1 << 19),
200 	IB_DEVICE_XRC				= (1 << 20),
201 
202 	/*
203 	 * This device supports the IB "base memory management extension",
204 	 * which includes support for fast registrations (IB_WR_REG_MR,
205 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
206 	 * also be set by any iWarp device which must support FRs to comply
207 	 * to the iWarp verbs spec.  iWarp devices also support the
208 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
209 	 * stag.
210 	 */
211 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
212 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
213 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
214 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
215 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
216 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
217 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
218 	/*
219 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
220 	 * support execution of WQEs that involve synchronization
221 	 * of I/O operations with single completion queue managed
222 	 * by hardware.
223 	 */
224 	IB_DEVICE_CROSS_CHANNEL			= (1 << 27),
225 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
226 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
227 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
228 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
229 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
230 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
231 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
232 	IB_DEVICE_RDMA_NETDEV_OPA_VNIC		= (1ULL << 35),
233 	/* The device supports padding incoming writes to cacheline. */
234 	IB_DEVICE_PCI_WRITE_END_PADDING		= (1ULL << 36),
235 };
236 
237 enum ib_signature_prot_cap {
238 	IB_PROT_T10DIF_TYPE_1 = 1,
239 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
240 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
241 };
242 
243 enum ib_signature_guard_cap {
244 	IB_GUARD_T10DIF_CRC	= 1,
245 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
246 };
247 
248 enum ib_atomic_cap {
249 	IB_ATOMIC_NONE,
250 	IB_ATOMIC_HCA,
251 	IB_ATOMIC_GLOB
252 };
253 
254 enum ib_odp_general_cap_bits {
255 	IB_ODP_SUPPORT		= 1 << 0,
256 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
257 };
258 
259 enum ib_odp_transport_cap_bits {
260 	IB_ODP_SUPPORT_SEND	= 1 << 0,
261 	IB_ODP_SUPPORT_RECV	= 1 << 1,
262 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
263 	IB_ODP_SUPPORT_READ	= 1 << 3,
264 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
265 };
266 
267 struct ib_odp_caps {
268 	uint64_t general_caps;
269 	struct {
270 		uint32_t  rc_odp_caps;
271 		uint32_t  uc_odp_caps;
272 		uint32_t  ud_odp_caps;
273 	} per_transport_caps;
274 };
275 
276 struct ib_rss_caps {
277 	/* Corresponding bit will be set if qp type from
278 	 * 'enum ib_qp_type' is supported, e.g.
279 	 * supported_qpts |= 1 << IB_QPT_UD
280 	 */
281 	u32 supported_qpts;
282 	u32 max_rwq_indirection_tables;
283 	u32 max_rwq_indirection_table_size;
284 };
285 
286 enum ib_tm_cap_flags {
287 	/*  Support tag matching on RC transport */
288 	IB_TM_CAP_RC		    = 1 << 0,
289 };
290 
291 struct ib_tm_caps {
292 	/* Max size of RNDV header */
293 	u32 max_rndv_hdr_size;
294 	/* Max number of entries in tag matching list */
295 	u32 max_num_tags;
296 	/* From enum ib_tm_cap_flags */
297 	u32 flags;
298 	/* Max number of outstanding list operations */
299 	u32 max_ops;
300 	/* Max number of SGE in tag matching entry */
301 	u32 max_sge;
302 };
303 
304 struct ib_cq_init_attr {
305 	unsigned int	cqe;
306 	int		comp_vector;
307 	u32		flags;
308 };
309 
310 enum ib_cq_attr_mask {
311 	IB_CQ_MODERATE = 1 << 0,
312 };
313 
314 struct ib_cq_caps {
315 	u16     max_cq_moderation_count;
316 	u16     max_cq_moderation_period;
317 };
318 
319 struct ib_device_attr {
320 	u64			fw_ver;
321 	__be64			sys_image_guid;
322 	u64			max_mr_size;
323 	u64			page_size_cap;
324 	u32			vendor_id;
325 	u32			vendor_part_id;
326 	u32			hw_ver;
327 	int			max_qp;
328 	int			max_qp_wr;
329 	u64			device_cap_flags;
330 	int			max_sge;
331 	int			max_sge_rd;
332 	int			max_cq;
333 	int			max_cqe;
334 	int			max_mr;
335 	int			max_pd;
336 	int			max_qp_rd_atom;
337 	int			max_ee_rd_atom;
338 	int			max_res_rd_atom;
339 	int			max_qp_init_rd_atom;
340 	int			max_ee_init_rd_atom;
341 	enum ib_atomic_cap	atomic_cap;
342 	enum ib_atomic_cap	masked_atomic_cap;
343 	int			max_ee;
344 	int			max_rdd;
345 	int			max_mw;
346 	int			max_raw_ipv6_qp;
347 	int			max_raw_ethy_qp;
348 	int			max_mcast_grp;
349 	int			max_mcast_qp_attach;
350 	int			max_total_mcast_qp_attach;
351 	int			max_ah;
352 	int			max_fmr;
353 	int			max_map_per_fmr;
354 	int			max_srq;
355 	int			max_srq_wr;
356 	int			max_srq_sge;
357 	unsigned int		max_fast_reg_page_list_len;
358 	u16			max_pkeys;
359 	u8			local_ca_ack_delay;
360 	int			sig_prot_cap;
361 	int			sig_guard_cap;
362 	struct ib_odp_caps	odp_caps;
363 	uint64_t		timestamp_mask;
364 	uint64_t		hca_core_clock; /* in KHZ */
365 	struct ib_rss_caps	rss_caps;
366 	u32			max_wq_type_rq;
367 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
368 	struct ib_tm_caps	tm_caps;
369 	struct ib_cq_caps       cq_caps;
370 };
371 
372 enum ib_mtu {
373 	IB_MTU_256  = 1,
374 	IB_MTU_512  = 2,
375 	IB_MTU_1024 = 3,
376 	IB_MTU_2048 = 4,
377 	IB_MTU_4096 = 5
378 };
379 
380 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
381 {
382 	switch (mtu) {
383 	case IB_MTU_256:  return  256;
384 	case IB_MTU_512:  return  512;
385 	case IB_MTU_1024: return 1024;
386 	case IB_MTU_2048: return 2048;
387 	case IB_MTU_4096: return 4096;
388 	default: 	  return -1;
389 	}
390 }
391 
392 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
393 {
394 	if (mtu >= 4096)
395 		return IB_MTU_4096;
396 	else if (mtu >= 2048)
397 		return IB_MTU_2048;
398 	else if (mtu >= 1024)
399 		return IB_MTU_1024;
400 	else if (mtu >= 512)
401 		return IB_MTU_512;
402 	else
403 		return IB_MTU_256;
404 }
405 
406 enum ib_port_state {
407 	IB_PORT_NOP		= 0,
408 	IB_PORT_DOWN		= 1,
409 	IB_PORT_INIT		= 2,
410 	IB_PORT_ARMED		= 3,
411 	IB_PORT_ACTIVE		= 4,
412 	IB_PORT_ACTIVE_DEFER	= 5
413 };
414 
415 enum ib_port_cap_flags {
416 	IB_PORT_SM				= 1 <<  1,
417 	IB_PORT_NOTICE_SUP			= 1 <<  2,
418 	IB_PORT_TRAP_SUP			= 1 <<  3,
419 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
420 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
421 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
422 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
423 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
424 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
425 	IB_PORT_SM_DISABLED			= 1 << 10,
426 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
427 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
428 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
429 	IB_PORT_CM_SUP				= 1 << 16,
430 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
431 	IB_PORT_REINIT_SUP			= 1 << 18,
432 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
433 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
434 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
435 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
436 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
437 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
438 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
439 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
440 };
441 
442 enum ib_port_width {
443 	IB_WIDTH_1X	= 1,
444 	IB_WIDTH_4X	= 2,
445 	IB_WIDTH_8X	= 4,
446 	IB_WIDTH_12X	= 8
447 };
448 
449 static inline int ib_width_enum_to_int(enum ib_port_width width)
450 {
451 	switch (width) {
452 	case IB_WIDTH_1X:  return  1;
453 	case IB_WIDTH_4X:  return  4;
454 	case IB_WIDTH_8X:  return  8;
455 	case IB_WIDTH_12X: return 12;
456 	default: 	  return -1;
457 	}
458 }
459 
460 enum ib_port_speed {
461 	IB_SPEED_SDR	= 1,
462 	IB_SPEED_DDR	= 2,
463 	IB_SPEED_QDR	= 4,
464 	IB_SPEED_FDR10	= 8,
465 	IB_SPEED_FDR	= 16,
466 	IB_SPEED_EDR	= 32,
467 	IB_SPEED_HDR	= 64
468 };
469 
470 /**
471  * struct rdma_hw_stats
472  * @timestamp - Used by the core code to track when the last update was
473  * @lifespan - Used by the core code to determine how old the counters
474  *   should be before being updated again.  Stored in jiffies, defaults
475  *   to 10 milliseconds, drivers can override the default be specifying
476  *   their own value during their allocation routine.
477  * @name - Array of pointers to static names used for the counters in
478  *   directory.
479  * @num_counters - How many hardware counters there are.  If name is
480  *   shorter than this number, a kernel oops will result.  Driver authors
481  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
482  *   in their code to prevent this.
483  * @value - Array of u64 counters that are accessed by the sysfs code and
484  *   filled in by the drivers get_stats routine
485  */
486 struct rdma_hw_stats {
487 	unsigned long	timestamp;
488 	unsigned long	lifespan;
489 	const char * const *names;
490 	int		num_counters;
491 	u64		value[];
492 };
493 
494 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
495 /**
496  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
497  *   for drivers.
498  * @names - Array of static const char *
499  * @num_counters - How many elements in array
500  * @lifespan - How many milliseconds between updates
501  */
502 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
503 		const char * const *names, int num_counters,
504 		unsigned long lifespan)
505 {
506 	struct rdma_hw_stats *stats;
507 
508 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
509 			GFP_KERNEL);
510 	if (!stats)
511 		return NULL;
512 	stats->names = names;
513 	stats->num_counters = num_counters;
514 	stats->lifespan = msecs_to_jiffies(lifespan);
515 
516 	return stats;
517 }
518 
519 
520 /* Define bits for the various functionality this port needs to be supported by
521  * the core.
522  */
523 /* Management                           0x00000FFF */
524 #define RDMA_CORE_CAP_IB_MAD            0x00000001
525 #define RDMA_CORE_CAP_IB_SMI            0x00000002
526 #define RDMA_CORE_CAP_IB_CM             0x00000004
527 #define RDMA_CORE_CAP_IW_CM             0x00000008
528 #define RDMA_CORE_CAP_IB_SA             0x00000010
529 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
530 
531 /* Address format                       0x000FF000 */
532 #define RDMA_CORE_CAP_AF_IB             0x00001000
533 #define RDMA_CORE_CAP_ETH_AH            0x00002000
534 #define RDMA_CORE_CAP_OPA_AH            0x00004000
535 
536 /* Protocol                             0xFFF00000 */
537 #define RDMA_CORE_CAP_PROT_IB           0x00100000
538 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
539 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
540 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
541 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
542 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
543 
544 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
545 					| RDMA_CORE_CAP_IB_MAD \
546 					| RDMA_CORE_CAP_IB_SMI \
547 					| RDMA_CORE_CAP_IB_CM  \
548 					| RDMA_CORE_CAP_IB_SA  \
549 					| RDMA_CORE_CAP_AF_IB)
550 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
551 					| RDMA_CORE_CAP_IB_MAD  \
552 					| RDMA_CORE_CAP_IB_CM   \
553 					| RDMA_CORE_CAP_AF_IB   \
554 					| RDMA_CORE_CAP_ETH_AH)
555 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
556 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
557 					| RDMA_CORE_CAP_IB_MAD  \
558 					| RDMA_CORE_CAP_IB_CM   \
559 					| RDMA_CORE_CAP_AF_IB   \
560 					| RDMA_CORE_CAP_ETH_AH)
561 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
562 					| RDMA_CORE_CAP_IW_CM)
563 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
564 					| RDMA_CORE_CAP_OPA_MAD)
565 
566 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
567 
568 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
569 
570 struct ib_port_attr {
571 	u64			subnet_prefix;
572 	enum ib_port_state	state;
573 	enum ib_mtu		max_mtu;
574 	enum ib_mtu		active_mtu;
575 	int			gid_tbl_len;
576 	u32			port_cap_flags;
577 	u32			max_msg_sz;
578 	u32			bad_pkey_cntr;
579 	u32			qkey_viol_cntr;
580 	u16			pkey_tbl_len;
581 	u32			sm_lid;
582 	u32			lid;
583 	u8			lmc;
584 	u8			max_vl_num;
585 	u8			sm_sl;
586 	u8			subnet_timeout;
587 	u8			init_type_reply;
588 	u8			active_width;
589 	u8			active_speed;
590 	u8                      phys_state;
591 	bool			grh_required;
592 };
593 
594 enum ib_device_modify_flags {
595 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
596 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
597 };
598 
599 #define IB_DEVICE_NODE_DESC_MAX 64
600 
601 struct ib_device_modify {
602 	u64	sys_image_guid;
603 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
604 };
605 
606 enum ib_port_modify_flags {
607 	IB_PORT_SHUTDOWN		= 1,
608 	IB_PORT_INIT_TYPE		= (1<<2),
609 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
610 	IB_PORT_OPA_MASK_CHG		= (1<<4)
611 };
612 
613 struct ib_port_modify {
614 	u32	set_port_cap_mask;
615 	u32	clr_port_cap_mask;
616 	u8	init_type;
617 };
618 
619 enum ib_event_type {
620 	IB_EVENT_CQ_ERR,
621 	IB_EVENT_QP_FATAL,
622 	IB_EVENT_QP_REQ_ERR,
623 	IB_EVENT_QP_ACCESS_ERR,
624 	IB_EVENT_COMM_EST,
625 	IB_EVENT_SQ_DRAINED,
626 	IB_EVENT_PATH_MIG,
627 	IB_EVENT_PATH_MIG_ERR,
628 	IB_EVENT_DEVICE_FATAL,
629 	IB_EVENT_PORT_ACTIVE,
630 	IB_EVENT_PORT_ERR,
631 	IB_EVENT_LID_CHANGE,
632 	IB_EVENT_PKEY_CHANGE,
633 	IB_EVENT_SM_CHANGE,
634 	IB_EVENT_SRQ_ERR,
635 	IB_EVENT_SRQ_LIMIT_REACHED,
636 	IB_EVENT_QP_LAST_WQE_REACHED,
637 	IB_EVENT_CLIENT_REREGISTER,
638 	IB_EVENT_GID_CHANGE,
639 	IB_EVENT_WQ_FATAL,
640 };
641 
642 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
643 
644 struct ib_event {
645 	struct ib_device	*device;
646 	union {
647 		struct ib_cq	*cq;
648 		struct ib_qp	*qp;
649 		struct ib_srq	*srq;
650 		struct ib_wq	*wq;
651 		u8		port_num;
652 	} element;
653 	enum ib_event_type	event;
654 };
655 
656 struct ib_event_handler {
657 	struct ib_device *device;
658 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
659 	struct list_head  list;
660 };
661 
662 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
663 	do {							\
664 		(_ptr)->device  = _device;			\
665 		(_ptr)->handler = _handler;			\
666 		INIT_LIST_HEAD(&(_ptr)->list);			\
667 	} while (0)
668 
669 struct ib_global_route {
670 	union ib_gid	dgid;
671 	u32		flow_label;
672 	u8		sgid_index;
673 	u8		hop_limit;
674 	u8		traffic_class;
675 };
676 
677 struct ib_grh {
678 	__be32		version_tclass_flow;
679 	__be16		paylen;
680 	u8		next_hdr;
681 	u8		hop_limit;
682 	union ib_gid	sgid;
683 	union ib_gid	dgid;
684 };
685 
686 union rdma_network_hdr {
687 	struct ib_grh ibgrh;
688 	struct {
689 		/* The IB spec states that if it's IPv4, the header
690 		 * is located in the last 20 bytes of the header.
