xref: /openbmc/linux/include/rdma/ib_verbs.h (revision 80483c3a)
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 <asm/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 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
211 	/*
212 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
213 	 * support execution of WQEs that involve synchronization
214 	 * of I/O operations with single completion queue managed
215 	 * by hardware.
216 	 */
217 	IB_DEVICE_CROSS_CHANNEL		= (1 << 27),
218 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
219 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
220 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
221 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
222 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
223 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
224 };
225 
226 enum ib_signature_prot_cap {
227 	IB_PROT_T10DIF_TYPE_1 = 1,
228 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
229 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
230 };
231 
232 enum ib_signature_guard_cap {
233 	IB_GUARD_T10DIF_CRC	= 1,
234 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
235 };
236 
237 enum ib_atomic_cap {
238 	IB_ATOMIC_NONE,
239 	IB_ATOMIC_HCA,
240 	IB_ATOMIC_GLOB
241 };
242 
243 enum ib_odp_general_cap_bits {
244 	IB_ODP_SUPPORT = 1 << 0,
245 };
246 
247 enum ib_odp_transport_cap_bits {
248 	IB_ODP_SUPPORT_SEND	= 1 << 0,
249 	IB_ODP_SUPPORT_RECV	= 1 << 1,
250 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
251 	IB_ODP_SUPPORT_READ	= 1 << 3,
252 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
253 };
254 
255 struct ib_odp_caps {
256 	uint64_t general_caps;
257 	struct {
258 		uint32_t  rc_odp_caps;
259 		uint32_t  uc_odp_caps;
260 		uint32_t  ud_odp_caps;
261 	} per_transport_caps;
262 };
263 
264 enum ib_cq_creation_flags {
265 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
266 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
267 };
268 
269 struct ib_cq_init_attr {
270 	unsigned int	cqe;
271 	int		comp_vector;
272 	u32		flags;
273 };
274 
275 struct ib_device_attr {
276 	u64			fw_ver;
277 	__be64			sys_image_guid;
278 	u64			max_mr_size;
279 	u64			page_size_cap;
280 	u32			vendor_id;
281 	u32			vendor_part_id;
282 	u32			hw_ver;
283 	int			max_qp;
284 	int			max_qp_wr;
285 	u64			device_cap_flags;
286 	int			max_sge;
287 	int			max_sge_rd;
288 	int			max_cq;
289 	int			max_cqe;
290 	int			max_mr;
291 	int			max_pd;
292 	int			max_qp_rd_atom;
293 	int			max_ee_rd_atom;
294 	int			max_res_rd_atom;
295 	int			max_qp_init_rd_atom;
296 	int			max_ee_init_rd_atom;
297 	enum ib_atomic_cap	atomic_cap;
298 	enum ib_atomic_cap	masked_atomic_cap;
299 	int			max_ee;
300 	int			max_rdd;
301 	int			max_mw;
302 	int			max_raw_ipv6_qp;
303 	int			max_raw_ethy_qp;
304 	int			max_mcast_grp;
305 	int			max_mcast_qp_attach;
306 	int			max_total_mcast_qp_attach;
307 	int			max_ah;
308 	int			max_fmr;
309 	int			max_map_per_fmr;
310 	int			max_srq;
311 	int			max_srq_wr;
312 	int			max_srq_sge;
313 	unsigned int		max_fast_reg_page_list_len;
314 	u16			max_pkeys;
315 	u8			local_ca_ack_delay;
316 	int			sig_prot_cap;
317 	int			sig_guard_cap;
318 	struct ib_odp_caps	odp_caps;
319 	uint64_t		timestamp_mask;
320 	uint64_t		hca_core_clock; /* in KHZ */
321 };
322 
323 enum ib_mtu {
324 	IB_MTU_256  = 1,
325 	IB_MTU_512  = 2,
326 	IB_MTU_1024 = 3,
327 	IB_MTU_2048 = 4,
328 	IB_MTU_4096 = 5
329 };
330 
331 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
332 {
333 	switch (mtu) {
334 	case IB_MTU_256:  return  256;
335 	case IB_MTU_512:  return  512;
336 	case IB_MTU_1024: return 1024;
337 	case IB_MTU_2048: return 2048;
338 	case IB_MTU_4096: return 4096;
339 	default: 	  return -1;
340 	}
341 }
342 
343 enum ib_port_state {
344 	IB_PORT_NOP		= 0,
345 	IB_PORT_DOWN		= 1,
346 	IB_PORT_INIT		= 2,
347 	IB_PORT_ARMED		= 3,
348 	IB_PORT_ACTIVE		= 4,
349 	IB_PORT_ACTIVE_DEFER	= 5
350 };
351 
352 enum ib_port_cap_flags {
353 	IB_PORT_SM				= 1 <<  1,
354 	IB_PORT_NOTICE_SUP			= 1 <<  2,
355 	IB_PORT_TRAP_SUP			= 1 <<  3,
356 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
357 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
358 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
359 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
360 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
361 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
362 	IB_PORT_SM_DISABLED			= 1 << 10,
363 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
364 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
365 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
366 	IB_PORT_CM_SUP				= 1 << 16,
367 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
368 	IB_PORT_REINIT_SUP			= 1 << 18,
369 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
370 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
371 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
372 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
373 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
374 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
375 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
376 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
377 };
378 
379 enum ib_port_width {
380 	IB_WIDTH_1X	= 1,
381 	IB_WIDTH_4X	= 2,
382 	IB_WIDTH_8X	= 4,
383 	IB_WIDTH_12X	= 8
384 };
385 
386 static inline int ib_width_enum_to_int(enum ib_port_width width)
387 {
388 	switch (width) {
389 	case IB_WIDTH_1X:  return  1;
390 	case IB_WIDTH_4X:  return  4;
391 	case IB_WIDTH_8X:  return  8;
392 	case IB_WIDTH_12X: return 12;
393 	default: 	  return -1;
394 	}
395 }
396 
397 enum ib_port_speed {
398 	IB_SPEED_SDR	= 1,
399 	IB_SPEED_DDR	= 2,
400 	IB_SPEED_QDR	= 4,
401 	IB_SPEED_FDR10	= 8,
402 	IB_SPEED_FDR	= 16,
403 	IB_SPEED_EDR	= 32
404 };
405 
406 /**
407  * struct rdma_hw_stats
408  * @timestamp - Used by the core code to track when the last update was
409  * @lifespan - Used by the core code to determine how old the counters
410  *   should be before being updated again.  Stored in jiffies, defaults
411  *   to 10 milliseconds, drivers can override the default be specifying
412  *   their own value during their allocation routine.
413  * @name - Array of pointers to static names used for the counters in
414  *   directory.
415  * @num_counters - How many hardware counters there are.  If name is
416  *   shorter than this number, a kernel oops will result.  Driver authors
417  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
418  *   in their code to prevent this.
419  * @value - Array of u64 counters that are accessed by the sysfs code and
420  *   filled in by the drivers get_stats routine
421  */
422 struct rdma_hw_stats {
423 	unsigned long	timestamp;
424 	unsigned long	lifespan;
425 	const char * const *names;
426 	int		num_counters;
427 	u64		value[];
428 };
429 
430 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
431 /**
432  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
433  *   for drivers.
434  * @names - Array of static const char *
435  * @num_counters - How many elements in array
436  * @lifespan - How many milliseconds between updates
437  */
438 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
439 		const char * const *names, int num_counters,
440 		unsigned long lifespan)
441 {
442 	struct rdma_hw_stats *stats;
443 
444 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
445 			GFP_KERNEL);
446 	if (!stats)
447 		return NULL;
448 	stats->names = names;
449 	stats->num_counters = num_counters;
450 	stats->lifespan = msecs_to_jiffies(lifespan);
451 
452 	return stats;
453 }
454 
455 
456 /* Define bits for the various functionality this port needs to be supported by
457  * the core.
458  */
459 /* Management                           0x00000FFF */
460 #define RDMA_CORE_CAP_IB_MAD            0x00000001
461 #define RDMA_CORE_CAP_IB_SMI            0x00000002
462 #define RDMA_CORE_CAP_IB_CM             0x00000004
463 #define RDMA_CORE_CAP_IW_CM             0x00000008
464 #define RDMA_CORE_CAP_IB_SA             0x00000010
465 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
466 
467 /* Address format                       0x000FF000 */
468 #define RDMA_CORE_CAP_AF_IB             0x00001000
469 #define RDMA_CORE_CAP_ETH_AH            0x00002000
470 
471 /* Protocol                             0xFFF00000 */
472 #define RDMA_CORE_CAP_PROT_IB           0x00100000
473 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
474 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
475 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
476 
477 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
478 					| RDMA_CORE_CAP_IB_MAD \
479 					| RDMA_CORE_CAP_IB_SMI \
480 					| RDMA_CORE_CAP_IB_CM  \
481 					| RDMA_CORE_CAP_IB_SA  \
482 					| RDMA_CORE_CAP_AF_IB)
483 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
484 					| RDMA_CORE_CAP_IB_MAD  \
485 					| RDMA_CORE_CAP_IB_CM   \
486 					| RDMA_CORE_CAP_AF_IB   \
487 					| RDMA_CORE_CAP_ETH_AH)
488 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
489 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
490 					| RDMA_CORE_CAP_IB_MAD  \
491 					| RDMA_CORE_CAP_IB_CM   \
492 					| RDMA_CORE_CAP_AF_IB   \
493 					| RDMA_CORE_CAP_ETH_AH)
494 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
495 					| RDMA_CORE_CAP_IW_CM)
496 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
497 					| RDMA_CORE_CAP_OPA_MAD)
498 
499 struct ib_port_attr {
500 	u64			subnet_prefix;
501 	enum ib_port_state	state;
502 	enum ib_mtu		max_mtu;
503 	enum ib_mtu		active_mtu;
504 	int			gid_tbl_len;
505 	u32			port_cap_flags;
506 	u32			max_msg_sz;
507 	u32			bad_pkey_cntr;
508 	u32			qkey_viol_cntr;
509 	u16			pkey_tbl_len;
510 	u16			lid;
511 	u16			sm_lid;
512 	u8			lmc;
513 	u8			max_vl_num;
514 	u8			sm_sl;
515 	u8			subnet_timeout;
516 	u8			init_type_reply;
517 	u8			active_width;
518 	u8			active_speed;
519 	u8                      phys_state;
520 	bool			grh_required;
521 };
522 
523 enum ib_device_modify_flags {
524 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
525 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
526 };
527 
528 struct ib_device_modify {
529 	u64	sys_image_guid;
530 	char	node_desc[64];
531 };
532 
533 enum ib_port_modify_flags {
534 	IB_PORT_SHUTDOWN		= 1,
535 	IB_PORT_INIT_TYPE		= (1<<2),
536 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
537 };
538 
539 struct ib_port_modify {
540 	u32	set_port_cap_mask;
541 	u32	clr_port_cap_mask;
542 	u8	init_type;
543 };
544 
545 enum ib_event_type {
546 	IB_EVENT_CQ_ERR,
547 	IB_EVENT_QP_FATAL,
548 	IB_EVENT_QP_REQ_ERR,
549 	IB_EVENT_QP_ACCESS_ERR,
550 	IB_EVENT_COMM_EST,
551 	IB_EVENT_SQ_DRAINED,
552 	IB_EVENT_PATH_MIG,
553 	IB_EVENT_PATH_MIG_ERR,
554 	IB_EVENT_DEVICE_FATAL,
555 	IB_EVENT_PORT_ACTIVE,
556 	IB_EVENT_PORT_ERR,
557 	IB_EVENT_LID_CHANGE,
558 	IB_EVENT_PKEY_CHANGE,
559 	IB_EVENT_SM_CHANGE,
560 	IB_EVENT_SRQ_ERR,
561 	IB_EVENT_SRQ_LIMIT_REACHED,
562 	IB_EVENT_QP_LAST_WQE_REACHED,
563 	IB_EVENT_CLIENT_REREGISTER,
564 	IB_EVENT_GID_CHANGE,
565 	IB_EVENT_WQ_FATAL,
566 };
567 
568 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
569 
570 struct ib_event {
571 	struct ib_device	*device;
572 	union {
573 		struct ib_cq	*cq;
574 		struct ib_qp	*qp;
575 		struct ib_srq	*srq;
576 		struct ib_wq	*wq;
577 		u8		port_num;
578 	} element;
579 	enum ib_event_type	event;
580 };
581 
582 struct ib_event_handler {
583 	struct ib_device *device;
584 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
585 	struct list_head  list;
586 };
587 
588 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
589 	do {							\
590 		(_ptr)->device  = _device;			\
591 		(_ptr)->handler = _handler;			\
592 		INIT_LIST_HEAD(&(_ptr)->list);			\
593 	} while (0)
594 
595 struct ib_global_route {
596 	union ib_gid	dgid;
597 	u32		flow_label;
598 	u8		sgid_index;
599 	u8		hop_limit;
600 	u8		traffic_class;
601 };
602 
603 struct ib_grh {
604 	__be32		version_tclass_flow;
605 	__be16		paylen;
606 	u8		next_hdr;
607 	u8		hop_limit;
608 	union ib_gid	sgid;
609 	union ib_gid	dgid;
610 };
611 
612 union rdma_network_hdr {
613 	struct ib_grh ibgrh;
614 	struct {
615 		/* The IB spec states that if it's IPv4, the header
616 		 * is located in the last 20 bytes of the header.
