xref: /openbmc/linux/include/rdma/ib_verbs.h (revision 4e1a33b1)
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 
59 #include <linux/if_link.h>
60 #include <linux/atomic.h>
61 #include <linux/mmu_notifier.h>
62 #include <linux/uaccess.h>
63 
64 extern struct workqueue_struct *ib_wq;
65 extern struct workqueue_struct *ib_comp_wq;
66 
67 union ib_gid {
68 	u8	raw[16];
69 	struct {
70 		__be64	subnet_prefix;
71 		__be64	interface_id;
72 	} global;
73 };
74 
75 extern union ib_gid zgid;
76 
77 enum ib_gid_type {
78 	/* If link layer is Ethernet, this is RoCE V1 */
79 	IB_GID_TYPE_IB        = 0,
80 	IB_GID_TYPE_ROCE      = 0,
81 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
82 	IB_GID_TYPE_SIZE
83 };
84 
85 #define ROCE_V2_UDP_DPORT      4791
86 struct ib_gid_attr {
87 	enum ib_gid_type	gid_type;
88 	struct net_device	*ndev;
89 };
90 
91 enum rdma_node_type {
92 	/* IB values map to NodeInfo:NodeType. */
93 	RDMA_NODE_IB_CA 	= 1,
94 	RDMA_NODE_IB_SWITCH,
95 	RDMA_NODE_IB_ROUTER,
96 	RDMA_NODE_RNIC,
97 	RDMA_NODE_USNIC,
98 	RDMA_NODE_USNIC_UDP,
99 };
100 
101 enum {
102 	/* set the local administered indication */
103 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
104 };
105 
106 enum rdma_transport_type {
107 	RDMA_TRANSPORT_IB,
108 	RDMA_TRANSPORT_IWARP,
109 	RDMA_TRANSPORT_USNIC,
110 	RDMA_TRANSPORT_USNIC_UDP
111 };
112 
113 enum rdma_protocol_type {
114 	RDMA_PROTOCOL_IB,
115 	RDMA_PROTOCOL_IBOE,
116 	RDMA_PROTOCOL_IWARP,
117 	RDMA_PROTOCOL_USNIC_UDP
118 };
119 
120 __attribute_const__ enum rdma_transport_type
121 rdma_node_get_transport(enum rdma_node_type node_type);
122 
123 enum rdma_network_type {
124 	RDMA_NETWORK_IB,
125 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
126 	RDMA_NETWORK_IPV4,
127 	RDMA_NETWORK_IPV6
128 };
129 
130 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
131 {
132 	if (network_type == RDMA_NETWORK_IPV4 ||
133 	    network_type == RDMA_NETWORK_IPV6)
134 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
135 
136 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
137 	return IB_GID_TYPE_IB;
138 }
139 
140 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
141 							    union ib_gid *gid)
142 {
143 	if (gid_type == IB_GID_TYPE_IB)
144 		return RDMA_NETWORK_IB;
145 
146 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
147 		return RDMA_NETWORK_IPV4;
148 	else
149 		return RDMA_NETWORK_IPV6;
150 }
151 
152 enum rdma_link_layer {
153 	IB_LINK_LAYER_UNSPECIFIED,
154 	IB_LINK_LAYER_INFINIBAND,
155 	IB_LINK_LAYER_ETHERNET,
156 };
157 
158 enum ib_device_cap_flags {
159 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
160 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
161 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
162 	IB_DEVICE_RAW_MULTI			= (1 << 3),
163 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
164 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
165 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
166 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
167 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
168 	IB_DEVICE_INIT_TYPE			= (1 << 9),
169 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
170 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
171 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
172 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
173 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
174 
175 	/*
176 	 * This device supports a per-device lkey or stag that can be
177 	 * used without performing a memory registration for the local
178 	 * memory.  Note that ULPs should never check this flag, but
179 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
180 	 * which will always contain a usable lkey.
181 	 */
182 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
183 	IB_DEVICE_RESERVED /* old SEND_W_INV */	= (1 << 16),
184 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
185 	/*
186 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
187 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
188 	 * messages and can verify the validity of checksum for
189 	 * incoming messages.  Setting this flag implies that the
190 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
191 	 */
192 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
193 	IB_DEVICE_UD_TSO			= (1 << 19),
194 	IB_DEVICE_XRC				= (1 << 20),
195 
196 	/*
197 	 * This device supports the IB "base memory management extension",
198 	 * which includes support for fast registrations (IB_WR_REG_MR,
199 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
200 	 * also be set by any iWarp device which must support FRs to comply
201 	 * to the iWarp verbs spec.  iWarp devices also support the
202 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
203 	 * stag.
204 	 */
205 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
206 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
207 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
208 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
209 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
210 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
211 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
212 	/*
213 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
214 	 * support execution of WQEs that involve synchronization
215 	 * of I/O operations with single completion queue managed
216 	 * by hardware.
217 	 */
218 	IB_DEVICE_CROSS_CHANNEL		= (1 << 27),
219 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
220 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
221 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
222 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
223 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
224 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
225 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
226 };
227 
228 enum ib_signature_prot_cap {
229 	IB_PROT_T10DIF_TYPE_1 = 1,
230 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
231 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
232 };
233 
234 enum ib_signature_guard_cap {
235 	IB_GUARD_T10DIF_CRC	= 1,
236 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
237 };
238 
239 enum ib_atomic_cap {
240 	IB_ATOMIC_NONE,
241 	IB_ATOMIC_HCA,
242 	IB_ATOMIC_GLOB
243 };
244 
245 enum ib_odp_general_cap_bits {
246 	IB_ODP_SUPPORT		= 1 << 0,
247 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
248 };
249 
250 enum ib_odp_transport_cap_bits {
251 	IB_ODP_SUPPORT_SEND	= 1 << 0,
252 	IB_ODP_SUPPORT_RECV	= 1 << 1,
253 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
254 	IB_ODP_SUPPORT_READ	= 1 << 3,
255 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
256 };
257 
258 struct ib_odp_caps {
259 	uint64_t general_caps;
260 	struct {
261 		uint32_t  rc_odp_caps;
262 		uint32_t  uc_odp_caps;
263 		uint32_t  ud_odp_caps;
264 	} per_transport_caps;
265 };
266 
267 struct ib_rss_caps {
268 	/* Corresponding bit will be set if qp type from
269 	 * 'enum ib_qp_type' is supported, e.g.
270 	 * supported_qpts |= 1 << IB_QPT_UD
271 	 */
272 	u32 supported_qpts;
273 	u32 max_rwq_indirection_tables;
274 	u32 max_rwq_indirection_table_size;
275 };
276 
277 enum ib_cq_creation_flags {
278 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
279 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
280 };
281 
282 struct ib_cq_init_attr {
283 	unsigned int	cqe;
284 	int		comp_vector;
285 	u32		flags;
286 };
287 
288 struct ib_device_attr {
289 	u64			fw_ver;
290 	__be64			sys_image_guid;
291 	u64			max_mr_size;
292 	u64			page_size_cap;
293 	u32			vendor_id;
294 	u32			vendor_part_id;
295 	u32			hw_ver;
296 	int			max_qp;
297 	int			max_qp_wr;
298 	u64			device_cap_flags;
299 	int			max_sge;
300 	int			max_sge_rd;
301 	int			max_cq;
302 	int			max_cqe;
303 	int			max_mr;
304 	int			max_pd;
305 	int			max_qp_rd_atom;
306 	int			max_ee_rd_atom;
307 	int			max_res_rd_atom;
308 	int			max_qp_init_rd_atom;
309 	int			max_ee_init_rd_atom;
310 	enum ib_atomic_cap	atomic_cap;
311 	enum ib_atomic_cap	masked_atomic_cap;
312 	int			max_ee;
313 	int			max_rdd;
314 	int			max_mw;
315 	int			max_raw_ipv6_qp;
316 	int			max_raw_ethy_qp;
317 	int			max_mcast_grp;
318 	int			max_mcast_qp_attach;
319 	int			max_total_mcast_qp_attach;
320 	int			max_ah;
321 	int			max_fmr;
322 	int			max_map_per_fmr;
323 	int			max_srq;
324 	int			max_srq_wr;
325 	int			max_srq_sge;
326 	unsigned int		max_fast_reg_page_list_len;
327 	u16			max_pkeys;
328 	u8			local_ca_ack_delay;
329 	int			sig_prot_cap;
330 	int			sig_guard_cap;
331 	struct ib_odp_caps	odp_caps;
332 	uint64_t		timestamp_mask;
333 	uint64_t		hca_core_clock; /* in KHZ */
334 	struct ib_rss_caps	rss_caps;
335 	u32			max_wq_type_rq;
336 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
337 };
338 
339 enum ib_mtu {
340 	IB_MTU_256  = 1,
341 	IB_MTU_512  = 2,
342 	IB_MTU_1024 = 3,
343 	IB_MTU_2048 = 4,
344 	IB_MTU_4096 = 5
345 };
346 
347 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
348 {
349 	switch (mtu) {
350 	case IB_MTU_256:  return  256;
351 	case IB_MTU_512:  return  512;
352 	case IB_MTU_1024: return 1024;
353 	case IB_MTU_2048: return 2048;
354 	case IB_MTU_4096: return 4096;
355 	default: 	  return -1;
356 	}
357 }
358 
359 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
360 {
361 	if (mtu >= 4096)
362 		return IB_MTU_4096;
363 	else if (mtu >= 2048)
364 		return IB_MTU_2048;
365 	else if (mtu >= 1024)
366 		return IB_MTU_1024;
367 	else if (mtu >= 512)
368 		return IB_MTU_512;
369 	else
370 		return IB_MTU_256;
371 }
372 
373 enum ib_port_state {
374 	IB_PORT_NOP		= 0,
375 	IB_PORT_DOWN		= 1,
376 	IB_PORT_INIT		= 2,
377 	IB_PORT_ARMED		= 3,
378 	IB_PORT_ACTIVE		= 4,
379 	IB_PORT_ACTIVE_DEFER	= 5
380 };
381 
382 enum ib_port_cap_flags {
383 	IB_PORT_SM				= 1 <<  1,
384 	IB_PORT_NOTICE_SUP			= 1 <<  2,
385 	IB_PORT_TRAP_SUP			= 1 <<  3,
386 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
387 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
388 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
389 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
390 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
391 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
392 	IB_PORT_SM_DISABLED			= 1 << 10,
393 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
394 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
395 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
396 	IB_PORT_CM_SUP				= 1 << 16,
397 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
398 	IB_PORT_REINIT_SUP			= 1 << 18,
399 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
400 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
401 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
402 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
403 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
404 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
405 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
406 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
407 };
408 
409 enum ib_port_width {
410 	IB_WIDTH_1X	= 1,
411 	IB_WIDTH_4X	= 2,
412 	IB_WIDTH_8X	= 4,
413 	IB_WIDTH_12X	= 8
414 };
415 
416 static inline int ib_width_enum_to_int(enum ib_port_width width)
417 {
418 	switch (width) {
419 	case IB_WIDTH_1X:  return  1;
420 	case IB_WIDTH_4X:  return  4;
421 	case IB_WIDTH_8X:  return  8;
422 	case IB_WIDTH_12X: return 12;
423 	default: 	  return -1;
424 	}
425 }
426 
427 enum ib_port_speed {
428 	IB_SPEED_SDR	= 1,
429 	IB_SPEED_DDR	= 2,
430 	IB_SPEED_QDR	= 4,
431 	IB_SPEED_FDR10	= 8,
432 	IB_SPEED_FDR	= 16,
433 	IB_SPEED_EDR	= 32
434 };
435 
436 /**
437  * struct rdma_hw_stats
438  * @timestamp - Used by the core code to track when the last update was
439  * @lifespan - Used by the core code to determine how old the counters
440  *   should be before being updated again.  Stored in jiffies, defaults
441  *   to 10 milliseconds, drivers can override the default be specifying
442  *   their own value during their allocation routine.
443  * @name - Array of pointers to static names used for the counters in
444  *   directory.
445  * @num_counters - How many hardware counters there are.  If name is
446  *   shorter than this number, a kernel oops will result.  Driver authors
447  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
448  *   in their code to prevent this.
449  * @value - Array of u64 counters that are accessed by the sysfs code and
450  *   filled in by the drivers get_stats routine
451  */
452 struct rdma_hw_stats {
453 	unsigned long	timestamp;
454 	unsigned long	lifespan;
455 	const char * const *names;
456 	int		num_counters;
457 	u64		value[];
458 };
459 
460 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
461 /**
462  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
463  *   for drivers.
464  * @names - Array of static const char *
465  * @num_counters - How many elements in array
466  * @lifespan - How many milliseconds between updates
467  */
468 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
469 		const char * const *names, int num_counters,
470 		unsigned long lifespan)
471 {
472 	struct rdma_hw_stats *stats;
473 
474 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
475 			GFP_KERNEL);
476 	if (!stats)
477 		return NULL;
478 	stats->names = names;
479 	stats->num_counters = num_counters;
480 	stats->lifespan = msecs_to_jiffies(lifespan);
481 
482 	return stats;
483 }
484 
485 
486 /* Define bits for the various functionality this port needs to be supported by
487  * the core.