691 		 */
692 		u8		reserved[20];
693 		struct iphdr	roce4grh;
694 	};
695 };
696 
697 #define IB_QPN_MASK		0xFFFFFF
698 
699 enum {
700 	IB_MULTICAST_QPN = 0xffffff
701 };
702 
703 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
704 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
705 
706 enum ib_ah_flags {
707 	IB_AH_GRH	= 1
708 };
709 
710 enum ib_rate {
711 	IB_RATE_PORT_CURRENT = 0,
712 	IB_RATE_2_5_GBPS = 2,
713 	IB_RATE_5_GBPS   = 5,
714 	IB_RATE_10_GBPS  = 3,
715 	IB_RATE_20_GBPS  = 6,
716 	IB_RATE_30_GBPS  = 4,
717 	IB_RATE_40_GBPS  = 7,
718 	IB_RATE_60_GBPS  = 8,
719 	IB_RATE_80_GBPS  = 9,
720 	IB_RATE_120_GBPS = 10,
721 	IB_RATE_14_GBPS  = 11,
722 	IB_RATE_56_GBPS  = 12,
723 	IB_RATE_112_GBPS = 13,
724 	IB_RATE_168_GBPS = 14,
725 	IB_RATE_25_GBPS  = 15,
726 	IB_RATE_100_GBPS = 16,
727 	IB_RATE_200_GBPS = 17,
728 	IB_RATE_300_GBPS = 18
729 };
730 
731 /**
732  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
733  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
734  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
735  * @rate: rate to convert.
736  */
737 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
738 
739 /**
740  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
741  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
742  * @rate: rate to convert.
743  */
744 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
745 
746 
747 /**
748  * enum ib_mr_type - memory region type
749  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
750  *                            normal registration
751  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
752  *                            signature operations (data-integrity
753  *                            capable regions)
754  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
755  *                            register any arbitrary sg lists (without
756  *                            the normal mr constraints - see
757  *                            ib_map_mr_sg)
758  */
759 enum ib_mr_type {
760 	IB_MR_TYPE_MEM_REG,
761 	IB_MR_TYPE_SIGNATURE,
762 	IB_MR_TYPE_SG_GAPS,
763 };
764 
765 /**
766  * Signature types
767  * IB_SIG_TYPE_NONE: Unprotected.
768  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
769  */
770 enum ib_signature_type {
771 	IB_SIG_TYPE_NONE,
772 	IB_SIG_TYPE_T10_DIF,
773 };
774 
775 /**
776  * Signature T10-DIF block-guard types
777  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
778  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
779  */
780 enum ib_t10_dif_bg_type {
781 	IB_T10DIF_CRC,
782 	IB_T10DIF_CSUM
783 };
784 
785 /**
786  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
787  *     domain.
788  * @bg_type: T10-DIF block guard type (CRC|CSUM)
789  * @pi_interval: protection information interval.
790  * @bg: seed of guard computation.
791  * @app_tag: application tag of guard block
792  * @ref_tag: initial guard block reference tag.
793  * @ref_remap: Indicate wethear the reftag increments each block
794  * @app_escape: Indicate to skip block check if apptag=0xffff
795  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
796  * @apptag_check_mask: check bitmask of application tag.
797  */
798 struct ib_t10_dif_domain {
799 	enum ib_t10_dif_bg_type bg_type;
800 	u16			pi_interval;
801 	u16			bg;
802 	u16			app_tag;
803 	u32			ref_tag;
804 	bool			ref_remap;
805 	bool			app_escape;
806 	bool			ref_escape;
807 	u16			apptag_check_mask;
808 };
809 
810 /**
811  * struct ib_sig_domain - Parameters for signature domain
812  * @sig_type: specific signauture type
813  * @sig: union of all signature domain attributes that may
814  *     be used to set domain layout.
815  */
816 struct ib_sig_domain {
817 	enum ib_signature_type sig_type;
818 	union {
819 		struct ib_t10_dif_domain dif;
820 	} sig;
821 };
822 
823 /**
824  * struct ib_sig_attrs - Parameters for signature handover operation
825  * @check_mask: bitmask for signature byte check (8 bytes)
826  * @mem: memory domain layout desciptor.
827  * @wire: wire domain layout desciptor.
828  */
829 struct ib_sig_attrs {
830 	u8			check_mask;
831 	struct ib_sig_domain	mem;
832 	struct ib_sig_domain	wire;
833 };
834 
835 enum ib_sig_err_type {
836 	IB_SIG_BAD_GUARD,
837 	IB_SIG_BAD_REFTAG,
838 	IB_SIG_BAD_APPTAG,
839 };
840 
841 /**
842  * struct ib_sig_err - signature error descriptor
843  */
844 struct ib_sig_err {
845 	enum ib_sig_err_type	err_type;
846 	u32			expected;
847 	u32			actual;
848 	u64			sig_err_offset;
849 	u32			key;
850 };
851 
852 enum ib_mr_status_check {
853 	IB_MR_CHECK_SIG_STATUS = 1,
854 };
855 
856 /**
857  * struct ib_mr_status - Memory region status container
858  *
859  * @fail_status: Bitmask of MR checks status. For each
860  *     failed check a corresponding status bit is set.
861  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
862  *     failure.
863  */
864 struct ib_mr_status {
865 	u32		    fail_status;
866 	struct ib_sig_err   sig_err;
867 };
868 
869 /**
870  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
871  * enum.
872  * @mult: multiple to convert.
873  */
874 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
875 
876 enum rdma_ah_attr_type {
877 	RDMA_AH_ATTR_TYPE_UNDEFINED,
878 	RDMA_AH_ATTR_TYPE_IB,
879 	RDMA_AH_ATTR_TYPE_ROCE,
880 	RDMA_AH_ATTR_TYPE_OPA,
881 };
882 
883 struct ib_ah_attr {
884 	u16			dlid;
885 	u8			src_path_bits;
886 };
887 
888 struct roce_ah_attr {
889 	u8			dmac[ETH_ALEN];
890 };
891 
892 struct opa_ah_attr {
893 	u32			dlid;
894 	u8			src_path_bits;
895 	bool			make_grd;
896 };
897 
898 struct rdma_ah_attr {
899 	struct ib_global_route	grh;
900 	u8			sl;
901 	u8			static_rate;
902 	u8			port_num;
903 	u8			ah_flags;
904 	enum rdma_ah_attr_type type;
905 	union {
906 		struct ib_ah_attr ib;
907 		struct roce_ah_attr roce;
908 		struct opa_ah_attr opa;
909 	};
910 };
911 
912 enum ib_wc_status {
913 	IB_WC_SUCCESS,
914 	IB_WC_LOC_LEN_ERR,
915 	IB_WC_LOC_QP_OP_ERR,
916 	IB_WC_LOC_EEC_OP_ERR,
917 	IB_WC_LOC_PROT_ERR,
918 	IB_WC_WR_FLUSH_ERR,
919 	IB_WC_MW_BIND_ERR,
920 	IB_WC_BAD_RESP_ERR,
921 	IB_WC_LOC_ACCESS_ERR,
922 	IB_WC_REM_INV_REQ_ERR,
923 	IB_WC_REM_ACCESS_ERR,
924 	IB_WC_REM_OP_ERR,
925 	IB_WC_RETRY_EXC_ERR,
926 	IB_WC_RNR_RETRY_EXC_ERR,
927 	IB_WC_LOC_RDD_VIOL_ERR,
928 	IB_WC_REM_INV_RD_REQ_ERR,
929 	IB_WC_REM_ABORT_ERR,
930 	IB_WC_INV_EECN_ERR,
931 	IB_WC_INV_EEC_STATE_ERR,
932 	IB_WC_FATAL_ERR,
933 	IB_WC_RESP_TIMEOUT_ERR,
934 	IB_WC_GENERAL_ERR
935 };
936 
937 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
938 
939 enum ib_wc_opcode {
940 	IB_WC_SEND,
941 	IB_WC_RDMA_WRITE,
942 	IB_WC_RDMA_READ,
943 	IB_WC_COMP_SWAP,
944 	IB_WC_FETCH_ADD,
945 	IB_WC_LSO,
946 	IB_WC_LOCAL_INV,
947 	IB_WC_REG_MR,
948 	IB_WC_MASKED_COMP_SWAP,
949 	IB_WC_MASKED_FETCH_ADD,
950 /*
951  * Set value of IB_WC_RECV so consumers can test if a completion is a
952  * receive by testing (opcode & IB_WC_RECV).
953  */
954 	IB_WC_RECV			= 1 << 7,
955 	IB_WC_RECV_RDMA_WITH_IMM
956 };
957 
958 enum ib_wc_flags {
959 	IB_WC_GRH		= 1,
960 	IB_WC_WITH_IMM		= (1<<1),
961 	IB_WC_WITH_INVALIDATE	= (1<<2),
962 	IB_WC_IP_CSUM_OK	= (1<<3),
963 	IB_WC_WITH_SMAC		= (1<<4),
964 	IB_WC_WITH_VLAN		= (1<<5),
965 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
966 };
967 
968 struct ib_wc {
969 	union {
970 		u64		wr_id;
971 		struct ib_cqe	*wr_cqe;
972 	};
973 	enum ib_wc_status	status;
974 	enum ib_wc_opcode	opcode;
975 	u32			vendor_err;
976 	u32			byte_len;
977 	struct ib_qp	       *qp;
978 	union {
979 		__be32		imm_data;
980 		u32		invalidate_rkey;
981 	} ex;
982 	u32			src_qp;
983 	u32			slid;
984 	int			wc_flags;
985 	u16			pkey_index;
986 	u8			sl;
987 	u8			dlid_path_bits;
988 	u8			port_num;	/* valid only for DR SMPs on switches */
989 	u8			smac[ETH_ALEN];
990 	u16			vlan_id;
991 	u8			network_hdr_type;
992 };
993 
994 enum ib_cq_notify_flags {
995 	IB_CQ_SOLICITED			= 1 << 0,
996 	IB_CQ_NEXT_COMP			= 1 << 1,
997 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
998 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
999 };
1000 
1001 enum ib_srq_type {
1002 	IB_SRQT_BASIC,
1003 	IB_SRQT_XRC,
1004 	IB_SRQT_TM,
1005 };
1006 
1007 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1008 {
1009 	return srq_type == IB_SRQT_XRC ||
1010 	       srq_type == IB_SRQT_TM;
1011 }
1012 
1013 enum ib_srq_attr_mask {
1014 	IB_SRQ_MAX_WR	= 1 << 0,
1015 	IB_SRQ_LIMIT	= 1 << 1,
1016 };
1017 
1018 struct ib_srq_attr {
1019 	u32	max_wr;
1020 	u32	max_sge;
1021 	u32	srq_limit;
1022 };
1023 
1024 struct ib_srq_init_attr {
1025 	void		      (*event_handler)(struct ib_event *, void *);
1026 	void		       *srq_context;
1027 	struct ib_srq_attr	attr;
1028 	enum ib_srq_type	srq_type;
1029 
1030 	struct {
1031 		struct ib_cq   *cq;
1032 		union {
1033 			struct {
1034 				struct ib_xrcd *xrcd;
1035 			} xrc;
1036 
1037 			struct {
1038 				u32		max_num_tags;
1039 			} tag_matching;
1040 		};
1041 	} ext;
1042 };
1043 
1044 struct ib_qp_cap {
1045 	u32	max_send_wr;
1046 	u32	max_recv_wr;
1047 	u32	max_send_sge;
1048 	u32	max_recv_sge;
1049 	u32	max_inline_data;
1050 
1051 	/*
1052 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1053 	 * ib_create_qp() will calculate the right amount of neededed WRs
1054 	 * and MRs based on this.
1055 	 */
1056 	u32	max_rdma_ctxs;
1057 };
1058 
1059 enum ib_sig_type {
1060 	IB_SIGNAL_ALL_WR,
1061 	IB_SIGNAL_REQ_WR
1062 };
1063 
1064 enum ib_qp_type {
1065 	/*
1066 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1067 	 * here (and in that order) since the MAD layer uses them as
1068 	 * indices into a 2-entry table.
1069 	 */
1070 	IB_QPT_SMI,
1071 	IB_QPT_GSI,
1072 
1073 	IB_QPT_RC,
1074 	IB_QPT_UC,
1075 	IB_QPT_UD,
1076 	IB_QPT_RAW_IPV6,
1077 	IB_QPT_RAW_ETHERTYPE,
1078 	IB_QPT_RAW_PACKET = 8,
1079 	IB_QPT_XRC_INI = 9,
1080 	IB_QPT_XRC_TGT,
1081 	IB_QPT_MAX,
1082 	IB_QPT_DRIVER = 0xFF,
1083 	/* Reserve a range for qp types internal to the low level driver.
1084 	 * These qp types will not be visible at the IB core layer, so the
1085 	 * IB_QPT_MAX usages should not be affected in the core layer
1086 	 */
1087 	IB_QPT_RESERVED1 = 0x1000,
1088 	IB_QPT_RESERVED2,
1089 	IB_QPT_RESERVED3,
1090 	IB_QPT_RESERVED4,
1091 	IB_QPT_RESERVED5,
1092 	IB_QPT_RESERVED6,
1093 	IB_QPT_RESERVED7,
1094 	IB_QPT_RESERVED8,
1095 	IB_QPT_RESERVED9,
1096 	IB_QPT_RESERVED10,
1097 };
1098 
1099 enum ib_qp_create_flags {
1100 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1101 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1102 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1103 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1104 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1105 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1106 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1107 	/* FREE					= 1 << 7, */
1108 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1109 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1110 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1111 	IB_QP_CREATE_PCI_WRITE_END_PADDING	= 1 << 11,
1112 	/* reserve bits 26-31 for low level drivers' internal use */
1113 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1114 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1115 };
1116 
1117 /*
1118  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1119  * callback to destroy the passed in QP.
1120  */
1121 
1122 struct ib_qp_init_attr {
1123 	void                  (*event_handler)(struct ib_event *, void *);
1124 	void		       *qp_context;
1125 	struct ib_cq	       *send_cq;
1126 	struct ib_cq	       *recv_cq;
1127 	struct ib_srq	       *srq;
1128 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1129 	struct ib_qp_cap	cap;
1130 	enum ib_sig_type	sq_sig_type;
1131 	enum ib_qp_type		qp_type;
1132 	enum ib_qp_create_flags	create_flags;
1133 
1134 	/*
1135 	 * Only needed for special QP types, or when using the RW API.