617 		 */
618 		u8		reserved[20];
619 		struct iphdr	roce4grh;
620 	};
621 };
622 
623 enum {
624 	IB_MULTICAST_QPN = 0xffffff
625 };
626 
627 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
628 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
629 
630 enum ib_ah_flags {
631 	IB_AH_GRH	= 1
632 };
633 
634 enum ib_rate {
635 	IB_RATE_PORT_CURRENT = 0,
636 	IB_RATE_2_5_GBPS = 2,
637 	IB_RATE_5_GBPS   = 5,
638 	IB_RATE_10_GBPS  = 3,
639 	IB_RATE_20_GBPS  = 6,
640 	IB_RATE_30_GBPS  = 4,
641 	IB_RATE_40_GBPS  = 7,
642 	IB_RATE_60_GBPS  = 8,
643 	IB_RATE_80_GBPS  = 9,
644 	IB_RATE_120_GBPS = 10,
645 	IB_RATE_14_GBPS  = 11,
646 	IB_RATE_56_GBPS  = 12,
647 	IB_RATE_112_GBPS = 13,
648 	IB_RATE_168_GBPS = 14,
649 	IB_RATE_25_GBPS  = 15,
650 	IB_RATE_100_GBPS = 16,
651 	IB_RATE_200_GBPS = 17,
652 	IB_RATE_300_GBPS = 18
653 };
654 
655 /**
656  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
657  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
658  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
659  * @rate: rate to convert.
660  */
661 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
662 
663 /**
664  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
665  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
666  * @rate: rate to convert.
667  */
668 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
669 
670 
671 /**
672  * enum ib_mr_type - memory region type
673  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
674  *                            normal registration
675  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
676  *                            signature operations (data-integrity
677  *                            capable regions)
678  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
679  *                            register any arbitrary sg lists (without
680  *                            the normal mr constraints - see
681  *                            ib_map_mr_sg)
682  */
683 enum ib_mr_type {
684 	IB_MR_TYPE_MEM_REG,
685 	IB_MR_TYPE_SIGNATURE,
686 	IB_MR_TYPE_SG_GAPS,
687 };
688 
689 /**
690  * Signature types
691  * IB_SIG_TYPE_NONE: Unprotected.
692  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
693  */
694 enum ib_signature_type {
695 	IB_SIG_TYPE_NONE,
696 	IB_SIG_TYPE_T10_DIF,
697 };
698 
699 /**
700  * Signature T10-DIF block-guard types
701  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
702  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
703  */
704 enum ib_t10_dif_bg_type {
705 	IB_T10DIF_CRC,
706 	IB_T10DIF_CSUM
707 };
708 
709 /**
710  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
711  *     domain.
712  * @bg_type: T10-DIF block guard type (CRC|CSUM)
713  * @pi_interval: protection information interval.
714  * @bg: seed of guard computation.
715  * @app_tag: application tag of guard block
716  * @ref_tag: initial guard block reference tag.
717  * @ref_remap: Indicate wethear the reftag increments each block
718  * @app_escape: Indicate to skip block check if apptag=0xffff
719  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
720  * @apptag_check_mask: check bitmask of application tag.
721  */
722 struct ib_t10_dif_domain {
723 	enum ib_t10_dif_bg_type bg_type;
724 	u16			pi_interval;
725 	u16			bg;
726 	u16			app_tag;
727 	u32			ref_tag;
728 	bool			ref_remap;
729 	bool			app_escape;
730 	bool			ref_escape;
731 	u16			apptag_check_mask;
732 };
733 
734 /**
735  * struct ib_sig_domain - Parameters for signature domain
736  * @sig_type: specific signauture type
737  * @sig: union of all signature domain attributes that may
738  *     be used to set domain layout.
739  */
740 struct ib_sig_domain {
741 	enum ib_signature_type sig_type;
742 	union {
743 		struct ib_t10_dif_domain dif;
744 	} sig;
745 };
746 
747 /**
748  * struct ib_sig_attrs - Parameters for signature handover operation
749  * @check_mask: bitmask for signature byte check (8 bytes)
750  * @mem: memory domain layout desciptor.
751  * @wire: wire domain layout desciptor.
752  */
753 struct ib_sig_attrs {
754 	u8			check_mask;
755 	struct ib_sig_domain	mem;
756 	struct ib_sig_domain	wire;
757 };
758 
759 enum ib_sig_err_type {
760 	IB_SIG_BAD_GUARD,
761 	IB_SIG_BAD_REFTAG,
762 	IB_SIG_BAD_APPTAG,
763 };
764 
765 /**
766  * struct ib_sig_err - signature error descriptor
767  */
768 struct ib_sig_err {
769 	enum ib_sig_err_type	err_type;
770 	u32			expected;
771 	u32			actual;
772 	u64			sig_err_offset;
773 	u32			key;
774 };
775 
776 enum ib_mr_status_check {
777 	IB_MR_CHECK_SIG_STATUS = 1,
778 };
779 
780 /**
781  * struct ib_mr_status - Memory region status container
782  *
783  * @fail_status: Bitmask of MR checks status. For each
784  *     failed check a corresponding status bit is set.
785  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
786  *     failure.
787  */
788 struct ib_mr_status {
789 	u32		    fail_status;
790 	struct ib_sig_err   sig_err;
791 };
792 
793 /**
794  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
795  * enum.
796  * @mult: multiple to convert.
797  */
798 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
799 
800 struct ib_ah_attr {
801 	struct ib_global_route	grh;
802 	u16			dlid;
803 	u8			sl;
804 	u8			src_path_bits;
805 	u8			static_rate;
806 	u8			ah_flags;
807 	u8			port_num;
808 	u8			dmac[ETH_ALEN];
809 };
810 
811 enum ib_wc_status {
812 	IB_WC_SUCCESS,
813 	IB_WC_LOC_LEN_ERR,
814 	IB_WC_LOC_QP_OP_ERR,
815 	IB_WC_LOC_EEC_OP_ERR,
816 	IB_WC_LOC_PROT_ERR,
817 	IB_WC_WR_FLUSH_ERR,
818 	IB_WC_MW_BIND_ERR,
819 	IB_WC_BAD_RESP_ERR,
820 	IB_WC_LOC_ACCESS_ERR,
821 	IB_WC_REM_INV_REQ_ERR,
822 	IB_WC_REM_ACCESS_ERR,
823 	IB_WC_REM_OP_ERR,
824 	IB_WC_RETRY_EXC_ERR,
825 	IB_WC_RNR_RETRY_EXC_ERR,
826 	IB_WC_LOC_RDD_VIOL_ERR,
827 	IB_WC_REM_INV_RD_REQ_ERR,
828 	IB_WC_REM_ABORT_ERR,
829 	IB_WC_INV_EECN_ERR,
830 	IB_WC_INV_EEC_STATE_ERR,
831 	IB_WC_FATAL_ERR,
832 	IB_WC_RESP_TIMEOUT_ERR,
833 	IB_WC_GENERAL_ERR
834 };
835 
836 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
837 
838 enum ib_wc_opcode {
839 	IB_WC_SEND,
840 	IB_WC_RDMA_WRITE,
841 	IB_WC_RDMA_READ,
842 	IB_WC_COMP_SWAP,
843 	IB_WC_FETCH_ADD,
844 	IB_WC_LSO,
845 	IB_WC_LOCAL_INV,
846 	IB_WC_REG_MR,
847 	IB_WC_MASKED_COMP_SWAP,
848 	IB_WC_MASKED_FETCH_ADD,
849 /*
850  * Set value of IB_WC_RECV so consumers can test if a completion is a
851  * receive by testing (opcode & IB_WC_RECV).
852  */
853 	IB_WC_RECV			= 1 << 7,
854 	IB_WC_RECV_RDMA_WITH_IMM
855 };
856 
857 enum ib_wc_flags {
858 	IB_WC_GRH		= 1,
859 	IB_WC_WITH_IMM		= (1<<1),
860 	IB_WC_WITH_INVALIDATE	= (1<<2),
861 	IB_WC_IP_CSUM_OK	= (1<<3),
862 	IB_WC_WITH_SMAC		= (1<<4),
863 	IB_WC_WITH_VLAN		= (1<<5),
864 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
865 };
866 
867 struct ib_wc {
868 	union {
869 		u64		wr_id;
870 		struct ib_cqe	*wr_cqe;
871 	};
872 	enum ib_wc_status	status;
873 	enum ib_wc_opcode	opcode;
874 	u32			vendor_err;
875 	u32			byte_len;
876 	struct ib_qp	       *qp;
877 	union {
878 		__be32		imm_data;
879 		u32		invalidate_rkey;
880 	} ex;
881 	u32			src_qp;
882 	int			wc_flags;
883 	u16			pkey_index;
884 	u16			slid;
885 	u8			sl;
886 	u8			dlid_path_bits;
887 	u8			port_num;	/* valid only for DR SMPs on switches */
888 	u8			smac[ETH_ALEN];
889 	u16			vlan_id;
890 	u8			network_hdr_type;
891 };
892 
893 enum ib_cq_notify_flags {
894 	IB_CQ_SOLICITED			= 1 << 0,
895 	IB_CQ_NEXT_COMP			= 1 << 1,
896 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
897 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
898 };
899 
900 enum ib_srq_type {
901 	IB_SRQT_BASIC,
902 	IB_SRQT_XRC
903 };
904 
905 enum ib_srq_attr_mask {
906 	IB_SRQ_MAX_WR	= 1 << 0,
907 	IB_SRQ_LIMIT	= 1 << 1,
908 };
909 
910 struct ib_srq_attr {
911 	u32	max_wr;
912 	u32	max_sge;
913 	u32	srq_limit;
914 };
915 
916 struct ib_srq_init_attr {
917 	void		      (*event_handler)(struct ib_event *, void *);
918 	void		       *srq_context;
919 	struct ib_srq_attr	attr;
920 	enum ib_srq_type	srq_type;
921 
922 	union {
923 		struct {
924 			struct ib_xrcd *xrcd;
925 			struct ib_cq   *cq;
926 		} xrc;
927 	} ext;
928 };
929 
930 struct ib_qp_cap {
931 	u32	max_send_wr;
932 	u32	max_recv_wr;
933 	u32	max_send_sge;
934 	u32	max_recv_sge;
935 	u32	max_inline_data;
936 
937 	/*
938 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
939 	 * ib_create_qp() will calculate the right amount of neededed WRs
940 	 * and MRs based on this.
941 	 */
942 	u32	max_rdma_ctxs;
943 };
944 
945 enum ib_sig_type {
946 	IB_SIGNAL_ALL_WR,
947 	IB_SIGNAL_REQ_WR
948 };
949 
950 enum ib_qp_type {
951 	/*
952 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
953 	 * here (and in that order) since the MAD layer uses them as
954 	 * indices into a 2-entry table.
955 	 */
956 	IB_QPT_SMI,
957 	IB_QPT_GSI,
958 
959 	IB_QPT_RC,
960 	IB_QPT_UC,
961 	IB_QPT_UD,
962 	IB_QPT_RAW_IPV6,
963 	IB_QPT_RAW_ETHERTYPE,
964 	IB_QPT_RAW_PACKET = 8,
965 	IB_QPT_XRC_INI = 9,
966 	IB_QPT_XRC_TGT,
967 	IB_QPT_MAX,
968 	/* Reserve a range for qp types internal to the low level driver.
969 	 * These qp types will not be visible at the IB core layer, so the
970 	 * IB_QPT_MAX usages should not be affected in the core layer
971 	 */
972 	IB_QPT_RESERVED1 = 0x1000,
973 	IB_QPT_RESERVED2,
974 	IB_QPT_RESERVED3,
975 	IB_QPT_RESERVED4,
976 	IB_QPT_RESERVED5,
977 	IB_QPT_RESERVED6,
978 	IB_QPT_RESERVED7,
979 	IB_QPT_RESERVED8,
980 	IB_QPT_RESERVED9,
981 	IB_QPT_RESERVED10,
982 };
983 
984 enum ib_qp_create_flags {
985 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
986 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
987 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
988 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
989 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
990 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
991 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
992 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
993 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
994 	/* reserve bits 26-31 for low level drivers' internal use */
995 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
996 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
997 };
998 
999 /*
1000  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1001  * callback to destroy the passed in QP.
1002  */
1003 
1004 struct ib_qp_init_attr {
1005 	void                  (*event_handler)(struct ib_event *, void *);
1006 	void		       *qp_context;
1007 	struct ib_cq	       *send_cq;
1008 	struct ib_cq	       *recv_cq;
1009 	struct ib_srq	       *srq;
1010 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1011 	struct ib_qp_cap	cap;
1012 	enum ib_sig_type	sq_sig_type;
1013 	enum ib_qp_type		qp_type;
1014 	enum ib_qp_create_flags	create_flags;
1015 
1016 	/*
1017 	 * Only needed for special QP types, or when using the RW API.