488  */
489 /* Management                           0x00000FFF */
490 #define RDMA_CORE_CAP_IB_MAD            0x00000001
491 #define RDMA_CORE_CAP_IB_SMI            0x00000002
492 #define RDMA_CORE_CAP_IB_CM             0x00000004
493 #define RDMA_CORE_CAP_IW_CM             0x00000008
494 #define RDMA_CORE_CAP_IB_SA             0x00000010
495 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
496 
497 /* Address format                       0x000FF000 */
498 #define RDMA_CORE_CAP_AF_IB             0x00001000
499 #define RDMA_CORE_CAP_ETH_AH            0x00002000
500 
501 /* Protocol                             0xFFF00000 */
502 #define RDMA_CORE_CAP_PROT_IB           0x00100000
503 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
504 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
505 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
506 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
507 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
508 
509 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
510 					| RDMA_CORE_CAP_IB_MAD \
511 					| RDMA_CORE_CAP_IB_SMI \
512 					| RDMA_CORE_CAP_IB_CM  \
513 					| RDMA_CORE_CAP_IB_SA  \
514 					| RDMA_CORE_CAP_AF_IB)
515 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
516 					| RDMA_CORE_CAP_IB_MAD  \
517 					| RDMA_CORE_CAP_IB_CM   \
518 					| RDMA_CORE_CAP_AF_IB   \
519 					| RDMA_CORE_CAP_ETH_AH)
520 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
521 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
522 					| RDMA_CORE_CAP_IB_MAD  \
523 					| RDMA_CORE_CAP_IB_CM   \
524 					| RDMA_CORE_CAP_AF_IB   \
525 					| RDMA_CORE_CAP_ETH_AH)
526 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
527 					| RDMA_CORE_CAP_IW_CM)
528 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
529 					| RDMA_CORE_CAP_OPA_MAD)
530 
531 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
532 
533 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
534 
535 struct ib_port_attr {
536 	u64			subnet_prefix;
537 	enum ib_port_state	state;
538 	enum ib_mtu		max_mtu;
539 	enum ib_mtu		active_mtu;
540 	int			gid_tbl_len;
541 	u32			port_cap_flags;
542 	u32			max_msg_sz;
543 	u32			bad_pkey_cntr;
544 	u32			qkey_viol_cntr;
545 	u16			pkey_tbl_len;
546 	u16			lid;
547 	u16			sm_lid;
548 	u8			lmc;
549 	u8			max_vl_num;
550 	u8			sm_sl;
551 	u8			subnet_timeout;
552 	u8			init_type_reply;
553 	u8			active_width;
554 	u8			active_speed;
555 	u8                      phys_state;
556 	bool			grh_required;
557 };
558 
559 enum ib_device_modify_flags {
560 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
561 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
562 };
563 
564 #define IB_DEVICE_NODE_DESC_MAX 64
565 
566 struct ib_device_modify {
567 	u64	sys_image_guid;
568 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
569 };
570 
571 enum ib_port_modify_flags {
572 	IB_PORT_SHUTDOWN		= 1,
573 	IB_PORT_INIT_TYPE		= (1<<2),
574 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
575 };
576 
577 struct ib_port_modify {
578 	u32	set_port_cap_mask;
579 	u32	clr_port_cap_mask;
580 	u8	init_type;
581 };
582 
583 enum ib_event_type {
584 	IB_EVENT_CQ_ERR,
585 	IB_EVENT_QP_FATAL,
586 	IB_EVENT_QP_REQ_ERR,
587 	IB_EVENT_QP_ACCESS_ERR,
588 	IB_EVENT_COMM_EST,
589 	IB_EVENT_SQ_DRAINED,
590 	IB_EVENT_PATH_MIG,
591 	IB_EVENT_PATH_MIG_ERR,
592 	IB_EVENT_DEVICE_FATAL,
593 	IB_EVENT_PORT_ACTIVE,
594 	IB_EVENT_PORT_ERR,
595 	IB_EVENT_LID_CHANGE,
596 	IB_EVENT_PKEY_CHANGE,
597 	IB_EVENT_SM_CHANGE,
598 	IB_EVENT_SRQ_ERR,
599 	IB_EVENT_SRQ_LIMIT_REACHED,
600 	IB_EVENT_QP_LAST_WQE_REACHED,
601 	IB_EVENT_CLIENT_REREGISTER,
602 	IB_EVENT_GID_CHANGE,
603 	IB_EVENT_WQ_FATAL,
604 };
605 
606 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
607 
608 struct ib_event {
609 	struct ib_device	*device;
610 	union {
611 		struct ib_cq	*cq;
612 		struct ib_qp	*qp;
613 		struct ib_srq	*srq;
614 		struct ib_wq	*wq;
615 		u8		port_num;
616 	} element;
617 	enum ib_event_type	event;
618 };
619 
620 struct ib_event_handler {
621 	struct ib_device *device;
622 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
623 	struct list_head  list;
624 };
625 
626 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
627 	do {							\
628 		(_ptr)->device  = _device;			\
629 		(_ptr)->handler = _handler;			\
630 		INIT_LIST_HEAD(&(_ptr)->list);			\
631 	} while (0)
632 
633 struct ib_global_route {
634 	union ib_gid	dgid;
635 	u32		flow_label;
636 	u8		sgid_index;
637 	u8		hop_limit;
638 	u8		traffic_class;
639 };
640 
641 struct ib_grh {
642 	__be32		version_tclass_flow;
643 	__be16		paylen;
644 	u8		next_hdr;
645 	u8		hop_limit;
646 	union ib_gid	sgid;
647 	union ib_gid	dgid;
648 };
649 
650 union rdma_network_hdr {
651 	struct ib_grh ibgrh;
652 	struct {
653 		/* The IB spec states that if it's IPv4, the header
654 		 * is located in the last 20 bytes of the header.
655 		 */
656 		u8		reserved[20];
657 		struct iphdr	roce4grh;
658 	};
659 };
660 
661 enum {
662 	IB_MULTICAST_QPN = 0xffffff
663 };
664 
665 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
666 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
667 
668 enum ib_ah_flags {
669 	IB_AH_GRH	= 1
670 };
671 
672 enum ib_rate {
673 	IB_RATE_PORT_CURRENT = 0,
674 	IB_RATE_2_5_GBPS = 2,
675 	IB_RATE_5_GBPS   = 5,
676 	IB_RATE_10_GBPS  = 3,
677 	IB_RATE_20_GBPS  = 6,
678 	IB_RATE_30_GBPS  = 4,
679 	IB_RATE_40_GBPS  = 7,
680 	IB_RATE_60_GBPS  = 8,
681 	IB_RATE_80_GBPS  = 9,
682 	IB_RATE_120_GBPS = 10,
683 	IB_RATE_14_GBPS  = 11,
684 	IB_RATE_56_GBPS  = 12,
685 	IB_RATE_112_GBPS = 13,
686 	IB_RATE_168_GBPS = 14,
687 	IB_RATE_25_GBPS  = 15,
688 	IB_RATE_100_GBPS = 16,
689 	IB_RATE_200_GBPS = 17,
690 	IB_RATE_300_GBPS = 18
691 };
692 
693 /**
694  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
695  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
696  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
697  * @rate: rate to convert.
698  */
699 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
700 
701 /**
702  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
703  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
704  * @rate: rate to convert.
705  */
706 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
707 
708 
709 /**
710  * enum ib_mr_type - memory region type
711  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
712  *                            normal registration
713  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
714  *                            signature operations (data-integrity
715  *                            capable regions)
716  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
717  *                            register any arbitrary sg lists (without
718  *                            the normal mr constraints - see
719  *                            ib_map_mr_sg)
720  */
721 enum ib_mr_type {
722 	IB_MR_TYPE_MEM_REG,
723 	IB_MR_TYPE_SIGNATURE,
724 	IB_MR_TYPE_SG_GAPS,
725 };
726 
727 /**
728  * Signature types
729  * IB_SIG_TYPE_NONE: Unprotected.
730  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
731  */
732 enum ib_signature_type {
733 	IB_SIG_TYPE_NONE,
734 	IB_SIG_TYPE_T10_DIF,
735 };
736 
737 /**
738  * Signature T10-DIF block-guard types
739  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
740  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
741  */
742 enum ib_t10_dif_bg_type {
743 	IB_T10DIF_CRC,
744 	IB_T10DIF_CSUM
745 };
746 
747 /**
748  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
749  *     domain.
750  * @bg_type: T10-DIF block guard type (CRC|CSUM)
751  * @pi_interval: protection information interval.
752  * @bg: seed of guard computation.
753  * @app_tag: application tag of guard block
754  * @ref_tag: initial guard block reference tag.
755  * @ref_remap: Indicate wethear the reftag increments each block
756  * @app_escape: Indicate to skip block check if apptag=0xffff
757  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
758  * @apptag_check_mask: check bitmask of application tag.
759  */
760 struct ib_t10_dif_domain {
761 	enum ib_t10_dif_bg_type bg_type;
762 	u16			pi_interval;
763 	u16			bg;
764 	u16			app_tag;
765 	u32			ref_tag;
766 	bool			ref_remap;
767 	bool			app_escape;
768 	bool			ref_escape;
769 	u16			apptag_check_mask;
770 };
771 
772 /**
773  * struct ib_sig_domain - Parameters for signature domain
774  * @sig_type: specific signauture type
775  * @sig: union of all signature domain attributes that may
776  *     be used to set domain layout.
777  */
778 struct ib_sig_domain {
779 	enum ib_signature_type sig_type;
780 	union {
781 		struct ib_t10_dif_domain dif;
782 	} sig;
783 };
784 
785 /**
786  * struct ib_sig_attrs - Parameters for signature handover operation
787  * @check_mask: bitmask for signature byte check (8 bytes)
788  * @mem: memory domain layout desciptor.
789  * @wire: wire domain layout desciptor.
790  */
791 struct ib_sig_attrs {
792 	u8			check_mask;
793 	struct ib_sig_domain	mem;
794 	struct ib_sig_domain	wire;
795 };
796 
797 enum ib_sig_err_type {
798 	IB_SIG_BAD_GUARD,
799 	IB_SIG_BAD_REFTAG,
800 	IB_SIG_BAD_APPTAG,
801 };
802 
803 /**
804  * struct ib_sig_err - signature error descriptor
805  */
806 struct ib_sig_err {
807 	enum ib_sig_err_type	err_type;
808 	u32			expected;
809 	u32			actual;
810 	u64			sig_err_offset;
811 	u32			key;
812 };
813 
814 enum ib_mr_status_check {
815 	IB_MR_CHECK_SIG_STATUS = 1,
816 };
817 
818 /**
819  * struct ib_mr_status - Memory region status container
820  *
821  * @fail_status: Bitmask of MR checks status. For each
822  *     failed check a corresponding status bit is set.
823  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
824  *     failure.
825  */
826 struct ib_mr_status {
827 	u32		    fail_status;
828 	struct ib_sig_err   sig_err;
829 };
830 
831 /**
832  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
833  * enum.
834  * @mult: multiple to convert.
835  */
836 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
837 
838 struct ib_ah_attr {
839 	struct ib_global_route	grh;
840 	u16			dlid;
841 	u8			sl;
842 	u8			src_path_bits;
843 	u8			static_rate;
844 	u8			ah_flags;
845 	u8			port_num;
846 	u8			dmac[ETH_ALEN];
847 };
848 
849 enum ib_wc_status {
850 	IB_WC_SUCCESS,
851 	IB_WC_LOC_LEN_ERR,
852 	IB_WC_LOC_QP_OP_ERR,
853 	IB_WC_LOC_EEC_OP_ERR,
854 	IB_WC_LOC_PROT_ERR,
855 	IB_WC_WR_FLUSH_ERR,
856 	IB_WC_MW_BIND_ERR,
857 	IB_WC_BAD_RESP_ERR,
858 	IB_WC_LOC_ACCESS_ERR,
859 	IB_WC_REM_INV_REQ_ERR,
860 	IB_WC_REM_ACCESS_ERR,
861 	IB_WC_REM_OP_ERR,
862 	IB_WC_RETRY_EXC_ERR,
863 	IB_WC_RNR_RETRY_EXC_ERR,
864 	IB_WC_LOC_RDD_VIOL_ERR,
865 	IB_WC_REM_INV_RD_REQ_ERR,
866 	IB_WC_REM_ABORT_ERR,
867 	IB_WC_INV_EECN_ERR,
868 	IB_WC_INV_EEC_STATE_ERR,
869 	IB_WC_FATAL_ERR,
870 	IB_WC_RESP_TIMEOUT_ERR,
871 	IB_WC_GENERAL_ERR
872 };
873 
874 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
875 
876 enum ib_wc_opcode {
877 	IB_WC_SEND,
878 	IB_WC_RDMA_WRITE,
879 	IB_WC_RDMA_READ,
880 	IB_WC_COMP_SWAP,
881 	IB_WC_FETCH_ADD,
882 	IB_WC_LSO,
883 	IB_WC_LOCAL_INV,
884 	IB_WC_REG_MR,
885 	IB_WC_MASKED_COMP_SWAP,
886 	IB_WC_MASKED_FETCH_ADD,
887 /*
888  * Set value of IB_WC_RECV so consumers can test if a completion is a
889  * receive by testing (opcode & IB_WC_RECV).
890  */
891 	IB_WC_RECV			= 1 << 7,
892 	IB_WC_RECV_RDMA_WITH_IMM
893 };
894 
895 enum ib_wc_flags {
896 	IB_WC_GRH		= 1,
897 	IB_WC_WITH_IMM		= (1<<1),
898 	IB_WC_WITH_INVALIDATE	= (1<<2),
899 	IB_WC_IP_CSUM_OK	= (1<<3),
900 	IB_WC_WITH_SMAC		= (1<<4),
901 	IB_WC_WITH_VLAN		= (1<<5),
902 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
903 };
904 
905 struct ib_wc {
906 	union {
907 		u64		wr_id;
908 		struct ib_cqe	*wr_cqe;
909 	};
910 	enum ib_wc_status	status;
911 	enum ib_wc_opcode	opcode;
912 	u32			vendor_err;
913 	u32			byte_len;
914 	struct ib_qp	       *qp;
915 	union {
916 		__be32		imm_data;
917 		u32		invalidate_rkey;
918 	} ex;
919 	u32			src_qp;
920 	int			wc_flags;
921 	u16			pkey_index;
922 	u16			slid;
923 	u8			sl;
924 	u8			dlid_path_bits;
925 	u8			port_num;	/* valid only for DR SMPs on switches */
926 	u8			smac[ETH_ALEN];
927 	u16			vlan_id;
928 	u8			network_hdr_type;
929 };
930 
931 enum ib_cq_notify_flags {
932 	IB_CQ_SOLICITED			= 1 << 0,
933 	IB_CQ_NEXT_COMP			= 1 << 1,
934 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
935 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
936 };
937 
938 enum ib_srq_type {
939 	IB_SRQT_BASIC,
940 	IB_SRQT_XRC
941 };
942 
943 enum ib_srq_attr_mask {
944 	IB_SRQ_MAX_WR	= 1 << 0,
945 	IB_SRQ_LIMIT	= 1 << 1,
946 };
947 
948 struct ib_srq_attr {
949 	u32	max_wr;
950 	u32	max_sge;
951 	u32	srq_limit;
952 };
953 
954 struct ib_srq_init_attr {
955 	void		      (*event_handler)(struct ib_event *, void *);
956 	void		       *srq_context;
957 	struct ib_srq_attr	attr;
958 	enum ib_srq_type	srq_type;
959 
960 	union {
961 		struct {
962 			struct ib_xrcd *xrcd;
963 			struct ib_cq   *cq;
964 		} xrc;
965 	} ext;
966 };
967 
968 struct ib_qp_cap {
969 	u32	max_send_wr;
970 	u32	max_recv_wr;
971 	u32	max_send_sge;
972 	u32	max_recv_sge;
973 	u32	max_inline_data;
974 
975 	/*
976 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
977 	 * ib_create_qp() will calculate the right amount of neededed WRs
978 	 * and MRs based on this.