1136 	 */
1137 	u8			port_num;
1138 	struct ib_rwq_ind_table *rwq_ind_tbl;
1139 	u32			source_qpn;
1140 };
1141 
1142 struct ib_qp_open_attr {
1143 	void                  (*event_handler)(struct ib_event *, void *);
1144 	void		       *qp_context;
1145 	u32			qp_num;
1146 	enum ib_qp_type		qp_type;
1147 };
1148 
1149 enum ib_rnr_timeout {
1150 	IB_RNR_TIMER_655_36 =  0,
1151 	IB_RNR_TIMER_000_01 =  1,
1152 	IB_RNR_TIMER_000_02 =  2,
1153 	IB_RNR_TIMER_000_03 =  3,
1154 	IB_RNR_TIMER_000_04 =  4,
1155 	IB_RNR_TIMER_000_06 =  5,
1156 	IB_RNR_TIMER_000_08 =  6,
1157 	IB_RNR_TIMER_000_12 =  7,
1158 	IB_RNR_TIMER_000_16 =  8,
1159 	IB_RNR_TIMER_000_24 =  9,
1160 	IB_RNR_TIMER_000_32 = 10,
1161 	IB_RNR_TIMER_000_48 = 11,
1162 	IB_RNR_TIMER_000_64 = 12,
1163 	IB_RNR_TIMER_000_96 = 13,
1164 	IB_RNR_TIMER_001_28 = 14,
1165 	IB_RNR_TIMER_001_92 = 15,
1166 	IB_RNR_TIMER_002_56 = 16,
1167 	IB_RNR_TIMER_003_84 = 17,
1168 	IB_RNR_TIMER_005_12 = 18,
1169 	IB_RNR_TIMER_007_68 = 19,
1170 	IB_RNR_TIMER_010_24 = 20,
1171 	IB_RNR_TIMER_015_36 = 21,
1172 	IB_RNR_TIMER_020_48 = 22,
1173 	IB_RNR_TIMER_030_72 = 23,
1174 	IB_RNR_TIMER_040_96 = 24,
1175 	IB_RNR_TIMER_061_44 = 25,
1176 	IB_RNR_TIMER_081_92 = 26,
1177 	IB_RNR_TIMER_122_88 = 27,
1178 	IB_RNR_TIMER_163_84 = 28,
1179 	IB_RNR_TIMER_245_76 = 29,
1180 	IB_RNR_TIMER_327_68 = 30,
1181 	IB_RNR_TIMER_491_52 = 31
1182 };
1183 
1184 enum ib_qp_attr_mask {
1185 	IB_QP_STATE			= 1,
1186 	IB_QP_CUR_STATE			= (1<<1),
1187 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1188 	IB_QP_ACCESS_FLAGS		= (1<<3),
1189 	IB_QP_PKEY_INDEX		= (1<<4),
1190 	IB_QP_PORT			= (1<<5),
1191 	IB_QP_QKEY			= (1<<6),
1192 	IB_QP_AV			= (1<<7),
1193 	IB_QP_PATH_MTU			= (1<<8),
1194 	IB_QP_TIMEOUT			= (1<<9),
1195 	IB_QP_RETRY_CNT			= (1<<10),
1196 	IB_QP_RNR_RETRY			= (1<<11),
1197 	IB_QP_RQ_PSN			= (1<<12),
1198 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1199 	IB_QP_ALT_PATH			= (1<<14),
1200 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1201 	IB_QP_SQ_PSN			= (1<<16),
1202 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1203 	IB_QP_PATH_MIG_STATE		= (1<<18),
1204 	IB_QP_CAP			= (1<<19),
1205 	IB_QP_DEST_QPN			= (1<<20),
1206 	IB_QP_RESERVED1			= (1<<21),
1207 	IB_QP_RESERVED2			= (1<<22),
1208 	IB_QP_RESERVED3			= (1<<23),
1209 	IB_QP_RESERVED4			= (1<<24),
1210 	IB_QP_RATE_LIMIT		= (1<<25),
1211 };
1212 
1213 enum ib_qp_state {
1214 	IB_QPS_RESET,
1215 	IB_QPS_INIT,
1216 	IB_QPS_RTR,
1217 	IB_QPS_RTS,
1218 	IB_QPS_SQD,
1219 	IB_QPS_SQE,
1220 	IB_QPS_ERR
1221 };
1222 
1223 enum ib_mig_state {
1224 	IB_MIG_MIGRATED,
1225 	IB_MIG_REARM,
1226 	IB_MIG_ARMED
1227 };
1228 
1229 enum ib_mw_type {
1230 	IB_MW_TYPE_1 = 1,
1231 	IB_MW_TYPE_2 = 2
1232 };
1233 
1234 struct ib_qp_attr {
1235 	enum ib_qp_state	qp_state;
1236 	enum ib_qp_state	cur_qp_state;
1237 	enum ib_mtu		path_mtu;
1238 	enum ib_mig_state	path_mig_state;
1239 	u32			qkey;
1240 	u32			rq_psn;
1241 	u32			sq_psn;
1242 	u32			dest_qp_num;
1243 	int			qp_access_flags;
1244 	struct ib_qp_cap	cap;
1245 	struct rdma_ah_attr	ah_attr;
1246 	struct rdma_ah_attr	alt_ah_attr;
1247 	u16			pkey_index;
1248 	u16			alt_pkey_index;
1249 	u8			en_sqd_async_notify;
1250 	u8			sq_draining;
1251 	u8			max_rd_atomic;
1252 	u8			max_dest_rd_atomic;
1253 	u8			min_rnr_timer;
1254 	u8			port_num;
1255 	u8			timeout;
1256 	u8			retry_cnt;
1257 	u8			rnr_retry;
1258 	u8			alt_port_num;
1259 	u8			alt_timeout;
1260 	u32			rate_limit;
1261 };
1262 
1263 enum ib_wr_opcode {
1264 	IB_WR_RDMA_WRITE,
1265 	IB_WR_RDMA_WRITE_WITH_IMM,
1266 	IB_WR_SEND,
1267 	IB_WR_SEND_WITH_IMM,
1268 	IB_WR_RDMA_READ,
1269 	IB_WR_ATOMIC_CMP_AND_SWP,
1270 	IB_WR_ATOMIC_FETCH_AND_ADD,
1271 	IB_WR_LSO,
1272 	IB_WR_SEND_WITH_INV,
1273 	IB_WR_RDMA_READ_WITH_INV,
1274 	IB_WR_LOCAL_INV,
1275 	IB_WR_REG_MR,
1276 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1277 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1278 	IB_WR_REG_SIG_MR,
1279 	/* reserve values for low level drivers' internal use.
1280 	 * These values will not be used at all in the ib core layer.
1281 	 */
1282 	IB_WR_RESERVED1 = 0xf0,
1283 	IB_WR_RESERVED2,
1284 	IB_WR_RESERVED3,
1285 	IB_WR_RESERVED4,
1286 	IB_WR_RESERVED5,
1287 	IB_WR_RESERVED6,
1288 	IB_WR_RESERVED7,
1289 	IB_WR_RESERVED8,
1290 	IB_WR_RESERVED9,
1291 	IB_WR_RESERVED10,
1292 };
1293 
1294 enum ib_send_flags {
1295 	IB_SEND_FENCE		= 1,
1296 	IB_SEND_SIGNALED	= (1<<1),
1297 	IB_SEND_SOLICITED	= (1<<2),
1298 	IB_SEND_INLINE		= (1<<3),
1299 	IB_SEND_IP_CSUM		= (1<<4),
1300 
1301 	/* reserve bits 26-31 for low level drivers' internal use */
1302 	IB_SEND_RESERVED_START	= (1 << 26),
1303 	IB_SEND_RESERVED_END	= (1 << 31),
1304 };
1305 
1306 struct ib_sge {
1307 	u64	addr;
1308 	u32	length;
1309 	u32	lkey;
1310 };
1311 
1312 struct ib_cqe {
1313 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1314 };
1315 
1316 struct ib_send_wr {
1317 	struct ib_send_wr      *next;
1318 	union {
1319 		u64		wr_id;
1320 		struct ib_cqe	*wr_cqe;
1321 	};
1322 	struct ib_sge	       *sg_list;
1323 	int			num_sge;
1324 	enum ib_wr_opcode	opcode;
1325 	int			send_flags;
1326 	union {
1327 		__be32		imm_data;
1328 		u32		invalidate_rkey;
1329 	} ex;
1330 };
1331 
1332 struct ib_rdma_wr {
1333 	struct ib_send_wr	wr;
1334 	u64			remote_addr;
1335 	u32			rkey;
1336 };
1337 
1338 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1339 {
1340 	return container_of(wr, struct ib_rdma_wr, wr);
1341 }
1342 
1343 struct ib_atomic_wr {
1344 	struct ib_send_wr	wr;
1345 	u64			remote_addr;
1346 	u64			compare_add;
1347 	u64			swap;
1348 	u64			compare_add_mask;
1349 	u64			swap_mask;
1350 	u32			rkey;
1351 };
1352 
1353 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1354 {
1355 	return container_of(wr, struct ib_atomic_wr, wr);
1356 }
1357 
1358 struct ib_ud_wr {
1359 	struct ib_send_wr	wr;
1360 	struct ib_ah		*ah;
1361 	void			*header;
1362 	int			hlen;
1363 	int			mss;
1364 	u32			remote_qpn;
1365 	u32			remote_qkey;
1366 	u16			pkey_index; /* valid for GSI only */
1367 	u8			port_num;   /* valid for DR SMPs on switch only */
1368 };
1369 
1370 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1371 {
1372 	return container_of(wr, struct ib_ud_wr, wr);
1373 }
1374 
1375 struct ib_reg_wr {
1376 	struct ib_send_wr	wr;
1377 	struct ib_mr		*mr;
1378 	u32			key;
1379 	int			access;
1380 };
1381 
1382 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1383 {
1384 	return container_of(wr, struct ib_reg_wr, wr);
1385 }
1386 
1387 struct ib_sig_handover_wr {
1388 	struct ib_send_wr	wr;
1389 	struct ib_sig_attrs    *sig_attrs;
1390 	struct ib_mr	       *sig_mr;
1391 	int			access_flags;
1392 	struct ib_sge	       *prot;
1393 };
1394 
1395 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1396 {
1397 	return container_of(wr, struct ib_sig_handover_wr, wr);
1398 }
1399 
1400 struct ib_recv_wr {
1401 	struct ib_recv_wr      *next;
1402 	union {
1403 		u64		wr_id;
1404 		struct ib_cqe	*wr_cqe;
1405 	};
1406 	struct ib_sge	       *sg_list;
1407 	int			num_sge;
1408 };
1409 
1410 enum ib_access_flags {
1411 	IB_ACCESS_LOCAL_WRITE	= 1,
1412 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1413 	IB_ACCESS_REMOTE_READ	= (1<<2),
1414 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1415 	IB_ACCESS_MW_BIND	= (1<<4),
1416 	IB_ZERO_BASED		= (1<<5),
1417 	IB_ACCESS_ON_DEMAND     = (1<<6),
1418 	IB_ACCESS_HUGETLB	= (1<<7),
1419 };
1420 
1421 /*
1422  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1423  * are hidden here instead of a uapi header!
1424  */
1425 enum ib_mr_rereg_flags {
1426 	IB_MR_REREG_TRANS	= 1,
1427 	IB_MR_REREG_PD		= (1<<1),
1428 	IB_MR_REREG_ACCESS	= (1<<2),
1429 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1430 };
1431 
1432 struct ib_fmr_attr {
1433 	int	max_pages;
1434 	int	max_maps;
1435 	u8	page_shift;
1436 };
1437 
1438 struct ib_umem;
1439 
1440 enum rdma_remove_reason {
1441 	/* Userspace requested uobject deletion. Call could fail */
1442 	RDMA_REMOVE_DESTROY,
1443 	/* Context deletion. This call should delete the actual object itself */
1444 	RDMA_REMOVE_CLOSE,
1445 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1446 	RDMA_REMOVE_DRIVER_REMOVE,
1447 	/* Context is being cleaned-up, but commit was just completed */
1448 	RDMA_REMOVE_DURING_CLEANUP,
1449 };
1450 
1451 struct ib_rdmacg_object {
1452 #ifdef CONFIG_CGROUP_RDMA
1453 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1454 #endif
1455 };
1456 
1457 struct ib_ucontext {
1458 	struct ib_device       *device;
1459 	struct ib_uverbs_file  *ufile;
1460 	int			closing;
1461 
1462 	/* locking the uobjects_list */
1463 	struct mutex		uobjects_lock;
1464 	struct list_head	uobjects;
1465 	/* protects cleanup process from other actions */
1466 	struct rw_semaphore	cleanup_rwsem;
1467 	enum rdma_remove_reason cleanup_reason;
1468 
1469 	struct pid             *tgid;
1470 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1471 	struct rb_root_cached   umem_tree;
1472 	/*
1473 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1474 	 * mmu notifiers registration.
1475 	 */
1476 	struct rw_semaphore	umem_rwsem;
1477 	void (*invalidate_range)(struct ib_umem *umem,
1478 				 unsigned long start, unsigned long end);
1479 
1480 	struct mmu_notifier	mn;
1481 	atomic_t		notifier_count;
1482 	/* A list of umems that don't have private mmu notifier counters yet. */
1483 	struct list_head	no_private_counters;
1484 	int                     odp_mrs_count;
1485 #endif
1486 
1487 	struct ib_rdmacg_object	cg_obj;
1488 };
1489 
1490 struct ib_uobject {
1491 	u64			user_handle;	/* handle given to us by userspace */
1492 	struct ib_ucontext     *context;	/* associated user context */
1493 	void		       *object;		/* containing object */
1494 	struct list_head	list;		/* link to context's list */
1495 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1496 	int			id;		/* index into kernel idr */
1497 	struct kref		ref;
1498 	atomic_t		usecnt;		/* protects exclusive access */
1499 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1500 
1501 	const struct uverbs_obj_type *type;
1502 };
1503 
1504 struct ib_uobject_file {
1505 	struct ib_uobject	uobj;
1506 	/* ufile contains the lock between context release and file close */
1507 	struct ib_uverbs_file	*ufile;
1508 };
1509 
1510 struct ib_udata {
1511 	const void __user *inbuf;
1512 	void __user *outbuf;
1513 	size_t       inlen;
1514 	size_t       outlen;
1515 };
1516 
1517 struct ib_pd {
1518 	u32			local_dma_lkey;
1519 	u32			flags;
1520 	struct ib_device       *device;
1521 	struct ib_uobject      *uobject;
1522 	atomic_t          	usecnt; /* count all resources */
1523 
1524 	u32			unsafe_global_rkey;
1525 
1526 	/*
1527 	 * Implementation details of the RDMA core, don't use in drivers:
1528 	 */
1529 	struct ib_mr	       *__internal_mr;
1530 	struct rdma_restrack_entry res;
1531 };
1532 
1533 struct ib_xrcd {
1534 	struct ib_device       *device;
1535 	atomic_t		usecnt; /* count all exposed resources */
1536 	struct inode	       *inode;
1537 
1538 	struct mutex		tgt_qp_mutex;
1539 	struct list_head	tgt_qp_list;
1540 	/*
1541 	 * Implementation details of the RDMA core, don't use in drivers:
1542 	 */
1543 	struct rdma_restrack_entry res;
1544 };
1545 
1546 struct ib_ah {
1547 	struct ib_device	*device;
1548 	struct ib_pd		*pd;
1549 	struct ib_uobject	*uobject;
1550 	enum rdma_ah_attr_type	type;
1551 };
1552 
1553 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1554 
1555 enum ib_poll_context {
1556 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1557 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1558 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1559 };
1560 
1561 struct ib_cq {
1562 	struct ib_device       *device;
1563 	struct ib_uobject      *uobject;
1564 	ib_comp_handler   	comp_handler;
1565 	void                  (*event_handler)(struct ib_event *, void *);
1566 	void                   *cq_context;
1567 	int               	cqe;
1568 	atomic_t          	usecnt; /* count number of work queues */
1569 	enum ib_poll_context	poll_ctx;
1570 	struct ib_wc		*wc;
1571 	union {
1572 		struct irq_poll		iop;
1573 		struct work_struct	work;
1574 	};
1575 	/*
1576 	 * Implementation details of the RDMA core, don't use in drivers:
1577 	 */
1578 	struct rdma_restrack_entry res;
1579 };
1580 
1581 struct ib_srq {
1582 	struct ib_device       *device;
1583 	struct ib_pd	       *pd;
1584 	struct ib_uobject      *uobject;
1585 	void		      (*event_handler)(struct ib_event *, void *);
1586 	void		       *srq_context;
1587 	enum ib_srq_type	srq_type;
1588 	atomic_t		usecnt;
1589 
1590 	struct {
1591 		struct ib_cq   *cq;
1592 		union {
1593 			struct {
1594 				struct ib_xrcd *xrcd;
1595 				u32		srq_num;
1596 			} xrc;
1597 		};
1598 	} ext;
1599 };
1600 
1601 enum ib_raw_packet_caps {
1602 	/* Strip cvlan from incoming packet and report it in the matching work
1603 	 * completion is supported.