1018 	 */
1019 	u8			port_num;
1020 	struct ib_rwq_ind_table *rwq_ind_tbl;
1021 };
1022 
1023 struct ib_qp_open_attr {
1024 	void                  (*event_handler)(struct ib_event *, void *);
1025 	void		       *qp_context;
1026 	u32			qp_num;
1027 	enum ib_qp_type		qp_type;
1028 };
1029 
1030 enum ib_rnr_timeout {
1031 	IB_RNR_TIMER_655_36 =  0,
1032 	IB_RNR_TIMER_000_01 =  1,
1033 	IB_RNR_TIMER_000_02 =  2,
1034 	IB_RNR_TIMER_000_03 =  3,
1035 	IB_RNR_TIMER_000_04 =  4,
1036 	IB_RNR_TIMER_000_06 =  5,
1037 	IB_RNR_TIMER_000_08 =  6,
1038 	IB_RNR_TIMER_000_12 =  7,
1039 	IB_RNR_TIMER_000_16 =  8,
1040 	IB_RNR_TIMER_000_24 =  9,
1041 	IB_RNR_TIMER_000_32 = 10,
1042 	IB_RNR_TIMER_000_48 = 11,
1043 	IB_RNR_TIMER_000_64 = 12,
1044 	IB_RNR_TIMER_000_96 = 13,
1045 	IB_RNR_TIMER_001_28 = 14,
1046 	IB_RNR_TIMER_001_92 = 15,
1047 	IB_RNR_TIMER_002_56 = 16,
1048 	IB_RNR_TIMER_003_84 = 17,
1049 	IB_RNR_TIMER_005_12 = 18,
1050 	IB_RNR_TIMER_007_68 = 19,
1051 	IB_RNR_TIMER_010_24 = 20,
1052 	IB_RNR_TIMER_015_36 = 21,
1053 	IB_RNR_TIMER_020_48 = 22,
1054 	IB_RNR_TIMER_030_72 = 23,
1055 	IB_RNR_TIMER_040_96 = 24,
1056 	IB_RNR_TIMER_061_44 = 25,
1057 	IB_RNR_TIMER_081_92 = 26,
1058 	IB_RNR_TIMER_122_88 = 27,
1059 	IB_RNR_TIMER_163_84 = 28,
1060 	IB_RNR_TIMER_245_76 = 29,
1061 	IB_RNR_TIMER_327_68 = 30,
1062 	IB_RNR_TIMER_491_52 = 31
1063 };
1064 
1065 enum ib_qp_attr_mask {
1066 	IB_QP_STATE			= 1,
1067 	IB_QP_CUR_STATE			= (1<<1),
1068 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1069 	IB_QP_ACCESS_FLAGS		= (1<<3),
1070 	IB_QP_PKEY_INDEX		= (1<<4),
1071 	IB_QP_PORT			= (1<<5),
1072 	IB_QP_QKEY			= (1<<6),
1073 	IB_QP_AV			= (1<<7),
1074 	IB_QP_PATH_MTU			= (1<<8),
1075 	IB_QP_TIMEOUT			= (1<<9),
1076 	IB_QP_RETRY_CNT			= (1<<10),
1077 	IB_QP_RNR_RETRY			= (1<<11),
1078 	IB_QP_RQ_PSN			= (1<<12),
1079 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1080 	IB_QP_ALT_PATH			= (1<<14),
1081 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1082 	IB_QP_SQ_PSN			= (1<<16),
1083 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1084 	IB_QP_PATH_MIG_STATE		= (1<<18),
1085 	IB_QP_CAP			= (1<<19),
1086 	IB_QP_DEST_QPN			= (1<<20),
1087 	IB_QP_RESERVED1			= (1<<21),
1088 	IB_QP_RESERVED2			= (1<<22),
1089 	IB_QP_RESERVED3			= (1<<23),
1090 	IB_QP_RESERVED4			= (1<<24),
1091 };
1092 
1093 enum ib_qp_state {
1094 	IB_QPS_RESET,
1095 	IB_QPS_INIT,
1096 	IB_QPS_RTR,
1097 	IB_QPS_RTS,
1098 	IB_QPS_SQD,
1099 	IB_QPS_SQE,
1100 	IB_QPS_ERR
1101 };
1102 
1103 enum ib_mig_state {
1104 	IB_MIG_MIGRATED,
1105 	IB_MIG_REARM,
1106 	IB_MIG_ARMED
1107 };
1108 
1109 enum ib_mw_type {
1110 	IB_MW_TYPE_1 = 1,
1111 	IB_MW_TYPE_2 = 2
1112 };
1113 
1114 struct ib_qp_attr {
1115 	enum ib_qp_state	qp_state;
1116 	enum ib_qp_state	cur_qp_state;
1117 	enum ib_mtu		path_mtu;
1118 	enum ib_mig_state	path_mig_state;
1119 	u32			qkey;
1120 	u32			rq_psn;
1121 	u32			sq_psn;
1122 	u32			dest_qp_num;
1123 	int			qp_access_flags;
1124 	struct ib_qp_cap	cap;
1125 	struct ib_ah_attr	ah_attr;
1126 	struct ib_ah_attr	alt_ah_attr;
1127 	u16			pkey_index;
1128 	u16			alt_pkey_index;
1129 	u8			en_sqd_async_notify;
1130 	u8			sq_draining;
1131 	u8			max_rd_atomic;
1132 	u8			max_dest_rd_atomic;
1133 	u8			min_rnr_timer;
1134 	u8			port_num;
1135 	u8			timeout;
1136 	u8			retry_cnt;
1137 	u8			rnr_retry;
1138 	u8			alt_port_num;
1139 	u8			alt_timeout;
1140 };
1141 
1142 enum ib_wr_opcode {
1143 	IB_WR_RDMA_WRITE,
1144 	IB_WR_RDMA_WRITE_WITH_IMM,
1145 	IB_WR_SEND,
1146 	IB_WR_SEND_WITH_IMM,
1147 	IB_WR_RDMA_READ,
1148 	IB_WR_ATOMIC_CMP_AND_SWP,
1149 	IB_WR_ATOMIC_FETCH_AND_ADD,
1150 	IB_WR_LSO,
1151 	IB_WR_SEND_WITH_INV,
1152 	IB_WR_RDMA_READ_WITH_INV,
1153 	IB_WR_LOCAL_INV,
1154 	IB_WR_REG_MR,
1155 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1156 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1157 	IB_WR_REG_SIG_MR,
1158 	/* reserve values for low level drivers' internal use.
1159 	 * These values will not be used at all in the ib core layer.
1160 	 */
1161 	IB_WR_RESERVED1 = 0xf0,
1162 	IB_WR_RESERVED2,
1163 	IB_WR_RESERVED3,
1164 	IB_WR_RESERVED4,
1165 	IB_WR_RESERVED5,
1166 	IB_WR_RESERVED6,
1167 	IB_WR_RESERVED7,
1168 	IB_WR_RESERVED8,
1169 	IB_WR_RESERVED9,
1170 	IB_WR_RESERVED10,
1171 };
1172 
1173 enum ib_send_flags {
1174 	IB_SEND_FENCE		= 1,
1175 	IB_SEND_SIGNALED	= (1<<1),
1176 	IB_SEND_SOLICITED	= (1<<2),
1177 	IB_SEND_INLINE		= (1<<3),
1178 	IB_SEND_IP_CSUM		= (1<<4),
1179 
1180 	/* reserve bits 26-31 for low level drivers' internal use */
1181 	IB_SEND_RESERVED_START	= (1 << 26),
1182 	IB_SEND_RESERVED_END	= (1 << 31),
1183 };
1184 
1185 struct ib_sge {
1186 	u64	addr;
1187 	u32	length;
1188 	u32	lkey;
1189 };
1190 
1191 struct ib_cqe {
1192 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1193 };
1194 
1195 struct ib_send_wr {
1196 	struct ib_send_wr      *next;
1197 	union {
1198 		u64		wr_id;
1199 		struct ib_cqe	*wr_cqe;
1200 	};
1201 	struct ib_sge	       *sg_list;
1202 	int			num_sge;
1203 	enum ib_wr_opcode	opcode;
1204 	int			send_flags;
1205 	union {
1206 		__be32		imm_data;
1207 		u32		invalidate_rkey;
1208 	} ex;
1209 };
1210 
1211 struct ib_rdma_wr {
1212 	struct ib_send_wr	wr;
1213 	u64			remote_addr;
1214 	u32			rkey;
1215 };
1216 
1217 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1218 {
1219 	return container_of(wr, struct ib_rdma_wr, wr);
1220 }
1221 
1222 struct ib_atomic_wr {
1223 	struct ib_send_wr	wr;
1224 	u64			remote_addr;
1225 	u64			compare_add;
1226 	u64			swap;
1227 	u64			compare_add_mask;
1228 	u64			swap_mask;
1229 	u32			rkey;
1230 };
1231 
1232 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1233 {
1234 	return container_of(wr, struct ib_atomic_wr, wr);
1235 }
1236 
1237 struct ib_ud_wr {
1238 	struct ib_send_wr	wr;
1239 	struct ib_ah		*ah;
1240 	void			*header;
1241 	int			hlen;
1242 	int			mss;
1243 	u32			remote_qpn;
1244 	u32			remote_qkey;
1245 	u16			pkey_index; /* valid for GSI only */
1246 	u8			port_num;   /* valid for DR SMPs on switch only */
1247 };
1248 
1249 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1250 {
1251 	return container_of(wr, struct ib_ud_wr, wr);
1252 }
1253 
1254 struct ib_reg_wr {
1255 	struct ib_send_wr	wr;
1256 	struct ib_mr		*mr;
1257 	u32			key;
1258 	int			access;
1259 };
1260 
1261 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1262 {
1263 	return container_of(wr, struct ib_reg_wr, wr);
1264 }
1265 
1266 struct ib_sig_handover_wr {
1267 	struct ib_send_wr	wr;
1268 	struct ib_sig_attrs    *sig_attrs;
1269 	struct ib_mr	       *sig_mr;
1270 	int			access_flags;
1271 	struct ib_sge	       *prot;
1272 };
1273 
1274 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1275 {
1276 	return container_of(wr, struct ib_sig_handover_wr, wr);
1277 }
1278 
1279 struct ib_recv_wr {
1280 	struct ib_recv_wr      *next;
1281 	union {
1282 		u64		wr_id;
1283 		struct ib_cqe	*wr_cqe;
1284 	};
1285 	struct ib_sge	       *sg_list;
1286 	int			num_sge;
1287 };
1288 
1289 enum ib_access_flags {
1290 	IB_ACCESS_LOCAL_WRITE	= 1,
1291 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1292 	IB_ACCESS_REMOTE_READ	= (1<<2),
1293 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1294 	IB_ACCESS_MW_BIND	= (1<<4),
1295 	IB_ZERO_BASED		= (1<<5),
1296 	IB_ACCESS_ON_DEMAND     = (1<<6),
1297 };
1298 
1299 /*
1300  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1301  * are hidden here instead of a uapi header!
1302  */
1303 enum ib_mr_rereg_flags {
1304 	IB_MR_REREG_TRANS	= 1,
1305 	IB_MR_REREG_PD		= (1<<1),
1306 	IB_MR_REREG_ACCESS	= (1<<2),
1307 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1308 };
1309 
1310 struct ib_fmr_attr {
1311 	int	max_pages;
1312 	int	max_maps;
1313 	u8	page_shift;
1314 };
1315 
1316 struct ib_umem;
1317 
1318 struct ib_ucontext {
1319 	struct ib_device       *device;
1320 	struct list_head	pd_list;
1321 	struct list_head	mr_list;
1322 	struct list_head	mw_list;
1323 	struct list_head	cq_list;
1324 	struct list_head	qp_list;
1325 	struct list_head	srq_list;
1326 	struct list_head	ah_list;
1327 	struct list_head	xrcd_list;
1328 	struct list_head	rule_list;
1329 	struct list_head	wq_list;
1330 	struct list_head	rwq_ind_tbl_list;
1331 	int			closing;
1332 
1333 	struct pid             *tgid;
1334 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1335 	struct rb_root      umem_tree;
1336 	/*
1337 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1338 	 * mmu notifiers registration.