979 	 */
980 	u32	max_rdma_ctxs;
981 };
982 
983 enum ib_sig_type {
984 	IB_SIGNAL_ALL_WR,
985 	IB_SIGNAL_REQ_WR
986 };
987 
988 enum ib_qp_type {
989 	/*
990 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
991 	 * here (and in that order) since the MAD layer uses them as
992 	 * indices into a 2-entry table.
993 	 */
994 	IB_QPT_SMI,
995 	IB_QPT_GSI,
996 
997 	IB_QPT_RC,
998 	IB_QPT_UC,
999 	IB_QPT_UD,
1000 	IB_QPT_RAW_IPV6,
1001 	IB_QPT_RAW_ETHERTYPE,
1002 	IB_QPT_RAW_PACKET = 8,
1003 	IB_QPT_XRC_INI = 9,
1004 	IB_QPT_XRC_TGT,
1005 	IB_QPT_MAX,
1006 	/* Reserve a range for qp types internal to the low level driver.
1007 	 * These qp types will not be visible at the IB core layer, so the
1008 	 * IB_QPT_MAX usages should not be affected in the core layer
1009 	 */
1010 	IB_QPT_RESERVED1 = 0x1000,
1011 	IB_QPT_RESERVED2,
1012 	IB_QPT_RESERVED3,
1013 	IB_QPT_RESERVED4,
1014 	IB_QPT_RESERVED5,
1015 	IB_QPT_RESERVED6,
1016 	IB_QPT_RESERVED7,
1017 	IB_QPT_RESERVED8,
1018 	IB_QPT_RESERVED9,
1019 	IB_QPT_RESERVED10,
1020 };
1021 
1022 enum ib_qp_create_flags {
1023 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1024 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1025 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1026 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1027 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1028 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1029 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1030 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
1031 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1032 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1033 	/* reserve bits 26-31 for low level drivers' internal use */
1034 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1035 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1036 };
1037 
1038 /*
1039  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1040  * callback to destroy the passed in QP.
1041  */
1042 
1043 struct ib_qp_init_attr {
1044 	void                  (*event_handler)(struct ib_event *, void *);
1045 	void		       *qp_context;
1046 	struct ib_cq	       *send_cq;
1047 	struct ib_cq	       *recv_cq;
1048 	struct ib_srq	       *srq;
1049 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1050 	struct ib_qp_cap	cap;
1051 	enum ib_sig_type	sq_sig_type;
1052 	enum ib_qp_type		qp_type;
1053 	enum ib_qp_create_flags	create_flags;
1054 
1055 	/*
1056 	 * Only needed for special QP types, or when using the RW API.
1057 	 */
1058 	u8			port_num;
1059 	struct ib_rwq_ind_table *rwq_ind_tbl;
1060 };
1061 
1062 struct ib_qp_open_attr {
1063 	void                  (*event_handler)(struct ib_event *, void *);
1064 	void		       *qp_context;
1065 	u32			qp_num;
1066 	enum ib_qp_type		qp_type;
1067 };
1068 
1069 enum ib_rnr_timeout {
1070 	IB_RNR_TIMER_655_36 =  0,
1071 	IB_RNR_TIMER_000_01 =  1,
1072 	IB_RNR_TIMER_000_02 =  2,
1073 	IB_RNR_TIMER_000_03 =  3,
1074 	IB_RNR_TIMER_000_04 =  4,
1075 	IB_RNR_TIMER_000_06 =  5,
1076 	IB_RNR_TIMER_000_08 =  6,
1077 	IB_RNR_TIMER_000_12 =  7,
1078 	IB_RNR_TIMER_000_16 =  8,
1079 	IB_RNR_TIMER_000_24 =  9,
1080 	IB_RNR_TIMER_000_32 = 10,
1081 	IB_RNR_TIMER_000_48 = 11,
1082 	IB_RNR_TIMER_000_64 = 12,
1083 	IB_RNR_TIMER_000_96 = 13,
1084 	IB_RNR_TIMER_001_28 = 14,
1085 	IB_RNR_TIMER_001_92 = 15,
1086 	IB_RNR_TIMER_002_56 = 16,
1087 	IB_RNR_TIMER_003_84 = 17,
1088 	IB_RNR_TIMER_005_12 = 18,
1089 	IB_RNR_TIMER_007_68 = 19,
1090 	IB_RNR_TIMER_010_24 = 20,
1091 	IB_RNR_TIMER_015_36 = 21,
1092 	IB_RNR_TIMER_020_48 = 22,
1093 	IB_RNR_TIMER_030_72 = 23,
1094 	IB_RNR_TIMER_040_96 = 24,
1095 	IB_RNR_TIMER_061_44 = 25,
1096 	IB_RNR_TIMER_081_92 = 26,
1097 	IB_RNR_TIMER_122_88 = 27,
1098 	IB_RNR_TIMER_163_84 = 28,
1099 	IB_RNR_TIMER_245_76 = 29,
1100 	IB_RNR_TIMER_327_68 = 30,
1101 	IB_RNR_TIMER_491_52 = 31
1102 };
1103 
1104 enum ib_qp_attr_mask {
1105 	IB_QP_STATE			= 1,
1106 	IB_QP_CUR_STATE			= (1<<1),
1107 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1108 	IB_QP_ACCESS_FLAGS		= (1<<3),
1109 	IB_QP_PKEY_INDEX		= (1<<4),
1110 	IB_QP_PORT			= (1<<5),
1111 	IB_QP_QKEY			= (1<<6),
1112 	IB_QP_AV			= (1<<7),
1113 	IB_QP_PATH_MTU			= (1<<8),
1114 	IB_QP_TIMEOUT			= (1<<9),
1115 	IB_QP_RETRY_CNT			= (1<<10),
1116 	IB_QP_RNR_RETRY			= (1<<11),
1117 	IB_QP_RQ_PSN			= (1<<12),
1118 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1119 	IB_QP_ALT_PATH			= (1<<14),
1120 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1121 	IB_QP_SQ_PSN			= (1<<16),
1122 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1123 	IB_QP_PATH_MIG_STATE		= (1<<18),
1124 	IB_QP_CAP			= (1<<19),
1125 	IB_QP_DEST_QPN			= (1<<20),
1126 	IB_QP_RESERVED1			= (1<<21),
1127 	IB_QP_RESERVED2			= (1<<22),
1128 	IB_QP_RESERVED3			= (1<<23),
1129 	IB_QP_RESERVED4			= (1<<24),
1130 	IB_QP_RATE_LIMIT		= (1<<25),
1131 };
1132 
1133 enum ib_qp_state {
1134 	IB_QPS_RESET,
1135 	IB_QPS_INIT,
1136 	IB_QPS_RTR,
1137 	IB_QPS_RTS,
1138 	IB_QPS_SQD,
1139 	IB_QPS_SQE,
1140 	IB_QPS_ERR
1141 };
1142 
1143 enum ib_mig_state {
1144 	IB_MIG_MIGRATED,
1145 	IB_MIG_REARM,
1146 	IB_MIG_ARMED
1147 };
1148 
1149 enum ib_mw_type {
1150 	IB_MW_TYPE_1 = 1,
1151 	IB_MW_TYPE_2 = 2
1152 };
1153 
1154 struct ib_qp_attr {
1155 	enum ib_qp_state	qp_state;
1156 	enum ib_qp_state	cur_qp_state;
1157 	enum ib_mtu		path_mtu;
1158 	enum ib_mig_state	path_mig_state;
1159 	u32			qkey;
1160 	u32			rq_psn;
1161 	u32			sq_psn;
1162 	u32			dest_qp_num;
1163 	int			qp_access_flags;
1164 	struct ib_qp_cap	cap;
1165 	struct ib_ah_attr	ah_attr;
1166 	struct ib_ah_attr	alt_ah_attr;
1167 	u16			pkey_index;
1168 	u16			alt_pkey_index;
1169 	u8			en_sqd_async_notify;
1170 	u8			sq_draining;
1171 	u8			max_rd_atomic;
1172 	u8			max_dest_rd_atomic;
1173 	u8			min_rnr_timer;
1174 	u8			port_num;
1175 	u8			timeout;
1176 	u8			retry_cnt;
1177 	u8			rnr_retry;
1178 	u8			alt_port_num;
1179 	u8			alt_timeout;
1180 	u32			rate_limit;
1181 };
1182 
1183 enum ib_wr_opcode {
1184 	IB_WR_RDMA_WRITE,
1185 	IB_WR_RDMA_WRITE_WITH_IMM,
1186 	IB_WR_SEND,
1187 	IB_WR_SEND_WITH_IMM,
1188 	IB_WR_RDMA_READ,
1189 	IB_WR_ATOMIC_CMP_AND_SWP,
1190 	IB_WR_ATOMIC_FETCH_AND_ADD,
1191 	IB_WR_LSO,
1192 	IB_WR_SEND_WITH_INV,
1193 	IB_WR_RDMA_READ_WITH_INV,
1194 	IB_WR_LOCAL_INV,
1195 	IB_WR_REG_MR,
1196 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1197 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1198 	IB_WR_REG_SIG_MR,
1199 	/* reserve values for low level drivers' internal use.
1200 	 * These values will not be used at all in the ib core layer.
1201 	 */
1202 	IB_WR_RESERVED1 = 0xf0,
1203 	IB_WR_RESERVED2,
1204 	IB_WR_RESERVED3,
1205 	IB_WR_RESERVED4,
1206 	IB_WR_RESERVED5,
1207 	IB_WR_RESERVED6,
1208 	IB_WR_RESERVED7,
1209 	IB_WR_RESERVED8,
1210 	IB_WR_RESERVED9,
1211 	IB_WR_RESERVED10,
1212 };
1213 
1214 enum ib_send_flags {
1215 	IB_SEND_FENCE		= 1,
1216 	IB_SEND_SIGNALED	= (1<<1),
1217 	IB_SEND_SOLICITED	= (1<<2),
1218 	IB_SEND_INLINE		= (1<<3),
1219 	IB_SEND_IP_CSUM		= (1<<4),
1220 
1221 	/* reserve bits 26-31 for low level drivers' internal use */
1222 	IB_SEND_RESERVED_START	= (1 << 26),
1223 	IB_SEND_RESERVED_END	= (1 << 31),
1224 };
1225 
1226 struct ib_sge {
1227 	u64	addr;
1228 	u32	length;
1229 	u32	lkey;
1230 };
1231 
1232 struct ib_cqe {
1233 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1234 };
1235 
1236 struct ib_send_wr {
1237 	struct ib_send_wr      *next;
1238 	union {
1239 		u64		wr_id;
1240 		struct ib_cqe	*wr_cqe;
1241 	};
1242 	struct ib_sge	       *sg_list;
1243 	int			num_sge;
1244 	enum ib_wr_opcode	opcode;
1245 	int			send_flags;
1246 	union {
1247 		__be32		imm_data;
1248 		u32		invalidate_rkey;
1249 	} ex;
1250 };
1251 
1252 struct ib_rdma_wr {
1253 	struct ib_send_wr	wr;
1254 	u64			remote_addr;
1255 	u32			rkey;
1256 };
1257 
1258 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1259 {
1260 	return container_of(wr, struct ib_rdma_wr, wr);
1261 }
1262 
1263 struct ib_atomic_wr {
1264 	struct ib_send_wr	wr;
1265 	u64			remote_addr;
1266 	u64			compare_add;
1267 	u64			swap;
1268 	u64			compare_add_mask;
1269 	u64			swap_mask;
1270 	u32			rkey;
1271 };
1272 
1273 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1274 {
1275 	return container_of(wr, struct ib_atomic_wr, wr);
1276 }
1277 
1278 struct ib_ud_wr {
1279 	struct ib_send_wr	wr;
1280 	struct ib_ah		*ah;
1281 	void			*header;
1282 	int			hlen;
1283 	int			mss;
1284 	u32			remote_qpn;
1285 	u32			remote_qkey;
1286 	u16			pkey_index; /* valid for GSI only */
1287 	u8			port_num;   /* valid for DR SMPs on switch only */
1288 };
1289 
1290 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1291 {
1292 	return container_of(wr, struct ib_ud_wr, wr);
1293 }
1294 
1295 struct ib_reg_wr {
1296 	struct ib_send_wr	wr;
1297 	struct ib_mr		*mr;
1298 	u32			key;
1299 	int			access;
1300 };
1301 
1302 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1303 {
1304 	return container_of(wr, struct ib_reg_wr, wr);
1305 }
1306 
1307 struct ib_sig_handover_wr {
1308 	struct ib_send_wr	wr;
1309 	struct ib_sig_attrs    *sig_attrs;
1310 	struct ib_mr	       *sig_mr;
1311 	int			access_flags;
1312 	struct ib_sge	       *prot;
1313 };
1314 
1315 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1316 {
1317 	return container_of(wr, struct ib_sig_handover_wr, wr);
1318 }
1319 
1320 struct ib_recv_wr {
1321 	struct ib_recv_wr      *next;
1322 	union {
1323 		u64		wr_id;
1324 		struct ib_cqe	*wr_cqe;
1325 	};
1326 	struct ib_sge	       *sg_list;
1327 	int			num_sge;
1328 };
1329 
1330 enum ib_access_flags {
1331 	IB_ACCESS_LOCAL_WRITE	= 1,
1332 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1333 	IB_ACCESS_REMOTE_READ	= (1<<2),
1334 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1335 	IB_ACCESS_MW_BIND	= (1<<4),
1336 	IB_ZERO_BASED		= (1<<5),
1337 	IB_ACCESS_ON_DEMAND     = (1<<6),
1338 };
1339 
1340 /*
1341  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1342  * are hidden here instead of a uapi header!
