• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
3  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
4  * Copyright (c) 2004 Intel Corporation.  All rights reserved.
5  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
6  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
7  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
9  *
10  * This software is available to you under a choice of one of two
11  * licenses.  You may choose to be licensed under the terms of the GNU
12  * General Public License (GPL) Version 2, available from the file
13  * COPYING in the main directory of this source tree, or the
14  * OpenIB.org BSD license below:
15  *
16  *     Redistribution and use in source and binary forms, with or
17  *     without modification, are permitted provided that the following
18  *     conditions are met:
19  *
20  *      - Redistributions of source code must retain the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer.
23  *
24  *      - Redistributions in binary form must reproduce the above
25  *        copyright notice, this list of conditions and the following
26  *        disclaimer in the documentation and/or other materials
27  *        provided with the distribution.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36  * SOFTWARE.
37  */
38 
39 #if !defined(IB_VERBS_H)
40 #define IB_VERBS_H
41 
42 #include <linux/types.h>
43 #include <linux/device.h>
44 #include <linux/mm.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
54 #include <net/ipv6.h>
55 #include <net/ip.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/netdevice.h>
59 
60 #include <linux/if_link.h>
61 #include <linux/atomic.h>
62 #include <linux/mmu_notifier.h>
63 #include <linux/uaccess.h>
64 #include <linux/cgroup_rdma.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 
67 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
68 
69 extern struct workqueue_struct *ib_wq;
70 extern struct workqueue_struct *ib_comp_wq;
71 extern struct workqueue_struct *ib_comp_unbound_wq;
72 
73 union ib_gid {
74 	u8	raw[16];
75 	struct {
76 		__be64	subnet_prefix;
77 		__be64	interface_id;
78 	} global;
79 };
80 
81 extern union ib_gid zgid;
82 
83 enum ib_gid_type {
84 	/* If link layer is Ethernet, this is RoCE V1 */
85 	IB_GID_TYPE_IB        = 0,
86 	IB_GID_TYPE_ROCE      = 0,
87 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
88 	IB_GID_TYPE_SIZE
89 };
90 
91 #define ROCE_V2_UDP_DPORT      4791
92 struct ib_gid_attr {
93 	enum ib_gid_type	gid_type;
94 	struct net_device	*ndev;
95 };
96 
97 enum rdma_node_type {
98 	/* IB values map to NodeInfo:NodeType. */
99 	RDMA_NODE_IB_CA 	= 1,
100 	RDMA_NODE_IB_SWITCH,
101 	RDMA_NODE_IB_ROUTER,
102 	RDMA_NODE_RNIC,
103 	RDMA_NODE_USNIC,
104 	RDMA_NODE_USNIC_UDP,
105 };
106 
107 enum {
108 	/* set the local administered indication */
109 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
110 };
111 
112 enum rdma_transport_type {
113 	RDMA_TRANSPORT_IB,
114 	RDMA_TRANSPORT_IWARP,
115 	RDMA_TRANSPORT_USNIC,
116 	RDMA_TRANSPORT_USNIC_UDP
117 };
118 
119 enum rdma_protocol_type {
120 	RDMA_PROTOCOL_IB,
121 	RDMA_PROTOCOL_IBOE,
122 	RDMA_PROTOCOL_IWARP,
123 	RDMA_PROTOCOL_USNIC_UDP
124 };
125 
126 __attribute_const__ enum rdma_transport_type
127 rdma_node_get_transport(enum rdma_node_type node_type);
128 
129 enum rdma_network_type {
130 	RDMA_NETWORK_IB,
131 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
132 	RDMA_NETWORK_IPV4,
133 	RDMA_NETWORK_IPV6
134 };
135 
ib_network_to_gid_type(enum rdma_network_type network_type)136 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
137 {
138 	if (network_type == RDMA_NETWORK_IPV4 ||
139 	    network_type == RDMA_NETWORK_IPV6)
140 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
141 
142 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
143 	return IB_GID_TYPE_IB;
144 }
145 
ib_gid_to_network_type(enum ib_gid_type gid_type,union ib_gid * gid)146 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
147 							    union ib_gid *gid)
148 {
149 	if (gid_type == IB_GID_TYPE_IB)
150 		return RDMA_NETWORK_IB;
151 
152 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
153 		return RDMA_NETWORK_IPV4;
154 	else
155 		return RDMA_NETWORK_IPV6;
156 }
157 
158 enum rdma_link_layer {
159 	IB_LINK_LAYER_UNSPECIFIED,
160 	IB_LINK_LAYER_INFINIBAND,
161 	IB_LINK_LAYER_ETHERNET,
162 };
163 
164 enum ib_device_cap_flags {
165 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
166 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
167 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
168 	IB_DEVICE_RAW_MULTI			= (1 << 3),
169 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
170 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
171 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
172 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
173 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
174 	/* Not in use, former INIT_TYPE		= (1 << 9),*/
175 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
176 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
177 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
178 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
179 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
180 
181 	/*
182 	 * This device supports a per-device lkey or stag that can be
183 	 * used without performing a memory registration for the local
184 	 * memory.  Note that ULPs should never check this flag, but
185 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
186 	 * which will always contain a usable lkey.
187 	 */
188 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
189 	/* Reserved, old SEND_W_INV		= (1 << 16),*/
190 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
191 	/*
192 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
193 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
194 	 * messages and can verify the validity of checksum for
195 	 * incoming messages.  Setting this flag implies that the
196 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
197 	 */
198 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
199 	IB_DEVICE_UD_TSO			= (1 << 19),
200 	IB_DEVICE_XRC				= (1 << 20),
201 
202 	/*
203 	 * This device supports the IB "base memory management extension",
204 	 * which includes support for fast registrations (IB_WR_REG_MR,
205 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
206 	 * also be set by any iWarp device which must support FRs to comply
207 	 * to the iWarp verbs spec.  iWarp devices also support the
208 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
209 	 * stag.
210 	 */
211 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
212 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
213 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
214 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
215 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
216 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
217 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
218 	/*
219 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
220 	 * support execution of WQEs that involve synchronization
221 	 * of I/O operations with single completion queue managed
222 	 * by hardware.
223 	 */
224 	IB_DEVICE_CROSS_CHANNEL			= (1 << 27),
225 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
226 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
227 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
228 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
229 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
230 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
231 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
232 	IB_DEVICE_RDMA_NETDEV_OPA_VNIC		= (1ULL << 35),
233 };
234 
235 enum ib_signature_prot_cap {
236 	IB_PROT_T10DIF_TYPE_1 = 1,
237 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
238 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
239 };
240 
241 enum ib_signature_guard_cap {
242 	IB_GUARD_T10DIF_CRC	= 1,
243 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
244 };
245 
246 enum ib_atomic_cap {
247 	IB_ATOMIC_NONE,
248 	IB_ATOMIC_HCA,
249 	IB_ATOMIC_GLOB
250 };
251 
252 enum ib_odp_general_cap_bits {
253 	IB_ODP_SUPPORT		= 1 << 0,
254 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
255 };
256 
257 enum ib_odp_transport_cap_bits {
258 	IB_ODP_SUPPORT_SEND	= 1 << 0,
259 	IB_ODP_SUPPORT_RECV	= 1 << 1,
260 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
261 	IB_ODP_SUPPORT_READ	= 1 << 3,
262 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
263 };
264 
265 struct ib_odp_caps {
266 	uint64_t general_caps;
267 	struct {
268 		uint32_t  rc_odp_caps;
269 		uint32_t  uc_odp_caps;
270 		uint32_t  ud_odp_caps;
271 	} per_transport_caps;
272 };
273 
274 struct ib_rss_caps {
275 	/* Corresponding bit will be set if qp type from
276 	 * 'enum ib_qp_type' is supported, e.g.
277 	 * supported_qpts |= 1 << IB_QPT_UD
278 	 */
279 	u32 supported_qpts;
280 	u32 max_rwq_indirection_tables;
281 	u32 max_rwq_indirection_table_size;
282 };
283 
284 enum ib_tm_cap_flags {
285 	/*  Support tag matching on RC transport */
286 	IB_TM_CAP_RC		    = 1 << 0,
287 };
288 
289 struct ib_tm_caps {
290 	/* Max size of RNDV header */
291 	u32 max_rndv_hdr_size;
292 	/* Max number of entries in tag matching list */
293 	u32 max_num_tags;
294 	/* From enum ib_tm_cap_flags */
295 	u32 flags;
296 	/* Max number of outstanding list operations */
297 	u32 max_ops;
298 	/* Max number of SGE in tag matching entry */
299 	u32 max_sge;
300 };
301 
302 enum ib_cq_creation_flags {
303 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
304 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
305 };
306 
307 struct ib_cq_init_attr {
308 	unsigned int	cqe;
309 	u32		comp_vector;
310 	u32		flags;
311 };
312 
313 struct ib_device_attr {
314 	u64			fw_ver;
315 	__be64			sys_image_guid;
316 	u64			max_mr_size;
317 	u64			page_size_cap;
318 	u32			vendor_id;
319 	u32			vendor_part_id;
320 	u32			hw_ver;
321 	int			max_qp;
322 	int			max_qp_wr;
323 	u64			device_cap_flags;
324 	int			max_sge;
325 	int			max_sge_rd;
326 	int			max_cq;
327 	int			max_cqe;
328 	int			max_mr;
329 	int			max_pd;
330 	int			max_qp_rd_atom;
331 	int			max_ee_rd_atom;
332 	int			max_res_rd_atom;
333 	int			max_qp_init_rd_atom;
334 	int			max_ee_init_rd_atom;
335 	enum ib_atomic_cap	atomic_cap;
336 	enum ib_atomic_cap	masked_atomic_cap;
337 	int			max_ee;
338 	int			max_rdd;
339 	int			max_mw;
340 	int			max_raw_ipv6_qp;
341 	int			max_raw_ethy_qp;
342 	int			max_mcast_grp;
343 	int			max_mcast_qp_attach;
344 	int			max_total_mcast_qp_attach;
345 	int			max_ah;
346 	int			max_fmr;
347 	int			max_map_per_fmr;
348 	int			max_srq;
349 	int			max_srq_wr;
350 	int			max_srq_sge;
351 	unsigned int		max_fast_reg_page_list_len;
352 	u16			max_pkeys;
353 	u8			local_ca_ack_delay;
354 	int			sig_prot_cap;
355 	int			sig_guard_cap;
356 	struct ib_odp_caps	odp_caps;
357 	uint64_t		timestamp_mask;
358 	uint64_t		hca_core_clock; /* in KHZ */
359 	struct ib_rss_caps	rss_caps;
360 	u32			max_wq_type_rq;
361 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
362 	struct ib_tm_caps	tm_caps;
363 };
364 
365 enum ib_mtu {
366 	IB_MTU_256  = 1,
367 	IB_MTU_512  = 2,
368 	IB_MTU_1024 = 3,
369 	IB_MTU_2048 = 4,
370 	IB_MTU_4096 = 5
371 };
372 
ib_mtu_enum_to_int(enum ib_mtu mtu)373 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
374 {
375 	switch (mtu) {
376 	case IB_MTU_256:  return  256;
377 	case IB_MTU_512:  return  512;
378 	case IB_MTU_1024: return 1024;
379 	case IB_MTU_2048: return 2048;
380 	case IB_MTU_4096: return 4096;
381 	default: 	  return -1;
382 	}
383 }
384 
ib_mtu_int_to_enum(int mtu)385 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
386 {
387 	if (mtu >= 4096)
388 		return IB_MTU_4096;
389 	else if (mtu >= 2048)
390 		return IB_MTU_2048;
391 	else if (mtu >= 1024)
392 		return IB_MTU_1024;
393 	else if (mtu >= 512)
394 		return IB_MTU_512;
395 	else
396 		return IB_MTU_256;
397 }
398 
399 enum ib_port_state {
400 	IB_PORT_NOP		= 0,
401 	IB_PORT_DOWN		= 1,
402 	IB_PORT_INIT		= 2,
403 	IB_PORT_ARMED		= 3,
404 	IB_PORT_ACTIVE		= 4,
405 	IB_PORT_ACTIVE_DEFER	= 5
406 };
407 
408 enum ib_port_cap_flags {
409 	IB_PORT_SM				= 1 <<  1,
410 	IB_PORT_NOTICE_SUP			= 1 <<  2,
411 	IB_PORT_TRAP_SUP			= 1 <<  3,
412 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
413 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
414 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
415 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
416 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
417 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
418 	IB_PORT_SM_DISABLED			= 1 << 10,
419 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
420 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
421 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
422 	IB_PORT_CM_SUP				= 1 << 16,
423 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
424 	IB_PORT_REINIT_SUP			= 1 << 18,
425 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
426 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
427 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
428 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
429 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
430 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
431 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
432 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
433 };
434 
435 enum ib_port_width {
436 	IB_WIDTH_1X	= 1,
437 	IB_WIDTH_4X	= 2,
438 	IB_WIDTH_8X	= 4,
439 	IB_WIDTH_12X	= 8
440 };
441 
ib_width_enum_to_int(enum ib_port_width width)442 static inline int ib_width_enum_to_int(enum ib_port_width width)
443 {
444 	switch (width) {
445 	case IB_WIDTH_1X:  return  1;
446 	case IB_WIDTH_4X:  return  4;
447 	case IB_WIDTH_8X:  return  8;
448 	case IB_WIDTH_12X: return 12;
449 	default: 	  return -1;
450 	}
451 }
452 
453 enum ib_port_speed {
454 	IB_SPEED_SDR	= 1,
455 	IB_SPEED_DDR	= 2,
456 	IB_SPEED_QDR	= 4,
457 	IB_SPEED_FDR10	= 8,
458 	IB_SPEED_FDR	= 16,
459 	IB_SPEED_EDR	= 32,
460 	IB_SPEED_HDR	= 64
461 };
462 
463 /**
464  * struct rdma_hw_stats
465  * @timestamp - Used by the core code to track when the last update was
466  * @lifespan - Used by the core code to determine how old the counters
467  *   should be before being updated again.  Stored in jiffies, defaults
468  *   to 10 milliseconds, drivers can override the default be specifying
469  *   their own value during their allocation routine.