1339 	 */
1340 	struct rw_semaphore	umem_rwsem;
1341 	void (*invalidate_range)(struct ib_umem *umem,
1342 				 unsigned long start, unsigned long end);
1343 
1344 	struct mmu_notifier	mn;
1345 	atomic_t		notifier_count;
1346 	/* A list of umems that don't have private mmu notifier counters yet. */
1347 	struct list_head	no_private_counters;
1348 	int                     odp_mrs_count;
1349 #endif
1350 };
1351 
1352 struct ib_uobject {
1353 	u64			user_handle;	/* handle given to us by userspace */
1354 	struct ib_ucontext     *context;	/* associated user context */
1355 	void		       *object;		/* containing object */
1356 	struct list_head	list;		/* link to context's list */
1357 	int			id;		/* index into kernel idr */
1358 	struct kref		ref;
1359 	struct rw_semaphore	mutex;		/* protects .live */
1360 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1361 	int			live;
1362 };
1363 
1364 struct ib_udata {
1365 	const void __user *inbuf;
1366 	void __user *outbuf;
1367 	size_t       inlen;
1368 	size_t       outlen;
1369 };
1370 
1371 struct ib_pd {
1372 	u32			local_dma_lkey;
1373 	struct ib_device       *device;
1374 	struct ib_uobject      *uobject;
1375 	atomic_t          	usecnt; /* count all resources */
1376 	struct ib_mr	       *local_mr;
1377 };
1378 
1379 struct ib_xrcd {
1380 	struct ib_device       *device;
1381 	atomic_t		usecnt; /* count all exposed resources */
1382 	struct inode	       *inode;
1383 
1384 	struct mutex		tgt_qp_mutex;
1385 	struct list_head	tgt_qp_list;
1386 };
1387 
1388 struct ib_ah {
1389 	struct ib_device	*device;
1390 	struct ib_pd		*pd;
1391 	struct ib_uobject	*uobject;
1392 };
1393 
1394 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1395 
1396 enum ib_poll_context {
1397 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1398 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1399 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1400 };
1401 
1402 struct ib_cq {
1403 	struct ib_device       *device;
1404 	struct ib_uobject      *uobject;
1405 	ib_comp_handler   	comp_handler;
1406 	void                  (*event_handler)(struct ib_event *, void *);
1407 	void                   *cq_context;
1408 	int               	cqe;
1409 	atomic_t          	usecnt; /* count number of work queues */
1410 	enum ib_poll_context	poll_ctx;
1411 	struct ib_wc		*wc;
1412 	union {
1413 		struct irq_poll		iop;
1414 		struct work_struct	work;
1415 	};
1416 };
1417 
1418 struct ib_srq {
1419 	struct ib_device       *device;
1420 	struct ib_pd	       *pd;
1421 	struct ib_uobject      *uobject;
1422 	void		      (*event_handler)(struct ib_event *, void *);
1423 	void		       *srq_context;
1424 	enum ib_srq_type	srq_type;
1425 	atomic_t		usecnt;
1426 
1427 	union {
1428 		struct {
1429 			struct ib_xrcd *xrcd;
1430 			struct ib_cq   *cq;
1431 			u32		srq_num;
1432 		} xrc;
1433 	} ext;
1434 };
1435 
1436 enum ib_wq_type {
1437 	IB_WQT_RQ
1438 };
1439 
1440 enum ib_wq_state {
1441 	IB_WQS_RESET,
1442 	IB_WQS_RDY,
1443 	IB_WQS_ERR
1444 };
1445 
1446 struct ib_wq {
1447 	struct ib_device       *device;
1448 	struct ib_uobject      *uobject;
1449 	void		    *wq_context;
1450 	void		    (*event_handler)(struct ib_event *, void *);
1451 	struct ib_pd	       *pd;
1452 	struct ib_cq	       *cq;
1453 	u32		wq_num;
1454 	enum ib_wq_state       state;
1455 	enum ib_wq_type	wq_type;
1456 	atomic_t		usecnt;
1457 };
1458 
1459 struct ib_wq_init_attr {
1460 	void		       *wq_context;
1461 	enum ib_wq_type	wq_type;
1462 	u32		max_wr;
1463 	u32		max_sge;
1464 	struct	ib_cq	       *cq;
1465 	void		    (*event_handler)(struct ib_event *, void *);
1466 };
1467 
1468 enum ib_wq_attr_mask {
1469 	IB_WQ_STATE	= 1 << 0,
1470 	IB_WQ_CUR_STATE	= 1 << 1,
1471 };
1472 
1473 struct ib_wq_attr {
1474 	enum	ib_wq_state	wq_state;
1475 	enum	ib_wq_state	curr_wq_state;
1476 };
1477 
1478 struct ib_rwq_ind_table {
1479 	struct ib_device	*device;
1480 	struct ib_uobject      *uobject;
1481 	atomic_t		usecnt;
1482 	u32		ind_tbl_num;
1483 	u32		log_ind_tbl_size;
1484 	struct ib_wq	**ind_tbl;
1485 };
1486 
1487 struct ib_rwq_ind_table_init_attr {
1488 	u32		log_ind_tbl_size;
1489 	/* Each entry is a pointer to Receive Work Queue */
1490 	struct ib_wq	**ind_tbl;
1491 };
1492 
1493 /*
1494  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1495  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1496  */
1497 struct ib_qp {
1498 	struct ib_device       *device;
1499 	struct ib_pd	       *pd;
1500 	struct ib_cq	       *send_cq;
1501 	struct ib_cq	       *recv_cq;
1502 	spinlock_t		mr_lock;
1503 	int			mrs_used;
1504 	struct list_head	rdma_mrs;
1505 	struct list_head	sig_mrs;
1506 	struct ib_srq	       *srq;
1507 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1508 	struct list_head	xrcd_list;
1509 
1510 	/* count times opened, mcast attaches, flow attaches */
1511 	atomic_t		usecnt;
1512 	struct list_head	open_list;
1513 	struct ib_qp           *real_qp;
1514 	struct ib_uobject      *uobject;
1515 	void                  (*event_handler)(struct ib_event *, void *);
1516 	void		       *qp_context;
1517 	u32			qp_num;
1518 	u32			max_write_sge;
1519 	u32			max_read_sge;
1520 	enum ib_qp_type		qp_type;
1521 	struct ib_rwq_ind_table *rwq_ind_tbl;
1522 };
1523 
1524 struct ib_mr {
1525 	struct ib_device  *device;
1526 	struct ib_pd	  *pd;
1527 	u32		   lkey;
1528 	u32		   rkey;
1529 	u64		   iova;
1530 	u32		   length;
1531 	unsigned int	   page_size;
1532 	bool		   need_inval;
1533 	union {
1534 		struct ib_uobject	*uobject;	/* user */
1535 		struct list_head	qp_entry;	/* FR */
1536 	};
1537 };
1538 
1539 struct ib_mw {
1540 	struct ib_device	*device;
1541 	struct ib_pd		*pd;
1542 	struct ib_uobject	*uobject;
1543 	u32			rkey;
1544 	enum ib_mw_type         type;
1545 };
1546 
1547 struct ib_fmr {
1548 	struct ib_device	*device;
1549 	struct ib_pd		*pd;
1550 	struct list_head	list;
1551 	u32			lkey;
1552 	u32			rkey;
1553 };
1554 
1555 /* Supported steering options */
1556 enum ib_flow_attr_type {
1557 	/* steering according to rule specifications */
1558 	IB_FLOW_ATTR_NORMAL		= 0x0,
1559 	/* default unicast and multicast rule -
1560 	 * receive all Eth traffic which isn't steered to any QP
1561 	 */
1562 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1563 	/* default multicast rule -
1564 	 * receive all Eth multicast traffic which isn't steered to any QP
1565 	 */
1566 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1567 	/* sniffer rule - receive all port traffic */
1568 	IB_FLOW_ATTR_SNIFFER		= 0x3
1569 };
1570 
1571 /* Supported steering header types */
1572 enum ib_flow_spec_type {
1573 	/* L2 headers*/
1574 	IB_FLOW_SPEC_ETH	= 0x20,
1575 	IB_FLOW_SPEC_IB		= 0x22,
1576 	/* L3 header*/
1577 	IB_FLOW_SPEC_IPV4	= 0x30,
1578 	IB_FLOW_SPEC_IPV6	= 0x31,
1579 	/* L4 headers*/
1580 	IB_FLOW_SPEC_TCP	= 0x40,
1581 	IB_FLOW_SPEC_UDP	= 0x41
1582 };
1583 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1584 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1585 
1586 /* Flow steering rule priority is set according to it's domain.
1587  * Lower domain value means higher priority.
1588  */
1589 enum ib_flow_domain {
1590 	IB_FLOW_DOMAIN_USER,
1591 	IB_FLOW_DOMAIN_ETHTOOL,
1592 	IB_FLOW_DOMAIN_RFS,
1593 	IB_FLOW_DOMAIN_NIC,
1594 	IB_FLOW_DOMAIN_NUM /* Must be last */
1595 };
1596 
1597 enum ib_flow_flags {
1598 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1599 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1600 };
1601 
1602 struct ib_flow_eth_filter {
1603 	u8	dst_mac[6];
1604 	u8	src_mac[6];
1605 	__be16	ether_type;
1606 	__be16	vlan_tag;
1607 };
1608 
1609 struct ib_flow_spec_eth {
1610 	enum ib_flow_spec_type	  type;
1611 	u16			  size;
1612 	struct ib_flow_eth_filter val;
1613 	struct ib_flow_eth_filter mask;
1614 };
1615 
1616 struct ib_flow_ib_filter {
1617 	__be16 dlid;
1618 	__u8   sl;
1619 };
1620 
1621 struct ib_flow_spec_ib {
1622 	enum ib_flow_spec_type	 type;
1623 	u16			 size;
1624 	struct ib_flow_ib_filter val;
1625 	struct ib_flow_ib_filter mask;
1626 };
1627 
1628 struct ib_flow_ipv4_filter {
1629 	__be32	src_ip;
1630 	__be32	dst_ip;
1631 };
1632 
1633 struct ib_flow_spec_ipv4 {
1634 	enum ib_flow_spec_type	   type;
1635 	u16			   size;
1636 	struct ib_flow_ipv4_filter val;
1637 	struct ib_flow_ipv4_filter mask;
1638 };
1639 
1640 struct ib_flow_ipv6_filter {
1641 	u8	src_ip[16];
1642 	u8	dst_ip[16];
1643 };
1644 
1645 struct ib_flow_spec_ipv6 {
1646 	enum ib_flow_spec_type	   type;
1647 	u16			   size;
1648 	struct ib_flow_ipv6_filter val;
1649 	struct ib_flow_ipv6_filter mask;
1650 };
1651 
1652 struct ib_flow_tcp_udp_filter {
1653 	__be16	dst_port;
1654 	__be16	src_port;
1655 };
1656 
1657 struct ib_flow_spec_tcp_udp {
1658 	enum ib_flow_spec_type	      type;
1659 	u16			      size;
1660 	struct ib_flow_tcp_udp_filter val;
1661 	struct ib_flow_tcp_udp_filter mask;
1662 };
1663 
1664 union ib_flow_spec {
1665 	struct {
1666 		enum ib_flow_spec_type	type;
1667 		u16			size;
1668 	};
1669 	struct ib_flow_spec_eth		eth;
1670 	struct ib_flow_spec_ib		ib;
1671 	struct ib_flow_spec_ipv4        ipv4;
1672 	struct ib_flow_spec_tcp_udp	tcp_udp;
1673 	struct ib_flow_spec_ipv6        ipv6;
1674 };
1675 
1676 struct ib_flow_attr {
1677 	enum ib_flow_attr_type type;
1678 	u16	     size;
1679 	u16	     priority;
1680 	u32	     flags;
1681 	u8	     num_of_specs;
1682 	u8	     port;
1683 	/* Following are the optional layers according to user request
1684 	 * struct ib_flow_spec_xxx
1685 	 * struct ib_flow_spec_yyy
1686 	 */
1687 };
1688 
1689 struct ib_flow {
1690 	struct ib_qp		*qp;
1691 	struct ib_uobject	*uobject;
1692 };
1693 
1694 struct ib_mad_hdr;
1695 struct ib_grh;
1696 
1697 enum ib_process_mad_flags {
1698 	IB_MAD_IGNORE_MKEY	= 1,
1699 	IB_MAD_IGNORE_BKEY	= 2,
1700 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1701 };
1702 
1703 enum ib_mad_result {
1704 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1705 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1706 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1707 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1708 };
1709 
1710 #define IB_DEVICE_NAME_MAX 64
1711 
1712 struct ib_cache {
1713 	rwlock_t                lock;
1714 	struct ib_event_handler event_handler;
1715 	struct ib_pkey_cache  **pkey_cache;
1716 	struct ib_gid_table   **gid_cache;
1717 	u8                     *lmc_cache;
1718 };
1719 
1720 struct ib_dma_mapping_ops {
1721 	int		(*mapping_error)(struct ib_device *dev,
1722 					 u64 dma_addr);
1723 	u64		(*map_single)(struct ib_device *dev,
1724 				      void *ptr, size_t size,
1725 				      enum dma_data_direction direction);
1726 	void		(*unmap_single)(struct ib_device *dev,
1727 					u64 addr, size_t size,
1728 					enum dma_data_direction direction);
1729 	u64		(*map_page)(struct ib_device *dev,
1730 				    struct page *page, unsigned long offset,
1731 				    size_t size,
1732 				    enum dma_data_direction direction);
1733 	void		(*unmap_page)(struct ib_device *dev,
1734 				      u64 addr, size_t size,
1735 				      enum dma_data_direction direction);
1736 	int		(*map_sg)(struct ib_device *dev,
1737 				  struct scatterlist *sg, int nents,
1738 				  enum dma_data_direction direction);
1739 	void		(*unmap_sg)(struct ib_device *dev,
1740 				    struct scatterlist *sg, int nents,
1741 				    enum dma_data_direction direction);
1742 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1743 					       u64 dma_handle,
1744 					       size_t size,
1745 					       enum dma_data_direction dir);
1746 	void		(*sync_single_for_device)(struct ib_device *dev,
1747 						  u64 dma_handle,
1748 						  size_t size,
1749 						  enum dma_data_direction dir);
1750 	void		*(*alloc_coherent)(struct ib_device *dev,
1751 					   size_t size,
1752 					   u64 *dma_handle,
1753 					   gfp_t flag);
1754 	void		(*free_coherent)(struct ib_device *dev,
1755 					 size_t size, void *cpu_addr,
1756 					 u64 dma_handle);
1757 };
1758 
1759 struct iw_cm_verbs;
1760 
1761 struct ib_port_immutable {
1762 	int                           pkey_tbl_len;
1763 	int                           gid_tbl_len;
1764 	u32                           core_cap_flags;
1765 	u32                           max_mad_size;
1766 };
1767 
1768 struct ib_device {
1769 	struct device                *dma_device;
1770 
1771 	char                          name[IB_DEVICE_NAME_MAX];
1772 
1773 	struct list_head              event_handler_list;
1774 	spinlock_t                    event_handler_lock;
1775 
1776 	spinlock_t                    client_data_lock;
1777 	struct list_head              core_list;
1778 	/* Access to the client_data_list is protected by the client_data_lock
1779 	 * spinlock and the lists_rwsem read-write semaphore */
1780 	struct list_head              client_data_list;
1781 
1782 	struct ib_cache               cache;
1783 	/**
1784 	 * port_immutable is indexed by port number
1785 	 */
1786 	struct ib_port_immutable     *port_immutable;
1787 
1788 	int			      num_comp_vectors;
1789 
1790 	struct iw_cm_verbs	     *iwcm;
1791 
1792 	/**
1793 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1794 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
1795 	 *   core when the device is removed.  A lifespan of -1 in the return
1796 	 *   struct tells the core to set a default lifespan.