1343  */
1344 enum ib_mr_rereg_flags {
1345 	IB_MR_REREG_TRANS	= 1,
1346 	IB_MR_REREG_PD		= (1<<1),
1347 	IB_MR_REREG_ACCESS	= (1<<2),
1348 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1349 };
1350 
1351 struct ib_fmr_attr {
1352 	int	max_pages;
1353 	int	max_maps;
1354 	u8	page_shift;
1355 };
1356 
1357 struct ib_umem;
1358 
1359 struct ib_ucontext {
1360 	struct ib_device       *device;
1361 	struct list_head	pd_list;
1362 	struct list_head	mr_list;
1363 	struct list_head	mw_list;
1364 	struct list_head	cq_list;
1365 	struct list_head	qp_list;
1366 	struct list_head	srq_list;
1367 	struct list_head	ah_list;
1368 	struct list_head	xrcd_list;
1369 	struct list_head	rule_list;
1370 	struct list_head	wq_list;
1371 	struct list_head	rwq_ind_tbl_list;
1372 	int			closing;
1373 
1374 	struct pid             *tgid;
1375 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1376 	struct rb_root      umem_tree;
1377 	/*
1378 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1379 	 * mmu notifiers registration.
1380 	 */
1381 	struct rw_semaphore	umem_rwsem;
1382 	void (*invalidate_range)(struct ib_umem *umem,
1383 				 unsigned long start, unsigned long end);
1384 
1385 	struct mmu_notifier	mn;
1386 	atomic_t		notifier_count;
1387 	/* A list of umems that don't have private mmu notifier counters yet. */
1388 	struct list_head	no_private_counters;
1389 	int                     odp_mrs_count;
1390 #endif
1391 };
1392 
1393 struct ib_uobject {
1394 	u64			user_handle;	/* handle given to us by userspace */
1395 	struct ib_ucontext     *context;	/* associated user context */
1396 	void		       *object;		/* containing object */
1397 	struct list_head	list;		/* link to context's list */
1398 	int			id;		/* index into kernel idr */
1399 	struct kref		ref;
1400 	struct rw_semaphore	mutex;		/* protects .live */
1401 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1402 	int			live;
1403 };
1404 
1405 struct ib_udata {
1406 	const void __user *inbuf;
1407 	void __user *outbuf;
1408 	size_t       inlen;
1409 	size_t       outlen;
1410 };
1411 
1412 struct ib_pd {
1413 	u32			local_dma_lkey;
1414 	u32			flags;
1415 	struct ib_device       *device;
1416 	struct ib_uobject      *uobject;
1417 	atomic_t          	usecnt; /* count all resources */
1418 
1419 	u32			unsafe_global_rkey;
1420 
1421 	/*
1422 	 * Implementation details of the RDMA core, don't use in drivers:
1423 	 */
1424 	struct ib_mr	       *__internal_mr;
1425 };
1426 
1427 struct ib_xrcd {
1428 	struct ib_device       *device;
1429 	atomic_t		usecnt; /* count all exposed resources */
1430 	struct inode	       *inode;
1431 
1432 	struct mutex		tgt_qp_mutex;
1433 	struct list_head	tgt_qp_list;
1434 };
1435 
1436 struct ib_ah {
1437 	struct ib_device	*device;
1438 	struct ib_pd		*pd;
1439 	struct ib_uobject	*uobject;
1440 };
1441 
1442 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1443 
1444 enum ib_poll_context {
1445 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1446 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1447 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1448 };
1449 
1450 struct ib_cq {
1451 	struct ib_device       *device;
1452 	struct ib_uobject      *uobject;
1453 	ib_comp_handler   	comp_handler;
1454 	void                  (*event_handler)(struct ib_event *, void *);
1455 	void                   *cq_context;
1456 	int               	cqe;
1457 	atomic_t          	usecnt; /* count number of work queues */
1458 	enum ib_poll_context	poll_ctx;
1459 	struct ib_wc		*wc;
1460 	union {
1461 		struct irq_poll		iop;
1462 		struct work_struct	work;
1463 	};
1464 };
1465 
1466 struct ib_srq {
1467 	struct ib_device       *device;
1468 	struct ib_pd	       *pd;
1469 	struct ib_uobject      *uobject;
1470 	void		      (*event_handler)(struct ib_event *, void *);
1471 	void		       *srq_context;
1472 	enum ib_srq_type	srq_type;
1473 	atomic_t		usecnt;
1474 
1475 	union {
1476 		struct {
1477 			struct ib_xrcd *xrcd;
1478 			struct ib_cq   *cq;
1479 			u32		srq_num;
1480 		} xrc;
1481 	} ext;
1482 };
1483 
1484 enum ib_raw_packet_caps {
1485 	/* Strip cvlan from incoming packet and report it in the matching work
1486 	 * completion is supported.
1487 	 */
1488 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1489 	/* Scatter FCS field of an incoming packet to host memory is supported.
1490 	 */
1491 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1492 	/* Checksum offloads are supported (for both send and receive). */
1493 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1494 };
1495 
1496 enum ib_wq_type {
1497 	IB_WQT_RQ
1498 };
1499 
1500 enum ib_wq_state {
1501 	IB_WQS_RESET,
1502 	IB_WQS_RDY,
1503 	IB_WQS_ERR
1504 };
1505 
1506 struct ib_wq {
1507 	struct ib_device       *device;
1508 	struct ib_uobject      *uobject;
1509 	void		    *wq_context;
1510 	void		    (*event_handler)(struct ib_event *, void *);
1511 	struct ib_pd	       *pd;
1512 	struct ib_cq	       *cq;
1513 	u32		wq_num;
1514 	enum ib_wq_state       state;
1515 	enum ib_wq_type	wq_type;
1516 	atomic_t		usecnt;
1517 };
1518 
1519 enum ib_wq_flags {
1520 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1521 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1522 };
1523 
1524 struct ib_wq_init_attr {
1525 	void		       *wq_context;
1526 	enum ib_wq_type	wq_type;
1527 	u32		max_wr;
1528 	u32		max_sge;
1529 	struct	ib_cq	       *cq;
1530 	void		    (*event_handler)(struct ib_event *, void *);
1531 	u32		create_flags; /* Use enum ib_wq_flags */
1532 };
1533 
1534 enum ib_wq_attr_mask {
1535 	IB_WQ_STATE		= 1 << 0,
1536 	IB_WQ_CUR_STATE		= 1 << 1,
1537 	IB_WQ_FLAGS		= 1 << 2,
1538 };
1539 
1540 struct ib_wq_attr {
1541 	enum	ib_wq_state	wq_state;
1542 	enum	ib_wq_state	curr_wq_state;
1543 	u32			flags; /* Use enum ib_wq_flags */
1544 	u32			flags_mask; /* Use enum ib_wq_flags */
1545 };
1546 
1547 struct ib_rwq_ind_table {
1548 	struct ib_device	*device;
1549 	struct ib_uobject      *uobject;
1550 	atomic_t		usecnt;
1551 	u32		ind_tbl_num;
1552 	u32		log_ind_tbl_size;
1553 	struct ib_wq	**ind_tbl;
1554 };
1555 
1556 struct ib_rwq_ind_table_init_attr {
1557 	u32		log_ind_tbl_size;
1558 	/* Each entry is a pointer to Receive Work Queue */
1559 	struct ib_wq	**ind_tbl;
1560 };
1561 
1562 /*
1563  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1564  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1565  */
1566 struct ib_qp {
1567 	struct ib_device       *device;
1568 	struct ib_pd	       *pd;
1569 	struct ib_cq	       *send_cq;
1570 	struct ib_cq	       *recv_cq;
1571 	spinlock_t		mr_lock;
1572 	int			mrs_used;
1573 	struct list_head	rdma_mrs;
1574 	struct list_head	sig_mrs;
1575 	struct ib_srq	       *srq;
1576 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1577 	struct list_head	xrcd_list;
1578 
1579 	/* count times opened, mcast attaches, flow attaches */
1580 	atomic_t		usecnt;
1581 	struct list_head	open_list;
1582 	struct ib_qp           *real_qp;
1583 	struct ib_uobject      *uobject;
1584 	void                  (*event_handler)(struct ib_event *, void *);
1585 	void		       *qp_context;
1586 	u32			qp_num;
1587 	u32			max_write_sge;
1588 	u32			max_read_sge;
1589 	enum ib_qp_type		qp_type;
1590 	struct ib_rwq_ind_table *rwq_ind_tbl;
1591 };
1592 
1593 struct ib_mr {
1594 	struct ib_device  *device;
1595 	struct ib_pd	  *pd;
1596 	u32		   lkey;
1597 	u32		   rkey;
1598 	u64		   iova;
1599 	u32		   length;
1600 	unsigned int	   page_size;
1601 	bool		   need_inval;
1602 	union {
1603 		struct ib_uobject	*uobject;	/* user */
1604 		struct list_head	qp_entry;	/* FR */
1605 	};
1606 };
1607 
1608 struct ib_mw {
1609 	struct ib_device	*device;
1610 	struct ib_pd		*pd;
1611 	struct ib_uobject	*uobject;
1612 	u32			rkey;
1613 	enum ib_mw_type         type;
1614 };
1615 
1616 struct ib_fmr {
1617 	struct ib_device	*device;
1618 	struct ib_pd		*pd;
1619 	struct list_head	list;
1620 	u32			lkey;
1621 	u32			rkey;
1622 };
1623 
1624 /* Supported steering options */
1625 enum ib_flow_attr_type {
1626 	/* steering according to rule specifications */
1627 	IB_FLOW_ATTR_NORMAL		= 0x0,
1628 	/* default unicast and multicast rule -
1629 	 * receive all Eth traffic which isn't steered to any QP
1630 	 */
1631 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1632 	/* default multicast rule -
1633 	 * receive all Eth multicast traffic which isn't steered to any QP
1634 	 */
1635 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1636 	/* sniffer rule - receive all port traffic */
1637 	IB_FLOW_ATTR_SNIFFER		= 0x3
1638 };
1639 
1640 /* Supported steering header types */
1641 enum ib_flow_spec_type {
1642 	/* L2 headers*/
1643 	IB_FLOW_SPEC_ETH		= 0x20,
1644 	IB_FLOW_SPEC_IB			= 0x22,
1645 	/* L3 header*/
1646 	IB_FLOW_SPEC_IPV4		= 0x30,
1647 	IB_FLOW_SPEC_IPV6		= 0x31,
1648 	/* L4 headers*/
1649 	IB_FLOW_SPEC_TCP		= 0x40,
1650 	IB_FLOW_SPEC_UDP		= 0x41,
1651 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1652 	IB_FLOW_SPEC_INNER		= 0x100,
1653 	/* Actions */
1654 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1655 };
1656 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1657 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1658 
1659 /* Flow steering rule priority is set according to it's domain.
1660  * Lower domain value means higher priority.