470  * @name - Array of pointers to static names used for the counters in
471  *   directory.
472  * @num_counters - How many hardware counters there are.  If name is
473  *   shorter than this number, a kernel oops will result.  Driver authors
474  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
475  *   in their code to prevent this.
476  * @value - Array of u64 counters that are accessed by the sysfs code and
477  *   filled in by the drivers get_stats routine
478  */
479 struct rdma_hw_stats {
480 	unsigned long	timestamp;
481 	unsigned long	lifespan;
482 	const char * const *names;
483 	int		num_counters;
484 	u64		value[];
485 };
486 
487 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
488 /**
489  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
490  *   for drivers.
491  * @names - Array of static const char *
492  * @num_counters - How many elements in array
493  * @lifespan - How many milliseconds between updates
494  */
rdma_alloc_hw_stats_struct(const char * const * names,int num_counters,unsigned long lifespan)495 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
496 		const char * const *names, int num_counters,
497 		unsigned long lifespan)
498 {
499 	struct rdma_hw_stats *stats;
500 
501 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
502 			GFP_KERNEL);
503 	if (!stats)
504 		return NULL;
505 	stats->names = names;
506 	stats->num_counters = num_counters;
507 	stats->lifespan = msecs_to_jiffies(lifespan);
508 
509 	return stats;
510 }
511 
512 
513 /* Define bits for the various functionality this port needs to be supported by
514  * the core.
515  */
516 /* Management                           0x00000FFF */
517 #define RDMA_CORE_CAP_IB_MAD            0x00000001
518 #define RDMA_CORE_CAP_IB_SMI            0x00000002
519 #define RDMA_CORE_CAP_IB_CM             0x00000004
520 #define RDMA_CORE_CAP_IW_CM             0x00000008
521 #define RDMA_CORE_CAP_IB_SA             0x00000010
522 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
523 
524 /* Address format                       0x000FF000 */
525 #define RDMA_CORE_CAP_AF_IB             0x00001000
526 #define RDMA_CORE_CAP_ETH_AH            0x00002000
527 #define RDMA_CORE_CAP_OPA_AH            0x00004000
528 
529 /* Protocol                             0xFFF00000 */
530 #define RDMA_CORE_CAP_PROT_IB           0x00100000
531 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
532 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
533 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
534 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
535 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
536 
537 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
538 					| RDMA_CORE_CAP_IB_MAD \
539 					| RDMA_CORE_CAP_IB_SMI \
540 					| RDMA_CORE_CAP_IB_CM  \
541 					| RDMA_CORE_CAP_IB_SA  \
542 					| RDMA_CORE_CAP_AF_IB)
543 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
544 					| RDMA_CORE_CAP_IB_MAD  \
545 					| RDMA_CORE_CAP_IB_CM   \
546 					| RDMA_CORE_CAP_AF_IB   \
547 					| RDMA_CORE_CAP_ETH_AH)
548 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
549 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
550 					| RDMA_CORE_CAP_IB_MAD  \
551 					| RDMA_CORE_CAP_IB_CM   \
552 					| RDMA_CORE_CAP_AF_IB   \
553 					| RDMA_CORE_CAP_ETH_AH)
554 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
555 					| RDMA_CORE_CAP_IW_CM)
556 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
557 					| RDMA_CORE_CAP_OPA_MAD)
558 
559 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
560 
561 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
562 
563 struct ib_port_attr {
564 	u64			subnet_prefix;
565 	enum ib_port_state	state;
566 	enum ib_mtu		max_mtu;
567 	enum ib_mtu		active_mtu;
568 	int			gid_tbl_len;
569 	u32			port_cap_flags;
570 	u32			max_msg_sz;
571 	u32			bad_pkey_cntr;
572 	u32			qkey_viol_cntr;
573 	u16			pkey_tbl_len;
574 	u32			sm_lid;
575 	u32			lid;
576 	u8			lmc;
577 	u8			max_vl_num;
578 	u8			sm_sl;
579 	u8			subnet_timeout;
580 	u8			init_type_reply;
581 	u8			active_width;
582 	u8			active_speed;
583 	u8                      phys_state;
584 	bool			grh_required;
585 };
586 
587 enum ib_device_modify_flags {
588 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
589 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
590 };
591 
592 #define IB_DEVICE_NODE_DESC_MAX 64
593 
594 struct ib_device_modify {
595 	u64	sys_image_guid;
596 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
597 };
598 
599 enum ib_port_modify_flags {
600 	IB_PORT_SHUTDOWN		= 1,
601 	IB_PORT_INIT_TYPE		= (1<<2),
602 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
603 	IB_PORT_OPA_MASK_CHG		= (1<<4)
604 };
605 
606 struct ib_port_modify {
607 	u32	set_port_cap_mask;
608 	u32	clr_port_cap_mask;
609 	u8	init_type;
610 };
611 
612 enum ib_event_type {
613 	IB_EVENT_CQ_ERR,
614 	IB_EVENT_QP_FATAL,
615 	IB_EVENT_QP_REQ_ERR,
616 	IB_EVENT_QP_ACCESS_ERR,
617 	IB_EVENT_COMM_EST,
618 	IB_EVENT_SQ_DRAINED,
619 	IB_EVENT_PATH_MIG,
620 	IB_EVENT_PATH_MIG_ERR,
621 	IB_EVENT_DEVICE_FATAL,
622 	IB_EVENT_PORT_ACTIVE,
623 	IB_EVENT_PORT_ERR,
624 	IB_EVENT_LID_CHANGE,
625 	IB_EVENT_PKEY_CHANGE,
626 	IB_EVENT_SM_CHANGE,
627 	IB_EVENT_SRQ_ERR,
628 	IB_EVENT_SRQ_LIMIT_REACHED,
629 	IB_EVENT_QP_LAST_WQE_REACHED,
630 	IB_EVENT_CLIENT_REREGISTER,
631 	IB_EVENT_GID_CHANGE,
632 	IB_EVENT_WQ_FATAL,
633 };
634 
635 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
636 
637 struct ib_event {
638 	struct ib_device	*device;
639 	union {
640 		struct ib_cq	*cq;
641 		struct ib_qp	*qp;
642 		struct ib_srq	*srq;
643 		struct ib_wq	*wq;
644 		u8		port_num;
645 	} element;
646 	enum ib_event_type	event;
647 };
648 
649 struct ib_event_handler {
650 	struct ib_device *device;
651 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
652 	struct list_head  list;
653 };
654 
655 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
656 	do {							\
657 		(_ptr)->device  = _device;			\
658 		(_ptr)->handler = _handler;			\
659 		INIT_LIST_HEAD(&(_ptr)->list);			\
660 	} while (0)
661 
662 struct ib_global_route {
663 	union ib_gid	dgid;
664 	u32		flow_label;
665 	u8		sgid_index;
666 	u8		hop_limit;
667 	u8		traffic_class;
668 };
669 
670 struct ib_grh {
671 	__be32		version_tclass_flow;
672 	__be16		paylen;
673 	u8		next_hdr;
674 	u8		hop_limit;
675 	union ib_gid	sgid;
676 	union ib_gid	dgid;
677 };
678 
679 union rdma_network_hdr {
680 	struct ib_grh ibgrh;
681 	struct {
682 		/* The IB spec states that if it's IPv4, the header
683 		 * is located in the last 20 bytes of the header.
684 		 */
685 		u8		reserved[20];
686 		struct iphdr	roce4grh;
687 	};
688 };
689 
690 #define IB_QPN_MASK		0xFFFFFF
691 
692 enum {
693 	IB_MULTICAST_QPN = 0xffffff
694 };
695 
696 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
697 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
698 
699 enum ib_ah_flags {
700 	IB_AH_GRH	= 1
701 };
702 
703 enum ib_rate {
704 	IB_RATE_PORT_CURRENT = 0,
705 	IB_RATE_2_5_GBPS = 2,
706 	IB_RATE_5_GBPS   = 5,
707 	IB_RATE_10_GBPS  = 3,
708 	IB_RATE_20_GBPS  = 6,
709 	IB_RATE_30_GBPS  = 4,
710 	IB_RATE_40_GBPS  = 7,
711 	IB_RATE_60_GBPS  = 8,
712 	IB_RATE_80_GBPS  = 9,
713 	IB_RATE_120_GBPS = 10,
714 	IB_RATE_14_GBPS  = 11,
715 	IB_RATE_56_GBPS  = 12,
716 	IB_RATE_112_GBPS = 13,
717 	IB_RATE_168_GBPS = 14,
718 	IB_RATE_25_GBPS  = 15,
719 	IB_RATE_100_GBPS = 16,
720 	IB_RATE_200_GBPS = 17,
721 	IB_RATE_300_GBPS = 18
722 };
723 
724 /**
725  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
726  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
727  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
728  * @rate: rate to convert.
729  */
730 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
731 
732 /**
733  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
734  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
735  * @rate: rate to convert.
736  */
737 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
738 
739 
740 /**
741  * enum ib_mr_type - memory region type
742  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
743  *                            normal registration
744  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
745  *                            signature operations (data-integrity
746  *                            capable regions)
747  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
748  *                            register any arbitrary sg lists (without
749  *                            the normal mr constraints - see
750  *                            ib_map_mr_sg)
751  */
752 enum ib_mr_type {
753 	IB_MR_TYPE_MEM_REG,
754 	IB_MR_TYPE_SIGNATURE,
755 	IB_MR_TYPE_SG_GAPS,
756 };
757 
758 /**
759  * Signature types
760  * IB_SIG_TYPE_NONE: Unprotected.
761  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
762  */
763 enum ib_signature_type {
764 	IB_SIG_TYPE_NONE,
765 	IB_SIG_TYPE_T10_DIF,
766 };
767 
768 /**
769  * Signature T10-DIF block-guard types
770  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
771  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
772  */
773 enum ib_t10_dif_bg_type {
774 	IB_T10DIF_CRC,
775 	IB_T10DIF_CSUM
776 };
777 
778 /**
779  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
780  *     domain.
781  * @bg_type: T10-DIF block guard type (CRC|CSUM)
782  * @pi_interval: protection information interval.
783  * @bg: seed of guard computation.
784  * @app_tag: application tag of guard block
785  * @ref_tag: initial guard block reference tag.
786  * @ref_remap: Indicate wethear the reftag increments each block
787  * @app_escape: Indicate to skip block check if apptag=0xffff
788  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
789  * @apptag_check_mask: check bitmask of application tag.
790  */
791 struct ib_t10_dif_domain {
792 	enum ib_t10_dif_bg_type bg_type;
793 	u16			pi_interval;
794 	u16			bg;
795 	u16			app_tag;
796 	u32			ref_tag;
797 	bool			ref_remap;
798 	bool			app_escape;
799 	bool			ref_escape;
800 	u16			apptag_check_mask;
801 };
802 
803 /**
804  * struct ib_sig_domain - Parameters for signature domain
805  * @sig_type: specific signauture type
806  * @sig: union of all signature domain attributes that may
807  *     be used to set domain layout.
808  */
809 struct ib_sig_domain {
810 	enum ib_signature_type sig_type;
811 	union {
812 		struct ib_t10_dif_domain dif;
813 	} sig;
814 };
815 
816 /**
817  * struct ib_sig_attrs - Parameters for signature handover operation
818  * @check_mask: bitmask for signature byte check (8 bytes)
819  * @mem: memory domain layout desciptor.
820  * @wire: wire domain layout desciptor.