1797 	 */
1798 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
1799 						     u8 port_num);
1800 	/**
1801 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1802 	 * @index - The index in the value array we wish to have updated, or
1803 	 *   num_counters if we want all stats updated
1804 	 * Return codes -
1805 	 *   < 0 - Error, no counters updated
1806 	 *   index - Updated the single counter pointed to by index
1807 	 *   num_counters - Updated all counters (will reset the timestamp
1808 	 *     and prevent further calls for lifespan milliseconds)
1809 	 * Drivers are allowed to update all counters in leiu of just the
1810 	 *   one given in index at their option
1811 	 */
1812 	int		           (*get_hw_stats)(struct ib_device *device,
1813 						   struct rdma_hw_stats *stats,
1814 						   u8 port, int index);
1815 	int		           (*query_device)(struct ib_device *device,
1816 						   struct ib_device_attr *device_attr,
1817 						   struct ib_udata *udata);
1818 	int		           (*query_port)(struct ib_device *device,
1819 						 u8 port_num,
1820 						 struct ib_port_attr *port_attr);
1821 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1822 						     u8 port_num);
1823 	/* When calling get_netdev, the HW vendor's driver should return the
1824 	 * net device of device @device at port @port_num or NULL if such
1825 	 * a net device doesn't exist. The vendor driver should call dev_hold
1826 	 * on this net device. The HW vendor's device driver must guarantee
1827 	 * that this function returns NULL before the net device reaches
1828 	 * NETDEV_UNREGISTER_FINAL state.
1829 	 */
1830 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1831 						 u8 port_num);
1832 	int		           (*query_gid)(struct ib_device *device,
1833 						u8 port_num, int index,
1834 						union ib_gid *gid);
1835 	/* When calling add_gid, the HW vendor's driver should
1836 	 * add the gid of device @device at gid index @index of
1837 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1838 	 * the network device related to this gid is available
1839 	 * at @attr. @context allows the HW vendor driver to store extra
1840 	 * information together with a GID entry. The HW vendor may allocate
1841 	 * memory to contain this information and store it in @context when a
1842 	 * new GID entry is written to. Params are consistent until the next
1843 	 * call of add_gid or delete_gid. The function should return 0 on
1844 	 * success or error otherwise. The function could be called
1845 	 * concurrently for different ports. This function is only called
1846 	 * when roce_gid_table is used.
1847 	 */
1848 	int		           (*add_gid)(struct ib_device *device,
1849 					      u8 port_num,
1850 					      unsigned int index,
1851 					      const union ib_gid *gid,
1852 					      const struct ib_gid_attr *attr,
1853 					      void **context);
1854 	/* When calling del_gid, the HW vendor's driver should delete the
1855 	 * gid of device @device at gid index @index of port @port_num.
1856 	 * Upon the deletion of a GID entry, the HW vendor must free any
1857 	 * allocated memory. The caller will clear @context afterwards.
1858 	 * This function is only called when roce_gid_table is used.
1859 	 */
1860 	int		           (*del_gid)(struct ib_device *device,
1861 					      u8 port_num,
1862 					      unsigned int index,
1863 					      void **context);
1864 	int		           (*query_pkey)(struct ib_device *device,
1865 						 u8 port_num, u16 index, u16 *pkey);
1866 	int		           (*modify_device)(struct ib_device *device,
1867 						    int device_modify_mask,
1868 						    struct ib_device_modify *device_modify);
1869 	int		           (*modify_port)(struct ib_device *device,
1870 						  u8 port_num, int port_modify_mask,
1871 						  struct ib_port_modify *port_modify);
1872 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1873 						     struct ib_udata *udata);
1874 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1875 	int                        (*mmap)(struct ib_ucontext *context,
1876 					   struct vm_area_struct *vma);
1877 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1878 					       struct ib_ucontext *context,
1879 					       struct ib_udata *udata);
1880 	int                        (*dealloc_pd)(struct ib_pd *pd);
1881 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1882 						struct ib_ah_attr *ah_attr);
1883 	int                        (*modify_ah)(struct ib_ah *ah,
1884 						struct ib_ah_attr *ah_attr);
1885 	int                        (*query_ah)(struct ib_ah *ah,
1886 					       struct ib_ah_attr *ah_attr);
1887 	int                        (*destroy_ah)(struct ib_ah *ah);
1888 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1889 						 struct ib_srq_init_attr *srq_init_attr,
1890 						 struct ib_udata *udata);
1891 	int                        (*modify_srq)(struct ib_srq *srq,
1892 						 struct ib_srq_attr *srq_attr,
1893 						 enum ib_srq_attr_mask srq_attr_mask,
1894 						 struct ib_udata *udata);
1895 	int                        (*query_srq)(struct ib_srq *srq,
1896 						struct ib_srq_attr *srq_attr);
1897 	int                        (*destroy_srq)(struct ib_srq *srq);
1898 	int                        (*post_srq_recv)(struct ib_srq *srq,
1899 						    struct ib_recv_wr *recv_wr,
1900 						    struct ib_recv_wr **bad_recv_wr);
1901 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1902 						struct ib_qp_init_attr *qp_init_attr,
1903 						struct ib_udata *udata);
1904 	int                        (*modify_qp)(struct ib_qp *qp,
1905 						struct ib_qp_attr *qp_attr,
1906 						int qp_attr_mask,
1907 						struct ib_udata *udata);
1908 	int                        (*query_qp)(struct ib_qp *qp,
1909 					       struct ib_qp_attr *qp_attr,
1910 					       int qp_attr_mask,
1911 					       struct ib_qp_init_attr *qp_init_attr);
1912 	int                        (*destroy_qp)(struct ib_qp *qp);
1913 	int                        (*post_send)(struct ib_qp *qp,
1914 						struct ib_send_wr *send_wr,
1915 						struct ib_send_wr **bad_send_wr);
1916 	int                        (*post_recv)(struct ib_qp *qp,
1917 						struct ib_recv_wr *recv_wr,
1918 						struct ib_recv_wr **bad_recv_wr);
1919 	struct ib_cq *             (*create_cq)(struct ib_device *device,
1920 						const struct ib_cq_init_attr *attr,
1921 						struct ib_ucontext *context,
1922 						struct ib_udata *udata);
1923 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1924 						u16 cq_period);
1925 	int                        (*destroy_cq)(struct ib_cq *cq);
1926 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1927 						struct ib_udata *udata);
1928 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1929 					      struct ib_wc *wc);
1930 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1931 	int                        (*req_notify_cq)(struct ib_cq *cq,
1932 						    enum ib_cq_notify_flags flags);
1933 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1934 						      int wc_cnt);
1935 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1936 						 int mr_access_flags);
1937 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1938 						  u64 start, u64 length,
1939 						  u64 virt_addr,
1940 						  int mr_access_flags,
1941 						  struct ib_udata *udata);
1942 	int			   (*rereg_user_mr)(struct ib_mr *mr,
1943 						    int flags,
1944 						    u64 start, u64 length,
1945 						    u64 virt_addr,
1946 						    int mr_access_flags,
1947 						    struct ib_pd *pd,
1948 						    struct ib_udata *udata);
1949 	int                        (*dereg_mr)(struct ib_mr *mr);
1950 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
1951 					       enum ib_mr_type mr_type,
1952 					       u32 max_num_sg);
1953 	int                        (*map_mr_sg)(struct ib_mr *mr,
1954 						struct scatterlist *sg,
1955 						int sg_nents,
1956 						unsigned int *sg_offset);
1957 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
1958 					       enum ib_mw_type type,
1959 					       struct ib_udata *udata);
1960 	int                        (*dealloc_mw)(struct ib_mw *mw);
1961 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1962 						int mr_access_flags,
1963 						struct ib_fmr_attr *fmr_attr);
1964 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1965 						   u64 *page_list, int list_len,
1966 						   u64 iova);
1967 	int		           (*unmap_fmr)(struct list_head *fmr_list);
1968 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1969 	int                        (*attach_mcast)(struct ib_qp *qp,
1970 						   union ib_gid *gid,
1971 						   u16 lid);
1972 	int                        (*detach_mcast)(struct ib_qp *qp,
1973 						   union ib_gid *gid,
1974 						   u16 lid);
1975 	int                        (*process_mad)(struct ib_device *device,
1976 						  int process_mad_flags,
1977 						  u8 port_num,
1978 						  const struct ib_wc *in_wc,
1979 						  const struct ib_grh *in_grh,
1980 						  const struct ib_mad_hdr *in_mad,
1981 						  size_t in_mad_size,
1982 						  struct ib_mad_hdr *out_mad,
1983 						  size_t *out_mad_size,
1984 						  u16 *out_mad_pkey_index);
1985 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
1986 						 struct ib_ucontext *ucontext,
1987 						 struct ib_udata *udata);
1988 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1989 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
1990 						  struct ib_flow_attr
1991 						  *flow_attr,
1992 						  int domain);
1993 	int			   (*destroy_flow)(struct ib_flow *flow_id);
1994 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1995 						      struct ib_mr_status *mr_status);
1996 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
1997 	void			   (*drain_rq)(struct ib_qp *qp);
1998 	void			   (*drain_sq)(struct ib_qp *qp);
1999 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2000 							int state);
2001 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2002 						   struct ifla_vf_info *ivf);
2003 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2004 						   struct ifla_vf_stats *stats);
2005 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2006 						  int type);
2007 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2008 						struct ib_wq_init_attr *init_attr,
2009 						struct ib_udata *udata);
2010 	int			   (*destroy_wq)(struct ib_wq *wq);
2011 	int			   (*modify_wq)(struct ib_wq *wq,
2012 						struct ib_wq_attr *attr,
2013 						u32 wq_attr_mask,
2014 						struct ib_udata *udata);
2015 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2016 							   struct ib_rwq_ind_table_init_attr *init_attr,
2017 							   struct ib_udata *udata);
2018 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2019 	struct ib_dma_mapping_ops   *dma_ops;
2020 
2021 	struct module               *owner;
2022 	struct device                dev;
2023 	struct kobject               *ports_parent;
2024 	struct list_head             port_list;
2025 
2026 	enum {
2027 		IB_DEV_UNINITIALIZED,
2028 		IB_DEV_REGISTERED,
2029 		IB_DEV_UNREGISTERED
2030 	}                            reg_state;
2031 
2032 	int			     uverbs_abi_ver;
2033 	u64			     uverbs_cmd_mask;
2034 	u64			     uverbs_ex_cmd_mask;
2035 
2036 	char			     node_desc[64];
2037 	__be64			     node_guid;
2038 	u32			     local_dma_lkey;
2039 	u16                          is_switch:1;
2040 	u8                           node_type;
2041 	u8                           phys_port_cnt;
2042 	struct ib_device_attr        attrs;
2043 	struct attribute_group	     *hw_stats_ag;
2044 	struct rdma_hw_stats         *hw_stats;
2045 
2046 	/**
2047 	 * The following mandatory functions are used only at device
2048 	 * registration.  Keep functions such as these at the end of this
2049 	 * structure to avoid cache line misses when accessing struct ib_device
2050 	 * in fast paths.