1604 	 */
1605 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1606 	/* Scatter FCS field of an incoming packet to host memory is supported.
1607 	 */
1608 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1609 	/* Checksum offloads are supported (for both send and receive). */
1610 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1611 	/* When a packet is received for an RQ with no receive WQEs, the
1612 	 * packet processing is delayed.
1613 	 */
1614 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1615 };
1616 
1617 enum ib_wq_type {
1618 	IB_WQT_RQ
1619 };
1620 
1621 enum ib_wq_state {
1622 	IB_WQS_RESET,
1623 	IB_WQS_RDY,
1624 	IB_WQS_ERR
1625 };
1626 
1627 struct ib_wq {
1628 	struct ib_device       *device;
1629 	struct ib_uobject      *uobject;
1630 	void		    *wq_context;
1631 	void		    (*event_handler)(struct ib_event *, void *);
1632 	struct ib_pd	       *pd;
1633 	struct ib_cq	       *cq;
1634 	u32		wq_num;
1635 	enum ib_wq_state       state;
1636 	enum ib_wq_type	wq_type;
1637 	atomic_t		usecnt;
1638 };
1639 
1640 enum ib_wq_flags {
1641 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1642 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1643 	IB_WQ_FLAGS_DELAY_DROP		= 1 << 2,
1644 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1645 };
1646 
1647 struct ib_wq_init_attr {
1648 	void		       *wq_context;
1649 	enum ib_wq_type	wq_type;
1650 	u32		max_wr;
1651 	u32		max_sge;
1652 	struct	ib_cq	       *cq;
1653 	void		    (*event_handler)(struct ib_event *, void *);
1654 	u32		create_flags; /* Use enum ib_wq_flags */
1655 };
1656 
1657 enum ib_wq_attr_mask {
1658 	IB_WQ_STATE		= 1 << 0,
1659 	IB_WQ_CUR_STATE		= 1 << 1,
1660 	IB_WQ_FLAGS		= 1 << 2,
1661 };
1662 
1663 struct ib_wq_attr {
1664 	enum	ib_wq_state	wq_state;
1665 	enum	ib_wq_state	curr_wq_state;
1666 	u32			flags; /* Use enum ib_wq_flags */
1667 	u32			flags_mask; /* Use enum ib_wq_flags */
1668 };
1669 
1670 struct ib_rwq_ind_table {
1671 	struct ib_device	*device;
1672 	struct ib_uobject      *uobject;
1673 	atomic_t		usecnt;
1674 	u32		ind_tbl_num;
1675 	u32		log_ind_tbl_size;
1676 	struct ib_wq	**ind_tbl;
1677 };
1678 
1679 struct ib_rwq_ind_table_init_attr {
1680 	u32		log_ind_tbl_size;
1681 	/* Each entry is a pointer to Receive Work Queue */
1682 	struct ib_wq	**ind_tbl;
1683 };
1684 
1685 enum port_pkey_state {
1686 	IB_PORT_PKEY_NOT_VALID = 0,
1687 	IB_PORT_PKEY_VALID = 1,
1688 	IB_PORT_PKEY_LISTED = 2,
1689 };
1690 
1691 struct ib_qp_security;
1692 
1693 struct ib_port_pkey {
1694 	enum port_pkey_state	state;
1695 	u16			pkey_index;
1696 	u8			port_num;
1697 	struct list_head	qp_list;
1698 	struct list_head	to_error_list;
1699 	struct ib_qp_security  *sec;
1700 };
1701 
1702 struct ib_ports_pkeys {
1703 	struct ib_port_pkey	main;
1704 	struct ib_port_pkey	alt;
1705 };
1706 
1707 struct ib_qp_security {
1708 	struct ib_qp	       *qp;
1709 	struct ib_device       *dev;
1710 	/* Hold this mutex when changing port and pkey settings. */
1711 	struct mutex		mutex;
1712 	struct ib_ports_pkeys  *ports_pkeys;
1713 	/* A list of all open shared QP handles.  Required to enforce security
1714 	 * properly for all users of a shared QP.
1715 	 */
1716 	struct list_head        shared_qp_list;
1717 	void                   *security;
1718 	bool			destroying;
1719 	atomic_t		error_list_count;
1720 	struct completion	error_complete;
1721 	int			error_comps_pending;
1722 };
1723 
1724 /*
1725  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1726  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1727  */
1728 struct ib_qp {
1729 	struct ib_device       *device;
1730 	struct ib_pd	       *pd;
1731 	struct ib_cq	       *send_cq;
1732 	struct ib_cq	       *recv_cq;
1733 	spinlock_t		mr_lock;
1734 	int			mrs_used;
1735 	struct list_head	rdma_mrs;
1736 	struct list_head	sig_mrs;
1737 	struct ib_srq	       *srq;
1738 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1739 	struct list_head	xrcd_list;
1740 
1741 	/* count times opened, mcast attaches, flow attaches */
1742 	atomic_t		usecnt;
1743 	struct list_head	open_list;
1744 	struct ib_qp           *real_qp;
1745 	struct ib_uobject      *uobject;
1746 	void                  (*event_handler)(struct ib_event *, void *);
1747 	void		       *qp_context;
1748 	u32			qp_num;
1749 	u32			max_write_sge;
1750 	u32			max_read_sge;
1751 	enum ib_qp_type		qp_type;
1752 	struct ib_rwq_ind_table *rwq_ind_tbl;
1753 	struct ib_qp_security  *qp_sec;
1754 	u8			port;
1755 
1756 	/*
1757 	 * Implementation details of the RDMA core, don't use in drivers:
1758 	 */
1759 	struct rdma_restrack_entry     res;
1760 };
1761 
1762 struct ib_mr {
1763 	struct ib_device  *device;
1764 	struct ib_pd	  *pd;
1765 	u32		   lkey;
1766 	u32		   rkey;
1767 	u64		   iova;
1768 	u64		   length;
1769 	unsigned int	   page_size;
1770 	bool		   need_inval;
1771 	union {
1772 		struct ib_uobject	*uobject;	/* user */
1773 		struct list_head	qp_entry;	/* FR */
1774 	};
1775 };
1776 
1777 struct ib_mw {
1778 	struct ib_device	*device;
1779 	struct ib_pd		*pd;
1780 	struct ib_uobject	*uobject;
1781 	u32			rkey;
1782 	enum ib_mw_type         type;
1783 };
1784 
1785 struct ib_fmr {
1786 	struct ib_device	*device;
1787 	struct ib_pd		*pd;
1788 	struct list_head	list;
1789 	u32			lkey;
1790 	u32			rkey;
1791 };
1792 
1793 /* Supported steering options */
1794 enum ib_flow_attr_type {
1795 	/* steering according to rule specifications */
1796 	IB_FLOW_ATTR_NORMAL		= 0x0,
1797 	/* default unicast and multicast rule -
1798 	 * receive all Eth traffic which isn't steered to any QP
1799 	 */
1800 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1801 	/* default multicast rule -
1802 	 * receive all Eth multicast traffic which isn't steered to any QP
1803 	 */
1804 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1805 	/* sniffer rule - receive all port traffic */
1806 	IB_FLOW_ATTR_SNIFFER		= 0x3
1807 };
1808 
1809 /* Supported steering header types */
1810 enum ib_flow_spec_type {
1811 	/* L2 headers*/
1812 	IB_FLOW_SPEC_ETH		= 0x20,
1813 	IB_FLOW_SPEC_IB			= 0x22,
1814 	/* L3 header*/
1815 	IB_FLOW_SPEC_IPV4		= 0x30,
1816 	IB_FLOW_SPEC_IPV6		= 0x31,
1817 	/* L4 headers*/
1818 	IB_FLOW_SPEC_TCP		= 0x40,
1819 	IB_FLOW_SPEC_UDP		= 0x41,
1820 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1821 	IB_FLOW_SPEC_INNER		= 0x100,
1822 	/* Actions */
1823 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1824 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1825 };
1826 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1827 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1828 
1829 /* Flow steering rule priority is set according to it's domain.
1830  * Lower domain value means higher priority.
1831  */
1832 enum ib_flow_domain {
1833 	IB_FLOW_DOMAIN_USER,
1834 	IB_FLOW_DOMAIN_ETHTOOL,
1835 	IB_FLOW_DOMAIN_RFS,
1836 	IB_FLOW_DOMAIN_NIC,
1837 	IB_FLOW_DOMAIN_NUM /* Must be last */
1838 };
1839 
1840 enum ib_flow_flags {
1841 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1842 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1843 };
1844 
1845 struct ib_flow_eth_filter {
1846 	u8	dst_mac[6];
1847 	u8	src_mac[6];
1848 	__be16	ether_type;
1849 	__be16	vlan_tag;
1850 	/* Must be last */
1851 	u8	real_sz[0];
1852 };
1853 
1854 struct ib_flow_spec_eth {
1855 	u32			  type;
1856 	u16			  size;
1857 	struct ib_flow_eth_filter val;
1858 	struct ib_flow_eth_filter mask;
1859 };
1860 
1861 struct ib_flow_ib_filter {
1862 	__be16 dlid;
1863 	__u8   sl;
1864 	/* Must be last */
1865 	u8	real_sz[0];
1866 };
1867 
1868 struct ib_flow_spec_ib {
1869 	u32			 type;
1870 	u16			 size;
1871 	struct ib_flow_ib_filter val;
1872 	struct ib_flow_ib_filter mask;
1873 };
1874 
1875 /* IPv4 header flags */
1876 enum ib_ipv4_flags {
1877 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1878 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1879 				    last have this flag set */
1880 };
1881 
1882 struct ib_flow_ipv4_filter {
1883 	__be32	src_ip;
1884 	__be32	dst_ip;
1885 	u8	proto;
1886 	u8	tos;
1887 	u8	ttl;
1888 	u8	flags;
1889 	/* Must be last */
1890 	u8	real_sz[0];
1891 };
1892 
1893 struct ib_flow_spec_ipv4 {
1894 	u32			   type;
1895 	u16			   size;
1896 	struct ib_flow_ipv4_filter val;
1897 	struct ib_flow_ipv4_filter mask;
1898 };
1899 
1900 struct ib_flow_ipv6_filter {
1901 	u8	src_ip[16];
1902 	u8	dst_ip[16];
1903 	__be32	flow_label;
1904 	u8	next_hdr;
1905 	u8	traffic_class;
1906 	u8	hop_limit;
1907 	/* Must be last */
1908 	u8	real_sz[0];
1909 };
1910 
1911 struct ib_flow_spec_ipv6 {
1912 	u32			   type;
1913 	u16			   size;
1914 	struct ib_flow_ipv6_filter val;
1915 	struct ib_flow_ipv6_filter mask;
1916 };
1917 
1918 struct ib_flow_tcp_udp_filter {
1919 	__be16	dst_port;
1920 	__be16	src_port;
1921 	/* Must be last */
1922 	u8	real_sz[0];
1923 };
1924 
1925 struct ib_flow_spec_tcp_udp {
1926 	u32			      type;
1927 	u16			      size;
1928 	struct ib_flow_tcp_udp_filter val;
1929 	struct ib_flow_tcp_udp_filter mask;
1930 };
1931 
1932 struct ib_flow_tunnel_filter {
1933 	__be32	tunnel_id;
1934 	u8	real_sz[0];
1935 };
1936 
1937 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1938  * the tunnel_id from val has the vni value
1939  */
1940 struct ib_flow_spec_tunnel {
1941 	u32			      type;
1942 	u16			      size;
1943 	struct ib_flow_tunnel_filter  val;
1944 	struct ib_flow_tunnel_filter  mask;
1945 };
1946 
1947 struct ib_flow_spec_action_tag {
1948 	enum ib_flow_spec_type	      type;
1949 	u16			      size;
1950 	u32                           tag_id;
1951 };
1952 
1953 struct ib_flow_spec_action_drop {
1954 	enum ib_flow_spec_type	      type;
1955 	u16			      size;
1956 };
1957 
1958 union ib_flow_spec {
1959 	struct {
1960 		u32			type;
1961 		u16			size;
1962 	};
1963 	struct ib_flow_spec_eth		eth;
1964 	struct ib_flow_spec_ib		ib;
1965 	struct ib_flow_spec_ipv4        ipv4;
1966 	struct ib_flow_spec_tcp_udp	tcp_udp;
1967 	struct ib_flow_spec_ipv6        ipv6;
1968 	struct ib_flow_spec_tunnel      tunnel;
1969 	struct ib_flow_spec_action_tag  flow_tag;
1970 	struct ib_flow_spec_action_drop drop;
1971 };
1972 
1973 struct ib_flow_attr {
1974 	enum ib_flow_attr_type type;
1975 	u16	     size;
1976 	u16	     priority;
1977 	u32	     flags;
1978 	u8	     num_of_specs;
1979 	u8	     port;
1980 	/* Following are the optional layers according to user request
1981 	 * struct ib_flow_spec_xxx
1982 	 * struct ib_flow_spec_yyy
1983 	 */
1984 };
1985 
1986 struct ib_flow {
1987 	struct ib_qp		*qp;
1988 	struct ib_uobject	*uobject;
1989 };
1990 
1991 struct ib_mad_hdr;
1992 struct ib_grh;
1993 
1994 enum ib_process_mad_flags {
1995 	IB_MAD_IGNORE_MKEY	= 1,
1996 	IB_MAD_IGNORE_BKEY	= 2,
1997 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1998 };
1999 
2000 enum ib_mad_result {
2001 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2002 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2003 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2004 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2005 };
2006 
2007 struct ib_port_cache {
2008 	u64		      subnet_prefix;
2009 	struct ib_pkey_cache  *pkey;
2010 	struct ib_gid_table   *gid;
2011 	u8                     lmc;
2012 	enum ib_port_state     port_state;
2013 };
2014 
2015 struct ib_cache {
2016 	rwlock_t                lock;
2017 	struct ib_event_handler event_handler;
2018 	struct ib_port_cache   *ports;
2019 };
2020 
2021 struct iw_cm_verbs;
2022 
2023 struct ib_port_immutable {
2024 	int                           pkey_tbl_len;
2025 	int                           gid_tbl_len;
2026 	u32                           core_cap_flags;
2027 	u32                           max_mad_size;
2028 };
2029 
2030 /* rdma netdev type - specifies protocol type */
2031 enum rdma_netdev_t {
2032 	RDMA_NETDEV_OPA_VNIC,
2033 	RDMA_NETDEV_IPOIB,
2034 };
2035 
2036 /**
2037  * struct rdma_netdev - rdma netdev
2038  * For cases where netstack interfacing is required.