821  */
822 struct ib_sig_attrs {
823 	u8			check_mask;
824 	struct ib_sig_domain	mem;
825 	struct ib_sig_domain	wire;
826 };
827 
828 enum ib_sig_err_type {
829 	IB_SIG_BAD_GUARD,
830 	IB_SIG_BAD_REFTAG,
831 	IB_SIG_BAD_APPTAG,
832 };
833 
834 /**
835  * struct ib_sig_err - signature error descriptor
836  */
837 struct ib_sig_err {
838 	enum ib_sig_err_type	err_type;
839 	u32			expected;
840 	u32			actual;
841 	u64			sig_err_offset;
842 	u32			key;
843 };
844 
845 enum ib_mr_status_check {
846 	IB_MR_CHECK_SIG_STATUS = 1,
847 };
848 
849 /**
850  * struct ib_mr_status - Memory region status container
851  *
852  * @fail_status: Bitmask of MR checks status. For each
853  *     failed check a corresponding status bit is set.
854  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
855  *     failure.
856  */
857 struct ib_mr_status {
858 	u32		    fail_status;
859 	struct ib_sig_err   sig_err;
860 };
861 
862 /**
863  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
864  * enum.
865  * @mult: multiple to convert.
866  */
867 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
868 
869 enum rdma_ah_attr_type {
870 	RDMA_AH_ATTR_TYPE_UNDEFINED,
871 	RDMA_AH_ATTR_TYPE_IB,
872 	RDMA_AH_ATTR_TYPE_ROCE,
873 	RDMA_AH_ATTR_TYPE_OPA,
874 };
875 
876 struct ib_ah_attr {
877 	u16			dlid;
878 	u8			src_path_bits;
879 };
880 
881 struct roce_ah_attr {
882 	u8			dmac[ETH_ALEN];
883 };
884 
885 struct opa_ah_attr {
886 	u32			dlid;
887 	u8			src_path_bits;
888 	bool			make_grd;
889 };
890 
891 struct rdma_ah_attr {
892 	struct ib_global_route	grh;
893 	u8			sl;
894 	u8			static_rate;
895 	u8			port_num;
896 	u8			ah_flags;
897 	enum rdma_ah_attr_type type;
898 	union {
899 		struct ib_ah_attr ib;
900 		struct roce_ah_attr roce;
901 		struct opa_ah_attr opa;
902 	};
903 };
904 
905 enum ib_wc_status {
906 	IB_WC_SUCCESS,
907 	IB_WC_LOC_LEN_ERR,
908 	IB_WC_LOC_QP_OP_ERR,
909 	IB_WC_LOC_EEC_OP_ERR,
910 	IB_WC_LOC_PROT_ERR,
911 	IB_WC_WR_FLUSH_ERR,
912 	IB_WC_MW_BIND_ERR,
913 	IB_WC_BAD_RESP_ERR,
914 	IB_WC_LOC_ACCESS_ERR,
915 	IB_WC_REM_INV_REQ_ERR,
916 	IB_WC_REM_ACCESS_ERR,
917 	IB_WC_REM_OP_ERR,
918 	IB_WC_RETRY_EXC_ERR,
919 	IB_WC_RNR_RETRY_EXC_ERR,
920 	IB_WC_LOC_RDD_VIOL_ERR,
921 	IB_WC_REM_INV_RD_REQ_ERR,
922 	IB_WC_REM_ABORT_ERR,
923 	IB_WC_INV_EECN_ERR,
924 	IB_WC_INV_EEC_STATE_ERR,
925 	IB_WC_FATAL_ERR,
926 	IB_WC_RESP_TIMEOUT_ERR,
927 	IB_WC_GENERAL_ERR
928 };
929 
930 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
931 
932 enum ib_wc_opcode {
933 	IB_WC_SEND,
934 	IB_WC_RDMA_WRITE,
935 	IB_WC_RDMA_READ,
936 	IB_WC_COMP_SWAP,
937 	IB_WC_FETCH_ADD,
938 	IB_WC_LSO,
939 	IB_WC_LOCAL_INV,
940 	IB_WC_REG_MR,
941 	IB_WC_MASKED_COMP_SWAP,
942 	IB_WC_MASKED_FETCH_ADD,
943 /*
944  * Set value of IB_WC_RECV so consumers can test if a completion is a
945  * receive by testing (opcode & IB_WC_RECV).
946  */
947 	IB_WC_RECV			= 1 << 7,
948 	IB_WC_RECV_RDMA_WITH_IMM
949 };
950 
951 enum ib_wc_flags {
952 	IB_WC_GRH		= 1,
953 	IB_WC_WITH_IMM		= (1<<1),
954 	IB_WC_WITH_INVALIDATE	= (1<<2),
955 	IB_WC_IP_CSUM_OK	= (1<<3),
956 	IB_WC_WITH_SMAC		= (1<<4),
957 	IB_WC_WITH_VLAN		= (1<<5),
958 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
959 };
960 
961 struct ib_wc {
962 	union {
963 		u64		wr_id;
964 		struct ib_cqe	*wr_cqe;
965 	};
966 	enum ib_wc_status	status;
967 	enum ib_wc_opcode	opcode;
968 	u32			vendor_err;
969 	u32			byte_len;
970 	struct ib_qp	       *qp;
971 	union {
972 		__be32		imm_data;
973 		u32		invalidate_rkey;
974 	} ex;
975 	u32			src_qp;
976 	u32			slid;
977 	int			wc_flags;
978 	u16			pkey_index;
979 	u8			sl;
980 	u8			dlid_path_bits;
981 	u8			port_num;	/* valid only for DR SMPs on switches */
982 	u8			smac[ETH_ALEN];
983 	u16			vlan_id;
984 	u8			network_hdr_type;
985 };
986 
987 enum ib_cq_notify_flags {
988 	IB_CQ_SOLICITED			= 1 << 0,
989 	IB_CQ_NEXT_COMP			= 1 << 1,
990 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
991 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
992 };
993 
994 enum ib_srq_type {
995 	IB_SRQT_BASIC,
996 	IB_SRQT_XRC,
997 	IB_SRQT_TM,
998 };
999 
ib_srq_has_cq(enum ib_srq_type srq_type)1000 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1001 {
1002 	return srq_type == IB_SRQT_XRC ||
1003 	       srq_type == IB_SRQT_TM;
1004 }
1005 
1006 enum ib_srq_attr_mask {
1007 	IB_SRQ_MAX_WR	= 1 << 0,
1008 	IB_SRQ_LIMIT	= 1 << 1,
1009 };
1010 
1011 struct ib_srq_attr {
1012 	u32	max_wr;
1013 	u32	max_sge;
1014 	u32	srq_limit;
1015 };
1016 
1017 struct ib_srq_init_attr {
1018 	void		      (*event_handler)(struct ib_event *, void *);
1019 	void		       *srq_context;
1020 	struct ib_srq_attr	attr;
1021 	enum ib_srq_type	srq_type;
1022 
1023 	struct {
1024 		struct ib_cq   *cq;
1025 		union {
1026 			struct {
1027 				struct ib_xrcd *xrcd;
1028 			} xrc;
1029 
1030 			struct {
1031 				u32		max_num_tags;
1032 			} tag_matching;
1033 		};
1034 	} ext;
1035 };
1036 
1037 struct ib_qp_cap {
1038 	u32	max_send_wr;
1039 	u32	max_recv_wr;
1040 	u32	max_send_sge;
1041 	u32	max_recv_sge;
1042 	u32	max_inline_data;
1043 
1044 	/*
1045 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1046 	 * ib_create_qp() will calculate the right amount of neededed WRs
1047 	 * and MRs based on this.
1048 	 */
1049 	u32	max_rdma_ctxs;
1050 };
1051 
1052 enum ib_sig_type {
1053 	IB_SIGNAL_ALL_WR,
1054 	IB_SIGNAL_REQ_WR
1055 };
1056 
1057 enum ib_qp_type {
1058 	/*
1059 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1060 	 * here (and in that order) since the MAD layer uses them as
1061 	 * indices into a 2-entry table.
1062 	 */
1063 	IB_QPT_SMI,
1064 	IB_QPT_GSI,
1065 
1066 	IB_QPT_RC,
1067 	IB_QPT_UC,
1068 	IB_QPT_UD,
1069 	IB_QPT_RAW_IPV6,
1070 	IB_QPT_RAW_ETHERTYPE,
1071 	IB_QPT_RAW_PACKET = 8,
1072 	IB_QPT_XRC_INI = 9,
1073 	IB_QPT_XRC_TGT,
1074 	IB_QPT_MAX,
1075 	/* Reserve a range for qp types internal to the low level driver.
1076 	 * These qp types will not be visible at the IB core layer, so the
1077 	 * IB_QPT_MAX usages should not be affected in the core layer
1078 	 */
1079 	IB_QPT_RESERVED1 = 0x1000,
1080 	IB_QPT_RESERVED2,
1081 	IB_QPT_RESERVED3,
1082 	IB_QPT_RESERVED4,
1083 	IB_QPT_RESERVED5,
1084 	IB_QPT_RESERVED6,
1085 	IB_QPT_RESERVED7,
1086 	IB_QPT_RESERVED8,
1087 	IB_QPT_RESERVED9,
1088 	IB_QPT_RESERVED10,
1089 };
1090 
1091 enum ib_qp_create_flags {
1092 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1093 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1094 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1095 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1096 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1097 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1098 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1099 	/* FREE					= 1 << 7, */
1100 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1101 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1102 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1103 	/* reserve bits 26-31 for low level drivers' internal use */
1104 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1105 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1106 };
1107 
1108 /*
1109  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1110  * callback to destroy the passed in QP.
1111  */
1112 
1113 struct ib_qp_init_attr {
1114 	void                  (*event_handler)(struct ib_event *, void *);
1115 	void		       *qp_context;
1116 	struct ib_cq	       *send_cq;
1117 	struct ib_cq	       *recv_cq;
1118 	struct ib_srq	       *srq;
1119 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1120 	struct ib_qp_cap	cap;
1121 	enum ib_sig_type	sq_sig_type;
1122 	enum ib_qp_type		qp_type;
1123 	u32			create_flags;
1124 
1125 	/*
1126 	 * Only needed for special QP types, or when using the RW API.