2051 	 */
2052 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2053 	void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2054 };
2055 
2056 struct ib_client {
2057 	char  *name;
2058 	void (*add)   (struct ib_device *);
2059 	void (*remove)(struct ib_device *, void *client_data);
2060 
2061 	/* Returns the net_dev belonging to this ib_client and matching the
2062 	 * given parameters.
2063 	 * @dev:	 An RDMA device that the net_dev use for communication.
2064 	 * @port:	 A physical port number on the RDMA device.
2065 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2066 	 * @gid:	 A GID that the net_dev uses to communicate.
2067 	 * @addr:	 An IP address the net_dev is configured with.
2068 	 * @client_data: The device's client data set by ib_set_client_data().
2069 	 *
2070 	 * An ib_client that implements a net_dev on top of RDMA devices
2071 	 * (such as IP over IB) should implement this callback, allowing the
2072 	 * rdma_cm module to find the right net_dev for a given request.
2073 	 *
2074 	 * The caller is responsible for calling dev_put on the returned
2075 	 * netdev. */
2076 	struct net_device *(*get_net_dev_by_params)(
2077 			struct ib_device *dev,
2078 			u8 port,
2079 			u16 pkey,
2080 			const union ib_gid *gid,
2081 			const struct sockaddr *addr,
2082 			void *client_data);
2083 	struct list_head list;
2084 };
2085 
2086 struct ib_device *ib_alloc_device(size_t size);
2087 void ib_dealloc_device(struct ib_device *device);
2088 
2089 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2090 
2091 int ib_register_device(struct ib_device *device,
2092 		       int (*port_callback)(struct ib_device *,
2093 					    u8, struct kobject *));
2094 void ib_unregister_device(struct ib_device *device);
2095 
2096 int ib_register_client   (struct ib_client *client);
2097 void ib_unregister_client(struct ib_client *client);
2098 
2099 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2100 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2101 			 void *data);
2102 
2103 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2104 {
2105 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2106 }
2107 
2108 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2109 {
2110 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2111 }
2112 
2113 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2114 				       size_t offset,
2115 				       size_t len)
2116 {
2117 	const void __user *p = udata->inbuf + offset;
2118 	bool ret = false;
2119 	u8 *buf;
2120 
2121 	if (len > USHRT_MAX)
2122 		return false;
2123 
2124 	buf = kmalloc(len, GFP_KERNEL);
2125 	if (!buf)
2126 		return false;
2127 
2128 	if (copy_from_user(buf, p, len))
2129 		goto free;
2130 
2131 	ret = !memchr_inv(buf, 0, len);
2132 
2133 free:
2134 	kfree(buf);
2135 	return ret;
2136 }
2137 
2138 /**
2139  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2140  * contains all required attributes and no attributes not allowed for
2141  * the given QP state transition.
2142  * @cur_state: Current QP state
2143  * @next_state: Next QP state
2144  * @type: QP type
2145  * @mask: Mask of supplied QP attributes
2146  * @ll : link layer of port
2147  *
2148  * This function is a helper function that a low-level driver's
2149  * modify_qp method can use to validate the consumer's input.  It
2150  * checks that cur_state and next_state are valid QP states, that a
2151  * transition from cur_state to next_state is allowed by the IB spec,
2152  * and that the attribute mask supplied is allowed for the transition.
2153  */
2154 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2155 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2156 		       enum rdma_link_layer ll);
2157 
2158 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2159 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2160 void ib_dispatch_event(struct ib_event *event);
2161 
2162 int ib_query_port(struct ib_device *device,
2163 		  u8 port_num, struct ib_port_attr *port_attr);
2164 
2165 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2166 					       u8 port_num);
2167 
2168 /**
2169  * rdma_cap_ib_switch - Check if the device is IB switch
2170  * @device: Device to check
2171  *
2172  * Device driver is responsible for setting is_switch bit on
2173  * in ib_device structure at init time.
2174  *
2175  * Return: true if the device is IB switch.
2176  */
2177 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2178 {
2179 	return device->is_switch;
2180 }
2181 
2182 /**
2183  * rdma_start_port - Return the first valid port number for the device
2184  * specified
2185  *
2186  * @device: Device to be checked
2187  *
2188  * Return start port number
2189  */
2190 static inline u8 rdma_start_port(const struct ib_device *device)
2191 {
2192 	return rdma_cap_ib_switch(device) ? 0 : 1;
2193 }
2194 
2195 /**
2196  * rdma_end_port - Return the last valid port number for the device
2197  * specified
2198  *
2199  * @device: Device to be checked
2200  *
2201  * Return last port number
2202  */
2203 static inline u8 rdma_end_port(const struct ib_device *device)
2204 {
2205 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2206 }
2207 
2208 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2209 {
2210 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2211 }
2212 
2213 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2214 {
2215 	return device->port_immutable[port_num].core_cap_flags &
2216 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2217 }
2218 
2219 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2220 {
2221 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2222 }
2223 
2224 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2225 {
2226 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2227 }
2228 
2229 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2230 {
2231 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2232 }
2233 
2234 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2235 {
2236 	return rdma_protocol_ib(device, port_num) ||
2237 		rdma_protocol_roce(device, port_num);
2238 }
2239 
2240 /**
2241  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2242  * Management Datagrams.
2243  * @device: Device to check
2244  * @port_num: Port number to check
2245  *
2246  * Management Datagrams (MAD) are a required part of the InfiniBand
2247  * specification and are supported on all InfiniBand devices.  A slightly
2248  * extended version are also supported on OPA interfaces.
2249  *
2250  * Return: true if the port supports sending/receiving of MAD packets.
2251  */
2252 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2253 {
2254 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2255 }
2256 
2257 /**
2258  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2259  * Management Datagrams.
2260  * @device: Device to check
2261  * @port_num: Port number to check
2262  *
2263  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2264  * datagrams with their own versions.  These OPA MADs share many but not all of
2265  * the characteristics of InfiniBand MADs.
2266  *
2267  * OPA MADs differ in the following ways:
2268  *
2269  *    1) MADs are variable size up to 2K
2270  *       IBTA defined MADs remain fixed at 256 bytes
2271  *    2) OPA SMPs must carry valid PKeys
2272  *    3) OPA SMP packets are a different format
2273  *
2274  * Return: true if the port supports OPA MAD packet formats.
2275  */
2276 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2277 {
2278 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2279 		== RDMA_CORE_CAP_OPA_MAD;
2280 }
2281 
2282 /**
2283  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2284  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2285  * @device: Device to check
2286  * @port_num: Port number to check
2287  *
2288  * Each InfiniBand node is required to provide a Subnet Management Agent
2289  * that the subnet manager can access.  Prior to the fabric being fully
2290  * configured by the subnet manager, the SMA is accessed via a well known
2291  * interface called the Subnet Management Interface (SMI).  This interface
2292  * uses directed route packets to communicate with the SM to get around the
2293  * chicken and egg problem of the SM needing to know what's on the fabric
2294  * in order to configure the fabric, and needing to configure the fabric in
2295  * order to send packets to the devices on the fabric.  These directed
2296  * route packets do not need the fabric fully configured in order to reach
2297  * their destination.  The SMI is the only method allowed to send
2298  * directed route packets on an InfiniBand fabric.
2299  *
2300  * Return: true if the port provides an SMI.
2301  */
2302 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2303 {
2304 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2305 }
2306 
2307 /**
2308  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2309  * Communication Manager.
2310  * @device: Device to check
2311  * @port_num: Port number to check
2312  *
2313  * The InfiniBand Communication Manager is one of many pre-defined General
2314  * Service Agents (GSA) that are accessed via the General Service
2315  * Interface (GSI).  It's role is to facilitate establishment of connections
2316  * between nodes as well as other management related tasks for established
2317  * connections.
2318  *
2319  * Return: true if the port supports an IB CM (this does not guarantee that
2320  * a CM is actually running however).
2321  */
2322 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2323 {
2324 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2325 }
2326 
2327 /**
2328  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2329  * Communication Manager.
2330  * @device: Device to check
2331  * @port_num: Port number to check
2332  *
2333  * Similar to above, but specific to iWARP connections which have a different
2334  * managment protocol than InfiniBand.
2335  *
2336  * Return: true if the port supports an iWARP CM (this does not guarantee that
2337  * a CM is actually running however).
2338  */
2339 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2340 {
2341 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2342 }
2343 
2344 /**
2345  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2346  * Subnet Administration.
2347  * @device: Device to check
2348  * @port_num: Port number to check
2349  *
2350  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2351  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2352  * fabrics, devices should resolve routes to other hosts by contacting the
2353  * SA to query the proper route.
2354  *
2355  * Return: true if the port should act as a client to the fabric Subnet
2356  * Administration interface.  This does not imply that the SA service is
2357  * running locally.
2358  */
2359 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2360 {
2361 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2362 }
2363 
2364 /**
2365  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2366  * Multicast.
2367  * @device: Device to check
2368  * @port_num: Port number to check
2369  *
2370  * InfiniBand multicast registration is more complex than normal IPv4 or
2371  * IPv6 multicast registration.  Each Host Channel Adapter must register
2372  * with the Subnet Manager when it wishes to join a multicast group.  It
2373  * should do so only once regardless of how many queue pairs it subscribes
2374  * to this group.  And it should leave the group only after all queue pairs
2375  * attached to the group have been detached.
2376  *
2377  * Return: true if the port must undertake the additional adminstrative
2378  * overhead of registering/unregistering with the SM and tracking of the
2379  * total number of queue pairs attached to the multicast group.
2380  */
2381 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2382 {
2383 	return rdma_cap_ib_sa(device, port_num);
2384 }
2385 
2386 /**
2387  * rdma_cap_af_ib - Check if the port of device has the capability
2388  * Native Infiniband Address.
2389  * @device: Device to check
2390  * @port_num: Port number to check
2391  *
2392  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2393  * GID.  RoCE uses a different mechanism, but still generates a GID via
2394  * a prescribed mechanism and port specific data.
2395  *
2396  * Return: true if the port uses a GID address to identify devices on the
2397  * network.
2398  */
2399 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2400 {
2401 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2402 }
2403 
2404 /**
2405  * rdma_cap_eth_ah - Check if the port of device has the capability
2406  * Ethernet Address Handle.
2407  * @device: Device to check
2408  * @port_num: Port number to check
2409  *
2410  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2411  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2412  * port.  Normally, packet headers are generated by the sending host
2413  * adapter, but when sending connectionless datagrams, we must manually
2414  * inject the proper headers for the fabric we are communicating over.
2415  *
2416  * Return: true if we are running as a RoCE port and must force the
2417  * addition of a Global Route Header built from our Ethernet Address
2418  * Handle into our header list for connectionless packets.
2419  */
2420 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2421 {
2422 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2423 }
2424 
2425 /**
2426  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2427  *
2428  * @device: Device
2429  * @port_num: Port number
2430  *
2431  * This MAD size includes the MAD headers and MAD payload.  No other headers
2432  * are included.
2433  *
2434  * Return the max MAD size required by the Port.  Will return 0 if the port
2435  * does not support MADs
2436  */
2437 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2438 {
2439 	return device->port_immutable[port_num].max_mad_size;
2440 }
2441 
2442 /**
2443  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2444  * @device: Device to check
2445  * @port_num: Port number to check
2446  *
2447  * RoCE GID table mechanism manages the various GIDs for a device.
2448  *
2449  * NOTE: if allocating the port's GID table has failed, this call will still
2450  * return true, but any RoCE GID table API will fail.
2451  *
2452  * Return: true if the port uses RoCE GID table mechanism in order to manage
2453  * its GIDs.
2454  */
2455 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2456 					   u8 port_num)
2457 {
2458 	return rdma_protocol_roce(device, port_num) &&
2459 		device->add_gid && device->del_gid;
2460 }
2461 
2462 /*
2463  * Check if the device supports READ W/ INVALIDATE.
2464  */
2465 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2466 {
2467 	/*
2468 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2469 	 * has support for it yet.
2470 	 */
2471 	return rdma_protocol_iwarp(dev, port_num);
2472 }
2473 
2474 int ib_query_gid(struct ib_device *device,
2475 		 u8 port_num, int index, union ib_gid *gid,
2476 		 struct ib_gid_attr *attr);
2477 
2478 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2479 			 int state);
2480 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2481 		     struct ifla_vf_info *info);
2482 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2483 		    struct ifla_vf_stats *stats);
2484 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2485 		   int type);
2486 
2487 int ib_query_pkey(struct ib_device *device,
2488 		  u8 port_num, u16 index, u16 *pkey);
2489 
2490 int ib_modify_device(struct ib_device *device,
2491 		     int device_modify_mask,
2492 		     struct ib_device_modify *device_modify);
2493 
2494 int ib_modify_port(struct ib_device *device,
2495 		   u8 port_num, int port_modify_mask,
2496 		   struct ib_port_modify *port_modify);
2497 
2498 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2499 		enum ib_gid_type gid_type, struct net_device *ndev,
2500 		u8 *port_num, u16 *index);
2501 
2502 int ib_find_pkey(struct ib_device *device,
2503 		 u8 port_num, u16 pkey, u16 *index);
2504 
2505 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2506 
2507 void ib_dealloc_pd(struct ib_pd *pd);
2508 
2509 /**
2510  * ib_create_ah - Creates an address handle for the given address vector.