2039  */
2040 struct rdma_netdev {
2041 	void              *clnt_priv;
2042 	struct ib_device  *hca;
2043 	u8                 port_num;
2044 
2045 	/* cleanup function must be specified */
2046 	void (*free_rdma_netdev)(struct net_device *netdev);
2047 
2048 	/* control functions */
2049 	void (*set_id)(struct net_device *netdev, int id);
2050 	/* send packet */
2051 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2052 		    struct ib_ah *address, u32 dqpn);
2053 	/* multicast */
2054 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2055 			    union ib_gid *gid, u16 mlid,
2056 			    int set_qkey, u32 qkey);
2057 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2058 			    union ib_gid *gid, u16 mlid);
2059 };
2060 
2061 struct ib_port_pkey_list {
2062 	/* Lock to hold while modifying the list. */
2063 	spinlock_t                    list_lock;
2064 	struct list_head              pkey_list;
2065 };
2066 
2067 struct ib_device {
2068 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2069 	struct device                *dma_device;
2070 
2071 	char                          name[IB_DEVICE_NAME_MAX];
2072 
2073 	struct list_head              event_handler_list;
2074 	spinlock_t                    event_handler_lock;
2075 
2076 	spinlock_t                    client_data_lock;
2077 	struct list_head              core_list;
2078 	/* Access to the client_data_list is protected by the client_data_lock
2079 	 * spinlock and the lists_rwsem read-write semaphore */
2080 	struct list_head              client_data_list;
2081 
2082 	struct ib_cache               cache;
2083 	/**
2084 	 * port_immutable is indexed by port number
2085 	 */
2086 	struct ib_port_immutable     *port_immutable;
2087 
2088 	int			      num_comp_vectors;
2089 
2090 	struct ib_port_pkey_list     *port_pkey_list;
2091 
2092 	struct iw_cm_verbs	     *iwcm;
2093 
2094 	/**
2095 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2096 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2097 	 *   core when the device is removed.  A lifespan of -1 in the return
2098 	 *   struct tells the core to set a default lifespan.
2099 	 */
2100 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
2101 						     u8 port_num);
2102 	/**
2103 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2104 	 * @index - The index in the value array we wish to have updated, or
2105 	 *   num_counters if we want all stats updated
2106 	 * Return codes -
2107 	 *   < 0 - Error, no counters updated
2108 	 *   index - Updated the single counter pointed to by index
2109 	 *   num_counters - Updated all counters (will reset the timestamp
2110 	 *     and prevent further calls for lifespan milliseconds)
2111 	 * Drivers are allowed to update all counters in leiu of just the
2112 	 *   one given in index at their option
2113 	 */
2114 	int		           (*get_hw_stats)(struct ib_device *device,
2115 						   struct rdma_hw_stats *stats,
2116 						   u8 port, int index);
2117 	int		           (*query_device)(struct ib_device *device,
2118 						   struct ib_device_attr *device_attr,
2119 						   struct ib_udata *udata);
2120 	int		           (*query_port)(struct ib_device *device,
2121 						 u8 port_num,
2122 						 struct ib_port_attr *port_attr);
2123 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
2124 						     u8 port_num);
2125 	/* When calling get_netdev, the HW vendor's driver should return the
2126 	 * net device of device @device at port @port_num or NULL if such
2127 	 * a net device doesn't exist. The vendor driver should call dev_hold
2128 	 * on this net device. The HW vendor's device driver must guarantee
2129 	 * that this function returns NULL before the net device reaches
2130 	 * NETDEV_UNREGISTER_FINAL state.
2131 	 */
2132 	struct net_device	  *(*get_netdev)(struct ib_device *device,
2133 						 u8 port_num);
2134 	int		           (*query_gid)(struct ib_device *device,
2135 						u8 port_num, int index,
2136 						union ib_gid *gid);
2137 	/* When calling add_gid, the HW vendor's driver should
2138 	 * add the gid of device @device at gid index @index of
2139 	 * port @port_num to be @gid. Meta-info of that gid (for example,
2140 	 * the network device related to this gid is available
2141 	 * at @attr. @context allows the HW vendor driver to store extra
2142 	 * information together with a GID entry. The HW vendor may allocate
2143 	 * memory to contain this information and store it in @context when a
2144 	 * new GID entry is written to. Params are consistent until the next
2145 	 * call of add_gid or delete_gid. The function should return 0 on
2146 	 * success or error otherwise. The function could be called
2147 	 * concurrently for different ports. This function is only called
2148 	 * when roce_gid_table is used.
2149 	 */
2150 	int		           (*add_gid)(struct ib_device *device,
2151 					      u8 port_num,
2152 					      unsigned int index,
2153 					      const union ib_gid *gid,
2154 					      const struct ib_gid_attr *attr,
2155 					      void **context);
2156 	/* When calling del_gid, the HW vendor's driver should delete the
2157 	 * gid of device @device at gid index @index of port @port_num.
2158 	 * Upon the deletion of a GID entry, the HW vendor must free any
2159 	 * allocated memory. The caller will clear @context afterwards.
2160 	 * This function is only called when roce_gid_table is used.
2161 	 */
2162 	int		           (*del_gid)(struct ib_device *device,
2163 					      u8 port_num,
2164 					      unsigned int index,
2165 					      void **context);
2166 	int		           (*query_pkey)(struct ib_device *device,
2167 						 u8 port_num, u16 index, u16 *pkey);
2168 	int		           (*modify_device)(struct ib_device *device,
2169 						    int device_modify_mask,
2170 						    struct ib_device_modify *device_modify);
2171 	int		           (*modify_port)(struct ib_device *device,
2172 						  u8 port_num, int port_modify_mask,
2173 						  struct ib_port_modify *port_modify);
2174 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
2175 						     struct ib_udata *udata);
2176 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
2177 	int                        (*mmap)(struct ib_ucontext *context,
2178 					   struct vm_area_struct *vma);
2179 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
2180 					       struct ib_ucontext *context,
2181 					       struct ib_udata *udata);
2182 	int                        (*dealloc_pd)(struct ib_pd *pd);
2183 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
2184 						struct rdma_ah_attr *ah_attr,
2185 						struct ib_udata *udata);
2186 	int                        (*modify_ah)(struct ib_ah *ah,
2187 						struct rdma_ah_attr *ah_attr);
2188 	int                        (*query_ah)(struct ib_ah *ah,
2189 					       struct rdma_ah_attr *ah_attr);
2190 	int                        (*destroy_ah)(struct ib_ah *ah);
2191 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
2192 						 struct ib_srq_init_attr *srq_init_attr,
2193 						 struct ib_udata *udata);
2194 	int                        (*modify_srq)(struct ib_srq *srq,
2195 						 struct ib_srq_attr *srq_attr,
2196 						 enum ib_srq_attr_mask srq_attr_mask,
2197 						 struct ib_udata *udata);
2198 	int                        (*query_srq)(struct ib_srq *srq,
2199 						struct ib_srq_attr *srq_attr);
2200 	int                        (*destroy_srq)(struct ib_srq *srq);
2201 	int                        (*post_srq_recv)(struct ib_srq *srq,
2202 						    struct ib_recv_wr *recv_wr,
2203 						    struct ib_recv_wr **bad_recv_wr);
2204 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
2205 						struct ib_qp_init_attr *qp_init_attr,
2206 						struct ib_udata *udata);
2207 	int                        (*modify_qp)(struct ib_qp *qp,
2208 						struct ib_qp_attr *qp_attr,
2209 						int qp_attr_mask,
2210 						struct ib_udata *udata);
2211 	int                        (*query_qp)(struct ib_qp *qp,
2212 					       struct ib_qp_attr *qp_attr,
2213 					       int qp_attr_mask,
2214 					       struct ib_qp_init_attr *qp_init_attr);
2215 	int                        (*destroy_qp)(struct ib_qp *qp);
2216 	int                        (*post_send)(struct ib_qp *qp,
2217 						struct ib_send_wr *send_wr,
2218 						struct ib_send_wr **bad_send_wr);
2219 	int                        (*post_recv)(struct ib_qp *qp,
2220 						struct ib_recv_wr *recv_wr,
2221 						struct ib_recv_wr **bad_recv_wr);
2222 	struct ib_cq *             (*create_cq)(struct ib_device *device,
2223 						const struct ib_cq_init_attr *attr,
2224 						struct ib_ucontext *context,
2225 						struct ib_udata *udata);
2226 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2227 						u16 cq_period);
2228 	int                        (*destroy_cq)(struct ib_cq *cq);
2229 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
2230 						struct ib_udata *udata);
2231 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
2232 					      struct ib_wc *wc);
2233 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2234 	int                        (*req_notify_cq)(struct ib_cq *cq,
2235 						    enum ib_cq_notify_flags flags);
2236 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
2237 						      int wc_cnt);
2238 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
2239 						 int mr_access_flags);
2240 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
2241 						  u64 start, u64 length,
2242 						  u64 virt_addr,
2243 						  int mr_access_flags,
2244 						  struct ib_udata *udata);
2245 	int			   (*rereg_user_mr)(struct ib_mr *mr,
2246 						    int flags,
2247 						    u64 start, u64 length,
2248 						    u64 virt_addr,
2249 						    int mr_access_flags,
2250 						    struct ib_pd *pd,
2251 						    struct ib_udata *udata);
2252 	int                        (*dereg_mr)(struct ib_mr *mr);
2253 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2254 					       enum ib_mr_type mr_type,
2255 					       u32 max_num_sg);
2256 	int                        (*map_mr_sg)(struct ib_mr *mr,
2257 						struct scatterlist *sg,
2258 						int sg_nents,
2259 						unsigned int *sg_offset);
2260 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2261 					       enum ib_mw_type type,
2262 					       struct ib_udata *udata);
2263 	int                        (*dealloc_mw)(struct ib_mw *mw);
2264 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2265 						int mr_access_flags,
2266 						struct ib_fmr_attr *fmr_attr);
2267 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2268 						   u64 *page_list, int list_len,
2269 						   u64 iova);
2270 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2271 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2272 	int                        (*attach_mcast)(struct ib_qp *qp,
2273 						   union ib_gid *gid,
2274 						   u16 lid);
2275 	int                        (*detach_mcast)(struct ib_qp *qp,
2276 						   union ib_gid *gid,
2277 						   u16 lid);
2278 	int                        (*process_mad)(struct ib_device *device,
2279 						  int process_mad_flags,
2280 						  u8 port_num,
2281 						  const struct ib_wc *in_wc,
2282 						  const struct ib_grh *in_grh,
2283 						  const struct ib_mad_hdr *in_mad,
2284 						  size_t in_mad_size,
2285 						  struct ib_mad_hdr *out_mad,
2286 						  size_t *out_mad_size,
2287 						  u16 *out_mad_pkey_index);
2288 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2289 						 struct ib_ucontext *ucontext,
2290 						 struct ib_udata *udata);
2291 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2292 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2293 						  struct ib_flow_attr
2294 						  *flow_attr,
2295 						  int domain);
2296 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2297 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2298 						      struct ib_mr_status *mr_status);
2299 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2300 	void			   (*drain_rq)(struct ib_qp *qp);
2301 	void			   (*drain_sq)(struct ib_qp *qp);
2302 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2303 							int state);
2304 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2305 						   struct ifla_vf_info *ivf);
2306 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2307 						   struct ifla_vf_stats *stats);
2308 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2309 						  int type);
2310 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2311 						struct ib_wq_init_attr *init_attr,
2312 						struct ib_udata *udata);
2313 	int			   (*destroy_wq)(struct ib_wq *wq);
2314 	int			   (*modify_wq)(struct ib_wq *wq,
2315 						struct ib_wq_attr *attr,
2316 						u32 wq_attr_mask,
2317 						struct ib_udata *udata);
2318 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2319 							   struct ib_rwq_ind_table_init_attr *init_attr,
2320 							   struct ib_udata *udata);
2321 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2322 	/**
2323 	 * rdma netdev operation
2324 	 *
2325 	 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2326 	 * doesn't support the specified rdma netdev type.
2327 	 */
2328 	struct net_device *(*alloc_rdma_netdev)(
2329 					struct ib_device *device,
2330 					u8 port_num,
2331 					enum rdma_netdev_t type,
2332 					const char *name,
2333 					unsigned char name_assign_type,
2334 					void (*setup)(struct net_device *));
2335 
2336 	struct module               *owner;
2337 	struct device                dev;
2338 	struct kobject               *ports_parent;
2339 	struct list_head             port_list;
2340 
2341 	enum {
2342 		IB_DEV_UNINITIALIZED,
2343 		IB_DEV_REGISTERED,
2344 		IB_DEV_UNREGISTERED
2345 	}                            reg_state;
2346 
2347 	int			     uverbs_abi_ver;
2348 	u64			     uverbs_cmd_mask;
2349 	u64			     uverbs_ex_cmd_mask;
2350 
2351 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2352 	__be64			     node_guid;
2353 	u32			     local_dma_lkey;
2354 	u16                          is_switch:1;
2355 	u8                           node_type;
2356 	u8                           phys_port_cnt;
2357 	struct ib_device_attr        attrs;
2358 	struct attribute_group	     *hw_stats_ag;
2359 	struct rdma_hw_stats         *hw_stats;
2360 
2361 #ifdef CONFIG_CGROUP_RDMA
2362 	struct rdmacg_device         cg_device;
2363 #endif
2364 
2365 	u32                          index;
2366 	/*
2367 	 * Implementation details of the RDMA core, don't use in drivers
2368 	 */
2369 	struct rdma_restrack_root     res;
2370 
2371 	/**
2372 	 * The following mandatory functions are used only at device
2373 	 * registration.  Keep functions such as these at the end of this
2374 	 * structure to avoid cache line misses when accessing struct ib_device
2375 	 * in fast paths.
2376 	 */
2377 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2378 	void (*get_dev_fw_str)(struct ib_device *, char *str);
2379 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2380 						     int comp_vector);
2381 
2382 	struct uverbs_root_spec		*specs_root;
2383 };
2384 
2385 struct ib_client {
2386 	char  *name;
2387 	void (*add)   (struct ib_device *);
2388 	void (*remove)(struct ib_device *, void *client_data);
2389 
2390 	/* Returns the net_dev belonging to this ib_client and matching the
2391 	 * given parameters.