1661  */
1662 enum ib_flow_domain {
1663 	IB_FLOW_DOMAIN_USER,
1664 	IB_FLOW_DOMAIN_ETHTOOL,
1665 	IB_FLOW_DOMAIN_RFS,
1666 	IB_FLOW_DOMAIN_NIC,
1667 	IB_FLOW_DOMAIN_NUM /* Must be last */
1668 };
1669 
1670 enum ib_flow_flags {
1671 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1672 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1673 };
1674 
1675 struct ib_flow_eth_filter {
1676 	u8	dst_mac[6];
1677 	u8	src_mac[6];
1678 	__be16	ether_type;
1679 	__be16	vlan_tag;
1680 	/* Must be last */
1681 	u8	real_sz[0];
1682 };
1683 
1684 struct ib_flow_spec_eth {
1685 	u32			  type;
1686 	u16			  size;
1687 	struct ib_flow_eth_filter val;
1688 	struct ib_flow_eth_filter mask;
1689 };
1690 
1691 struct ib_flow_ib_filter {
1692 	__be16 dlid;
1693 	__u8   sl;
1694 	/* Must be last */
1695 	u8	real_sz[0];
1696 };
1697 
1698 struct ib_flow_spec_ib {
1699 	u32			 type;
1700 	u16			 size;
1701 	struct ib_flow_ib_filter val;
1702 	struct ib_flow_ib_filter mask;
1703 };
1704 
1705 /* IPv4 header flags */
1706 enum ib_ipv4_flags {
1707 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1708 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1709 				    last have this flag set */
1710 };
1711 
1712 struct ib_flow_ipv4_filter {
1713 	__be32	src_ip;
1714 	__be32	dst_ip;
1715 	u8	proto;
1716 	u8	tos;
1717 	u8	ttl;
1718 	u8	flags;
1719 	/* Must be last */
1720 	u8	real_sz[0];
1721 };
1722 
1723 struct ib_flow_spec_ipv4 {
1724 	u32			   type;
1725 	u16			   size;
1726 	struct ib_flow_ipv4_filter val;
1727 	struct ib_flow_ipv4_filter mask;
1728 };
1729 
1730 struct ib_flow_ipv6_filter {
1731 	u8	src_ip[16];
1732 	u8	dst_ip[16];
1733 	__be32	flow_label;
1734 	u8	next_hdr;
1735 	u8	traffic_class;
1736 	u8	hop_limit;
1737 	/* Must be last */
1738 	u8	real_sz[0];
1739 };
1740 
1741 struct ib_flow_spec_ipv6 {
1742 	u32			   type;
1743 	u16			   size;
1744 	struct ib_flow_ipv6_filter val;
1745 	struct ib_flow_ipv6_filter mask;
1746 };
1747 
1748 struct ib_flow_tcp_udp_filter {
1749 	__be16	dst_port;
1750 	__be16	src_port;
1751 	/* Must be last */
1752 	u8	real_sz[0];
1753 };
1754 
1755 struct ib_flow_spec_tcp_udp {
1756 	u32			      type;
1757 	u16			      size;
1758 	struct ib_flow_tcp_udp_filter val;
1759 	struct ib_flow_tcp_udp_filter mask;
1760 };
1761 
1762 struct ib_flow_tunnel_filter {
1763 	__be32	tunnel_id;
1764 	u8	real_sz[0];
1765 };
1766 
1767 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1768  * the tunnel_id from val has the vni value
1769  */
1770 struct ib_flow_spec_tunnel {
1771 	u32			      type;
1772 	u16			      size;
1773 	struct ib_flow_tunnel_filter  val;
1774 	struct ib_flow_tunnel_filter  mask;
1775 };
1776 
1777 struct ib_flow_spec_action_tag {
1778 	enum ib_flow_spec_type	      type;
1779 	u16			      size;
1780 	u32                           tag_id;
1781 };
1782 
1783 union ib_flow_spec {
1784 	struct {
1785 		u32			type;
1786 		u16			size;
1787 	};
1788 	struct ib_flow_spec_eth		eth;
1789 	struct ib_flow_spec_ib		ib;
1790 	struct ib_flow_spec_ipv4        ipv4;
1791 	struct ib_flow_spec_tcp_udp	tcp_udp;
1792 	struct ib_flow_spec_ipv6        ipv6;
1793 	struct ib_flow_spec_tunnel      tunnel;
1794 	struct ib_flow_spec_action_tag  flow_tag;
1795 };
1796 
1797 struct ib_flow_attr {
1798 	enum ib_flow_attr_type type;
1799 	u16	     size;
1800 	u16	     priority;
1801 	u32	     flags;
1802 	u8	     num_of_specs;
1803 	u8	     port;
1804 	/* Following are the optional layers according to user request
1805 	 * struct ib_flow_spec_xxx
1806 	 * struct ib_flow_spec_yyy
1807 	 */
1808 };
1809 
1810 struct ib_flow {
1811 	struct ib_qp		*qp;
1812 	struct ib_uobject	*uobject;
1813 };
1814 
1815 struct ib_mad_hdr;
1816 struct ib_grh;
1817 
1818 enum ib_process_mad_flags {
1819 	IB_MAD_IGNORE_MKEY	= 1,
1820 	IB_MAD_IGNORE_BKEY	= 2,
1821 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1822 };
1823 
1824 enum ib_mad_result {
1825 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1826 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1827 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1828 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1829 };
1830 
1831 #define IB_DEVICE_NAME_MAX 64
1832 
1833 struct ib_port_cache {
1834 	struct ib_pkey_cache  *pkey;
1835 	struct ib_gid_table   *gid;
1836 	u8                     lmc;
1837 	enum ib_port_state     port_state;
1838 };
1839 
1840 struct ib_cache {
1841 	rwlock_t                lock;
1842 	struct ib_event_handler event_handler;
1843 	struct ib_port_cache   *ports;
1844 };
1845 
1846 struct ib_dma_mapping_ops {
1847 	int		(*mapping_error)(struct ib_device *dev,
1848 					 u64 dma_addr);
1849 	u64		(*map_single)(struct ib_device *dev,
1850 				      void *ptr, size_t size,
1851 				      enum dma_data_direction direction);
1852 	void		(*unmap_single)(struct ib_device *dev,
1853 					u64 addr, size_t size,
1854 					enum dma_data_direction direction);
1855 	u64		(*map_page)(struct ib_device *dev,
1856 				    struct page *page, unsigned long offset,
1857 				    size_t size,
1858 				    enum dma_data_direction direction);
1859 	void		(*unmap_page)(struct ib_device *dev,
1860 				      u64 addr, size_t size,
1861 				      enum dma_data_direction direction);
1862 	int		(*map_sg)(struct ib_device *dev,
1863 				  struct scatterlist *sg, int nents,
1864 				  enum dma_data_direction direction);
1865 	void		(*unmap_sg)(struct ib_device *dev,
1866 				    struct scatterlist *sg, int nents,
1867 				    enum dma_data_direction direction);
1868 	int		(*map_sg_attrs)(struct ib_device *dev,
1869 					struct scatterlist *sg, int nents,
1870 					enum dma_data_direction direction,
1871 					unsigned long attrs);
1872 	void		(*unmap_sg_attrs)(struct ib_device *dev,
1873 					  struct scatterlist *sg, int nents,
1874 					  enum dma_data_direction direction,
1875 					  unsigned long attrs);
1876 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1877 					       u64 dma_handle,
1878 					       size_t size,
1879 					       enum dma_data_direction dir);
1880 	void		(*sync_single_for_device)(struct ib_device *dev,
1881 						  u64 dma_handle,
1882 						  size_t size,
1883 						  enum dma_data_direction dir);
1884 	void		*(*alloc_coherent)(struct ib_device *dev,
1885 					   size_t size,
1886 					   u64 *dma_handle,
1887 					   gfp_t flag);
1888 	void		(*free_coherent)(struct ib_device *dev,
1889 					 size_t size, void *cpu_addr,
1890 					 u64 dma_handle);
1891 };
1892 
1893 struct iw_cm_verbs;
1894 
1895 struct ib_port_immutable {
1896 	int                           pkey_tbl_len;
1897 	int                           gid_tbl_len;
1898 	u32                           core_cap_flags;
1899 	u32                           max_mad_size;
1900 };
1901 
1902 struct ib_device {
1903 	struct device                *dma_device;
1904 
1905 	char                          name[IB_DEVICE_NAME_MAX];
1906 
1907 	struct list_head              event_handler_list;
1908 	spinlock_t                    event_handler_lock;
1909 
1910 	spinlock_t                    client_data_lock;
1911 	struct list_head              core_list;
1912 	/* Access to the client_data_list is protected by the client_data_lock
1913 	 * spinlock and the lists_rwsem read-write semaphore */
1914 	struct list_head              client_data_list;
1915 
1916 	struct ib_cache               cache;
1917 	/**
1918 	 * port_immutable is indexed by port number
1919 	 */
1920 	struct ib_port_immutable     *port_immutable;
1921 
1922 	int			      num_comp_vectors;
1923 
1924 	struct iw_cm_verbs	     *iwcm;
1925 
1926 	/**
1927 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1928 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
1929 	 *   core when the device is removed.  A lifespan of -1 in the return
1930 	 *   struct tells the core to set a default lifespan.
1931 	 */
1932 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
1933 						     u8 port_num);
1934 	/**
1935 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1936 	 * @index - The index in the value array we wish to have updated, or
1937 	 *   num_counters if we want all stats updated
1938 	 * Return codes -
1939 	 *   < 0 - Error, no counters updated
1940 	 *   index - Updated the single counter pointed to by index
1941 	 *   num_counters - Updated all counters (will reset the timestamp
1942 	 *     and prevent further calls for lifespan milliseconds)
1943 	 * Drivers are allowed to update all counters in leiu of just the
1944 	 *   one given in index at their option
1945 	 */
1946 	int		           (*get_hw_stats)(struct ib_device *device,
1947 						   struct rdma_hw_stats *stats,
1948 						   u8 port, int index);
1949 	int		           (*query_device)(struct ib_device *device,
1950 						   struct ib_device_attr *device_attr,
1951 						   struct ib_udata *udata);
1952 	int		           (*query_port)(struct ib_device *device,
1953 						 u8 port_num,
1954 						 struct ib_port_attr *port_attr);
1955 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1956 						     u8 port_num);
1957 	/* When calling get_netdev, the HW vendor's driver should return the
1958 	 * net device of device @device at port @port_num or NULL if such
1959 	 * a net device doesn't exist. The vendor driver should call dev_hold
1960 	 * on this net device. The HW vendor's device driver must guarantee
1961 	 * that this function returns NULL before the net device reaches
1962 	 * NETDEV_UNREGISTER_FINAL state.
1963 	 */
1964 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1965 						 u8 port_num);
1966 	int		           (*query_gid)(struct ib_device *device,
1967 						u8 port_num, int index,
1968 						union ib_gid *gid);
1969 	/* When calling add_gid, the HW vendor's driver should
1970 	 * add the gid of device @device at gid index @index of
1971 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1972 	 * the network device related to this gid is available
1973 	 * at @attr. @context allows the HW vendor driver to store extra
1974 	 * information together with a GID entry. The HW vendor may allocate
1975 	 * memory to contain this information and store it in @context when a
1976 	 * new GID entry is written to. Params are consistent until the next
1977 	 * call of add_gid or delete_gid. The function should return 0 on
1978 	 * success or error otherwise. The function could be called
1979 	 * concurrently for different ports. This function is only called
1980 	 * when roce_gid_table is used.
1981 	 */
1982 	int		           (*add_gid)(struct ib_device *device,
1983 					      u8 port_num,
1984 					      unsigned int index,
1985 					      const union ib_gid *gid,
1986 					      const struct ib_gid_attr *attr,
1987 					      void **context);
1988 	/* When calling del_gid, the HW vendor's driver should delete the
1989 	 * gid of device @device at gid index @index of port @port_num.
1990 	 * Upon the deletion of a GID entry, the HW vendor must free any
1991 	 * allocated memory. The caller will clear @context afterwards.
1992 	 * This function is only called when roce_gid_table is used.
1993 	 */
1994 	int		           (*del_gid)(struct ib_device *device,
1995 					      u8 port_num,
1996 					      unsigned int index,
1997 					      void **context);
1998 	int		           (*query_pkey)(struct ib_device *device,
1999 						 u8 port_num, u16 index, u16 *pkey);
2000 	int		           (*modify_device)(struct ib_device *device,
2001 						    int device_modify_mask,
2002 						    struct ib_device_modify *device_modify);
2003 	int		           (*modify_port)(struct ib_device *device,
2004 						  u8 port_num, int port_modify_mask,
2005 						  struct ib_port_modify *port_modify);
2006 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
2007 						     struct ib_udata *udata);
2008 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
2009 	int                        (*mmap)(struct ib_ucontext *context,
2010 					   struct vm_area_struct *vma);
2011 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
2012 					       struct ib_ucontext *context,
2013 					       struct ib_udata *udata);
2014 	int                        (*dealloc_pd)(struct ib_pd *pd);
2015 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
2016 						struct ib_ah_attr *ah_attr,
2017 						struct ib_udata *udata);
2018 	int                        (*modify_ah)(struct ib_ah *ah,
2019 						struct ib_ah_attr *ah_attr);
2020 	int                        (*query_ah)(struct ib_ah *ah,
2021 					       struct ib_ah_attr *ah_attr);
2022 	int                        (*destroy_ah)(struct ib_ah *ah);
2023 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
2024 						 struct ib_srq_init_attr *srq_init_attr,
2025 						 struct ib_udata *udata);
2026 	int                        (*modify_srq)(struct ib_srq *srq,
2027 						 struct ib_srq_attr *srq_attr,
2028 						 enum ib_srq_attr_mask srq_attr_mask,
2029 						 struct ib_udata *udata);
2030 	int                        (*query_srq)(struct ib_srq *srq,
2031 						struct ib_srq_attr *srq_attr);
2032 	int                        (*destroy_srq)(struct ib_srq *srq);
2033 	int                        (*post_srq_recv)(struct ib_srq *srq,
2034 						    struct ib_recv_wr *recv_wr,
2035 						    struct ib_recv_wr **bad_recv_wr);
2036 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
2037 						struct ib_qp_init_attr *qp_init_attr,
2038 						struct ib_udata *udata);
2039 	int                        (*modify_qp)(struct ib_qp *qp,
2040 						struct ib_qp_attr *qp_attr,
2041 						int qp_attr_mask,
2042 						struct ib_udata *udata);
2043 	int                        (*query_qp)(struct ib_qp *qp,
2044 					       struct ib_qp_attr *qp_attr,
2045 					       int qp_attr_mask,
2046 					       struct ib_qp_init_attr *qp_init_attr);
2047 	int                        (*destroy_qp)(struct ib_qp *qp);
2048 	int                        (*post_send)(struct ib_qp *qp,
2049 						struct ib_send_wr *send_wr,
2050 						struct ib_send_wr **bad_send_wr);
2051 	int                        (*post_recv)(struct ib_qp *qp,
2052 						struct ib_recv_wr *recv_wr,
2053 						struct ib_recv_wr **bad_recv_wr);
2054 	struct ib_cq *             (*create_cq)(struct ib_device *device,
2055 						const struct ib_cq_init_attr *attr,
2056 						struct ib_ucontext *context,
2057 						struct ib_udata *udata);
2058 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2059 						u16 cq_period);
2060 	int                        (*destroy_cq)(struct ib_cq *cq);
2061 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
2062 						struct ib_udata *udata);
2063 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
2064 					      struct ib_wc *wc);
2065 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2066 	int                        (*req_notify_cq)(struct ib_cq *cq,
2067 						    enum ib_cq_notify_flags flags);
2068 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
2069 						      int wc_cnt);
2070 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
2071 						 int mr_access_flags);
2072 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
2073 						  u64 start, u64 length,
2074 						  u64 virt_addr,
2075 						  int mr_access_flags,
2076 						  struct ib_udata *udata);
2077 	int			   (*rereg_user_mr)(struct ib_mr *mr,
2078 						    int flags,
2079 						    u64 start, u64 length,
2080 						    u64 virt_addr,
2081 						    int mr_access_flags,
2082 						    struct ib_pd *pd,
2083 						    struct ib_udata *udata);
2084 	int                        (*dereg_mr)(struct ib_mr *mr);
2085 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2086 					       enum ib_mr_type mr_type,
2087 					       u32 max_num_sg);
2088 	int                        (*map_mr_sg)(struct ib_mr *mr,
2089 						struct scatterlist *sg,
2090 						int sg_nents,
2091 						unsigned int *sg_offset);
2092 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2093 					       enum ib_mw_type type,
2094 					       struct ib_udata *udata);
2095 	int                        (*dealloc_mw)(struct ib_mw *mw);
2096 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2097 						int mr_access_flags,
2098 						struct ib_fmr_attr *fmr_attr);
2099 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2100 						   u64 *page_list, int list_len,
2101 						   u64 iova);
2102 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2103 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2104 	int                        (*attach_mcast)(struct ib_qp *qp,
2105 						   union ib_gid *gid,
2106 						   u16 lid);
2107 	int                        (*detach_mcast)(struct ib_qp *qp,
2108 						   union ib_gid *gid,
2109 						   u16 lid);
2110 	int                        (*process_mad)(struct ib_device *device,
2111 						  int process_mad_flags,
2112 						  u8 port_num,
2113 						  const struct ib_wc *in_wc,
2114 						  const struct ib_grh *in_grh,
2115 						  const struct ib_mad_hdr *in_mad,
2116 						  size_t in_mad_size,
2117 						  struct ib_mad_hdr *out_mad,
2118 						  size_t *out_mad_size,
2119 						  u16 *out_mad_pkey_index);
2120 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2121 						 struct ib_ucontext *ucontext,
2122 						 struct ib_udata *udata);
2123 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2124 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2125 						  struct ib_flow_attr
2126 						  *flow_attr,
2127 						  int domain);
2128 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2129 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2130 						      struct ib_mr_status *mr_status);
2131 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2132 	void			   (*drain_rq)(struct ib_qp *qp);
2133 	void			   (*drain_sq)(struct ib_qp *qp);
2134 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2135 							int state);
2136 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2137 						   struct ifla_vf_info *ivf);
2138 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2139 						   struct ifla_vf_stats *stats);
2140 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2141 						  int type);
2142 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2143 						struct ib_wq_init_attr *init_attr,
2144 						struct ib_udata *udata);
2145 	int			   (*destroy_wq)(struct ib_wq *wq);
2146 	int			   (*modify_wq)(struct ib_wq *wq,
2147 						struct ib_wq_attr *attr,
2148 						u32 wq_attr_mask,
2149 						struct ib_udata *udata);
2150 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2151 							   struct ib_rwq_ind_table_init_attr *init_attr,
2152 							   struct ib_udata *udata);
2153 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2154 	struct ib_dma_mapping_ops   *dma_ops;
2155 
2156 	struct module               *owner;
2157 	struct device                dev;
2158 	struct kobject               *ports_parent;
2159 	struct list_head             port_list;
2160 
2161 	enum {
2162 		IB_DEV_UNINITIALIZED,
2163 		IB_DEV_REGISTERED,
2164 		IB_DEV_UNREGISTERED
2165 	}                            reg_state;
2166 
2167 	int			     uverbs_abi_ver;
2168 	u64			     uverbs_cmd_mask;
2169 	u64			     uverbs_ex_cmd_mask;
2170 
2171 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2172 	__be64			     node_guid;
2173 	u32			     local_dma_lkey;
2174 	u16                          is_switch:1;
2175 	u8                           node_type;
2176 	u8                           phys_port_cnt;
2177 	struct ib_device_attr        attrs;
2178 	struct attribute_group	     *hw_stats_ag;
2179 	struct rdma_hw_stats         *hw_stats;
2180 
2181 	/**
2182 	 * The following mandatory functions are used only at device
2183 	 * registration.  Keep functions such as these at the end of this
2184 	 * structure to avoid cache line misses when accessing struct ib_device
2185 	 * in fast paths.