1127 	 */
1128 	u8			port_num;
1129 	struct ib_rwq_ind_table *rwq_ind_tbl;
1130 	u32			source_qpn;
1131 };
1132 
1133 struct ib_qp_open_attr {
1134 	void                  (*event_handler)(struct ib_event *, void *);
1135 	void		       *qp_context;
1136 	u32			qp_num;
1137 	enum ib_qp_type		qp_type;
1138 };
1139 
1140 enum ib_rnr_timeout {
1141 	IB_RNR_TIMER_655_36 =  0,
1142 	IB_RNR_TIMER_000_01 =  1,
1143 	IB_RNR_TIMER_000_02 =  2,
1144 	IB_RNR_TIMER_000_03 =  3,
1145 	IB_RNR_TIMER_000_04 =  4,
1146 	IB_RNR_TIMER_000_06 =  5,
1147 	IB_RNR_TIMER_000_08 =  6,
1148 	IB_RNR_TIMER_000_12 =  7,
1149 	IB_RNR_TIMER_000_16 =  8,
1150 	IB_RNR_TIMER_000_24 =  9,
1151 	IB_RNR_TIMER_000_32 = 10,
1152 	IB_RNR_TIMER_000_48 = 11,
1153 	IB_RNR_TIMER_000_64 = 12,
1154 	IB_RNR_TIMER_000_96 = 13,
1155 	IB_RNR_TIMER_001_28 = 14,
1156 	IB_RNR_TIMER_001_92 = 15,
1157 	IB_RNR_TIMER_002_56 = 16,
1158 	IB_RNR_TIMER_003_84 = 17,
1159 	IB_RNR_TIMER_005_12 = 18,
1160 	IB_RNR_TIMER_007_68 = 19,
1161 	IB_RNR_TIMER_010_24 = 20,
1162 	IB_RNR_TIMER_015_36 = 21,
1163 	IB_RNR_TIMER_020_48 = 22,
1164 	IB_RNR_TIMER_030_72 = 23,
1165 	IB_RNR_TIMER_040_96 = 24,
1166 	IB_RNR_TIMER_061_44 = 25,
1167 	IB_RNR_TIMER_081_92 = 26,
1168 	IB_RNR_TIMER_122_88 = 27,
1169 	IB_RNR_TIMER_163_84 = 28,
1170 	IB_RNR_TIMER_245_76 = 29,
1171 	IB_RNR_TIMER_327_68 = 30,
1172 	IB_RNR_TIMER_491_52 = 31
1173 };
1174 
1175 enum ib_qp_attr_mask {
1176 	IB_QP_STATE			= 1,
1177 	IB_QP_CUR_STATE			= (1<<1),
1178 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1179 	IB_QP_ACCESS_FLAGS		= (1<<3),
1180 	IB_QP_PKEY_INDEX		= (1<<4),
1181 	IB_QP_PORT			= (1<<5),
1182 	IB_QP_QKEY			= (1<<6),
1183 	IB_QP_AV			= (1<<7),
1184 	IB_QP_PATH_MTU			= (1<<8),
1185 	IB_QP_TIMEOUT			= (1<<9),
1186 	IB_QP_RETRY_CNT			= (1<<10),
1187 	IB_QP_RNR_RETRY			= (1<<11),
1188 	IB_QP_RQ_PSN			= (1<<12),
1189 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1190 	IB_QP_ALT_PATH			= (1<<14),
1191 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1192 	IB_QP_SQ_PSN			= (1<<16),
1193 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1194 	IB_QP_PATH_MIG_STATE		= (1<<18),
1195 	IB_QP_CAP			= (1<<19),
1196 	IB_QP_DEST_QPN			= (1<<20),
1197 	IB_QP_RESERVED1			= (1<<21),
1198 	IB_QP_RESERVED2			= (1<<22),
1199 	IB_QP_RESERVED3			= (1<<23),
1200 	IB_QP_RESERVED4			= (1<<24),
1201 	IB_QP_RATE_LIMIT		= (1<<25),
1202 };
1203 
1204 enum ib_qp_state {
1205 	IB_QPS_RESET,
1206 	IB_QPS_INIT,
1207 	IB_QPS_RTR,
1208 	IB_QPS_RTS,
1209 	IB_QPS_SQD,
1210 	IB_QPS_SQE,
1211 	IB_QPS_ERR
1212 };
1213 
1214 enum ib_mig_state {
1215 	IB_MIG_MIGRATED,
1216 	IB_MIG_REARM,
1217 	IB_MIG_ARMED
1218 };
1219 
1220 enum ib_mw_type {
1221 	IB_MW_TYPE_1 = 1,
1222 	IB_MW_TYPE_2 = 2
1223 };
1224 
1225 struct ib_qp_attr {
1226 	enum ib_qp_state	qp_state;
1227 	enum ib_qp_state	cur_qp_state;
1228 	enum ib_mtu		path_mtu;
1229 	enum ib_mig_state	path_mig_state;
1230 	u32			qkey;
1231 	u32			rq_psn;
1232 	u32			sq_psn;
1233 	u32			dest_qp_num;
1234 	int			qp_access_flags;
1235 	struct ib_qp_cap	cap;
1236 	struct rdma_ah_attr	ah_attr;
1237 	struct rdma_ah_attr	alt_ah_attr;
1238 	u16			pkey_index;
1239 	u16			alt_pkey_index;
1240 	u8			en_sqd_async_notify;
1241 	u8			sq_draining;
1242 	u8			max_rd_atomic;
1243 	u8			max_dest_rd_atomic;
1244 	u8			min_rnr_timer;
1245 	u8			port_num;
1246 	u8			timeout;
1247 	u8			retry_cnt;
1248 	u8			rnr_retry;
1249 	u8			alt_port_num;
1250 	u8			alt_timeout;
1251 	u32			rate_limit;
1252 };
1253 
1254 enum ib_wr_opcode {
1255 	/* These are shared with userspace */
1256 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1257 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1258 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1259 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1260 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1261 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1262 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1263 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1264 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1265 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1266 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1267 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1268 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1269 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1270 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1271 
1272 	/* These are kernel only and can not be issued by userspace */
1273 	IB_WR_REG_MR = 0x20,
1274 	IB_WR_REG_SIG_MR,
1275 
1276 	/* reserve values for low level drivers' internal use.
1277 	 * These values will not be used at all in the ib core layer.
1278 	 */
1279 	IB_WR_RESERVED1 = 0xf0,
1280 	IB_WR_RESERVED2,
1281 	IB_WR_RESERVED3,
1282 	IB_WR_RESERVED4,
1283 	IB_WR_RESERVED5,
1284 	IB_WR_RESERVED6,
1285 	IB_WR_RESERVED7,
1286 	IB_WR_RESERVED8,
1287 	IB_WR_RESERVED9,
1288 	IB_WR_RESERVED10,
1289 };
1290 
1291 enum ib_send_flags {
1292 	IB_SEND_FENCE		= 1,
1293 	IB_SEND_SIGNALED	= (1<<1),
1294 	IB_SEND_SOLICITED	= (1<<2),
1295 	IB_SEND_INLINE		= (1<<3),
1296 	IB_SEND_IP_CSUM		= (1<<4),
1297 
1298 	/* reserve bits 26-31 for low level drivers' internal use */
1299 	IB_SEND_RESERVED_START	= (1 << 26),
1300 	IB_SEND_RESERVED_END	= (1 << 31),
1301 };
1302 
1303 struct ib_sge {
1304 	u64	addr;
1305 	u32	length;
1306 	u32	lkey;
1307 };
1308 
1309 struct ib_cqe {
1310 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1311 };
1312 
1313 struct ib_send_wr {
1314 	struct ib_send_wr      *next;
1315 	union {
1316 		u64		wr_id;
1317 		struct ib_cqe	*wr_cqe;
1318 	};
1319 	struct ib_sge	       *sg_list;
1320 	int			num_sge;
1321 	enum ib_wr_opcode	opcode;
1322 	int			send_flags;
1323 	union {
1324 		__be32		imm_data;
1325 		u32		invalidate_rkey;
1326 	} ex;
1327 };
1328 
1329 struct ib_rdma_wr {
1330 	struct ib_send_wr	wr;
1331 	u64			remote_addr;
1332 	u32			rkey;
1333 };
1334 
rdma_wr(struct ib_send_wr * wr)1335 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1336 {
1337 	return container_of(wr, struct ib_rdma_wr, wr);
1338 }
1339 
1340 struct ib_atomic_wr {
1341 	struct ib_send_wr	wr;
1342 	u64			remote_addr;
1343 	u64			compare_add;
1344 	u64			swap;
1345 	u64			compare_add_mask;
1346 	u64			swap_mask;
1347 	u32			rkey;
1348 };
1349 
atomic_wr(struct ib_send_wr * wr)1350 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1351 {
1352 	return container_of(wr, struct ib_atomic_wr, wr);
1353 }
1354 
1355 struct ib_ud_wr {
1356 	struct ib_send_wr	wr;
1357 	struct ib_ah		*ah;
1358 	void			*header;
1359 	int			hlen;
1360 	int			mss;
1361 	u32			remote_qpn;
1362 	u32			remote_qkey;
1363 	u16			pkey_index; /* valid for GSI only */
1364 	u8			port_num;   /* valid for DR SMPs on switch only */
1365 };
1366 
ud_wr(struct ib_send_wr * wr)1367 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1368 {
1369 	return container_of(wr, struct ib_ud_wr, wr);
1370 }
1371 
1372 struct ib_reg_wr {
1373 	struct ib_send_wr	wr;
1374 	struct ib_mr		*mr;
1375 	u32			key;
1376 	int			access;
1377 };
1378 
reg_wr(struct ib_send_wr * wr)1379 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1380 {
1381 	return container_of(wr, struct ib_reg_wr, wr);
1382 }
1383 
1384 struct ib_sig_handover_wr {
1385 	struct ib_send_wr	wr;
1386 	struct ib_sig_attrs    *sig_attrs;
1387 	struct ib_mr	       *sig_mr;
1388 	int			access_flags;
1389 	struct ib_sge	       *prot;
1390 };
1391 
sig_handover_wr(struct ib_send_wr * wr)1392 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1393 {
1394 	return container_of(wr, struct ib_sig_handover_wr, wr);
1395 }
1396 
1397 struct ib_recv_wr {
1398 	struct ib_recv_wr      *next;
1399 	union {
1400 		u64		wr_id;
1401 		struct ib_cqe	*wr_cqe;
1402 	};
1403 	struct ib_sge	       *sg_list;
1404 	int			num_sge;
1405 };
1406 
1407 enum ib_access_flags {
1408 	IB_ACCESS_LOCAL_WRITE	= 1,
1409 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1410 	IB_ACCESS_REMOTE_READ	= (1<<2),
1411 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1412 	IB_ACCESS_MW_BIND	= (1<<4),
1413 	IB_ZERO_BASED		= (1<<5),
1414 	IB_ACCESS_ON_DEMAND     = (1<<6),
1415 	IB_ACCESS_HUGETLB	= (1<<7),
1416 };
1417 
1418 /*
1419  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1420  * are hidden here instead of a uapi header!
1421  */
1422 enum ib_mr_rereg_flags {
1423 	IB_MR_REREG_TRANS	= 1,
1424 	IB_MR_REREG_PD		= (1<<1),
1425 	IB_MR_REREG_ACCESS	= (1<<2),
1426 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1427 };
1428 
1429 struct ib_fmr_attr {
1430 	int	max_pages;
1431 	int	max_maps;
1432 	u8	page_shift;
1433 };
1434 
1435 struct ib_umem;
1436 
1437 enum rdma_remove_reason {
1438 	/* Userspace requested uobject deletion. Call could fail */
1439 	RDMA_REMOVE_DESTROY,
1440 	/* Context deletion. This call should delete the actual object itself */
1441 	RDMA_REMOVE_CLOSE,
1442 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1443 	RDMA_REMOVE_DRIVER_REMOVE,
1444 	/* Context is being cleaned-up, but commit was just completed */
1445 	RDMA_REMOVE_DURING_CLEANUP,
1446 };
1447 
1448 struct ib_rdmacg_object {
1449 #ifdef CONFIG_CGROUP_RDMA
1450 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1451 #endif
1452 };
1453 
1454 struct ib_ucontext {
1455 	struct ib_device       *device;
1456 	struct ib_uverbs_file  *ufile;
1457 	int			closing;
1458 
1459 	/* locking the uobjects_list */
1460 	struct mutex		uobjects_lock;
1461 	struct list_head	uobjects;
1462 	/* protects cleanup process from other actions */
1463 	struct rw_semaphore	cleanup_rwsem;
1464 	enum rdma_remove_reason cleanup_reason;
1465 
1466 	struct pid             *tgid;
1467 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1468 	struct rb_root_cached   umem_tree;
1469 	/*
1470 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1471 	 * mmu notifiers registration.
1472 	 */
1473 	struct rw_semaphore	umem_rwsem;
1474 	void (*invalidate_range)(struct ib_umem *umem,
1475 				 unsigned long start, unsigned long end);
1476 
1477 	struct mmu_notifier	mn;
1478 	atomic_t		notifier_count;
1479 	/* A list of umems that don't have private mmu notifier counters yet. */
1480 	struct list_head	no_private_counters;
1481 	int                     odp_mrs_count;
1482 #endif
1483 
1484 	struct ib_rdmacg_object	cg_obj;
1485 };
1486 
1487 struct ib_uobject {
1488 	u64			user_handle;	/* handle given to us by userspace */
1489 	struct ib_ucontext     *context;	/* associated user context */
1490 	void		       *object;		/* containing object */
1491 	struct list_head	list;		/* link to context's list */
1492 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1493 	int			id;		/* index into kernel idr */
1494 	struct kref		ref;
1495 	atomic_t		usecnt;		/* protects exclusive access */
1496 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1497 
1498 	const struct uverbs_obj_type *type;
1499 };
1500 
1501 struct ib_uobject_file {
1502 	struct ib_uobject	uobj;
1503 	/* ufile contains the lock between context release and file close */
1504 	struct ib_uverbs_file	*ufile;
1505 };
1506 
1507 struct ib_udata {
1508 	const void __user *inbuf;
1509 	void __user *outbuf;
1510 	size_t       inlen;
1511 	size_t       outlen;
1512 };
1513 
1514 struct ib_pd {
1515 	u32			local_dma_lkey;
1516 	u32			flags;
1517 	struct ib_device       *device;
1518 	struct ib_uobject      *uobject;
1519 	atomic_t          	usecnt; /* count all resources */
1520 
1521 	u32			unsafe_global_rkey;
1522 
1523 	/*
1524 	 * Implementation details of the RDMA core, don't use in drivers:
1525 	 */
1526 	struct ib_mr	       *__internal_mr;
1527 };
1528 
1529 struct ib_xrcd {
1530 	struct ib_device       *device;
1531 	atomic_t		usecnt; /* count all exposed resources */
1532 	struct inode	       *inode;
1533 
1534 	struct mutex		tgt_qp_mutex;
1535 	struct list_head	tgt_qp_list;
1536 };
1537 
1538 struct ib_ah {
1539 	struct ib_device	*device;
1540 	struct ib_pd		*pd;
1541 	struct ib_uobject	*uobject;
1542 	enum rdma_ah_attr_type	type;
1543 };
1544 
1545 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1546 
1547 enum ib_poll_context {
1548 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1549 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1550 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1551 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1552 };
1553 
1554 struct ib_cq {
1555 	struct ib_device       *device;
1556 	struct ib_uobject      *uobject;
1557 	ib_comp_handler   	comp_handler;
1558 	void                  (*event_handler)(struct ib_event *, void *);
1559 	void                   *cq_context;
1560 	int               	cqe;
1561 	atomic_t          	usecnt; /* count number of work queues */
1562 	enum ib_poll_context	poll_ctx;
1563 	struct ib_wc		*wc;
1564 	union {
1565 		struct irq_poll		iop;
1566 		struct work_struct	work;
1567 	};
1568 	struct workqueue_struct *comp_wq;
1569 };
1570 
1571 struct ib_srq {
1572 	struct ib_device       *device;
1573 	struct ib_pd	       *pd;
1574 	struct ib_uobject      *uobject;
1575 	void		      (*event_handler)(struct ib_event *, void *);
1576 	void		       *srq_context;
1577 	enum ib_srq_type	srq_type;
1578 	atomic_t		usecnt;
1579 
1580 	struct {
1581 		struct ib_cq   *cq;
1582 		union {
1583 			struct {
1584 				struct ib_xrcd *xrcd;
1585 				u32		srq_num;
1586 			} xrc;
1587 		};
1588 	} ext;
1589 };
1590 
1591 enum ib_raw_packet_caps {
1592 	/* Strip cvlan from incoming packet and report it in the matching work
1593 	 * completion is supported.