2392 	 * @dev:	 An RDMA device that the net_dev use for communication.
2393 	 * @port:	 A physical port number on the RDMA device.
2394 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2395 	 * @gid:	 A GID that the net_dev uses to communicate.
2396 	 * @addr:	 An IP address the net_dev is configured with.
2397 	 * @client_data: The device's client data set by ib_set_client_data().
2398 	 *
2399 	 * An ib_client that implements a net_dev on top of RDMA devices
2400 	 * (such as IP over IB) should implement this callback, allowing the
2401 	 * rdma_cm module to find the right net_dev for a given request.
2402 	 *
2403 	 * The caller is responsible for calling dev_put on the returned
2404 	 * netdev. */
2405 	struct net_device *(*get_net_dev_by_params)(
2406 			struct ib_device *dev,
2407 			u8 port,
2408 			u16 pkey,
2409 			const union ib_gid *gid,
2410 			const struct sockaddr *addr,
2411 			void *client_data);
2412 	struct list_head list;
2413 };
2414 
2415 struct ib_device *ib_alloc_device(size_t size);
2416 void ib_dealloc_device(struct ib_device *device);
2417 
2418 void ib_get_device_fw_str(struct ib_device *device, char *str);
2419 
2420 int ib_register_device(struct ib_device *device,
2421 		       int (*port_callback)(struct ib_device *,
2422 					    u8, struct kobject *));
2423 void ib_unregister_device(struct ib_device *device);
2424 
2425 int ib_register_client   (struct ib_client *client);
2426 void ib_unregister_client(struct ib_client *client);
2427 
2428 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2429 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2430 			 void *data);
2431 
2432 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2433 {
2434 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2435 }
2436 
2437 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2438 {
2439 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2440 }
2441 
2442 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2443 				       size_t offset,
2444 				       size_t len)
2445 {
2446 	const void __user *p = udata->inbuf + offset;
2447 	bool ret;
2448 	u8 *buf;
2449 
2450 	if (len > USHRT_MAX)
2451 		return false;
2452 
2453 	buf = memdup_user(p, len);
2454 	if (IS_ERR(buf))
2455 		return false;
2456 
2457 	ret = !memchr_inv(buf, 0, len);
2458 	kfree(buf);
2459 	return ret;
2460 }
2461 
2462 /**
2463  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2464  * contains all required attributes and no attributes not allowed for
2465  * the given QP state transition.
2466  * @cur_state: Current QP state
2467  * @next_state: Next QP state
2468  * @type: QP type
2469  * @mask: Mask of supplied QP attributes
2470  * @ll : link layer of port
2471  *
2472  * This function is a helper function that a low-level driver's
2473  * modify_qp method can use to validate the consumer's input.  It
2474  * checks that cur_state and next_state are valid QP states, that a
2475  * transition from cur_state to next_state is allowed by the IB spec,
2476  * and that the attribute mask supplied is allowed for the transition.
2477  */
2478 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2479 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2480 		       enum rdma_link_layer ll);
2481 
2482 void ib_register_event_handler(struct ib_event_handler *event_handler);
2483 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2484 void ib_dispatch_event(struct ib_event *event);
2485 
2486 int ib_query_port(struct ib_device *device,
2487 		  u8 port_num, struct ib_port_attr *port_attr);
2488 
2489 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2490 					       u8 port_num);
2491 
2492 /**
2493  * rdma_cap_ib_switch - Check if the device is IB switch
2494  * @device: Device to check
2495  *
2496  * Device driver is responsible for setting is_switch bit on
2497  * in ib_device structure at init time.
2498  *
2499  * Return: true if the device is IB switch.
2500  */
2501 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2502 {
2503 	return device->is_switch;
2504 }
2505 
2506 /**
2507  * rdma_start_port - Return the first valid port number for the device
2508  * specified
2509  *
2510  * @device: Device to be checked
2511  *
2512  * Return start port number
2513  */
2514 static inline u8 rdma_start_port(const struct ib_device *device)
2515 {
2516 	return rdma_cap_ib_switch(device) ? 0 : 1;
2517 }
2518 
2519 /**
2520  * rdma_end_port - Return the last valid port number for the device
2521  * specified
2522  *
2523  * @device: Device to be checked
2524  *
2525  * Return last port number
2526  */
2527 static inline u8 rdma_end_port(const struct ib_device *device)
2528 {
2529 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2530 }
2531 
2532 static inline int rdma_is_port_valid(const struct ib_device *device,
2533 				     unsigned int port)
2534 {
2535 	return (port >= rdma_start_port(device) &&
2536 		port <= rdma_end_port(device));
2537 }
2538 
2539 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2540 {
2541 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2542 }
2543 
2544 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2545 {
2546 	return device->port_immutable[port_num].core_cap_flags &
2547 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2548 }
2549 
2550 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2551 {
2552 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2553 }
2554 
2555 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2556 {
2557 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2558 }
2559 
2560 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2561 {
2562 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2563 }
2564 
2565 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2566 {
2567 	return rdma_protocol_ib(device, port_num) ||
2568 		rdma_protocol_roce(device, port_num);
2569 }
2570 
2571 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2572 {
2573 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2574 }
2575 
2576 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2577 {
2578 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2579 }
2580 
2581 /**
2582  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2583  * Management Datagrams.
2584  * @device: Device to check
2585  * @port_num: Port number to check
2586  *
2587  * Management Datagrams (MAD) are a required part of the InfiniBand
2588  * specification and are supported on all InfiniBand devices.  A slightly
2589  * extended version are also supported on OPA interfaces.
2590  *
2591  * Return: true if the port supports sending/receiving of MAD packets.
2592  */
2593 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2594 {
2595 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2596 }
2597 
2598 /**
2599  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2600  * Management Datagrams.
2601  * @device: Device to check
2602  * @port_num: Port number to check
2603  *
2604  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2605  * datagrams with their own versions.  These OPA MADs share many but not all of
2606  * the characteristics of InfiniBand MADs.
2607  *
2608  * OPA MADs differ in the following ways:
2609  *
2610  *    1) MADs are variable size up to 2K
2611  *       IBTA defined MADs remain fixed at 256 bytes
2612  *    2) OPA SMPs must carry valid PKeys
2613  *    3) OPA SMP packets are a different format
2614  *
2615  * Return: true if the port supports OPA MAD packet formats.
2616  */
2617 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2618 {
2619 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2620 		== RDMA_CORE_CAP_OPA_MAD;
2621 }
2622 
2623 /**
2624  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2625  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2626  * @device: Device to check
2627  * @port_num: Port number to check
2628  *
2629  * Each InfiniBand node is required to provide a Subnet Management Agent
2630  * that the subnet manager can access.  Prior to the fabric being fully
2631  * configured by the subnet manager, the SMA is accessed via a well known
2632  * interface called the Subnet Management Interface (SMI).  This interface
2633  * uses directed route packets to communicate with the SM to get around the
2634  * chicken and egg problem of the SM needing to know what's on the fabric
2635  * in order to configure the fabric, and needing to configure the fabric in
2636  * order to send packets to the devices on the fabric.  These directed
2637  * route packets do not need the fabric fully configured in order to reach
2638  * their destination.  The SMI is the only method allowed to send
2639  * directed route packets on an InfiniBand fabric.
2640  *
2641  * Return: true if the port provides an SMI.
2642  */
2643 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2644 {
2645 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2646 }
2647 
2648 /**
2649  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2650  * Communication Manager.
2651  * @device: Device to check
2652  * @port_num: Port number to check
2653  *
2654  * The InfiniBand Communication Manager is one of many pre-defined General
2655  * Service Agents (GSA) that are accessed via the General Service
2656  * Interface (GSI).  It's role is to facilitate establishment of connections
2657  * between nodes as well as other management related tasks for established
2658  * connections.
2659  *
2660  * Return: true if the port supports an IB CM (this does not guarantee that
2661  * a CM is actually running however).
2662  */
2663 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2664 {
2665 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2666 }
2667 
2668 /**
2669  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2670  * Communication Manager.
2671  * @device: Device to check
2672  * @port_num: Port number to check
2673  *
2674  * Similar to above, but specific to iWARP connections which have a different
2675  * managment protocol than InfiniBand.
2676  *
2677  * Return: true if the port supports an iWARP CM (this does not guarantee that
2678  * a CM is actually running however).
2679  */
2680 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2681 {
2682 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2683 }
2684 
2685 /**
2686  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2687  * Subnet Administration.
2688  * @device: Device to check
2689  * @port_num: Port number to check
2690  *
2691  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2692  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2693  * fabrics, devices should resolve routes to other hosts by contacting the
2694  * SA to query the proper route.
2695  *
2696  * Return: true if the port should act as a client to the fabric Subnet
2697  * Administration interface.  This does not imply that the SA service is
2698  * running locally.
2699  */
2700 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2701 {
2702 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2703 }
2704 
2705 /**
2706  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2707  * Multicast.
2708  * @device: Device to check
2709  * @port_num: Port number to check
2710  *
2711  * InfiniBand multicast registration is more complex than normal IPv4 or
2712  * IPv6 multicast registration.  Each Host Channel Adapter must register
2713  * with the Subnet Manager when it wishes to join a multicast group.  It
2714  * should do so only once regardless of how many queue pairs it subscribes
2715  * to this group.  And it should leave the group only after all queue pairs
2716  * attached to the group have been detached.
2717  *
2718  * Return: true if the port must undertake the additional adminstrative
2719  * overhead of registering/unregistering with the SM and tracking of the
2720  * total number of queue pairs attached to the multicast group.
2721  */
2722 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2723 {
2724 	return rdma_cap_ib_sa(device, port_num);
2725 }
2726 
2727 /**
2728  * rdma_cap_af_ib - Check if the port of device has the capability
2729  * Native Infiniband Address.
2730  * @device: Device to check
2731  * @port_num: Port number to check
2732  *
2733  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2734  * GID.  RoCE uses a different mechanism, but still generates a GID via
2735  * a prescribed mechanism and port specific data.
2736  *
2737  * Return: true if the port uses a GID address to identify devices on the
2738  * network.
2739  */
2740 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2741 {
2742 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2743 }
2744 
2745 /**
2746  * rdma_cap_eth_ah - Check if the port of device has the capability
2747  * Ethernet Address Handle.
2748  * @device: Device to check
2749  * @port_num: Port number to check
2750  *
2751  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2752  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2753  * port.  Normally, packet headers are generated by the sending host
2754  * adapter, but when sending connectionless datagrams, we must manually
2755  * inject the proper headers for the fabric we are communicating over.
2756  *
2757  * Return: true if we are running as a RoCE port and must force the
2758  * addition of a Global Route Header built from our Ethernet Address
2759  * Handle into our header list for connectionless packets.
2760  */
2761 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2762 {
2763 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2764 }
2765 
2766 /**
2767  * rdma_cap_opa_ah - Check if the port of device supports
2768  * OPA Address handles
2769  * @device: Device to check
2770  * @port_num: Port number to check
2771  *
2772  * Return: true if we are running on an OPA device which supports
2773  * the extended OPA addressing.
2774  */
2775 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
2776 {
2777 	return (device->port_immutable[port_num].core_cap_flags &
2778 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
2779 }
2780 
2781 /**
2782  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2783  *
2784  * @device: Device
2785  * @port_num: Port number
2786  *
2787  * This MAD size includes the MAD headers and MAD payload.  No other headers
2788  * are included.
2789  *
2790  * Return the max MAD size required by the Port.  Will return 0 if the port
2791  * does not support MADs
2792  */
2793 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2794 {
2795 	return device->port_immutable[port_num].max_mad_size;
2796 }
2797 
2798 /**
2799  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2800  * @device: Device to check
2801  * @port_num: Port number to check
2802  *
2803  * RoCE GID table mechanism manages the various GIDs for a device.
2804  *
2805  * NOTE: if allocating the port's GID table has failed, this call will still
2806  * return true, but any RoCE GID table API will fail.
2807  *
2808  * Return: true if the port uses RoCE GID table mechanism in order to manage
2809  * its GIDs.
2810  */
2811 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2812 					   u8 port_num)
2813 {
2814 	return rdma_protocol_roce(device, port_num) &&
2815 		device->add_gid && device->del_gid;
2816 }
2817 
2818 /*
2819  * Check if the device supports READ W/ INVALIDATE.
2820  */
2821 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2822 {
2823 	/*
2824 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2825 	 * has support for it yet.
2826 	 */
2827 	return rdma_protocol_iwarp(dev, port_num);
2828 }
2829 
2830 int ib_query_gid(struct ib_device *device,
2831 		 u8 port_num, int index, union ib_gid *gid,
2832 		 struct ib_gid_attr *attr);
2833 
2834 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2835 			 int state);
2836 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2837 		     struct ifla_vf_info *info);
2838 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2839 		    struct ifla_vf_stats *stats);
2840 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2841 		   int type);
2842 
2843 int ib_query_pkey(struct ib_device *device,
2844 		  u8 port_num, u16 index, u16 *pkey);
2845 
2846 int ib_modify_device(struct ib_device *device,
2847 		     int device_modify_mask,
2848 		     struct ib_device_modify *device_modify);
2849 
2850 int ib_modify_port(struct ib_device *device,
2851 		   u8 port_num, int port_modify_mask,
2852 		   struct ib_port_modify *port_modify);
2853 
2854 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2855 		struct net_device *ndev, u8 *port_num, u16 *index);
2856 
2857 int ib_find_pkey(struct ib_device *device,
2858 		 u8 port_num, u16 pkey, u16 *index);
2859 
2860 enum ib_pd_flags {
2861 	/*
2862 	 * Create a memory registration for all memory in the system and place
2863 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2864 	 * ULPs to avoid the overhead of dynamic MRs.
2865 	 *
2866 	 * This flag is generally considered unsafe and must only be used in
2867 	 * extremly trusted environments.  Every use of it will log a warning
2868 	 * in the kernel log.
2869 	 */
2870 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2871 };
2872 
2873 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2874 		const char *caller);
2875 #define ib_alloc_pd(device, flags) \
2876 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
2877 void ib_dealloc_pd(struct ib_pd *pd);
2878 
2879 /**
2880  * rdma_create_ah - Creates an address handle for the given address vector.
2881  * @pd: The protection domain associated with the address handle.
2882  * @ah_attr: The attributes of the address vector.
2883  *
2884  * The address handle is used to reference a local or global destination
2885  * in all UD QP post sends.
2886  */
2887 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
2888 
2889 /**
2890  * rdma_create_user_ah - Creates an address handle for the given address vector.
2891  * It resolves destination mac address for ah attribute of RoCE type.
2892  * @pd: The protection domain associated with the address handle.
2893  * @ah_attr: The attributes of the address vector.
2894  * @udata: pointer to user's input output buffer information need by
2895  *         provider driver.
2896  *
2897  * It returns 0 on success and returns appropriate error code on error.
2898  * The address handle is used to reference a local or global destination
2899  * in all UD QP post sends.
2900  */
2901 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
2902 				  struct rdma_ah_attr *ah_attr,
2903 				  struct ib_udata *udata);
2904 /**
2905  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2906  *   work completion.