2186 	 */
2187 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2188 	void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2189 };
2190 
2191 struct ib_client {
2192 	char  *name;
2193 	void (*add)   (struct ib_device *);
2194 	void (*remove)(struct ib_device *, void *client_data);
2195 
2196 	/* Returns the net_dev belonging to this ib_client and matching the
2197 	 * given parameters.
2198 	 * @dev:	 An RDMA device that the net_dev use for communication.
2199 	 * @port:	 A physical port number on the RDMA device.
2200 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2201 	 * @gid:	 A GID that the net_dev uses to communicate.
2202 	 * @addr:	 An IP address the net_dev is configured with.
2203 	 * @client_data: The device's client data set by ib_set_client_data().
2204 	 *
2205 	 * An ib_client that implements a net_dev on top of RDMA devices
2206 	 * (such as IP over IB) should implement this callback, allowing the
2207 	 * rdma_cm module to find the right net_dev for a given request.
2208 	 *
2209 	 * The caller is responsible for calling dev_put on the returned
2210 	 * netdev. */
2211 	struct net_device *(*get_net_dev_by_params)(
2212 			struct ib_device *dev,
2213 			u8 port,
2214 			u16 pkey,
2215 			const union ib_gid *gid,
2216 			const struct sockaddr *addr,
2217 			void *client_data);
2218 	struct list_head list;
2219 };
2220 
2221 struct ib_device *ib_alloc_device(size_t size);
2222 void ib_dealloc_device(struct ib_device *device);
2223 
2224 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2225 
2226 int ib_register_device(struct ib_device *device,
2227 		       int (*port_callback)(struct ib_device *,
2228 					    u8, struct kobject *));
2229 void ib_unregister_device(struct ib_device *device);
2230 
2231 int ib_register_client   (struct ib_client *client);
2232 void ib_unregister_client(struct ib_client *client);
2233 
2234 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2235 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2236 			 void *data);
2237 
2238 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2239 {
2240 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2241 }
2242 
2243 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2244 {
2245 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2246 }
2247 
2248 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2249 				       size_t offset,
2250 				       size_t len)
2251 {
2252 	const void __user *p = udata->inbuf + offset;
2253 	bool ret;
2254 	u8 *buf;
2255 
2256 	if (len > USHRT_MAX)
2257 		return false;
2258 
2259 	buf = memdup_user(p, len);
2260 	if (IS_ERR(buf))
2261 		return false;
2262 
2263 	ret = !memchr_inv(buf, 0, len);
2264 	kfree(buf);
2265 	return ret;
2266 }
2267 
2268 /**
2269  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2270  * contains all required attributes and no attributes not allowed for
2271  * the given QP state transition.
2272  * @cur_state: Current QP state
2273  * @next_state: Next QP state
2274  * @type: QP type
2275  * @mask: Mask of supplied QP attributes
2276  * @ll : link layer of port
2277  *
2278  * This function is a helper function that a low-level driver's
2279  * modify_qp method can use to validate the consumer's input.  It
2280  * checks that cur_state and next_state are valid QP states, that a
2281  * transition from cur_state to next_state is allowed by the IB spec,
2282  * and that the attribute mask supplied is allowed for the transition.
2283  */
2284 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2285 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2286 		       enum rdma_link_layer ll);
2287 
2288 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2289 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2290 void ib_dispatch_event(struct ib_event *event);
2291 
2292 int ib_query_port(struct ib_device *device,
2293 		  u8 port_num, struct ib_port_attr *port_attr);
2294 
2295 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2296 					       u8 port_num);
2297 
2298 /**
2299  * rdma_cap_ib_switch - Check if the device is IB switch
2300  * @device: Device to check
2301  *
2302  * Device driver is responsible for setting is_switch bit on
2303  * in ib_device structure at init time.
2304  *
2305  * Return: true if the device is IB switch.
2306  */
2307 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2308 {
2309 	return device->is_switch;
2310 }
2311 
2312 /**
2313  * rdma_start_port - Return the first valid port number for the device
2314  * specified
2315  *
2316  * @device: Device to be checked
2317  *
2318  * Return start port number
2319  */
2320 static inline u8 rdma_start_port(const struct ib_device *device)
2321 {
2322 	return rdma_cap_ib_switch(device) ? 0 : 1;
2323 }
2324 
2325 /**
2326  * rdma_end_port - Return the last valid port number for the device
2327  * specified
2328  *
2329  * @device: Device to be checked
2330  *
2331  * Return last port number
2332  */
2333 static inline u8 rdma_end_port(const struct ib_device *device)
2334 {
2335 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2336 }
2337 
2338 static inline int rdma_is_port_valid(const struct ib_device *device,
2339 				     unsigned int port)
2340 {
2341 	return (port >= rdma_start_port(device) &&
2342 		port <= rdma_end_port(device));
2343 }
2344 
2345 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2346 {
2347 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2348 }
2349 
2350 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2351 {
2352 	return device->port_immutable[port_num].core_cap_flags &
2353 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2354 }
2355 
2356 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2357 {
2358 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2359 }
2360 
2361 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2362 {
2363 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2364 }
2365 
2366 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2367 {
2368 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2369 }
2370 
2371 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2372 {
2373 	return rdma_protocol_ib(device, port_num) ||
2374 		rdma_protocol_roce(device, port_num);
2375 }
2376 
2377 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2378 {
2379 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2380 }
2381 
2382 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2383 {
2384 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2385 }
2386 
2387 /**
2388  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2389  * Management Datagrams.
2390  * @device: Device to check
2391  * @port_num: Port number to check
2392  *
2393  * Management Datagrams (MAD) are a required part of the InfiniBand
2394  * specification and are supported on all InfiniBand devices.  A slightly
2395  * extended version are also supported on OPA interfaces.
2396  *
2397  * Return: true if the port supports sending/receiving of MAD packets.
2398  */
2399 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2400 {
2401 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2402 }
2403 
2404 /**
2405  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2406  * Management Datagrams.
2407  * @device: Device to check
2408  * @port_num: Port number to check
2409  *
2410  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2411  * datagrams with their own versions.  These OPA MADs share many but not all of
2412  * the characteristics of InfiniBand MADs.
2413  *
2414  * OPA MADs differ in the following ways:
2415  *
2416  *    1) MADs are variable size up to 2K
2417  *       IBTA defined MADs remain fixed at 256 bytes
2418  *    2) OPA SMPs must carry valid PKeys
2419  *    3) OPA SMP packets are a different format
2420  *
2421  * Return: true if the port supports OPA MAD packet formats.
2422  */
2423 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2424 {
2425 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2426 		== RDMA_CORE_CAP_OPA_MAD;
2427 }
2428 
2429 /**
2430  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2431  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2432  * @device: Device to check
2433  * @port_num: Port number to check
2434  *
2435  * Each InfiniBand node is required to provide a Subnet Management Agent
2436  * that the subnet manager can access.  Prior to the fabric being fully
2437  * configured by the subnet manager, the SMA is accessed via a well known
2438  * interface called the Subnet Management Interface (SMI).  This interface
2439  * uses directed route packets to communicate with the SM to get around the
2440  * chicken and egg problem of the SM needing to know what's on the fabric
2441  * in order to configure the fabric, and needing to configure the fabric in
2442  * order to send packets to the devices on the fabric.  These directed
2443  * route packets do not need the fabric fully configured in order to reach
2444  * their destination.  The SMI is the only method allowed to send
2445  * directed route packets on an InfiniBand fabric.
2446  *
2447  * Return: true if the port provides an SMI.
2448  */
2449 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2450 {
2451 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2452 }
2453 
2454 /**
2455  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2456  * Communication Manager.
2457  * @device: Device to check
2458  * @port_num: Port number to check
2459  *
2460  * The InfiniBand Communication Manager is one of many pre-defined General
2461  * Service Agents (GSA) that are accessed via the General Service
2462  * Interface (GSI).  It's role is to facilitate establishment of connections
2463  * between nodes as well as other management related tasks for established
2464  * connections.
2465  *
2466  * Return: true if the port supports an IB CM (this does not guarantee that
2467  * a CM is actually running however).
2468  */
2469 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2470 {
2471 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2472 }
2473 
2474 /**
2475  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2476  * Communication Manager.
2477  * @device: Device to check
2478  * @port_num: Port number to check
2479  *
2480  * Similar to above, but specific to iWARP connections which have a different
2481  * managment protocol than InfiniBand.
2482  *
2483  * Return: true if the port supports an iWARP CM (this does not guarantee that
2484  * a CM is actually running however).
2485  */
2486 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2487 {
2488 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2489 }
2490 
2491 /**
2492  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2493  * Subnet Administration.
2494  * @device: Device to check
2495  * @port_num: Port number to check
2496  *
2497  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2498  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2499  * fabrics, devices should resolve routes to other hosts by contacting the
2500  * SA to query the proper route.
2501  *
2502  * Return: true if the port should act as a client to the fabric Subnet
2503  * Administration interface.  This does not imply that the SA service is
2504  * running locally.
2505  */
2506 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2507 {
2508 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2509 }
2510 
2511 /**
2512  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2513  * Multicast.
2514  * @device: Device to check
2515  * @port_num: Port number to check
2516  *
2517  * InfiniBand multicast registration is more complex than normal IPv4 or
2518  * IPv6 multicast registration.  Each Host Channel Adapter must register
2519  * with the Subnet Manager when it wishes to join a multicast group.  It
2520  * should do so only once regardless of how many queue pairs it subscribes
2521  * to this group.  And it should leave the group only after all queue pairs
2522  * attached to the group have been detached.
2523  *
2524  * Return: true if the port must undertake the additional adminstrative
2525  * overhead of registering/unregistering with the SM and tracking of the
2526  * total number of queue pairs attached to the multicast group.
2527  */
2528 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2529 {
2530 	return rdma_cap_ib_sa(device, port_num);
2531 }
2532 
2533 /**
2534  * rdma_cap_af_ib - Check if the port of device has the capability
2535  * Native Infiniband Address.
2536  * @device: Device to check
2537  * @port_num: Port number to check
2538  *
2539  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2540  * GID.  RoCE uses a different mechanism, but still generates a GID via
2541  * a prescribed mechanism and port specific data.
2542  *
2543  * Return: true if the port uses a GID address to identify devices on the
2544  * network.
2545  */
2546 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2547 {
2548 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2549 }
2550 
2551 /**
2552  * rdma_cap_eth_ah - Check if the port of device has the capability
2553  * Ethernet Address Handle.
2554  * @device: Device to check
2555  * @port_num: Port number to check
2556  *
2557  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2558  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2559  * port.  Normally, packet headers are generated by the sending host
2560  * adapter, but when sending connectionless datagrams, we must manually
2561  * inject the proper headers for the fabric we are communicating over.
2562  *
2563  * Return: true if we are running as a RoCE port and must force the
2564  * addition of a Global Route Header built from our Ethernet Address
2565  * Handle into our header list for connectionless packets.
2566  */
2567 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2568 {
2569 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2570 }
2571 
2572 /**
2573  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2574  *
2575  * @device: Device
2576  * @port_num: Port number
2577  *
2578  * This MAD size includes the MAD headers and MAD payload.  No other headers
2579  * are included.
2580  *
2581  * Return the max MAD size required by the Port.  Will return 0 if the port
2582  * does not support MADs
2583  */
2584 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2585 {
2586 	return device->port_immutable[port_num].max_mad_size;
2587 }
2588 
2589 /**
2590  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2591  * @device: Device to check
2592  * @port_num: Port number to check
2593  *
2594  * RoCE GID table mechanism manages the various GIDs for a device.
2595  *
2596  * NOTE: if allocating the port's GID table has failed, this call will still
2597  * return true, but any RoCE GID table API will fail.
2598  *
2599  * Return: true if the port uses RoCE GID table mechanism in order to manage
2600  * its GIDs.