1594 	 */
1595 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1596 	/* Scatter FCS field of an incoming packet to host memory is supported.
1597 	 */
1598 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1599 	/* Checksum offloads are supported (for both send and receive). */
1600 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1601 	/* When a packet is received for an RQ with no receive WQEs, the
1602 	 * packet processing is delayed.
1603 	 */
1604 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1605 };
1606 
1607 enum ib_wq_type {
1608 	IB_WQT_RQ
1609 };
1610 
1611 enum ib_wq_state {
1612 	IB_WQS_RESET,
1613 	IB_WQS_RDY,
1614 	IB_WQS_ERR
1615 };
1616 
1617 struct ib_wq {
1618 	struct ib_device       *device;
1619 	struct ib_uobject      *uobject;
1620 	void		    *wq_context;
1621 	void		    (*event_handler)(struct ib_event *, void *);
1622 	struct ib_pd	       *pd;
1623 	struct ib_cq	       *cq;
1624 	u32		wq_num;
1625 	enum ib_wq_state       state;
1626 	enum ib_wq_type	wq_type;
1627 	atomic_t		usecnt;
1628 };
1629 
1630 enum ib_wq_flags {
1631 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1632 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1633 	IB_WQ_FLAGS_DELAY_DROP		= 1 << 2,
1634 };
1635 
1636 struct ib_wq_init_attr {
1637 	void		       *wq_context;
1638 	enum ib_wq_type	wq_type;
1639 	u32		max_wr;
1640 	u32		max_sge;
1641 	struct	ib_cq	       *cq;
1642 	void		    (*event_handler)(struct ib_event *, void *);
1643 	u32		create_flags; /* Use enum ib_wq_flags */
1644 };
1645 
1646 enum ib_wq_attr_mask {
1647 	IB_WQ_STATE		= 1 << 0,
1648 	IB_WQ_CUR_STATE		= 1 << 1,
1649 	IB_WQ_FLAGS		= 1 << 2,
1650 };
1651 
1652 struct ib_wq_attr {
1653 	enum	ib_wq_state	wq_state;
1654 	enum	ib_wq_state	curr_wq_state;
1655 	u32			flags; /* Use enum ib_wq_flags */
1656 	u32			flags_mask; /* Use enum ib_wq_flags */
1657 };
1658 
1659 struct ib_rwq_ind_table {
1660 	struct ib_device	*device;
1661 	struct ib_uobject      *uobject;
1662 	atomic_t		usecnt;
1663 	u32		ind_tbl_num;
1664 	u32		log_ind_tbl_size;
1665 	struct ib_wq	**ind_tbl;
1666 };
1667 
1668 struct ib_rwq_ind_table_init_attr {
1669 	u32		log_ind_tbl_size;
1670 	/* Each entry is a pointer to Receive Work Queue */
1671 	struct ib_wq	**ind_tbl;
1672 };
1673 
1674 enum port_pkey_state {
1675 	IB_PORT_PKEY_NOT_VALID = 0,
1676 	IB_PORT_PKEY_VALID = 1,
1677 	IB_PORT_PKEY_LISTED = 2,
1678 };
1679 
1680 struct ib_qp_security;
1681 
1682 struct ib_port_pkey {
1683 	enum port_pkey_state	state;
1684 	u16			pkey_index;
1685 	u8			port_num;
1686 	struct list_head	qp_list;
1687 	struct list_head	to_error_list;
1688 	struct ib_qp_security  *sec;
1689 };
1690 
1691 struct ib_ports_pkeys {
1692 	struct ib_port_pkey	main;
1693 	struct ib_port_pkey	alt;
1694 };
1695 
1696 struct ib_qp_security {
1697 	struct ib_qp	       *qp;
1698 	struct ib_device       *dev;
1699 	/* Hold this mutex when changing port and pkey settings. */
1700 	struct mutex		mutex;
1701 	struct ib_ports_pkeys  *ports_pkeys;
1702 	/* A list of all open shared QP handles.  Required to enforce security
1703 	 * properly for all users of a shared QP.
1704 	 */
1705 	struct list_head        shared_qp_list;
1706 	void                   *security;
1707 	bool			destroying;
1708 	atomic_t		error_list_count;
1709 	struct completion	error_complete;
1710 	int			error_comps_pending;
1711 };
1712 
1713 /*
1714  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1715  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1716  */
1717 struct ib_qp {
1718 	struct ib_device       *device;
1719 	struct ib_pd	       *pd;
1720 	struct ib_cq	       *send_cq;
1721 	struct ib_cq	       *recv_cq;
1722 	spinlock_t		mr_lock;
1723 	int			mrs_used;
1724 	struct list_head	rdma_mrs;
1725 	struct list_head	sig_mrs;
1726 	struct ib_srq	       *srq;
1727 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1728 	struct list_head	xrcd_list;
1729 
1730 	/* count times opened, mcast attaches, flow attaches */
1731 	atomic_t		usecnt;
1732 	struct list_head	open_list;
1733 	struct ib_qp           *real_qp;
1734 	struct ib_uobject      *uobject;
1735 	void                  (*event_handler)(struct ib_event *, void *);
1736 	void		       *qp_context;
1737 	u32			qp_num;
1738 	u32			max_write_sge;
1739 	u32			max_read_sge;
1740 	enum ib_qp_type		qp_type;
1741 	struct ib_rwq_ind_table *rwq_ind_tbl;
1742 	struct ib_qp_security  *qp_sec;
1743 	u8			port;
1744 };
1745 
1746 struct ib_mr {
1747 	struct ib_device  *device;
1748 	struct ib_pd	  *pd;
1749 	u32		   lkey;
1750 	u32		   rkey;
1751 	u64		   iova;
1752 	u64		   length;
1753 	unsigned int	   page_size;
1754 	bool		   need_inval;
1755 	union {
1756 		struct ib_uobject	*uobject;	/* user */
1757 		struct list_head	qp_entry;	/* FR */
1758 	};
1759 };
1760 
1761 struct ib_mw {
1762 	struct ib_device	*device;
1763 	struct ib_pd		*pd;
1764 	struct ib_uobject	*uobject;
1765 	u32			rkey;
1766 	enum ib_mw_type         type;
1767 };
1768 
1769 struct ib_fmr {
1770 	struct ib_device	*device;
1771 	struct ib_pd		*pd;
1772 	struct list_head	list;
1773 	u32			lkey;
1774 	u32			rkey;
1775 };
1776 
1777 /* Supported steering options */
1778 enum ib_flow_attr_type {
1779 	/* steering according to rule specifications */
1780 	IB_FLOW_ATTR_NORMAL		= 0x0,
1781 	/* default unicast and multicast rule -
1782 	 * receive all Eth traffic which isn't steered to any QP
1783 	 */
1784 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1785 	/* default multicast rule -
1786 	 * receive all Eth multicast traffic which isn't steered to any QP
1787 	 */
1788 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1789 	/* sniffer rule - receive all port traffic */
1790 	IB_FLOW_ATTR_SNIFFER		= 0x3
1791 };
1792 
1793 /* Supported steering header types */
1794 enum ib_flow_spec_type {
1795 	/* L2 headers*/
1796 	IB_FLOW_SPEC_ETH		= 0x20,
1797 	IB_FLOW_SPEC_IB			= 0x22,
1798 	/* L3 header*/
1799 	IB_FLOW_SPEC_IPV4		= 0x30,
1800 	IB_FLOW_SPEC_IPV6		= 0x31,
1801 	/* L4 headers*/
1802 	IB_FLOW_SPEC_TCP		= 0x40,
1803 	IB_FLOW_SPEC_UDP		= 0x41,
1804 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1805 	IB_FLOW_SPEC_INNER		= 0x100,
1806 	/* Actions */
1807 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1808 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1809 };
1810 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1811 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1812 
1813 /* Flow steering rule priority is set according to it's domain.
1814  * Lower domain value means higher priority.
1815  */
1816 enum ib_flow_domain {
1817 	IB_FLOW_DOMAIN_USER,
1818 	IB_FLOW_DOMAIN_ETHTOOL,
1819 	IB_FLOW_DOMAIN_RFS,
1820 	IB_FLOW_DOMAIN_NIC,
1821 	IB_FLOW_DOMAIN_NUM /* Must be last */
1822 };
1823 
1824 enum ib_flow_flags {
1825 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1826 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1827 };
1828 
1829 struct ib_flow_eth_filter {
1830 	u8	dst_mac[6];
1831 	u8	src_mac[6];
1832 	__be16	ether_type;
1833 	__be16	vlan_tag;
1834 	/* Must be last */
1835 	u8	real_sz[0];
1836 };
1837 
1838 struct ib_flow_spec_eth {
1839 	u32			  type;
1840 	u16			  size;
1841 	struct ib_flow_eth_filter val;
1842 	struct ib_flow_eth_filter mask;
1843 };
1844 
1845 struct ib_flow_ib_filter {
1846 	__be16 dlid;
1847 	__u8   sl;
1848 	/* Must be last */
1849 	u8	real_sz[0];
1850 };
1851 
1852 struct ib_flow_spec_ib {
1853 	u32			 type;
1854 	u16			 size;
1855 	struct ib_flow_ib_filter val;
1856 	struct ib_flow_ib_filter mask;
1857 };
1858 
1859 /* IPv4 header flags */
1860 enum ib_ipv4_flags {
1861 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1862 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1863 				    last have this flag set */
1864 };
1865 
1866 struct ib_flow_ipv4_filter {
1867 	__be32	src_ip;
1868 	__be32	dst_ip;
1869 	u8	proto;
1870 	u8	tos;
1871 	u8	ttl;
1872 	u8	flags;
1873 	/* Must be last */
1874 	u8	real_sz[0];
1875 };
1876 
1877 struct ib_flow_spec_ipv4 {
1878 	u32			   type;
1879 	u16			   size;
1880 	struct ib_flow_ipv4_filter val;
1881 	struct ib_flow_ipv4_filter mask;
1882 };
1883 
1884 struct ib_flow_ipv6_filter {
1885 	u8	src_ip[16];
1886 	u8	dst_ip[16];
1887 	__be32	flow_label;
1888 	u8	next_hdr;
1889 	u8	traffic_class;
1890 	u8	hop_limit;
1891 	/* Must be last */
1892 	u8	real_sz[0];
1893 };
1894 
1895 struct ib_flow_spec_ipv6 {
1896 	u32			   type;
1897 	u16			   size;
1898 	struct ib_flow_ipv6_filter val;
1899 	struct ib_flow_ipv6_filter mask;
1900 };
1901 
1902 struct ib_flow_tcp_udp_filter {
1903 	__be16	dst_port;
1904 	__be16	src_port;
1905 	/* Must be last */
1906 	u8	real_sz[0];
1907 };
1908 
1909 struct ib_flow_spec_tcp_udp {
1910 	u32			      type;
1911 	u16			      size;
1912 	struct ib_flow_tcp_udp_filter val;
1913 	struct ib_flow_tcp_udp_filter mask;
1914 };
1915 
1916 struct ib_flow_tunnel_filter {
1917 	__be32	tunnel_id;
1918 	u8	real_sz[0];
1919 };
1920 
1921 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1922  * the tunnel_id from val has the vni value
1923  */
1924 struct ib_flow_spec_tunnel {
1925 	u32			      type;
1926 	u16			      size;
1927 	struct ib_flow_tunnel_filter  val;
1928 	struct ib_flow_tunnel_filter  mask;
1929 };
1930 
1931 struct ib_flow_spec_action_tag {
1932 	enum ib_flow_spec_type	      type;
1933 	u16			      size;
1934 	u32                           tag_id;
1935 };
1936 
1937 struct ib_flow_spec_action_drop {
1938 	enum ib_flow_spec_type	      type;
1939 	u16			      size;
1940 };
1941 
1942 union ib_flow_spec {
1943 	struct {
1944 		u32			type;
1945 		u16			size;
1946 	};
1947 	struct ib_flow_spec_eth		eth;
1948 	struct ib_flow_spec_ib		ib;
1949 	struct ib_flow_spec_ipv4        ipv4;
1950 	struct ib_flow_spec_tcp_udp	tcp_udp;
1951 	struct ib_flow_spec_ipv6        ipv6;
1952 	struct ib_flow_spec_tunnel      tunnel;
1953 	struct ib_flow_spec_action_tag  flow_tag;
1954 	struct ib_flow_spec_action_drop drop;
1955 };
1956 
1957 struct ib_flow_attr {
1958 	enum ib_flow_attr_type type;
1959 	u16	     size;
1960 	u16	     priority;
1961 	u32	     flags;
1962 	u8	     num_of_specs;
1963 	u8	     port;
1964 	/* Following are the optional layers according to user request
1965 	 * struct ib_flow_spec_xxx
1966 	 * struct ib_flow_spec_yyy
1967 	 */
1968 };
1969 
1970 struct ib_flow {
1971 	struct ib_qp		*qp;
1972 	struct ib_uobject	*uobject;
1973 };
1974 
1975 struct ib_mad_hdr;
1976 struct ib_grh;
1977 
1978 enum ib_process_mad_flags {
1979 	IB_MAD_IGNORE_MKEY	= 1,
1980 	IB_MAD_IGNORE_BKEY	= 2,
1981 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1982 };
1983 
1984 enum ib_mad_result {
1985 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1986 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1987 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1988 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1989 };
1990 
1991 struct ib_port_cache {
1992 	u64		      subnet_prefix;
1993 	struct ib_pkey_cache  *pkey;
1994 	struct ib_gid_table   *gid;
1995 	u8                     lmc;
1996 	enum ib_port_state     port_state;
1997 };
1998 
1999 struct ib_cache {
2000 	rwlock_t                lock;
2001 	struct ib_event_handler event_handler;
2002 	struct ib_port_cache   *ports;
2003 };
2004 
2005 struct iw_cm_verbs;
2006 
2007 struct ib_port_immutable {
2008 	int                           pkey_tbl_len;
2009 	int                           gid_tbl_len;
2010 	u32                           core_cap_flags;
2011 	u32                           max_mad_size;
2012 };
2013 
2014 /* rdma netdev type - specifies protocol type */
2015 enum rdma_netdev_t {
2016 	RDMA_NETDEV_OPA_VNIC,
2017 	RDMA_NETDEV_IPOIB,
2018 };
2019 
2020 /**
2021  * struct rdma_netdev - rdma netdev
2022  * For cases where netstack interfacing is required.