2907  * @hdr: the L3 header to parse
2908  * @net_type: type of header to parse
2909  * @sgid: place to store source gid
2910  * @dgid: place to store destination gid
2911  */
2912 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2913 			      enum rdma_network_type net_type,
2914 			      union ib_gid *sgid, union ib_gid *dgid);
2915 
2916 /**
2917  * ib_get_rdma_header_version - Get the header version
2918  * @hdr: the L3 header to parse
2919  */
2920 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2921 
2922 /**
2923  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
2924  *   work completion.
2925  * @device: Device on which the received message arrived.
2926  * @port_num: Port on which the received message arrived.
2927  * @wc: Work completion associated with the received message.
2928  * @grh: References the received global route header.  This parameter is
2929  *   ignored unless the work completion indicates that the GRH is valid.
2930  * @ah_attr: Returned attributes that can be used when creating an address
2931  *   handle for replying to the message.
2932  */
2933 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
2934 			    const struct ib_wc *wc, const struct ib_grh *grh,
2935 			    struct rdma_ah_attr *ah_attr);
2936 
2937 /**
2938  * ib_create_ah_from_wc - Creates an address handle associated with the
2939  *   sender of the specified work completion.
2940  * @pd: The protection domain associated with the address handle.
2941  * @wc: Work completion information associated with a received message.
2942  * @grh: References the received global route header.  This parameter is
2943  *   ignored unless the work completion indicates that the GRH is valid.
2944  * @port_num: The outbound port number to associate with the address.
2945  *
2946  * The address handle is used to reference a local or global destination
2947  * in all UD QP post sends.
2948  */
2949 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2950 				   const struct ib_grh *grh, u8 port_num);
2951 
2952 /**
2953  * rdma_modify_ah - Modifies the address vector associated with an address
2954  *   handle.
2955  * @ah: The address handle to modify.
2956  * @ah_attr: The new address vector attributes to associate with the
2957  *   address handle.
2958  */
2959 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2960 
2961 /**
2962  * rdma_query_ah - Queries the address vector associated with an address
2963  *   handle.
2964  * @ah: The address handle to query.
2965  * @ah_attr: The address vector attributes associated with the address
2966  *   handle.
2967  */
2968 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2969 
2970 /**
2971  * rdma_destroy_ah - Destroys an address handle.
2972  * @ah: The address handle to destroy.
2973  */
2974 int rdma_destroy_ah(struct ib_ah *ah);
2975 
2976 /**
2977  * ib_create_srq - Creates a SRQ associated with the specified protection
2978  *   domain.
2979  * @pd: The protection domain associated with the SRQ.
2980  * @srq_init_attr: A list of initial attributes required to create the
2981  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2982  *   the actual capabilities of the created SRQ.
2983  *
2984  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2985  * requested size of the SRQ, and set to the actual values allocated
2986  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2987  * will always be at least as large as the requested values.
2988  */
2989 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2990 			     struct ib_srq_init_attr *srq_init_attr);
2991 
2992 /**
2993  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2994  * @srq: The SRQ to modify.
2995  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2996  *   the current values of selected SRQ attributes are returned.
2997  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2998  *   are being modified.
2999  *
3000  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3001  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3002  * the number of receives queued drops below the limit.
3003  */
3004 int ib_modify_srq(struct ib_srq *srq,
3005 		  struct ib_srq_attr *srq_attr,
3006 		  enum ib_srq_attr_mask srq_attr_mask);
3007 
3008 /**
3009  * ib_query_srq - Returns the attribute list and current values for the
3010  *   specified SRQ.
3011  * @srq: The SRQ to query.
3012  * @srq_attr: The attributes of the specified SRQ.
3013  */
3014 int ib_query_srq(struct ib_srq *srq,
3015 		 struct ib_srq_attr *srq_attr);
3016 
3017 /**
3018  * ib_destroy_srq - Destroys the specified SRQ.
3019  * @srq: The SRQ to destroy.
3020  */
3021 int ib_destroy_srq(struct ib_srq *srq);
3022 
3023 /**
3024  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3025  * @srq: The SRQ to post the work request on.
3026  * @recv_wr: A list of work requests to post on the receive queue.
3027  * @bad_recv_wr: On an immediate failure, this parameter will reference
3028  *   the work request that failed to be posted on the QP.
3029  */
3030 static inline int ib_post_srq_recv(struct ib_srq *srq,
3031 				   struct ib_recv_wr *recv_wr,
3032 				   struct ib_recv_wr **bad_recv_wr)
3033 {
3034 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
3035 }
3036 
3037 /**
3038  * ib_create_qp - Creates a QP associated with the specified protection
3039  *   domain.
3040  * @pd: The protection domain associated with the QP.
3041  * @qp_init_attr: A list of initial attributes required to create the
3042  *   QP.  If QP creation succeeds, then the attributes are updated to
3043  *   the actual capabilities of the created QP.
3044  */
3045 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3046 			   struct ib_qp_init_attr *qp_init_attr);
3047 
3048 /**
3049  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3050  * @qp: The QP to modify.
3051  * @attr: On input, specifies the QP attributes to modify.  On output,
3052  *   the current values of selected QP attributes are returned.
3053  * @attr_mask: A bit-mask used to specify which attributes of the QP
3054  *   are being modified.
3055  * @udata: pointer to user's input output buffer information
3056  *   are being modified.
3057  * It returns 0 on success and returns appropriate error code on error.
3058  */
3059 int ib_modify_qp_with_udata(struct ib_qp *qp,
3060 			    struct ib_qp_attr *attr,
3061 			    int attr_mask,
3062 			    struct ib_udata *udata);
3063 
3064 /**
3065  * ib_modify_qp - Modifies the attributes for the specified QP and then
3066  *   transitions the QP to the given state.
3067  * @qp: The QP to modify.
3068  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3069  *   the current values of selected QP attributes are returned.
3070  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3071  *   are being modified.
3072  */
3073 int ib_modify_qp(struct ib_qp *qp,
3074 		 struct ib_qp_attr *qp_attr,
3075 		 int qp_attr_mask);
3076 
3077 /**
3078  * ib_query_qp - Returns the attribute list and current values for the
3079  *   specified QP.
3080  * @qp: The QP to query.
3081  * @qp_attr: The attributes of the specified QP.
3082  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3083  * @qp_init_attr: Additional attributes of the selected QP.
3084  *
3085  * The qp_attr_mask may be used to limit the query to gathering only the
3086  * selected attributes.
3087  */
3088 int ib_query_qp(struct ib_qp *qp,
3089 		struct ib_qp_attr *qp_attr,
3090 		int qp_attr_mask,
3091 		struct ib_qp_init_attr *qp_init_attr);
3092 
3093 /**
3094  * ib_destroy_qp - Destroys the specified QP.
3095  * @qp: The QP to destroy.
3096  */
3097 int ib_destroy_qp(struct ib_qp *qp);
3098 
3099 /**
3100  * ib_open_qp - Obtain a reference to an existing sharable QP.
3101  * @xrcd - XRC domain
3102  * @qp_open_attr: Attributes identifying the QP to open.
3103  *
3104  * Returns a reference to a sharable QP.
3105  */
3106 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3107 			 struct ib_qp_open_attr *qp_open_attr);
3108 
3109 /**
3110  * ib_close_qp - Release an external reference to a QP.
3111  * @qp: The QP handle to release
3112  *
3113  * The opened QP handle is released by the caller.  The underlying
3114  * shared QP is not destroyed until all internal references are released.
3115  */
3116 int ib_close_qp(struct ib_qp *qp);
3117 
3118 /**
3119  * ib_post_send - Posts a list of work requests to the send queue of
3120  *   the specified QP.
3121  * @qp: The QP to post the work request on.
3122  * @send_wr: A list of work requests to post on the send queue.
3123  * @bad_send_wr: On an immediate failure, this parameter will reference
3124  *   the work request that failed to be posted on the QP.
3125  *
3126  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3127  * error is returned, the QP state shall not be affected,
3128  * ib_post_send() will return an immediate error after queueing any
3129  * earlier work requests in the list.
3130  */
3131 static inline int ib_post_send(struct ib_qp *qp,
3132 			       struct ib_send_wr *send_wr,
3133 			       struct ib_send_wr **bad_send_wr)
3134 {
3135 	return qp->device->post_send(qp, send_wr, bad_send_wr);
3136 }
3137 
3138 /**
3139  * ib_post_recv - Posts a list of work requests to the receive queue of
3140  *   the specified QP.
3141  * @qp: The QP to post the work request on.
3142  * @recv_wr: A list of work requests to post on the receive queue.
3143  * @bad_recv_wr: On an immediate failure, this parameter will reference
3144  *   the work request that failed to be posted on the QP.
3145  */
3146 static inline int ib_post_recv(struct ib_qp *qp,
3147 			       struct ib_recv_wr *recv_wr,
3148 			       struct ib_recv_wr **bad_recv_wr)
3149 {
3150 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
3151 }
3152 
3153 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,
3154 			    int nr_cqe, int comp_vector,
3155 			    enum ib_poll_context poll_ctx, const char *caller);
3156 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \
3157 	__ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME)
3158 
3159 void ib_free_cq(struct ib_cq *cq);
3160 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3161 
3162 /**
3163  * ib_create_cq - Creates a CQ on the specified device.
3164  * @device: The device on which to create the CQ.
3165  * @comp_handler: A user-specified callback that is invoked when a
3166  *   completion event occurs on the CQ.
3167  * @event_handler: A user-specified callback that is invoked when an
3168  *   asynchronous event not associated with a completion occurs on the CQ.
3169  * @cq_context: Context associated with the CQ returned to the user via
3170  *   the associated completion and event handlers.
3171  * @cq_attr: The attributes the CQ should be created upon.
3172  *
3173  * Users can examine the cq structure to determine the actual CQ size.
3174  */
3175 struct ib_cq *ib_create_cq(struct ib_device *device,
3176 			   ib_comp_handler comp_handler,
3177 			   void (*event_handler)(struct ib_event *, void *),
3178 			   void *cq_context,
3179 			   const struct ib_cq_init_attr *cq_attr);
3180 
3181 /**
3182  * ib_resize_cq - Modifies the capacity of the CQ.
3183  * @cq: The CQ to resize.
3184  * @cqe: The minimum size of the CQ.
3185  *
3186  * Users can examine the cq structure to determine the actual CQ size.
3187  */
3188 int ib_resize_cq(struct ib_cq *cq, int cqe);
3189 
3190 /**
3191  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3192  * @cq: The CQ to modify.
3193  * @cq_count: number of CQEs that will trigger an event
3194  * @cq_period: max period of time in usec before triggering an event
3195  *
3196  */
3197 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3198 
3199 /**
3200  * ib_destroy_cq - Destroys the specified CQ.
3201  * @cq: The CQ to destroy.
3202  */
3203 int ib_destroy_cq(struct ib_cq *cq);
3204 
3205 /**
3206  * ib_poll_cq - poll a CQ for completion(s)
3207  * @cq:the CQ being polled
3208  * @num_entries:maximum number of completions to return
3209  * @wc:array of at least @num_entries &struct ib_wc where completions
3210  *   will be returned
3211  *
3212  * Poll a CQ for (possibly multiple) completions.  If the return value
3213  * is < 0, an error occurred.  If the return value is >= 0, it is the
3214  * number of completions returned.  If the return value is
3215  * non-negative and < num_entries, then the CQ was emptied.
3216  */
3217 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3218 			     struct ib_wc *wc)
3219 {
3220 	return cq->device->poll_cq(cq, num_entries, wc);
3221 }
3222 
3223 /**
3224  * ib_peek_cq - Returns the number of unreaped completions currently
3225  *   on the specified CQ.
3226  * @cq: The CQ to peek.
3227  * @wc_cnt: A minimum number of unreaped completions to check for.
3228  *
3229  * If the number of unreaped completions is greater than or equal to wc_cnt,
3230  * this function returns wc_cnt, otherwise, it returns the actual number of
3231  * unreaped completions.
3232  */
3233 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
3234 
3235 /**
3236  * ib_req_notify_cq - Request completion notification on a CQ.
3237  * @cq: The CQ to generate an event for.
3238  * @flags:
3239  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3240  *   to request an event on the next solicited event or next work
3241  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3242  *   may also be |ed in to request a hint about missed events, as
3243  *   described below.
3244  *
3245  * Return Value:
3246  *    < 0 means an error occurred while requesting notification
3247  *   == 0 means notification was requested successfully, and if
3248  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3249  *        were missed and it is safe to wait for another event.  In
3250  *        this case is it guaranteed that any work completions added
3251  *        to the CQ since the last CQ poll will trigger a completion
3252  *        notification event.
3253  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3254  *        in.  It means that the consumer must poll the CQ again to
3255  *        make sure it is empty to avoid missing an event because of a
3256  *        race between requesting notification and an entry being
3257  *        added to the CQ.  This return value means it is possible
3258  *        (but not guaranteed) that a work completion has been added
3259  *        to the CQ since the last poll without triggering a
3260  *        completion notification event.
3261  */
3262 static inline int ib_req_notify_cq(struct ib_cq *cq,
3263 				   enum ib_cq_notify_flags flags)
3264 {
3265 	return cq->device->req_notify_cq(cq, flags);
3266 }
3267 
3268 /**
3269  * ib_req_ncomp_notif - Request completion notification when there are
3270  *   at least the specified number of unreaped completions on the CQ.
3271  * @cq: The CQ to generate an event for.
3272  * @wc_cnt: The number of unreaped completions that should be on the
3273  *   CQ before an event is generated.
3274  */
3275 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3276 {
3277 	return cq->device->req_ncomp_notif ?