2601  */
2602 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2603 					   u8 port_num)
2604 {
2605 	return rdma_protocol_roce(device, port_num) &&
2606 		device->add_gid && device->del_gid;
2607 }
2608 
2609 /*
2610  * Check if the device supports READ W/ INVALIDATE.
2611  */
2612 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2613 {
2614 	/*
2615 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2616 	 * has support for it yet.
2617 	 */
2618 	return rdma_protocol_iwarp(dev, port_num);
2619 }
2620 
2621 int ib_query_gid(struct ib_device *device,
2622 		 u8 port_num, int index, union ib_gid *gid,
2623 		 struct ib_gid_attr *attr);
2624 
2625 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2626 			 int state);
2627 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2628 		     struct ifla_vf_info *info);
2629 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2630 		    struct ifla_vf_stats *stats);
2631 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2632 		   int type);
2633 
2634 int ib_query_pkey(struct ib_device *device,
2635 		  u8 port_num, u16 index, u16 *pkey);
2636 
2637 int ib_modify_device(struct ib_device *device,
2638 		     int device_modify_mask,
2639 		     struct ib_device_modify *device_modify);
2640 
2641 int ib_modify_port(struct ib_device *device,
2642 		   u8 port_num, int port_modify_mask,
2643 		   struct ib_port_modify *port_modify);
2644 
2645 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2646 		enum ib_gid_type gid_type, struct net_device *ndev,
2647 		u8 *port_num, u16 *index);
2648 
2649 int ib_find_pkey(struct ib_device *device,
2650 		 u8 port_num, u16 pkey, u16 *index);
2651 
2652 enum ib_pd_flags {
2653 	/*
2654 	 * Create a memory registration for all memory in the system and place
2655 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2656 	 * ULPs to avoid the overhead of dynamic MRs.
2657 	 *
2658 	 * This flag is generally considered unsafe and must only be used in
2659 	 * extremly trusted environments.  Every use of it will log a warning
2660 	 * in the kernel log.
2661 	 */
2662 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2663 };
2664 
2665 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2666 		const char *caller);
2667 #define ib_alloc_pd(device, flags) \
2668 	__ib_alloc_pd((device), (flags), __func__)
2669 void ib_dealloc_pd(struct ib_pd *pd);
2670 
2671 /**
2672  * ib_create_ah - Creates an address handle for the given address vector.
2673  * @pd: The protection domain associated with the address handle.
2674  * @ah_attr: The attributes of the address vector.
2675  *
2676  * The address handle is used to reference a local or global destination
2677  * in all UD QP post sends.
2678  */
2679 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2680 
2681 /**
2682  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2683  *   work completion.
2684  * @hdr: the L3 header to parse
2685  * @net_type: type of header to parse
2686  * @sgid: place to store source gid
2687  * @dgid: place to store destination gid
2688  */
2689 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2690 			      enum rdma_network_type net_type,
2691 			      union ib_gid *sgid, union ib_gid *dgid);
2692 
2693 /**
2694  * ib_get_rdma_header_version - Get the header version
2695  * @hdr: the L3 header to parse
2696  */
2697 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2698 
2699 /**
2700  * ib_init_ah_from_wc - Initializes address handle attributes from a
2701  *   work completion.
2702  * @device: Device on which the received message arrived.
2703  * @port_num: Port on which the received message arrived.
2704  * @wc: Work completion associated with the received message.
2705  * @grh: References the received global route header.  This parameter is
2706  *   ignored unless the work completion indicates that the GRH is valid.
2707  * @ah_attr: Returned attributes that can be used when creating an address
2708  *   handle for replying to the message.
2709  */
2710 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2711 		       const struct ib_wc *wc, const struct ib_grh *grh,
2712 		       struct ib_ah_attr *ah_attr);
2713 
2714 /**
2715  * ib_create_ah_from_wc - Creates an address handle associated with the
2716  *   sender of the specified work completion.
2717  * @pd: The protection domain associated with the address handle.
2718  * @wc: Work completion information associated with a received message.
2719  * @grh: References the received global route header.  This parameter is
2720  *   ignored unless the work completion indicates that the GRH is valid.
2721  * @port_num: The outbound port number to associate with the address.
2722  *
2723  * The address handle is used to reference a local or global destination
2724  * in all UD QP post sends.
2725  */
2726 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2727 				   const struct ib_grh *grh, u8 port_num);
2728 
2729 /**
2730  * ib_modify_ah - Modifies the address vector associated with an address
2731  *   handle.
2732  * @ah: The address handle to modify.
2733  * @ah_attr: The new address vector attributes to associate with the
2734  *   address handle.
2735  */
2736 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2737 
2738 /**
2739  * ib_query_ah - Queries the address vector associated with an address
2740  *   handle.
2741  * @ah: The address handle to query.
2742  * @ah_attr: The address vector attributes associated with the address
2743  *   handle.
2744  */
2745 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2746 
2747 /**
2748  * ib_destroy_ah - Destroys an address handle.
2749  * @ah: The address handle to destroy.
2750  */
2751 int ib_destroy_ah(struct ib_ah *ah);
2752 
2753 /**
2754  * ib_create_srq - Creates a SRQ associated with the specified protection
2755  *   domain.
2756  * @pd: The protection domain associated with the SRQ.
2757  * @srq_init_attr: A list of initial attributes required to create the
2758  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2759  *   the actual capabilities of the created SRQ.
2760  *
2761  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2762  * requested size of the SRQ, and set to the actual values allocated
2763  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2764  * will always be at least as large as the requested values.
2765  */
2766 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2767 			     struct ib_srq_init_attr *srq_init_attr);
2768 
2769 /**
2770  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2771  * @srq: The SRQ to modify.
2772  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2773  *   the current values of selected SRQ attributes are returned.
2774  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2775  *   are being modified.
2776  *
2777  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2778  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2779  * the number of receives queued drops below the limit.
2780  */
2781 int ib_modify_srq(struct ib_srq *srq,
2782 		  struct ib_srq_attr *srq_attr,
2783 		  enum ib_srq_attr_mask srq_attr_mask);
2784 
2785 /**
2786  * ib_query_srq - Returns the attribute list and current values for the
2787  *   specified SRQ.
2788  * @srq: The SRQ to query.
2789  * @srq_attr: The attributes of the specified SRQ.
2790  */
2791 int ib_query_srq(struct ib_srq *srq,
2792 		 struct ib_srq_attr *srq_attr);
2793 
2794 /**
2795  * ib_destroy_srq - Destroys the specified SRQ.
2796  * @srq: The SRQ to destroy.
2797  */
2798 int ib_destroy_srq(struct ib_srq *srq);
2799 
2800 /**
2801  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2802  * @srq: The SRQ to post the work request on.
2803  * @recv_wr: A list of work requests to post on the receive queue.
2804  * @bad_recv_wr: On an immediate failure, this parameter will reference
2805  *   the work request that failed to be posted on the QP.
2806  */
2807 static inline int ib_post_srq_recv(struct ib_srq *srq,
2808 				   struct ib_recv_wr *recv_wr,
2809 				   struct ib_recv_wr **bad_recv_wr)
2810 {
2811 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2812 }
2813 
2814 /**
2815  * ib_create_qp - Creates a QP associated with the specified protection
2816  *   domain.
2817  * @pd: The protection domain associated with the QP.
2818  * @qp_init_attr: A list of initial attributes required to create the
2819  *   QP.  If QP creation succeeds, then the attributes are updated to
2820  *   the actual capabilities of the created QP.
2821  */
2822 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2823 			   struct ib_qp_init_attr *qp_init_attr);
2824 
2825 /**
2826  * ib_modify_qp - Modifies the attributes for the specified QP and then
2827  *   transitions the QP to the given state.
2828  * @qp: The QP to modify.
2829  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2830  *   the current values of selected QP attributes are returned.
2831  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2832  *   are being modified.
2833  */
2834 int ib_modify_qp(struct ib_qp *qp,
2835 		 struct ib_qp_attr *qp_attr,
2836 		 int qp_attr_mask);
2837 
2838 /**
2839  * ib_query_qp - Returns the attribute list and current values for the
2840  *   specified QP.
2841  * @qp: The QP to query.
2842  * @qp_attr: The attributes of the specified QP.
2843  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2844  * @qp_init_attr: Additional attributes of the selected QP.
2845  *
2846  * The qp_attr_mask may be used to limit the query to gathering only the
2847  * selected attributes.
2848  */
2849 int ib_query_qp(struct ib_qp *qp,
2850 		struct ib_qp_attr *qp_attr,
2851 		int qp_attr_mask,
2852 		struct ib_qp_init_attr *qp_init_attr);
2853 
2854 /**
2855  * ib_destroy_qp - Destroys the specified QP.
2856  * @qp: The QP to destroy.
2857  */
2858 int ib_destroy_qp(struct ib_qp *qp);
2859 
2860 /**
2861  * ib_open_qp - Obtain a reference to an existing sharable QP.
2862  * @xrcd - XRC domain
2863  * @qp_open_attr: Attributes identifying the QP to open.
2864  *
2865  * Returns a reference to a sharable QP.
2866  */
2867 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2868 			 struct ib_qp_open_attr *qp_open_attr);
2869 
2870 /**
2871  * ib_close_qp - Release an external reference to a QP.
2872  * @qp: The QP handle to release
2873  *
2874  * The opened QP handle is released by the caller.  The underlying
2875  * shared QP is not destroyed until all internal references are released.
2876  */
2877 int ib_close_qp(struct ib_qp *qp);
2878 
2879 /**
2880  * ib_post_send - Posts a list of work requests to the send queue of
2881  *   the specified QP.
2882  * @qp: The QP to post the work request on.
2883  * @send_wr: A list of work requests to post on the send queue.
2884  * @bad_send_wr: On an immediate failure, this parameter will reference
2885  *   the work request that failed to be posted on the QP.
2886  *
2887  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2888  * error is returned, the QP state shall not be affected,
2889  * ib_post_send() will return an immediate error after queueing any
2890  * earlier work requests in the list.
2891  */
2892 static inline int ib_post_send(struct ib_qp *qp,
2893 			       struct ib_send_wr *send_wr,
2894 			       struct ib_send_wr **bad_send_wr)
2895 {
2896 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2897 }
2898 
2899 /**
2900  * ib_post_recv - Posts a list of work requests to the receive queue of
2901  *   the specified QP.
2902  * @qp: The QP to post the work request on.
2903  * @recv_wr: A list of work requests to post on the receive queue.
2904  * @bad_recv_wr: On an immediate failure, this parameter will reference
2905  *   the work request that failed to be posted on the QP.
2906  */
2907 static inline int ib_post_recv(struct ib_qp *qp,
2908 			       struct ib_recv_wr *recv_wr,
2909 			       struct ib_recv_wr **bad_recv_wr)
2910 {
2911 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2912 }
2913 
2914 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2915 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2916 void ib_free_cq(struct ib_cq *cq);
2917 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2918 
2919 /**
2920  * ib_create_cq - Creates a CQ on the specified device.
2921  * @device: The device on which to create the CQ.
2922  * @comp_handler: A user-specified callback that is invoked when a
2923  *   completion event occurs on the CQ.
2924  * @event_handler: A user-specified callback that is invoked when an
2925  *   asynchronous event not associated with a completion occurs on the CQ.
2926  * @cq_context: Context associated with the CQ returned to the user via
2927  *   the associated completion and event handlers.
2928  * @cq_attr: The attributes the CQ should be created upon.
2929  *
2930  * Users can examine the cq structure to determine the actual CQ size.
2931  */
2932 struct ib_cq *ib_create_cq(struct ib_device *device,
2933 			   ib_comp_handler comp_handler,
2934 			   void (*event_handler)(struct ib_event *, void *),
2935 			   void *cq_context,
2936 			   const struct ib_cq_init_attr *cq_attr);
2937 
2938 /**
2939  * ib_resize_cq - Modifies the capacity of the CQ.
2940  * @cq: The CQ to resize.
2941  * @cqe: The minimum size of the CQ.
2942  *
2943  * Users can examine the cq structure to determine the actual CQ size.
2944  */
2945 int ib_resize_cq(struct ib_cq *cq, int cqe);
2946 
2947 /**
2948  * ib_modify_cq - Modifies moderation params of the CQ
2949  * @cq: The CQ to modify.
2950  * @cq_count: number of CQEs that will trigger an event
2951  * @cq_period: max period of time in usec before triggering an event
2952  *
2953  */
2954 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2955 
2956 /**
2957  * ib_destroy_cq - Destroys the specified CQ.
2958  * @cq: The CQ to destroy.
2959  */
2960 int ib_destroy_cq(struct ib_cq *cq);
2961 
2962 /**
2963  * ib_poll_cq - poll a CQ for completion(s)
2964  * @cq:the CQ being polled
2965  * @num_entries:maximum number of completions to return
2966  * @wc:array of at least @num_entries &struct ib_wc where completions
2967  *   will be returned
2968  *
2969  * Poll a CQ for (possibly multiple) completions.  If the return value
2970  * is < 0, an error occurred.  If the return value is >= 0, it is the
2971  * number of completions returned.  If the return value is
2972  * non-negative and < num_entries, then the CQ was emptied.
2973  */
2974 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2975 			     struct ib_wc *wc)
2976 {
2977 	return cq->device->poll_cq(cq, num_entries, wc);
2978 }
2979 
2980 /**
2981  * ib_peek_cq - Returns the number of unreaped completions currently
2982  *   on the specified CQ.
2983  * @cq: The CQ to peek.
2984  * @wc_cnt: A minimum number of unreaped completions to check for.
2985  *
2986  * If the number of unreaped completions is greater than or equal to wc_cnt,
2987  * this function returns wc_cnt, otherwise, it returns the actual number of
2988  * unreaped completions.
2989  */
2990 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2991 
2992 /**
2993  * ib_req_notify_cq - Request completion notification on a CQ.
2994  * @cq: The CQ to generate an event for.
2995  * @flags:
2996  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2997  *   to request an event on the next solicited event or next work
2998  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2999  *   may also be |ed in to request a hint about missed events, as
3000  *   described below.
3001  *
3002  * Return Value:
3003  *    < 0 means an error occurred while requesting notification
3004  *   == 0 means notification was requested successfully, and if
3005  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3006  *        were missed and it is safe to wait for another event.  In
3007  *        this case is it guaranteed that any work completions added
3008  *        to the CQ since the last CQ poll will trigger a completion
3009  *        notification event.