2023  */
2024 struct rdma_netdev {
2025 	void              *clnt_priv;
2026 	struct ib_device  *hca;
2027 	u8                 port_num;
2028 
2029 	/* cleanup function must be specified */
2030 	void (*free_rdma_netdev)(struct net_device *netdev);
2031 
2032 	/* control functions */
2033 	void (*set_id)(struct net_device *netdev, int id);
2034 	/* send packet */
2035 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2036 		    struct ib_ah *address, u32 dqpn);
2037 	/* multicast */
2038 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2039 			    union ib_gid *gid, u16 mlid,
2040 			    int set_qkey, u32 qkey);
2041 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2042 			    union ib_gid *gid, u16 mlid);
2043 };
2044 
2045 struct ib_port_pkey_list {
2046 	/* Lock to hold while modifying the list. */
2047 	spinlock_t                    list_lock;
2048 	struct list_head              pkey_list;
2049 };
2050 
2051 struct ib_device {
2052 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2053 	struct device                *dma_device;
2054 
2055 	char                          name[IB_DEVICE_NAME_MAX];
2056 
2057 	struct list_head              event_handler_list;
2058 	spinlock_t                    event_handler_lock;
2059 
2060 	spinlock_t                    client_data_lock;
2061 	struct list_head              core_list;
2062 	/* Access to the client_data_list is protected by the client_data_lock
2063 	 * spinlock and the lists_rwsem read-write semaphore */
2064 	struct list_head              client_data_list;
2065 
2066 	struct ib_cache               cache;
2067 	/**
2068 	 * port_immutable is indexed by port number
2069 	 */
2070 	struct ib_port_immutable     *port_immutable;
2071 
2072 	int			      num_comp_vectors;
2073 
2074 	struct ib_port_pkey_list     *port_pkey_list;
2075 
2076 	struct iw_cm_verbs	     *iwcm;
2077 
2078 	/**
2079 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2080 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2081 	 *   core when the device is removed.  A lifespan of -1 in the return
2082 	 *   struct tells the core to set a default lifespan.
2083 	 */
2084 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
2085 						     u8 port_num);
2086 	/**
2087 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2088 	 * @index - The index in the value array we wish to have updated, or
2089 	 *   num_counters if we want all stats updated
2090 	 * Return codes -
2091 	 *   < 0 - Error, no counters updated
2092 	 *   index - Updated the single counter pointed to by index
2093 	 *   num_counters - Updated all counters (will reset the timestamp
2094 	 *     and prevent further calls for lifespan milliseconds)
2095 	 * Drivers are allowed to update all counters in leiu of just the
2096 	 *   one given in index at their option
2097 	 */
2098 	int		           (*get_hw_stats)(struct ib_device *device,
2099 						   struct rdma_hw_stats *stats,
2100 						   u8 port, int index);
2101 	int		           (*query_device)(struct ib_device *device,
2102 						   struct ib_device_attr *device_attr,
2103 						   struct ib_udata *udata);
2104 	int		           (*query_port)(struct ib_device *device,
2105 						 u8 port_num,
2106 						 struct ib_port_attr *port_attr);
2107 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
2108 						     u8 port_num);
2109 	/* When calling get_netdev, the HW vendor's driver should return the
2110 	 * net device of device @device at port @port_num or NULL if such
2111 	 * a net device doesn't exist. The vendor driver should call dev_hold
2112 	 * on this net device. The HW vendor's device driver must guarantee
2113 	 * that this function returns NULL before the net device reaches
2114 	 * NETDEV_UNREGISTER_FINAL state.
2115 	 */
2116 	struct net_device	  *(*get_netdev)(struct ib_device *device,
2117 						 u8 port_num);
2118 	int		           (*query_gid)(struct ib_device *device,
2119 						u8 port_num, int index,
2120 						union ib_gid *gid);
2121 	/* When calling add_gid, the HW vendor's driver should
2122 	 * add the gid of device @device at gid index @index of
2123 	 * port @port_num to be @gid. Meta-info of that gid (for example,
2124 	 * the network device related to this gid is available
2125 	 * at @attr. @context allows the HW vendor driver to store extra
2126 	 * information together with a GID entry. The HW vendor may allocate
2127 	 * memory to contain this information and store it in @context when a
2128 	 * new GID entry is written to. Params are consistent until the next
2129 	 * call of add_gid or delete_gid. The function should return 0 on
2130 	 * success or error otherwise. The function could be called
2131 	 * concurrently for different ports. This function is only called
2132 	 * when roce_gid_table is used.
2133 	 */
2134 	int		           (*add_gid)(struct ib_device *device,
2135 					      u8 port_num,
2136 					      unsigned int index,
2137 					      const union ib_gid *gid,
2138 					      const struct ib_gid_attr *attr,
2139 					      void **context);
2140 	/* When calling del_gid, the HW vendor's driver should delete the
2141 	 * gid of device @device at gid index @index of port @port_num.
2142 	 * Upon the deletion of a GID entry, the HW vendor must free any
2143 	 * allocated memory. The caller will clear @context afterwards.
2144 	 * This function is only called when roce_gid_table is used.
2145 	 */
2146 	int		           (*del_gid)(struct ib_device *device,
2147 					      u8 port_num,
2148 					      unsigned int index,
2149 					      void **context);
2150 	int		           (*query_pkey)(struct ib_device *device,
2151 						 u8 port_num, u16 index, u16 *pkey);
2152 	int		           (*modify_device)(struct ib_device *device,
2153 						    int device_modify_mask,
2154 						    struct ib_device_modify *device_modify);
2155 	int		           (*modify_port)(struct ib_device *device,
2156 						  u8 port_num, int port_modify_mask,
2157 						  struct ib_port_modify *port_modify);
2158 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
2159 						     struct ib_udata *udata);
2160 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
2161 	int                        (*mmap)(struct ib_ucontext *context,
2162 					   struct vm_area_struct *vma);
2163 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
2164 					       struct ib_ucontext *context,
2165 					       struct ib_udata *udata);
2166 	int                        (*dealloc_pd)(struct ib_pd *pd);
2167 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
2168 						struct rdma_ah_attr *ah_attr,
2169 						struct ib_udata *udata);
2170 	int                        (*modify_ah)(struct ib_ah *ah,
2171 						struct rdma_ah_attr *ah_attr);
2172 	int                        (*query_ah)(struct ib_ah *ah,
2173 					       struct rdma_ah_attr *ah_attr);
2174 	int                        (*destroy_ah)(struct ib_ah *ah);
2175 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
2176 						 struct ib_srq_init_attr *srq_init_attr,
2177 						 struct ib_udata *udata);
2178 	int                        (*modify_srq)(struct ib_srq *srq,
2179 						 struct ib_srq_attr *srq_attr,
2180 						 enum ib_srq_attr_mask srq_attr_mask,
2181 						 struct ib_udata *udata);
2182 	int                        (*query_srq)(struct ib_srq *srq,
2183 						struct ib_srq_attr *srq_attr);
2184 	int                        (*destroy_srq)(struct ib_srq *srq);
2185 	int                        (*post_srq_recv)(struct ib_srq *srq,
2186 						    struct ib_recv_wr *recv_wr,
2187 						    struct ib_recv_wr **bad_recv_wr);
2188 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
2189 						struct ib_qp_init_attr *qp_init_attr,
2190 						struct ib_udata *udata);
2191 	int                        (*modify_qp)(struct ib_qp *qp,
2192 						struct ib_qp_attr *qp_attr,
2193 						int qp_attr_mask,
2194 						struct ib_udata *udata);
2195 	int                        (*query_qp)(struct ib_qp *qp,
2196 					       struct ib_qp_attr *qp_attr,
2197 					       int qp_attr_mask,
2198 					       struct ib_qp_init_attr *qp_init_attr);
2199 	int                        (*destroy_qp)(struct ib_qp *qp);
2200 	int                        (*post_send)(struct ib_qp *qp,
2201 						struct ib_send_wr *send_wr,
2202 						struct ib_send_wr **bad_send_wr);
2203 	int                        (*post_recv)(struct ib_qp *qp,
2204 						struct ib_recv_wr *recv_wr,
2205 						struct ib_recv_wr **bad_recv_wr);
2206 	struct ib_cq *             (*create_cq)(struct ib_device *device,
2207 						const struct ib_cq_init_attr *attr,
2208 						struct ib_ucontext *context,
2209 						struct ib_udata *udata);
2210 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2211 						u16 cq_period);
2212 	int                        (*destroy_cq)(struct ib_cq *cq);
2213 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
2214 						struct ib_udata *udata);
2215 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
2216 					      struct ib_wc *wc);
2217 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2218 	int                        (*req_notify_cq)(struct ib_cq *cq,
2219 						    enum ib_cq_notify_flags flags);
2220 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
2221 						      int wc_cnt);
2222 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
2223 						 int mr_access_flags);
2224 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
2225 						  u64 start, u64 length,
2226 						  u64 virt_addr,
2227 						  int mr_access_flags,
2228 						  struct ib_udata *udata);
2229 	int			   (*rereg_user_mr)(struct ib_mr *mr,
2230 						    int flags,
2231 						    u64 start, u64 length,
2232 						    u64 virt_addr,
2233 						    int mr_access_flags,
2234 						    struct ib_pd *pd,
2235 						    struct ib_udata *udata);
2236 	int                        (*dereg_mr)(struct ib_mr *mr);
2237 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2238 					       enum ib_mr_type mr_type,
2239 					       u32 max_num_sg);
2240 	int                        (*map_mr_sg)(struct ib_mr *mr,
2241 						struct scatterlist *sg,
2242 						int sg_nents,
2243 						unsigned int *sg_offset);
2244 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2245 					       enum ib_mw_type type,
2246 					       struct ib_udata *udata);
2247 	int                        (*dealloc_mw)(struct ib_mw *mw);
2248 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2249 						int mr_access_flags,
2250 						struct ib_fmr_attr *fmr_attr);
2251 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2252 						   u64 *page_list, int list_len,
2253 						   u64 iova);
2254 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2255 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2256 	int                        (*attach_mcast)(struct ib_qp *qp,
2257 						   union ib_gid *gid,
2258 						   u16 lid);
2259 	int                        (*detach_mcast)(struct ib_qp *qp,
2260 						   union ib_gid *gid,
2261 						   u16 lid);
2262 	int                        (*process_mad)(struct ib_device *device,
2263 						  int process_mad_flags,
2264 						  u8 port_num,
2265 						  const struct ib_wc *in_wc,
2266 						  const struct ib_grh *in_grh,
2267 						  const struct ib_mad_hdr *in_mad,
2268 						  size_t in_mad_size,
2269 						  struct ib_mad_hdr *out_mad,
2270 						  size_t *out_mad_size,
2271 						  u16 *out_mad_pkey_index);
2272 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2273 						 struct ib_ucontext *ucontext,
2274 						 struct ib_udata *udata);
2275 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2276 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2277 						  struct ib_flow_attr
2278 						  *flow_attr,
2279 						  int domain);
2280 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2281 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2282 						      struct ib_mr_status *mr_status);
2283 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2284 	void			   (*drain_rq)(struct ib_qp *qp);
2285 	void			   (*drain_sq)(struct ib_qp *qp);
2286 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2287 							int state);
2288 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2289 						   struct ifla_vf_info *ivf);
2290 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2291 						   struct ifla_vf_stats *stats);
2292 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2293 						  int type);
2294 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2295 						struct ib_wq_init_attr *init_attr,
2296 						struct ib_udata *udata);
2297 	int			   (*destroy_wq)(struct ib_wq *wq);
2298 	int			   (*modify_wq)(struct ib_wq *wq,
2299 						struct ib_wq_attr *attr,
2300 						u32 wq_attr_mask,
2301 						struct ib_udata *udata);
2302 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2303 							   struct ib_rwq_ind_table_init_attr *init_attr,
2304 							   struct ib_udata *udata);
2305 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2306 	/**
2307 	 * rdma netdev operation
2308 	 *
2309 	 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2310 	 * doesn't support the specified rdma netdev type.