3278 		cq->device->req_ncomp_notif(cq, wc_cnt) :
3279 		-ENOSYS;
3280 }
3281 
3282 /**
3283  * ib_dma_mapping_error - check a DMA addr for error
3284  * @dev: The device for which the dma_addr was created
3285  * @dma_addr: The DMA address to check
3286  */
3287 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3288 {
3289 	return dma_mapping_error(dev->dma_device, dma_addr);
3290 }
3291 
3292 /**
3293  * ib_dma_map_single - Map a kernel virtual address to DMA address
3294  * @dev: The device for which the dma_addr is to be created
3295  * @cpu_addr: The kernel virtual address
3296  * @size: The size of the region in bytes
3297  * @direction: The direction of the DMA
3298  */
3299 static inline u64 ib_dma_map_single(struct ib_device *dev,
3300 				    void *cpu_addr, size_t size,
3301 				    enum dma_data_direction direction)
3302 {
3303 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3304 }
3305 
3306 /**
3307  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3308  * @dev: The device for which the DMA address was created
3309  * @addr: The DMA address
3310  * @size: The size of the region in bytes
3311  * @direction: The direction of the DMA
3312  */
3313 static inline void ib_dma_unmap_single(struct ib_device *dev,
3314 				       u64 addr, size_t size,
3315 				       enum dma_data_direction direction)
3316 {
3317 	dma_unmap_single(dev->dma_device, addr, size, direction);
3318 }
3319 
3320 /**
3321  * ib_dma_map_page - Map a physical page to DMA address
3322  * @dev: The device for which the dma_addr is to be created
3323  * @page: The page to be mapped
3324  * @offset: The offset within the page
3325  * @size: The size of the region in bytes
3326  * @direction: The direction of the DMA
3327  */
3328 static inline u64 ib_dma_map_page(struct ib_device *dev,
3329 				  struct page *page,
3330 				  unsigned long offset,
3331 				  size_t size,
3332 					 enum dma_data_direction direction)
3333 {
3334 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3335 }
3336 
3337 /**
3338  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3339  * @dev: The device for which the DMA address was created
3340  * @addr: The DMA address
3341  * @size: The size of the region in bytes
3342  * @direction: The direction of the DMA
3343  */
3344 static inline void ib_dma_unmap_page(struct ib_device *dev,
3345 				     u64 addr, size_t size,
3346 				     enum dma_data_direction direction)
3347 {
3348 	dma_unmap_page(dev->dma_device, addr, size, direction);
3349 }
3350 
3351 /**
3352  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3353  * @dev: The device for which the DMA addresses are to be created
3354  * @sg: The array of scatter/gather entries
3355  * @nents: The number of scatter/gather entries
3356  * @direction: The direction of the DMA
3357  */
3358 static inline int ib_dma_map_sg(struct ib_device *dev,
3359 				struct scatterlist *sg, int nents,
3360 				enum dma_data_direction direction)
3361 {
3362 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3363 }
3364 
3365 /**
3366  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3367  * @dev: The device for which the DMA addresses were created
3368  * @sg: The array of scatter/gather entries
3369  * @nents: The number of scatter/gather entries
3370  * @direction: The direction of the DMA
3371  */
3372 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3373 				   struct scatterlist *sg, int nents,
3374 				   enum dma_data_direction direction)
3375 {
3376 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
3377 }
3378 
3379 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3380 				      struct scatterlist *sg, int nents,
3381 				      enum dma_data_direction direction,
3382 				      unsigned long dma_attrs)
3383 {
3384 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3385 				dma_attrs);
3386 }
3387 
3388 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3389 					 struct scatterlist *sg, int nents,
3390 					 enum dma_data_direction direction,
3391 					 unsigned long dma_attrs)
3392 {
3393 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3394 }
3395 /**
3396  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3397  * @dev: The device for which the DMA addresses were created
3398  * @sg: The scatter/gather entry
3399  *
3400  * Note: this function is obsolete. To do: change all occurrences of
3401  * ib_sg_dma_address() into sg_dma_address().
3402  */
3403 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3404 				    struct scatterlist *sg)
3405 {
3406 	return sg_dma_address(sg);
3407 }
3408 
3409 /**
3410  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3411  * @dev: The device for which the DMA addresses were created
3412  * @sg: The scatter/gather entry
3413  *
3414  * Note: this function is obsolete. To do: change all occurrences of
3415  * ib_sg_dma_len() into sg_dma_len().
3416  */
3417 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3418 					 struct scatterlist *sg)
3419 {
3420 	return sg_dma_len(sg);
3421 }
3422 
3423 /**
3424  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3425  * @dev: The device for which the DMA address was created
3426  * @addr: The DMA address
3427  * @size: The size of the region in bytes
3428  * @dir: The direction of the DMA
3429  */
3430 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3431 					      u64 addr,
3432 					      size_t size,
3433 					      enum dma_data_direction dir)
3434 {
3435 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3436 }
3437 
3438 /**
3439  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3440  * @dev: The device for which the DMA address was created
3441  * @addr: The DMA address
3442  * @size: The size of the region in bytes
3443  * @dir: The direction of the DMA
3444  */
3445 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3446 						 u64 addr,
3447 						 size_t size,
3448 						 enum dma_data_direction dir)
3449 {
3450 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3451 }
3452 
3453 /**
3454  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3455  * @dev: The device for which the DMA address is requested
3456  * @size: The size of the region to allocate in bytes
3457  * @dma_handle: A pointer for returning the DMA address of the region
3458  * @flag: memory allocator flags
3459  */
3460 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3461 					   size_t size,
3462 					   dma_addr_t *dma_handle,
3463 					   gfp_t flag)
3464 {
3465 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3466 }
3467 
3468 /**
3469  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3470  * @dev: The device for which the DMA addresses were allocated
3471  * @size: The size of the region
3472  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3473  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3474  */
3475 static inline void ib_dma_free_coherent(struct ib_device *dev,
3476 					size_t size, void *cpu_addr,
3477 					dma_addr_t dma_handle)
3478 {
3479 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3480 }
3481 
3482 /**
3483  * ib_dereg_mr - Deregisters a memory region and removes it from the
3484  *   HCA translation table.
3485  * @mr: The memory region to deregister.
3486  *
3487  * This function can fail, if the memory region has memory windows bound to it.
3488  */
3489 int ib_dereg_mr(struct ib_mr *mr);
3490 
3491 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3492 			  enum ib_mr_type mr_type,
3493 			  u32 max_num_sg);
3494 
3495 /**
3496  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3497  *   R_Key and L_Key.
3498  * @mr - struct ib_mr pointer to be updated.
3499  * @newkey - new key to be used.
3500  */
3501 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3502 {
3503 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3504 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3505 }
3506 
3507 /**
3508  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3509  * for calculating a new rkey for type 2 memory windows.
3510  * @rkey - the rkey to increment.
3511  */
3512 static inline u32 ib_inc_rkey(u32 rkey)
3513 {
3514 	const u32 mask = 0x000000ff;
3515 	return ((rkey + 1) & mask) | (rkey & ~mask);
3516 }
3517 
3518 /**
3519  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3520  * @pd: The protection domain associated with the unmapped region.
3521  * @mr_access_flags: Specifies the memory access rights.
3522  * @fmr_attr: Attributes of the unmapped region.
3523  *
3524  * A fast memory region must be mapped before it can be used as part of
3525  * a work request.
3526  */
3527 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3528 			    int mr_access_flags,
3529 			    struct ib_fmr_attr *fmr_attr);
3530 
3531 /**
3532  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3533  * @fmr: The fast memory region to associate with the pages.
3534  * @page_list: An array of physical pages to map to the fast memory region.
3535  * @list_len: The number of pages in page_list.
3536  * @iova: The I/O virtual address to use with the mapped region.
3537  */
3538 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3539 				  u64 *page_list, int list_len,
3540 				  u64 iova)
3541 {
3542 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3543 }
3544 
3545 /**
3546  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3547  * @fmr_list: A linked list of fast memory regions to unmap.
3548  */
3549 int ib_unmap_fmr(struct list_head *fmr_list);
3550 
3551 /**
3552  * ib_dealloc_fmr - Deallocates a fast memory region.
3553  * @fmr: The fast memory region to deallocate.
3554  */
3555 int ib_dealloc_fmr(struct ib_fmr *fmr);
3556 
3557 /**
3558  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3559  * @qp: QP to attach to the multicast group.  The QP must be type
3560  *   IB_QPT_UD.
3561  * @gid: Multicast group GID.
3562  * @lid: Multicast group LID in host byte order.
3563  *
3564  * In order to send and receive multicast packets, subnet
3565  * administration must have created the multicast group and configured
3566  * the fabric appropriately.  The port associated with the specified
3567  * QP must also be a member of the multicast group.
3568  */
3569 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3570 
3571 /**
3572  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3573  * @qp: QP to detach from the multicast group.
3574  * @gid: Multicast group GID.
3575  * @lid: Multicast group LID in host byte order.
3576  */
3577 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3578 
3579 /**
3580  * ib_alloc_xrcd - Allocates an XRC domain.
3581  * @device: The device on which to allocate the XRC domain.
3582  * @caller: Module name for kernel consumers
3583  */
3584 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
3585 #define ib_alloc_xrcd(device) \
3586 	__ib_alloc_xrcd((device), KBUILD_MODNAME)
3587 
3588 /**
3589  * ib_dealloc_xrcd - Deallocates an XRC domain.
3590  * @xrcd: The XRC domain to deallocate.
3591  */
3592 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3593 
3594 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3595 			       struct ib_flow_attr *flow_attr, int domain);
3596 int ib_destroy_flow(struct ib_flow *flow_id);
3597 
3598 static inline int ib_check_mr_access(int flags)
3599 {
3600 	/*
3601 	 * Local write permission is required if remote write or
3602 	 * remote atomic permission is also requested.
3603 	 */
3604 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3605 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3606 		return -EINVAL;
3607 
3608 	return 0;
3609 }
3610 
3611 /**
3612  * ib_check_mr_status: lightweight check of MR status.
3613  *     This routine may provide status checks on a selected
3614  *     ib_mr. first use is for signature status check.
3615  *
3616  * @mr: A memory region.
3617  * @check_mask: Bitmask of which checks to perform from
3618  *     ib_mr_status_check enumeration.
3619  * @mr_status: The container of relevant status checks.
3620  *     failed checks will be indicated in the status bitmask
3621  *     and the relevant info shall be in the error item.
3622  */
3623 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3624 		       struct ib_mr_status *mr_status);
3625 
3626 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3627 					    u16 pkey, const union ib_gid *gid,
3628 					    const struct sockaddr *addr);
3629 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3630 			   struct ib_wq_init_attr *init_attr);
3631 int ib_destroy_wq(struct ib_wq *wq);
3632 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3633 		 u32 wq_attr_mask);
3634 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3635 						 struct ib_rwq_ind_table_init_attr*
3636 						 wq_ind_table_init_attr);
3637 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3638 
3639 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3640 		 unsigned int *sg_offset, unsigned int page_size);
3641 
3642 static inline int
3643 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3644 		  unsigned int *sg_offset, unsigned int page_size)
3645 {
3646 	int n;
3647 
3648 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3649 	mr->iova = 0;
3650 
3651 	return n;
3652 }
3653 
3654 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3655 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3656 
3657 void ib_drain_rq(struct ib_qp *qp);
3658 void ib_drain_sq(struct ib_qp *qp);
3659 void ib_drain_qp(struct ib_qp *qp);
3660 
3661 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3662 
3663 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3664 {
3665 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3666 		return attr->roce.dmac;
3667 	return NULL;
3668 }
3669 
3670 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3671 {
3672 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3673 		attr->ib.dlid = (u16)dlid;
3674 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3675 		attr->opa.dlid = dlid;
3676 }
3677 
3678 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3679 {
3680 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3681 		return attr->ib.dlid;
3682 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3683 		return attr->opa.dlid;
3684 	return 0;
3685 }
3686 
3687 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3688 {
3689 	attr->sl = sl;
3690 }
3691 
3692 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3693 {
3694 	return attr->sl;
3695 }
3696 
3697 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3698 					 u8 src_path_bits)
3699 {
3700 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3701 		attr->ib.src_path_bits = src_path_bits;
3702 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3703 		attr->opa.src_path_bits = src_path_bits;
3704 }
3705 
3706 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3707 {
3708 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3709 		return attr->ib.src_path_bits;
3710 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3711 		return attr->opa.src_path_bits;
3712 	return 0;
3713 }
3714 
3715 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
3716 					bool make_grd)
3717 {
3718 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3719 		attr->opa.make_grd = make_grd;
3720 }
3721 
3722 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
3723 {
3724 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3725 		return attr->opa.make_grd;
3726 	return false;
3727 }
3728 
3729 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3730 {
3731 	attr->port_num = port_num;
3732 }
3733 
3734 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3735 {
3736 	return attr->port_num;
3737 }
3738 
3739 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3740 					   u8 static_rate)
3741 {
3742 	attr->static_rate = static_rate;
3743 }
3744 
3745 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3746 {
3747 	return attr->static_rate;
3748 }
3749 
3750 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3751 					enum ib_ah_flags flag)
3752 {
3753 	attr->ah_flags = flag;
3754 }
3755 
3756 static inline enum ib_ah_flags
3757 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3758 {
3759 	return attr->ah_flags;
3760 }
3761 
3762 static inline const struct ib_global_route
3763 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
3764 {
3765 	return &attr->grh;
3766 }
3767 
3768 /*To retrieve and modify the grh */
3769 static inline struct ib_global_route
3770 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
3771 {
3772 	return &attr->grh;
3773 }
3774 
3775 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
3776 {
3777 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3778 
3779 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
3780 }
3781 
3782 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
3783 					     __be64 prefix)
3784 {
3785 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3786 
3787 	grh->dgid.global.subnet_prefix = prefix;
3788 }
3789 
3790 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
3791 					    __be64 if_id)
3792 {
3793 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3794 
3795 	grh->dgid.global.interface_id = if_id;
3796 }
3797 
3798 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
3799 				   union ib_gid *dgid, u32 flow_label,
3800 				   u8 sgid_index, u8 hop_limit,
3801 				   u8 traffic_class)
3802 {
3803 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3804 
3805 	attr->ah_flags = IB_AH_GRH;
3806 	if (dgid)
3807 		grh->dgid = *dgid;
3808 	grh->flow_label = flow_label;
3809 	grh->sgid_index = sgid_index;
3810 	grh->hop_limit = hop_limit;
3811 	grh->traffic_class = traffic_class;
3812 }
3813 
3814 /**
3815  * rdma_ah_find_type - Return address handle type.
3816  *
3817  * @dev: Device to be checked
3818  * @port_num: Port number
3819  */
3820 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
3821 						       u8 port_num)
3822 {
3823 	if (rdma_protocol_roce(dev, port_num))
3824 		return RDMA_AH_ATTR_TYPE_ROCE;
3825 	if (rdma_protocol_ib(dev, port_num)) {
3826 		if (rdma_cap_opa_ah(dev, port_num))
3827 			return RDMA_AH_ATTR_TYPE_OPA;
3828 		return RDMA_AH_ATTR_TYPE_IB;
3829 	}
3830 
3831 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
3832 }
3833 
3834 /**
3835  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
3836  *     In the current implementation the only way to get
3837  *     get the 32bit lid is from other sources for OPA.
3838  *     For IB, lids will always be 16bits so cast the
3839  *     value accordingly.
3840  *
3841  * @lid: A 32bit LID
3842  */
3843 static inline u16 ib_lid_cpu16(u32 lid)
3844 {
3845 	WARN_ON_ONCE(lid & 0xFFFF0000);
3846 	return (u16)lid;
3847 }
3848 
3849 /**
3850  * ib_lid_be16 - Return lid in 16bit BE encoding.
3851  *
3852  * @lid: A 32bit LID
3853  */
3854 static inline __be16 ib_lid_be16(u32 lid)
3855 {
3856 	WARN_ON_ONCE(lid & 0xFFFF0000);
3857 	return cpu_to_be16((u16)lid);
3858 }
3859 
3860 /**
3861  * ib_get_vector_affinity - Get the affinity mappings of a given completion
3862  *   vector
3863  * @device:         the rdma device
3864  * @comp_vector:    index of completion vector
3865  *
3866  * Returns NULL on failure, otherwise a corresponding cpu map of the
3867  * completion vector (returns all-cpus map if the device driver doesn't
3868  * implement get_vector_affinity).
3869  */
3870 static inline const struct cpumask *
3871 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
3872 {
3873 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
3874 	    !device->get_vector_affinity)
3875 		return NULL;
3876 
3877 	return device->get_vector_affinity(device, comp_vector);
3878 
3879 }
3880 
3881 /**
3882  * rdma_roce_rescan_device - Rescan all of the network devices in the system
3883  * and add their gids, as needed, to the relevant RoCE devices.
3884  *
3885  * @device:         the rdma device
3886  */
3887 void rdma_roce_rescan_device(struct ib_device *ibdev);
3888 
3889 #endif /* IB_VERBS_H */
3890