2511  * @pd: The protection domain associated with the address handle.
2512  * @ah_attr: The attributes of the address vector.
2513  *
2514  * The address handle is used to reference a local or global destination
2515  * in all UD QP post sends.
2516  */
2517 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2518 
2519 /**
2520  * ib_init_ah_from_wc - Initializes address handle attributes from a
2521  *   work completion.
2522  * @device: Device on which the received message arrived.
2523  * @port_num: Port on which the received message arrived.
2524  * @wc: Work completion associated with the received message.
2525  * @grh: References the received global route header.  This parameter is
2526  *   ignored unless the work completion indicates that the GRH is valid.
2527  * @ah_attr: Returned attributes that can be used when creating an address
2528  *   handle for replying to the message.
2529  */
2530 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2531 		       const struct ib_wc *wc, const struct ib_grh *grh,
2532 		       struct ib_ah_attr *ah_attr);
2533 
2534 /**
2535  * ib_create_ah_from_wc - Creates an address handle associated with the
2536  *   sender of the specified work completion.
2537  * @pd: The protection domain associated with the address handle.
2538  * @wc: Work completion information associated with a received message.
2539  * @grh: References the received global route header.  This parameter is
2540  *   ignored unless the work completion indicates that the GRH is valid.
2541  * @port_num: The outbound port number to associate with the address.
2542  *
2543  * The address handle is used to reference a local or global destination
2544  * in all UD QP post sends.
2545  */
2546 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2547 				   const struct ib_grh *grh, u8 port_num);
2548 
2549 /**
2550  * ib_modify_ah - Modifies the address vector associated with an address
2551  *   handle.
2552  * @ah: The address handle to modify.
2553  * @ah_attr: The new address vector attributes to associate with the
2554  *   address handle.
2555  */
2556 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2557 
2558 /**
2559  * ib_query_ah - Queries the address vector associated with an address
2560  *   handle.
2561  * @ah: The address handle to query.
2562  * @ah_attr: The address vector attributes associated with the address
2563  *   handle.
2564  */
2565 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2566 
2567 /**
2568  * ib_destroy_ah - Destroys an address handle.
2569  * @ah: The address handle to destroy.
2570  */
2571 int ib_destroy_ah(struct ib_ah *ah);
2572 
2573 /**
2574  * ib_create_srq - Creates a SRQ associated with the specified protection
2575  *   domain.
2576  * @pd: The protection domain associated with the SRQ.
2577  * @srq_init_attr: A list of initial attributes required to create the
2578  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2579  *   the actual capabilities of the created SRQ.
2580  *
2581  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2582  * requested size of the SRQ, and set to the actual values allocated
2583  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2584  * will always be at least as large as the requested values.
2585  */
2586 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2587 			     struct ib_srq_init_attr *srq_init_attr);
2588 
2589 /**
2590  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2591  * @srq: The SRQ to modify.
2592  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2593  *   the current values of selected SRQ attributes are returned.
2594  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2595  *   are being modified.
2596  *
2597  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2598  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2599  * the number of receives queued drops below the limit.
2600  */
2601 int ib_modify_srq(struct ib_srq *srq,
2602 		  struct ib_srq_attr *srq_attr,
2603 		  enum ib_srq_attr_mask srq_attr_mask);
2604 
2605 /**
2606  * ib_query_srq - Returns the attribute list and current values for the
2607  *   specified SRQ.
2608  * @srq: The SRQ to query.
2609  * @srq_attr: The attributes of the specified SRQ.
2610  */
2611 int ib_query_srq(struct ib_srq *srq,
2612 		 struct ib_srq_attr *srq_attr);
2613 
2614 /**
2615  * ib_destroy_srq - Destroys the specified SRQ.
2616  * @srq: The SRQ to destroy.
2617  */
2618 int ib_destroy_srq(struct ib_srq *srq);
2619 
2620 /**
2621  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2622  * @srq: The SRQ to post the work request on.
2623  * @recv_wr: A list of work requests to post on the receive queue.
2624  * @bad_recv_wr: On an immediate failure, this parameter will reference
2625  *   the work request that failed to be posted on the QP.
2626  */
2627 static inline int ib_post_srq_recv(struct ib_srq *srq,
2628 				   struct ib_recv_wr *recv_wr,
2629 				   struct ib_recv_wr **bad_recv_wr)
2630 {
2631 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2632 }
2633 
2634 /**
2635  * ib_create_qp - Creates a QP associated with the specified protection
2636  *   domain.
2637  * @pd: The protection domain associated with the QP.
2638  * @qp_init_attr: A list of initial attributes required to create the
2639  *   QP.  If QP creation succeeds, then the attributes are updated to
2640  *   the actual capabilities of the created QP.
2641  */
2642 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2643 			   struct ib_qp_init_attr *qp_init_attr);
2644 
2645 /**
2646  * ib_modify_qp - Modifies the attributes for the specified QP and then
2647  *   transitions the QP to the given state.
2648  * @qp: The QP to modify.
2649  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2650  *   the current values of selected QP attributes are returned.
2651  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2652  *   are being modified.
2653  */
2654 int ib_modify_qp(struct ib_qp *qp,
2655 		 struct ib_qp_attr *qp_attr,
2656 		 int qp_attr_mask);
2657 
2658 /**
2659  * ib_query_qp - Returns the attribute list and current values for the
2660  *   specified QP.
2661  * @qp: The QP to query.
2662  * @qp_attr: The attributes of the specified QP.
2663  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2664  * @qp_init_attr: Additional attributes of the selected QP.
2665  *
2666  * The qp_attr_mask may be used to limit the query to gathering only the
2667  * selected attributes.
2668  */
2669 int ib_query_qp(struct ib_qp *qp,
2670 		struct ib_qp_attr *qp_attr,
2671 		int qp_attr_mask,
2672 		struct ib_qp_init_attr *qp_init_attr);
2673 
2674 /**
2675  * ib_destroy_qp - Destroys the specified QP.
2676  * @qp: The QP to destroy.
2677  */
2678 int ib_destroy_qp(struct ib_qp *qp);
2679 
2680 /**
2681  * ib_open_qp - Obtain a reference to an existing sharable QP.
2682  * @xrcd - XRC domain
2683  * @qp_open_attr: Attributes identifying the QP to open.
2684  *
2685  * Returns a reference to a sharable QP.
2686  */
2687 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2688 			 struct ib_qp_open_attr *qp_open_attr);
2689 
2690 /**
2691  * ib_close_qp - Release an external reference to a QP.
2692  * @qp: The QP handle to release
2693  *
2694  * The opened QP handle is released by the caller.  The underlying
2695  * shared QP is not destroyed until all internal references are released.
2696  */
2697 int ib_close_qp(struct ib_qp *qp);
2698 
2699 /**
2700  * ib_post_send - Posts a list of work requests to the send queue of
2701  *   the specified QP.
2702  * @qp: The QP to post the work request on.
2703  * @send_wr: A list of work requests to post on the send queue.
2704  * @bad_send_wr: On an immediate failure, this parameter will reference
2705  *   the work request that failed to be posted on the QP.
2706  *
2707  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2708  * error is returned, the QP state shall not be affected,
2709  * ib_post_send() will return an immediate error after queueing any
2710  * earlier work requests in the list.
2711  */
2712 static inline int ib_post_send(struct ib_qp *qp,
2713 			       struct ib_send_wr *send_wr,
2714 			       struct ib_send_wr **bad_send_wr)
2715 {
2716 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2717 }
2718 
2719 /**
2720  * ib_post_recv - Posts a list of work requests to the receive queue of
2721  *   the specified QP.
2722  * @qp: The QP to post the work request on.
2723  * @recv_wr: A list of work requests to post on the receive queue.
2724  * @bad_recv_wr: On an immediate failure, this parameter will reference
2725  *   the work request that failed to be posted on the QP.
2726  */
2727 static inline int ib_post_recv(struct ib_qp *qp,
2728 			       struct ib_recv_wr *recv_wr,
2729 			       struct ib_recv_wr **bad_recv_wr)
2730 {
2731 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2732 }
2733 
2734 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2735 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2736 void ib_free_cq(struct ib_cq *cq);
2737 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2738 
2739 /**
2740  * ib_create_cq - Creates a CQ on the specified device.
2741  * @device: The device on which to create the CQ.
2742  * @comp_handler: A user-specified callback that is invoked when a
2743  *   completion event occurs on the CQ.
2744  * @event_handler: A user-specified callback that is invoked when an
2745  *   asynchronous event not associated with a completion occurs on the CQ.
2746  * @cq_context: Context associated with the CQ returned to the user via
2747  *   the associated completion and event handlers.
2748  * @cq_attr: The attributes the CQ should be created upon.
2749  *
2750  * Users can examine the cq structure to determine the actual CQ size.
2751  */
2752 struct ib_cq *ib_create_cq(struct ib_device *device,
2753 			   ib_comp_handler comp_handler,
2754 			   void (*event_handler)(struct ib_event *, void *),
2755 			   void *cq_context,
2756 			   const struct ib_cq_init_attr *cq_attr);
2757 
2758 /**
2759  * ib_resize_cq - Modifies the capacity of the CQ.
2760  * @cq: The CQ to resize.
2761  * @cqe: The minimum size of the CQ.
2762  *
2763  * Users can examine the cq structure to determine the actual CQ size.
2764  */
2765 int ib_resize_cq(struct ib_cq *cq, int cqe);
2766 
2767 /**
2768  * ib_modify_cq - Modifies moderation params of the CQ
2769  * @cq: The CQ to modify.
2770  * @cq_count: number of CQEs that will trigger an event
2771  * @cq_period: max period of time in usec before triggering an event
2772  *
2773  */
2774 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2775 
2776 /**
2777  * ib_destroy_cq - Destroys the specified CQ.
2778  * @cq: The CQ to destroy.
2779  */
2780 int ib_destroy_cq(struct ib_cq *cq);
2781 
2782 /**
2783  * ib_poll_cq - poll a CQ for completion(s)
2784  * @cq:the CQ being polled
2785  * @num_entries:maximum number of completions to return
2786  * @wc:array of at least @num_entries &struct ib_wc where completions
2787  *   will be returned
2788  *
2789  * Poll a CQ for (possibly multiple) completions.  If the return value
2790  * is < 0, an error occurred.  If the return value is >= 0, it is the
2791  * number of completions returned.  If the return value is
2792  * non-negative and < num_entries, then the CQ was emptied.
2793  */
2794 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2795 			     struct ib_wc *wc)
2796 {
2797 	return cq->device->poll_cq(cq, num_entries, wc);
2798 }
2799 
2800 /**
2801  * ib_peek_cq - Returns the number of unreaped completions currently
2802  *   on the specified CQ.
2803  * @cq: The CQ to peek.
2804  * @wc_cnt: A minimum number of unreaped completions to check for.
2805  *
2806  * If the number of unreaped completions is greater than or equal to wc_cnt,
2807  * this function returns wc_cnt, otherwise, it returns the actual number of
2808  * unreaped completions.
2809  */
2810 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2811 
2812 /**
2813  * ib_req_notify_cq - Request completion notification on a CQ.
2814  * @cq: The CQ to generate an event for.
2815  * @flags:
2816  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2817  *   to request an event on the next solicited event or next work
2818  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2819  *   may also be |ed in to request a hint about missed events, as
2820  *   described below.
2821  *
2822  * Return Value:
2823  *    < 0 means an error occurred while requesting notification
2824  *   == 0 means notification was requested successfully, and if
2825  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2826  *        were missed and it is safe to wait for another event.  In
2827  *        this case is it guaranteed that any work completions added
2828  *        to the CQ since the last CQ poll will trigger a completion
2829  *        notification event.
2830  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2831  *        in.  It means that the consumer must poll the CQ again to
2832  *        make sure it is empty to avoid missing an event because of a
2833  *        race between requesting notification and an entry being
2834  *        added to the CQ.  This return value means it is possible
2835  *        (but not guaranteed) that a work completion has been added
2836  *        to the CQ since the last poll without triggering a
2837  *        completion notification event.
2838  */
2839 static inline int ib_req_notify_cq(struct ib_cq *cq,
2840 				   enum ib_cq_notify_flags flags)
2841 {
2842 	return cq->device->req_notify_cq(cq, flags);
2843 }
2844 
2845 /**
2846  * ib_req_ncomp_notif - Request completion notification when there are
2847  *   at least the specified number of unreaped completions on the CQ.
2848  * @cq: The CQ to generate an event for.
2849  * @wc_cnt: The number of unreaped completions that should be on the
2850  *   CQ before an event is generated.
2851  */
2852 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2853 {
2854 	return cq->device->req_ncomp_notif ?
2855 		cq->device->req_ncomp_notif(cq, wc_cnt) :
2856 		-ENOSYS;
2857 }
2858 
2859 /**
2860  * ib_get_dma_mr - Returns a memory region for system memory that is
2861  *   usable for DMA.