2311 	 */
2312 	struct net_device *(*alloc_rdma_netdev)(
2313 					struct ib_device *device,
2314 					u8 port_num,
2315 					enum rdma_netdev_t type,
2316 					const char *name,
2317 					unsigned char name_assign_type,
2318 					void (*setup)(struct net_device *));
2319 
2320 	struct module               *owner;
2321 	struct device                dev;
2322 	struct kobject               *ports_parent;
2323 	struct list_head             port_list;
2324 
2325 	enum {
2326 		IB_DEV_UNINITIALIZED,
2327 		IB_DEV_REGISTERED,
2328 		IB_DEV_UNREGISTERED
2329 	}                            reg_state;
2330 
2331 	int			     uverbs_abi_ver;
2332 	u64			     uverbs_cmd_mask;
2333 	u64			     uverbs_ex_cmd_mask;
2334 
2335 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2336 	__be64			     node_guid;
2337 	u32			     local_dma_lkey;
2338 	u16                          is_switch:1;
2339 	u8                           node_type;
2340 	u8                           phys_port_cnt;
2341 	struct ib_device_attr        attrs;
2342 	struct attribute_group	     *hw_stats_ag;
2343 	struct rdma_hw_stats         *hw_stats;
2344 
2345 #ifdef CONFIG_CGROUP_RDMA
2346 	struct rdmacg_device         cg_device;
2347 #endif
2348 
2349 	u32                          index;
2350 
2351 	/**
2352 	 * The following mandatory functions are used only at device
2353 	 * registration.  Keep functions such as these at the end of this
2354 	 * structure to avoid cache line misses when accessing struct ib_device
2355 	 * in fast paths.
2356 	 */
2357 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2358 	void (*get_dev_fw_str)(struct ib_device *, char *str);
2359 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2360 						     int comp_vector);
2361 
2362 	struct uverbs_root_spec		*specs_root;
2363 };
2364 
2365 struct ib_client {
2366 	char  *name;
2367 	void (*add)   (struct ib_device *);
2368 	void (*remove)(struct ib_device *, void *client_data);
2369 
2370 	/* Returns the net_dev belonging to this ib_client and matching the
2371 	 * given parameters.
2372 	 * @dev:	 An RDMA device that the net_dev use for communication.
2373 	 * @port:	 A physical port number on the RDMA device.
2374 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2375 	 * @gid:	 A GID that the net_dev uses to communicate.
2376 	 * @addr:	 An IP address the net_dev is configured with.
2377 	 * @client_data: The device's client data set by ib_set_client_data().
2378 	 *
2379 	 * An ib_client that implements a net_dev on top of RDMA devices
2380 	 * (such as IP over IB) should implement this callback, allowing the
2381 	 * rdma_cm module to find the right net_dev for a given request.
2382 	 *
2383 	 * The caller is responsible for calling dev_put on the returned
2384 	 * netdev. */
2385 	struct net_device *(*get_net_dev_by_params)(
2386 			struct ib_device *dev,
2387 			u8 port,
2388 			u16 pkey,
2389 			const union ib_gid *gid,
2390 			const struct sockaddr *addr,
2391 			void *client_data);
2392 	struct list_head list;
2393 };
2394 
2395 struct ib_device *ib_alloc_device(size_t size);
2396 void ib_dealloc_device(struct ib_device *device);
2397 
2398 void ib_get_device_fw_str(struct ib_device *device, char *str);
2399 
2400 int ib_register_device(struct ib_device *device,
2401 		       int (*port_callback)(struct ib_device *,
2402 					    u8, struct kobject *));
2403 void ib_unregister_device(struct ib_device *device);
2404 
2405 int ib_register_client   (struct ib_client *client);
2406 void ib_unregister_client(struct ib_client *client);
2407 
2408 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2409 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2410 			 void *data);
2411 
ib_copy_from_udata(void * dest,struct ib_udata * udata,size_t len)2412 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2413 {
2414 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2415 }
2416 
ib_copy_to_udata(struct ib_udata * udata,void * src,size_t len)2417 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2418 {
2419 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2420 }
2421 
ib_is_udata_cleared(struct ib_udata * udata,size_t offset,size_t len)2422 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2423 				       size_t offset,
2424 				       size_t len)
2425 {
2426 	const void __user *p = udata->inbuf + offset;
2427 	bool ret;
2428 	u8 *buf;
2429 
2430 	if (len > USHRT_MAX)
2431 		return false;
2432 
2433 	buf = memdup_user(p, len);
2434 	if (IS_ERR(buf))
2435 		return false;
2436 
2437 	ret = !memchr_inv(buf, 0, len);
2438 	kfree(buf);
2439 	return ret;
2440 }
2441 
2442 /**
2443  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2444  * contains all required attributes and no attributes not allowed for
2445  * the given QP state transition.
2446  * @cur_state: Current QP state
2447  * @next_state: Next QP state
2448  * @type: QP type
2449  * @mask: Mask of supplied QP attributes
2450  * @ll : link layer of port
2451  *
2452  * This function is a helper function that a low-level driver's
2453  * modify_qp method can use to validate the consumer's input.  It
2454  * checks that cur_state and next_state are valid QP states, that a
2455  * transition from cur_state to next_state is allowed by the IB spec,
2456  * and that the attribute mask supplied is allowed for the transition.
2457  */
2458 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2459 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2460 		       enum rdma_link_layer ll);
2461 
2462 void ib_register_event_handler(struct ib_event_handler *event_handler);
2463 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2464 void ib_dispatch_event(struct ib_event *event);
2465 
2466 int ib_query_port(struct ib_device *device,
2467 		  u8 port_num, struct ib_port_attr *port_attr);
2468 
2469 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2470 					       u8 port_num);
2471 
2472 /**
2473  * rdma_cap_ib_switch - Check if the device is IB switch
2474  * @device: Device to check
2475  *
2476  * Device driver is responsible for setting is_switch bit on
2477  * in ib_device structure at init time.
2478  *
2479  * Return: true if the device is IB switch.
2480  */
rdma_cap_ib_switch(const struct ib_device * device)2481 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2482 {
2483 	return device->is_switch;
2484 }
2485 
2486 /**
2487  * rdma_start_port - Return the first valid port number for the device
2488  * specified
2489  *
2490  * @device: Device to be checked
2491  *
2492  * Return start port number
2493  */
rdma_start_port(const struct ib_device * device)2494 static inline u8 rdma_start_port(const struct ib_device *device)
2495 {
2496 	return rdma_cap_ib_switch(device) ? 0 : 1;
2497 }
2498 
2499 /**
2500  * rdma_end_port - Return the last valid port number for the device
2501  * specified
2502  *
2503  * @device: Device to be checked
2504  *
2505  * Return last port number
2506  */
rdma_end_port(const struct ib_device * device)2507 static inline u8 rdma_end_port(const struct ib_device *device)
2508 {
2509 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2510 }
2511 
rdma_is_port_valid(const struct ib_device * device,unsigned int port)2512 static inline int rdma_is_port_valid(const struct ib_device *device,
2513 				     unsigned int port)
2514 {
2515 	return (port >= rdma_start_port(device) &&
2516 		port <= rdma_end_port(device));
2517 }
2518 
rdma_protocol_ib(const struct ib_device * device,u8 port_num)2519 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2520 {
2521 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2522 }
2523 
rdma_protocol_roce(const struct ib_device * device,u8 port_num)2524 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2525 {
2526 	return device->port_immutable[port_num].core_cap_flags &
2527 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2528 }
2529 
rdma_protocol_roce_udp_encap(const struct ib_device * device,u8 port_num)2530 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2531 {
2532 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2533 }
2534 
rdma_protocol_roce_eth_encap(const struct ib_device * device,u8 port_num)2535 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2536 {
2537 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2538 }
2539 
rdma_protocol_iwarp(const struct ib_device * device,u8 port_num)2540 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2541 {
2542 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2543 }
2544 
rdma_ib_or_roce(const struct ib_device * device,u8 port_num)2545 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2546 {
2547 	return rdma_protocol_ib(device, port_num) ||
2548 		rdma_protocol_roce(device, port_num);
2549 }
2550 
rdma_protocol_raw_packet(const struct ib_device * device,u8 port_num)2551 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2552 {
2553 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2554 }
2555 
rdma_protocol_usnic(const struct ib_device * device,u8 port_num)2556 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2557 {
2558 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2559 }
2560 
2561 /**
2562  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2563  * Management Datagrams.
2564  * @device: Device to check
2565  * @port_num: Port number to check
2566  *
2567  * Management Datagrams (MAD) are a required part of the InfiniBand
2568  * specification and are supported on all InfiniBand devices.  A slightly
2569  * extended version are also supported on OPA interfaces.
2570  *
2571  * Return: true if the port supports sending/receiving of MAD packets.
2572  */
rdma_cap_ib_mad(const struct ib_device * device,u8 port_num)2573 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2574 {
2575 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2576 }
2577 
2578 /**
2579  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2580  * Management Datagrams.
2581  * @device: Device to check
2582  * @port_num: Port number to check
2583  *
2584  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2585  * datagrams with their own versions.  These OPA MADs share many but not all of
2586  * the characteristics of InfiniBand MADs.
2587  *
2588  * OPA MADs differ in the following ways:
2589  *
2590  *    1) MADs are variable size up to 2K
2591  *       IBTA defined MADs remain fixed at 256 bytes
2592  *    2) OPA SMPs must carry valid PKeys
2593  *    3) OPA SMP packets are a different format
2594  *
2595  * Return: true if the port supports OPA MAD packet formats.
2596  */
rdma_cap_opa_mad(struct ib_device * device,u8 port_num)2597 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2598 {
2599 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2600 		== RDMA_CORE_CAP_OPA_MAD;
2601 }
2602 
2603 /**
2604  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2605  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2606  * @device: Device to check
2607  * @port_num: Port number to check
2608  *
2609  * Each InfiniBand node is required to provide a Subnet Management Agent
2610  * that the subnet manager can access.  Prior to the fabric being fully
2611  * configured by the subnet manager, the SMA is accessed via a well known
2612  * interface called the Subnet Management Interface (SMI).  This interface
2613  * uses directed route packets to communicate with the SM to get around the
2614  * chicken and egg problem of the SM needing to know what's on the fabric
2615  * in order to configure the fabric, and needing to configure the fabric in
2616  * order to send packets to the devices on the fabric.  These directed
2617  * route packets do not need the fabric fully configured in order to reach
2618  * their destination.  The SMI is the only method allowed to send
2619  * directed route packets on an InfiniBand fabric.
2620  *
2621  * Return: true if the port provides an SMI.
2622  */
rdma_cap_ib_smi(const struct ib_device * device,u8 port_num)2623 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2624 {
2625 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2626 }
2627 
2628 /**
2629  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2630  * Communication Manager.
2631  * @device: Device to check
2632  * @port_num: Port number to check
2633  *
2634  * The InfiniBand Communication Manager is one of many pre-defined General
2635  * Service Agents (GSA) that are accessed via the General Service
2636  * Interface (GSI).  It's role is to facilitate establishment of connections
2637  * between nodes as well as other management related tasks for established
2638  * connections.
2639  *
2640  * Return: true if the port supports an IB CM (this does not guarantee that
2641  * a CM is actually running however).
2642  */
rdma_cap_ib_cm(const struct ib_device * device,u8 port_num)2643 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2644 {
2645 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2646 }
2647 
2648 /**
2649  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2650  * Communication Manager.