3010  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3011  *        in.  It means that the consumer must poll the CQ again to
3012  *        make sure it is empty to avoid missing an event because of a
3013  *        race between requesting notification and an entry being
3014  *        added to the CQ.  This return value means it is possible
3015  *        (but not guaranteed) that a work completion has been added
3016  *        to the CQ since the last poll without triggering a
3017  *        completion notification event.
3018  */
3019 static inline int ib_req_notify_cq(struct ib_cq *cq,
3020 				   enum ib_cq_notify_flags flags)
3021 {
3022 	return cq->device->req_notify_cq(cq, flags);
3023 }
3024 
3025 /**
3026  * ib_req_ncomp_notif - Request completion notification when there are
3027  *   at least the specified number of unreaped completions on the CQ.
3028  * @cq: The CQ to generate an event for.
3029  * @wc_cnt: The number of unreaped completions that should be on the
3030  *   CQ before an event is generated.
3031  */
3032 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3033 {
3034 	return cq->device->req_ncomp_notif ?
3035 		cq->device->req_ncomp_notif(cq, wc_cnt) :
3036 		-ENOSYS;
3037 }
3038 
3039 /**
3040  * ib_dma_mapping_error - check a DMA addr for error
3041  * @dev: The device for which the dma_addr was created
3042  * @dma_addr: The DMA address to check
3043  */
3044 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3045 {
3046 	if (dev->dma_ops)
3047 		return dev->dma_ops->mapping_error(dev, dma_addr);
3048 	return dma_mapping_error(dev->dma_device, dma_addr);
3049 }
3050 
3051 /**
3052  * ib_dma_map_single - Map a kernel virtual address to DMA address
3053  * @dev: The device for which the dma_addr is to be created
3054  * @cpu_addr: The kernel virtual address
3055  * @size: The size of the region in bytes
3056  * @direction: The direction of the DMA
3057  */
3058 static inline u64 ib_dma_map_single(struct ib_device *dev,
3059 				    void *cpu_addr, size_t size,
3060 				    enum dma_data_direction direction)
3061 {
3062 	if (dev->dma_ops)
3063 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
3064 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3065 }
3066 
3067 /**
3068  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3069  * @dev: The device for which the DMA address was created
3070  * @addr: The DMA address
3071  * @size: The size of the region in bytes
3072  * @direction: The direction of the DMA
3073  */
3074 static inline void ib_dma_unmap_single(struct ib_device *dev,
3075 				       u64 addr, size_t size,
3076 				       enum dma_data_direction direction)
3077 {
3078 	if (dev->dma_ops)
3079 		dev->dma_ops->unmap_single(dev, addr, size, direction);
3080 	else
3081 		dma_unmap_single(dev->dma_device, addr, size, direction);
3082 }
3083 
3084 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
3085 					  void *cpu_addr, size_t size,
3086 					  enum dma_data_direction direction,
3087 					  unsigned long dma_attrs)
3088 {
3089 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
3090 				    direction, dma_attrs);
3091 }
3092 
3093 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
3094 					     u64 addr, size_t size,
3095 					     enum dma_data_direction direction,
3096 					     unsigned long dma_attrs)
3097 {
3098 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
3099 				      direction, dma_attrs);
3100 }
3101 
3102 /**
3103  * ib_dma_map_page - Map a physical page to DMA address
3104  * @dev: The device for which the dma_addr is to be created
3105  * @page: The page to be mapped
3106  * @offset: The offset within the page
3107  * @size: The size of the region in bytes
3108  * @direction: The direction of the DMA
3109  */
3110 static inline u64 ib_dma_map_page(struct ib_device *dev,
3111 				  struct page *page,
3112 				  unsigned long offset,
3113 				  size_t size,
3114 					 enum dma_data_direction direction)
3115 {
3116 	if (dev->dma_ops)
3117 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
3118 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3119 }
3120 
3121 /**
3122  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3123  * @dev: The device for which the DMA address was created
3124  * @addr: The DMA address
3125  * @size: The size of the region in bytes
3126  * @direction: The direction of the DMA
3127  */
3128 static inline void ib_dma_unmap_page(struct ib_device *dev,
3129 				     u64 addr, size_t size,
3130 				     enum dma_data_direction direction)
3131 {
3132 	if (dev->dma_ops)
3133 		dev->dma_ops->unmap_page(dev, addr, size, direction);
3134 	else
3135 		dma_unmap_page(dev->dma_device, addr, size, direction);
3136 }
3137 
3138 /**
3139  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3140  * @dev: The device for which the DMA addresses are to be created
3141  * @sg: The array of scatter/gather entries
3142  * @nents: The number of scatter/gather entries
3143  * @direction: The direction of the DMA
3144  */
3145 static inline int ib_dma_map_sg(struct ib_device *dev,
3146 				struct scatterlist *sg, int nents,
3147 				enum dma_data_direction direction)
3148 {
3149 	if (dev->dma_ops)
3150 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
3151 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3152 }
3153 
3154 /**
3155  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3156  * @dev: The device for which the DMA addresses were created
3157  * @sg: The array of scatter/gather entries
3158  * @nents: The number of scatter/gather entries
3159  * @direction: The direction of the DMA
3160  */
3161 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3162 				   struct scatterlist *sg, int nents,
3163 				   enum dma_data_direction direction)
3164 {
3165 	if (dev->dma_ops)
3166 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
3167 	else
3168 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
3169 }
3170 
3171 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3172 				      struct scatterlist *sg, int nents,
3173 				      enum dma_data_direction direction,
3174 				      unsigned long dma_attrs)
3175 {
3176 	if (dev->dma_ops)
3177 		return dev->dma_ops->map_sg_attrs(dev, sg, nents, direction,
3178 						  dma_attrs);
3179 	else
3180 		return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3181 					dma_attrs);
3182 }
3183 
3184 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3185 					 struct scatterlist *sg, int nents,
3186 					 enum dma_data_direction direction,
3187 					 unsigned long dma_attrs)
3188 {
3189 	if (dev->dma_ops)
3190 		return dev->dma_ops->unmap_sg_attrs(dev, sg, nents, direction,
3191 						  dma_attrs);
3192 	else
3193 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
3194 				   dma_attrs);
3195 }
3196 /**
3197  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3198  * @dev: The device for which the DMA addresses were created
3199  * @sg: The scatter/gather entry
3200  *
3201  * Note: this function is obsolete. To do: change all occurrences of
3202  * ib_sg_dma_address() into sg_dma_address().
3203  */
3204 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3205 				    struct scatterlist *sg)
3206 {
3207 	return sg_dma_address(sg);
3208 }
3209 
3210 /**
3211  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3212  * @dev: The device for which the DMA addresses were created
3213  * @sg: The scatter/gather entry
3214  *
3215  * Note: this function is obsolete. To do: change all occurrences of
3216  * ib_sg_dma_len() into sg_dma_len().
3217  */
3218 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3219 					 struct scatterlist *sg)
3220 {
3221 	return sg_dma_len(sg);
3222 }
3223 
3224 /**
3225  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3226  * @dev: The device for which the DMA address was created
3227  * @addr: The DMA address
3228  * @size: The size of the region in bytes
3229  * @dir: The direction of the DMA
3230  */
3231 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3232 					      u64 addr,
3233 					      size_t size,
3234 					      enum dma_data_direction dir)
3235 {
3236 	if (dev->dma_ops)
3237 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
3238 	else
3239 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3240 }
3241 
3242 /**
3243  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3244  * @dev: The device for which the DMA address was created
3245  * @addr: The DMA address
3246  * @size: The size of the region in bytes
3247  * @dir: The direction of the DMA
3248  */
3249 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3250 						 u64 addr,
3251 						 size_t size,
3252 						 enum dma_data_direction dir)
3253 {
3254 	if (dev->dma_ops)
3255 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
3256 	else
3257 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3258 }
3259 
3260 /**
3261  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3262  * @dev: The device for which the DMA address is requested
3263  * @size: The size of the region to allocate in bytes
3264  * @dma_handle: A pointer for returning the DMA address of the region
3265  * @flag: memory allocator flags
3266  */
3267 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3268 					   size_t size,
3269 					   u64 *dma_handle,
3270 					   gfp_t flag)
3271 {
3272 	if (dev->dma_ops)
3273 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
3274 	else {
3275 		dma_addr_t handle;
3276 		void *ret;
3277 
3278 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3279 		*dma_handle = handle;
3280 		return ret;
3281 	}
3282 }
3283 
3284 /**
3285  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3286  * @dev: The device for which the DMA addresses were allocated
3287  * @size: The size of the region
3288  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3289  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3290  */
3291 static inline void ib_dma_free_coherent(struct ib_device *dev,
3292 					size_t size, void *cpu_addr,
3293 					u64 dma_handle)
3294 {
3295 	if (dev->dma_ops)
3296 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3297 	else
3298 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3299 }
3300 
3301 /**
3302  * ib_dereg_mr - Deregisters a memory region and removes it from the
3303  *   HCA translation table.
3304  * @mr: The memory region to deregister.
3305  *
3306  * This function can fail, if the memory region has memory windows bound to it.
3307  */
3308 int ib_dereg_mr(struct ib_mr *mr);
3309 
3310 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3311 			  enum ib_mr_type mr_type,
3312 			  u32 max_num_sg);
3313 
3314 /**
3315  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3316  *   R_Key and L_Key.
3317  * @mr - struct ib_mr pointer to be updated.
3318  * @newkey - new key to be used.
3319  */
3320 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3321 {
3322 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3323 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3324 }
3325 
3326 /**
3327  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3328  * for calculating a new rkey for type 2 memory windows.
3329  * @rkey - the rkey to increment.
3330  */
3331 static inline u32 ib_inc_rkey(u32 rkey)
3332 {
3333 	const u32 mask = 0x000000ff;
3334 	return ((rkey + 1) & mask) | (rkey & ~mask);
3335 }
3336 
3337 /**
3338  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3339  * @pd: The protection domain associated with the unmapped region.
3340  * @mr_access_flags: Specifies the memory access rights.
3341  * @fmr_attr: Attributes of the unmapped region.
3342  *
3343  * A fast memory region must be mapped before it can be used as part of
3344  * a work request.
3345  */
3346 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3347 			    int mr_access_flags,
3348 			    struct ib_fmr_attr *fmr_attr);
3349 
3350 /**
3351  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3352  * @fmr: The fast memory region to associate with the pages.
3353  * @page_list: An array of physical pages to map to the fast memory region.
3354  * @list_len: The number of pages in page_list.
3355  * @iova: The I/O virtual address to use with the mapped region.
3356  */
3357 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3358 				  u64 *page_list, int list_len,
3359 				  u64 iova)
3360 {
3361 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3362 }
3363 
3364 /**
3365  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3366  * @fmr_list: A linked list of fast memory regions to unmap.
3367  */
3368 int ib_unmap_fmr(struct list_head *fmr_list);
3369 
3370 /**
3371  * ib_dealloc_fmr - Deallocates a fast memory region.
3372  * @fmr: The fast memory region to deallocate.
3373  */
3374 int ib_dealloc_fmr(struct ib_fmr *fmr);
3375 
3376 /**
3377  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3378  * @qp: QP to attach to the multicast group.  The QP must be type
3379  *   IB_QPT_UD.
3380  * @gid: Multicast group GID.
3381  * @lid: Multicast group LID in host byte order.
3382  *
3383  * In order to send and receive multicast packets, subnet
3384  * administration must have created the multicast group and configured
3385  * the fabric appropriately.  The port associated with the specified
3386  * QP must also be a member of the multicast group.
3387  */
3388 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3389 
3390 /**
3391  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3392  * @qp: QP to detach from the multicast group.
3393  * @gid: Multicast group GID.
3394  * @lid: Multicast group LID in host byte order.
3395  */
3396 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3397 
3398 /**
3399  * ib_alloc_xrcd - Allocates an XRC domain.
3400  * @device: The device on which to allocate the XRC domain.
3401  */
3402 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3403 
3404 /**
3405  * ib_dealloc_xrcd - Deallocates an XRC domain.
3406  * @xrcd: The XRC domain to deallocate.
3407  */
3408 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3409 
3410 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3411 			       struct ib_flow_attr *flow_attr, int domain);
3412 int ib_destroy_flow(struct ib_flow *flow_id);
3413 
3414 static inline int ib_check_mr_access(int flags)
3415 {
3416 	/*
3417 	 * Local write permission is required if remote write or
3418 	 * remote atomic permission is also requested.
3419 	 */
3420 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3421 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3422 		return -EINVAL;
3423 
3424 	return 0;
3425 }
3426 
3427 /**
3428  * ib_check_mr_status: lightweight check of MR status.
3429  *     This routine may provide status checks on a selected
3430  *     ib_mr. first use is for signature status check.
3431  *
3432  * @mr: A memory region.
3433  * @check_mask: Bitmask of which checks to perform from
3434  *     ib_mr_status_check enumeration.
3435  * @mr_status: The container of relevant status checks.
3436  *     failed checks will be indicated in the status bitmask
3437  *     and the relevant info shall be in the error item.
3438  */
3439 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3440 		       struct ib_mr_status *mr_status);
3441 
3442 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3443 					    u16 pkey, const union ib_gid *gid,
3444 					    const struct sockaddr *addr);
3445 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3446 			   struct ib_wq_init_attr *init_attr);
3447 int ib_destroy_wq(struct ib_wq *wq);
3448 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3449 		 u32 wq_attr_mask);
3450 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3451 						 struct ib_rwq_ind_table_init_attr*
3452 						 wq_ind_table_init_attr);
3453 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3454 
3455 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3456 		 unsigned int *sg_offset, unsigned int page_size);
3457 
3458 static inline int
3459 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3460 		  unsigned int *sg_offset, unsigned int page_size)
3461 {
3462 	int n;
3463 
3464 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3465 	mr->iova = 0;
3466 
3467 	return n;
3468 }
3469 
3470 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3471 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3472 
3473 void ib_drain_rq(struct ib_qp *qp);
3474 void ib_drain_sq(struct ib_qp *qp);
3475 void ib_drain_qp(struct ib_qp *qp);
3476 
3477 int ib_resolve_eth_dmac(struct ib_device *device,
3478 			struct ib_ah_attr *ah_attr);
3479 #endif /* IB_VERBS_H */
3480