2862  * @pd: The protection domain associated with the memory region.
2863  * @mr_access_flags: Specifies the memory access rights.
2864  *
2865  * Note that the ib_dma_*() functions defined below must be used
2866  * to create/destroy addresses used with the Lkey or Rkey returned
2867  * by ib_get_dma_mr().
2868  */
2869 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2870 
2871 /**
2872  * ib_dma_mapping_error - check a DMA addr for error
2873  * @dev: The device for which the dma_addr was created
2874  * @dma_addr: The DMA address to check
2875  */
2876 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2877 {
2878 	if (dev->dma_ops)
2879 		return dev->dma_ops->mapping_error(dev, dma_addr);
2880 	return dma_mapping_error(dev->dma_device, dma_addr);
2881 }
2882 
2883 /**
2884  * ib_dma_map_single - Map a kernel virtual address to DMA address
2885  * @dev: The device for which the dma_addr is to be created
2886  * @cpu_addr: The kernel virtual address
2887  * @size: The size of the region in bytes
2888  * @direction: The direction of the DMA
2889  */
2890 static inline u64 ib_dma_map_single(struct ib_device *dev,
2891 				    void *cpu_addr, size_t size,
2892 				    enum dma_data_direction direction)
2893 {
2894 	if (dev->dma_ops)
2895 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2896 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2897 }
2898 
2899 /**
2900  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2901  * @dev: The device for which the DMA address was created
2902  * @addr: The DMA address
2903  * @size: The size of the region in bytes
2904  * @direction: The direction of the DMA
2905  */
2906 static inline void ib_dma_unmap_single(struct ib_device *dev,
2907 				       u64 addr, size_t size,
2908 				       enum dma_data_direction direction)
2909 {
2910 	if (dev->dma_ops)
2911 		dev->dma_ops->unmap_single(dev, addr, size, direction);
2912 	else
2913 		dma_unmap_single(dev->dma_device, addr, size, direction);
2914 }
2915 
2916 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2917 					  void *cpu_addr, size_t size,
2918 					  enum dma_data_direction direction,
2919 					  unsigned long dma_attrs)
2920 {
2921 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2922 				    direction, dma_attrs);
2923 }
2924 
2925 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2926 					     u64 addr, size_t size,
2927 					     enum dma_data_direction direction,
2928 					     unsigned long dma_attrs)
2929 {
2930 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
2931 				      direction, dma_attrs);
2932 }
2933 
2934 /**
2935  * ib_dma_map_page - Map a physical page to DMA address
2936  * @dev: The device for which the dma_addr is to be created
2937  * @page: The page to be mapped
2938  * @offset: The offset within the page
2939  * @size: The size of the region in bytes
2940  * @direction: The direction of the DMA
2941  */
2942 static inline u64 ib_dma_map_page(struct ib_device *dev,
2943 				  struct page *page,
2944 				  unsigned long offset,
2945 				  size_t size,
2946 					 enum dma_data_direction direction)
2947 {
2948 	if (dev->dma_ops)
2949 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
2950 	return dma_map_page(dev->dma_device, page, offset, size, direction);
2951 }
2952 
2953 /**
2954  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2955  * @dev: The device for which the DMA address was created
2956  * @addr: The DMA address
2957  * @size: The size of the region in bytes
2958  * @direction: The direction of the DMA
2959  */
2960 static inline void ib_dma_unmap_page(struct ib_device *dev,
2961 				     u64 addr, size_t size,
2962 				     enum dma_data_direction direction)
2963 {
2964 	if (dev->dma_ops)
2965 		dev->dma_ops->unmap_page(dev, addr, size, direction);
2966 	else
2967 		dma_unmap_page(dev->dma_device, addr, size, direction);
2968 }
2969 
2970 /**
2971  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2972  * @dev: The device for which the DMA addresses are to be created
2973  * @sg: The array of scatter/gather entries
2974  * @nents: The number of scatter/gather entries
2975  * @direction: The direction of the DMA
2976  */
2977 static inline int ib_dma_map_sg(struct ib_device *dev,
2978 				struct scatterlist *sg, int nents,
2979 				enum dma_data_direction direction)
2980 {
2981 	if (dev->dma_ops)
2982 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
2983 	return dma_map_sg(dev->dma_device, sg, nents, direction);
2984 }
2985 
2986 /**
2987  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2988  * @dev: The device for which the DMA addresses were created
2989  * @sg: The array of scatter/gather entries
2990  * @nents: The number of scatter/gather entries
2991  * @direction: The direction of the DMA
2992  */
2993 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2994 				   struct scatterlist *sg, int nents,
2995 				   enum dma_data_direction direction)
2996 {
2997 	if (dev->dma_ops)
2998 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2999 	else
3000 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
3001 }
3002 
3003 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3004 				      struct scatterlist *sg, int nents,
3005 				      enum dma_data_direction direction,
3006 				      unsigned long dma_attrs)
3007 {
3008 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3009 				dma_attrs);
3010 }
3011 
3012 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3013 					 struct scatterlist *sg, int nents,
3014 					 enum dma_data_direction direction,
3015 					 unsigned long dma_attrs)
3016 {
3017 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3018 }
3019 /**
3020  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3021  * @dev: The device for which the DMA addresses were created
3022  * @sg: The scatter/gather entry
3023  *
3024  * Note: this function is obsolete. To do: change all occurrences of
3025  * ib_sg_dma_address() into sg_dma_address().
3026  */
3027 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3028 				    struct scatterlist *sg)
3029 {
3030 	return sg_dma_address(sg);
3031 }
3032 
3033 /**
3034  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3035  * @dev: The device for which the DMA addresses were created
3036  * @sg: The scatter/gather entry
3037  *
3038  * Note: this function is obsolete. To do: change all occurrences of
3039  * ib_sg_dma_len() into sg_dma_len().
3040  */
3041 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3042 					 struct scatterlist *sg)
3043 {
3044 	return sg_dma_len(sg);
3045 }
3046 
3047 /**
3048  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3049  * @dev: The device for which the DMA address was created
3050  * @addr: The DMA address
3051  * @size: The size of the region in bytes
3052  * @dir: The direction of the DMA
3053  */
3054 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3055 					      u64 addr,
3056 					      size_t size,
3057 					      enum dma_data_direction dir)
3058 {
3059 	if (dev->dma_ops)
3060 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
3061 	else
3062 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3063 }
3064 
3065 /**
3066  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3067  * @dev: The device for which the DMA address was created
3068  * @addr: The DMA address
3069  * @size: The size of the region in bytes
3070  * @dir: The direction of the DMA
3071  */
3072 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3073 						 u64 addr,
3074 						 size_t size,
3075 						 enum dma_data_direction dir)
3076 {
3077 	if (dev->dma_ops)
3078 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
3079 	else
3080 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3081 }
3082 
3083 /**
3084  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3085  * @dev: The device for which the DMA address is requested
3086  * @size: The size of the region to allocate in bytes
3087  * @dma_handle: A pointer for returning the DMA address of the region
3088  * @flag: memory allocator flags
3089  */
3090 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3091 					   size_t size,
3092 					   u64 *dma_handle,
3093 					   gfp_t flag)
3094 {
3095 	if (dev->dma_ops)
3096 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
3097 	else {
3098 		dma_addr_t handle;
3099 		void *ret;
3100 
3101 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3102 		*dma_handle = handle;
3103 		return ret;
3104 	}
3105 }
3106 
3107 /**
3108  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3109  * @dev: The device for which the DMA addresses were allocated
3110  * @size: The size of the region
3111  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3112  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3113  */
3114 static inline void ib_dma_free_coherent(struct ib_device *dev,
3115 					size_t size, void *cpu_addr,
3116 					u64 dma_handle)
3117 {
3118 	if (dev->dma_ops)
3119 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3120 	else
3121 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3122 }
3123 
3124 /**
3125  * ib_dereg_mr - Deregisters a memory region and removes it from the
3126  *   HCA translation table.
3127  * @mr: The memory region to deregister.
3128  *
3129  * This function can fail, if the memory region has memory windows bound to it.
3130  */
3131 int ib_dereg_mr(struct ib_mr *mr);
3132 
3133 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3134 			  enum ib_mr_type mr_type,
3135 			  u32 max_num_sg);
3136 
3137 /**
3138  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3139  *   R_Key and L_Key.
3140  * @mr - struct ib_mr pointer to be updated.
3141  * @newkey - new key to be used.
3142  */
3143 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3144 {
3145 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3146 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3147 }
3148 
3149 /**
3150  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3151  * for calculating a new rkey for type 2 memory windows.
3152  * @rkey - the rkey to increment.
3153  */
3154 static inline u32 ib_inc_rkey(u32 rkey)
3155 {
3156 	const u32 mask = 0x000000ff;
3157 	return ((rkey + 1) & mask) | (rkey & ~mask);
3158 }
3159 
3160 /**
3161  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3162  * @pd: The protection domain associated with the unmapped region.
3163  * @mr_access_flags: Specifies the memory access rights.
3164  * @fmr_attr: Attributes of the unmapped region.
3165  *
3166  * A fast memory region must be mapped before it can be used as part of
3167  * a work request.
3168  */
3169 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3170 			    int mr_access_flags,
3171 			    struct ib_fmr_attr *fmr_attr);
3172 
3173 /**
3174  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3175  * @fmr: The fast memory region to associate with the pages.
3176  * @page_list: An array of physical pages to map to the fast memory region.
3177  * @list_len: The number of pages in page_list.
3178  * @iova: The I/O virtual address to use with the mapped region.
3179  */
3180 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3181 				  u64 *page_list, int list_len,
3182 				  u64 iova)
3183 {
3184 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3185 }
3186 
3187 /**
3188  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3189  * @fmr_list: A linked list of fast memory regions to unmap.
3190  */
3191 int ib_unmap_fmr(struct list_head *fmr_list);
3192 
3193 /**
3194  * ib_dealloc_fmr - Deallocates a fast memory region.
3195  * @fmr: The fast memory region to deallocate.
3196  */
3197 int ib_dealloc_fmr(struct ib_fmr *fmr);
3198 
3199 /**
3200  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3201  * @qp: QP to attach to the multicast group.  The QP must be type
3202  *   IB_QPT_UD.
3203  * @gid: Multicast group GID.
3204  * @lid: Multicast group LID in host byte order.
3205  *
3206  * In order to send and receive multicast packets, subnet
3207  * administration must have created the multicast group and configured
3208  * the fabric appropriately.  The port associated with the specified
3209  * QP must also be a member of the multicast group.
3210  */
3211 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3212 
3213 /**
3214  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3215  * @qp: QP to detach from the multicast group.
3216  * @gid: Multicast group GID.
3217  * @lid: Multicast group LID in host byte order.
3218  */
3219 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3220 
3221 /**
3222  * ib_alloc_xrcd - Allocates an XRC domain.
3223  * @device: The device on which to allocate the XRC domain.
3224  */
3225 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3226 
3227 /**
3228  * ib_dealloc_xrcd - Deallocates an XRC domain.
3229  * @xrcd: The XRC domain to deallocate.
3230  */
3231 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3232 
3233 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3234 			       struct ib_flow_attr *flow_attr, int domain);
3235 int ib_destroy_flow(struct ib_flow *flow_id);
3236 
3237 static inline int ib_check_mr_access(int flags)
3238 {
3239 	/*
3240 	 * Local write permission is required if remote write or
3241 	 * remote atomic permission is also requested.
3242 	 */
3243 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3244 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3245 		return -EINVAL;
3246 
3247 	return 0;
3248 }
3249 
3250 /**
3251  * ib_check_mr_status: lightweight check of MR status.
3252  *     This routine may provide status checks on a selected
3253  *     ib_mr. first use is for signature status check.
3254  *
3255  * @mr: A memory region.
3256  * @check_mask: Bitmask of which checks to perform from
3257  *     ib_mr_status_check enumeration.
3258  * @mr_status: The container of relevant status checks.
3259  *     failed checks will be indicated in the status bitmask
3260  *     and the relevant info shall be in the error item.
3261  */
3262 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3263 		       struct ib_mr_status *mr_status);
3264 
3265 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3266 					    u16 pkey, const union ib_gid *gid,
3267 					    const struct sockaddr *addr);
3268 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3269 			   struct ib_wq_init_attr *init_attr);
3270 int ib_destroy_wq(struct ib_wq *wq);
3271 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3272 		 u32 wq_attr_mask);
3273 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3274 						 struct ib_rwq_ind_table_init_attr*
3275 						 wq_ind_table_init_attr);
3276 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3277 
3278 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3279 		 unsigned int *sg_offset, unsigned int page_size);
3280 
3281 static inline int
3282 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3283 		  unsigned int *sg_offset, unsigned int page_size)
3284 {
3285 	int n;
3286 
3287 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3288 	mr->iova = 0;
3289 
3290 	return n;
3291 }
3292 
3293 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3294 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3295 
3296 void ib_drain_rq(struct ib_qp *qp);
3297 void ib_drain_sq(struct ib_qp *qp);
3298 void ib_drain_qp(struct ib_qp *qp);
3299 #endif /* IB_VERBS_H */
3300