2651  * @device: Device to check
2652  * @port_num: Port number to check
2653  *
2654  * Similar to above, but specific to iWARP connections which have a different
2655  * managment protocol than InfiniBand.
2656  *
2657  * Return: true if the port supports an iWARP CM (this does not guarantee that
2658  * a CM is actually running however).
2659  */
rdma_cap_iw_cm(const struct ib_device * device,u8 port_num)2660 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2661 {
2662 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2663 }
2664 
2665 /**
2666  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2667  * Subnet Administration.
2668  * @device: Device to check
2669  * @port_num: Port number to check
2670  *
2671  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2672  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2673  * fabrics, devices should resolve routes to other hosts by contacting the
2674  * SA to query the proper route.
2675  *
2676  * Return: true if the port should act as a client to the fabric Subnet
2677  * Administration interface.  This does not imply that the SA service is
2678  * running locally.
2679  */
rdma_cap_ib_sa(const struct ib_device * device,u8 port_num)2680 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2681 {
2682 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2683 }
2684 
2685 /**
2686  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2687  * Multicast.
2688  * @device: Device to check
2689  * @port_num: Port number to check
2690  *
2691  * InfiniBand multicast registration is more complex than normal IPv4 or
2692  * IPv6 multicast registration.  Each Host Channel Adapter must register
2693  * with the Subnet Manager when it wishes to join a multicast group.  It
2694  * should do so only once regardless of how many queue pairs it subscribes
2695  * to this group.  And it should leave the group only after all queue pairs
2696  * attached to the group have been detached.
2697  *
2698  * Return: true if the port must undertake the additional adminstrative
2699  * overhead of registering/unregistering with the SM and tracking of the
2700  * total number of queue pairs attached to the multicast group.
2701  */
rdma_cap_ib_mcast(const struct ib_device * device,u8 port_num)2702 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2703 {
2704 	return rdma_cap_ib_sa(device, port_num);
2705 }
2706 
2707 /**
2708  * rdma_cap_af_ib - Check if the port of device has the capability
2709  * Native Infiniband Address.
2710  * @device: Device to check
2711  * @port_num: Port number to check
2712  *
2713  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2714  * GID.  RoCE uses a different mechanism, but still generates a GID via
2715  * a prescribed mechanism and port specific data.
2716  *
2717  * Return: true if the port uses a GID address to identify devices on the
2718  * network.
2719  */
rdma_cap_af_ib(const struct ib_device * device,u8 port_num)2720 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2721 {
2722 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2723 }
2724 
2725 /**
2726  * rdma_cap_eth_ah - Check if the port of device has the capability
2727  * Ethernet Address Handle.
2728  * @device: Device to check
2729  * @port_num: Port number to check
2730  *
2731  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2732  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2733  * port.  Normally, packet headers are generated by the sending host
2734  * adapter, but when sending connectionless datagrams, we must manually
2735  * inject the proper headers for the fabric we are communicating over.
2736  *
2737  * Return: true if we are running as a RoCE port and must force the
2738  * addition of a Global Route Header built from our Ethernet Address
2739  * Handle into our header list for connectionless packets.
2740  */
rdma_cap_eth_ah(const struct ib_device * device,u8 port_num)2741 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2742 {
2743 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2744 }
2745 
2746 /**
2747  * rdma_cap_opa_ah - Check if the port of device supports
2748  * OPA Address handles
2749  * @device: Device to check
2750  * @port_num: Port number to check
2751  *
2752  * Return: true if we are running on an OPA device which supports
2753  * the extended OPA addressing.
2754  */
rdma_cap_opa_ah(struct ib_device * device,u8 port_num)2755 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
2756 {
2757 	return (device->port_immutable[port_num].core_cap_flags &
2758 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
2759 }
2760 
2761 /**
2762  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2763  *
2764  * @device: Device
2765  * @port_num: Port number
2766  *
2767  * This MAD size includes the MAD headers and MAD payload.  No other headers
2768  * are included.
2769  *
2770  * Return the max MAD size required by the Port.  Will return 0 if the port
2771  * does not support MADs
2772  */
rdma_max_mad_size(const struct ib_device * device,u8 port_num)2773 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2774 {
2775 	return device->port_immutable[port_num].max_mad_size;
2776 }
2777 
2778 /**
2779  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2780  * @device: Device to check
2781  * @port_num: Port number to check
2782  *
2783  * RoCE GID table mechanism manages the various GIDs for a device.
2784  *
2785  * NOTE: if allocating the port's GID table has failed, this call will still
2786  * return true, but any RoCE GID table API will fail.
2787  *
2788  * Return: true if the port uses RoCE GID table mechanism in order to manage
2789  * its GIDs.
2790  */
rdma_cap_roce_gid_table(const struct ib_device * device,u8 port_num)2791 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2792 					   u8 port_num)
2793 {
2794 	return rdma_protocol_roce(device, port_num) &&
2795 		device->add_gid && device->del_gid;
2796 }
2797 
2798 /*
2799  * Check if the device supports READ W/ INVALIDATE.
2800  */
rdma_cap_read_inv(struct ib_device * dev,u32 port_num)2801 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2802 {
2803 	/*
2804 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2805 	 * has support for it yet.
2806 	 */
2807 	return rdma_protocol_iwarp(dev, port_num);
2808 }
2809 
2810 int ib_query_gid(struct ib_device *device,
2811 		 u8 port_num, int index, union ib_gid *gid,
2812 		 struct ib_gid_attr *attr);
2813 
2814 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2815 			 int state);
2816 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2817 		     struct ifla_vf_info *info);
2818 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2819 		    struct ifla_vf_stats *stats);
2820 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2821 		   int type);
2822 
2823 int ib_query_pkey(struct ib_device *device,
2824 		  u8 port_num, u16 index, u16 *pkey);
2825 
2826 int ib_modify_device(struct ib_device *device,
2827 		     int device_modify_mask,
2828 		     struct ib_device_modify *device_modify);
2829 
2830 int ib_modify_port(struct ib_device *device,
2831 		   u8 port_num, int port_modify_mask,
2832 		   struct ib_port_modify *port_modify);
2833 
2834 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2835 		enum ib_gid_type gid_type, struct net_device *ndev,
2836 		u8 *port_num, u16 *index);
2837 
2838 int ib_find_pkey(struct ib_device *device,
2839 		 u8 port_num, u16 pkey, u16 *index);
2840 
2841 enum ib_pd_flags {
2842 	/*
2843 	 * Create a memory registration for all memory in the system and place
2844 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2845 	 * ULPs to avoid the overhead of dynamic MRs.
2846 	 *
2847 	 * This flag is generally considered unsafe and must only be used in
2848 	 * extremly trusted environments.  Every use of it will log a warning
2849 	 * in the kernel log.
2850 	 */
2851 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2852 };
2853 
2854 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2855 		const char *caller);
2856 #define ib_alloc_pd(device, flags) \
2857 	__ib_alloc_pd((device), (flags), __func__)
2858 void ib_dealloc_pd(struct ib_pd *pd);
2859 
2860 /**
2861  * rdma_create_ah - Creates an address handle for the given address vector.
2862  * @pd: The protection domain associated with the address handle.
2863  * @ah_attr: The attributes of the address vector.
2864  *
2865  * The address handle is used to reference a local or global destination
2866  * in all UD QP post sends.
2867  */
2868 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
2869 
2870 /**
2871  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2872  *   work completion.
2873  * @hdr: the L3 header to parse
2874  * @net_type: type of header to parse
2875  * @sgid: place to store source gid
2876  * @dgid: place to store destination gid
2877  */
2878 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2879 			      enum rdma_network_type net_type,
2880 			      union ib_gid *sgid, union ib_gid *dgid);
2881 
2882 /**
2883  * ib_get_rdma_header_version - Get the header version
2884  * @hdr: the L3 header to parse
2885  */
2886 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2887 
2888 /**
2889  * ib_init_ah_from_wc - Initializes address handle attributes from a
2890  *   work completion.
2891  * @device: Device on which the received message arrived.
2892  * @port_num: Port on which the received message arrived.
2893  * @wc: Work completion associated with the received message.
2894  * @grh: References the received global route header.  This parameter is
2895  *   ignored unless the work completion indicates that the GRH is valid.
2896  * @ah_attr: Returned attributes that can be used when creating an address
2897  *   handle for replying to the message.
2898  */
2899 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2900 		       const struct ib_wc *wc, const struct ib_grh *grh,
2901 		       struct rdma_ah_attr *ah_attr);
2902 
2903 /**
2904  * ib_create_ah_from_wc - Creates an address handle associated with the
2905  *   sender of the specified work completion.
2906  * @pd: The protection domain associated with the address handle.
2907  * @wc: Work completion information associated with a received message.
2908  * @grh: References the received global route header.  This parameter is
2909  *   ignored unless the work completion indicates that the GRH is valid.
2910  * @port_num: The outbound port number to associate with the address.
2911  *
2912  * The address handle is used to reference a local or global destination
2913  * in all UD QP post sends.
2914  */
2915 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2916 				   const struct ib_grh *grh, u8 port_num);
2917 
2918 /**
2919  * rdma_modify_ah - Modifies the address vector associated with an address
2920  *   handle.
2921  * @ah: The address handle to modify.
2922  * @ah_attr: The new address vector attributes to associate with the
2923  *   address handle.
2924  */
2925 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2926 
2927 /**
2928  * rdma_query_ah - Queries the address vector associated with an address
2929  *   handle.
2930  * @ah: The address handle to query.
2931  * @ah_attr: The address vector attributes associated with the address
2932  *   handle.
2933  */
2934 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2935 
2936 /**
2937  * rdma_destroy_ah - Destroys an address handle.
2938  * @ah: The address handle to destroy.
2939  */
2940 int rdma_destroy_ah(struct ib_ah *ah);
2941 
2942 /**
2943  * ib_create_srq - Creates a SRQ associated with the specified protection
2944  *   domain.
2945  * @pd: The protection domain associated with the SRQ.
2946  * @srq_init_attr: A list of initial attributes required to create the
2947  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2948  *   the actual capabilities of the created SRQ.
2949  *
2950  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2951  * requested size of the SRQ, and set to the actual values allocated
2952  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2953  * will always be at least as large as the requested values.
2954  */
2955 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2956 			     struct ib_srq_init_attr *srq_init_attr);
2957 
2958 /**
2959  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2960  * @srq: The SRQ to modify.
2961  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2962  *   the current values of selected SRQ attributes are returned.
2963  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2964  *   are being modified.
2965  *
2966  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2967  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2968  * the number of receives queued drops below the limit.
2969  */
2970 int ib_modify_srq(struct ib_srq *srq,
2971 		  struct ib_srq_attr *srq_attr,
2972 		  enum ib_srq_attr_mask srq_attr_mask);
2973 
2974 /**
2975  * ib_query_srq - Returns the attribute list and current values for the
2976  *   specified SRQ.
2977  * @srq: The SRQ to query.
2978  * @srq_attr: The attributes of the specified SRQ.
2979  */
2980 int ib_query_srq(struct ib_srq *srq,
2981 		 struct ib_srq_attr *srq_attr);
2982 
2983 /**
2984  * ib_destroy_srq - Destroys the specified SRQ.
2985  * @srq: The SRQ to destroy.
2986  */
2987 int ib_destroy_srq(struct ib_srq *srq);
2988 
2989 /**
2990  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2991  * @srq: The SRQ to post the work request on.
2992  * @recv_wr: A list of work requests to post on the receive queue.
2993  * @bad_recv_wr: On an immediate failure, this parameter will reference
2994  *   the work request that failed to be posted on the QP.
2995  */
ib_post_srq_recv(struct ib_srq * srq,struct ib_recv_wr * recv_wr,struct ib_recv_wr ** bad_recv_wr)2996 static inline int ib_post_srq_recv(struct ib_srq *srq,
2997 				   struct ib_recv_wr *recv_wr,
2998 				   struct ib_recv_wr **bad_recv_wr)
2999 {
3000 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
3001 }
3002 
3003 /**
3004  * ib_create_qp - Creates a QP associated with the specified protection
3005  *   domain.
3006  * @pd: The protection domain associated with the QP.
3007  * @qp_init_attr: A list of initial attributes required to create the
3008  *   QP.  If QP creation succeeds, then the attributes are updated to
3009  *   the actual capabilities of the created QP.
3010  */
3011 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3012 			   struct ib_qp_init_attr *qp_init_attr);
3013 
3014 /**
3015  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3016  * @qp: The QP to modify.
3017  * @attr: On input, specifies the QP attributes to modify.  On output,
3018  *   the current values of selected QP attributes are returned.
3019  * @attr_mask: A bit-mask used to specify which attributes of the QP
3020  *   are being modified.
3021  * @udata: pointer to user's input output buffer information
3022  *   are being modified.
3023  * It returns 0 on success and returns appropriate error code on error.
3024  */
3025 int ib_modify_qp_with_udata(struct ib_qp *qp,
3026 			    struct ib_qp_attr *attr,
3027 			    int attr_mask,
3028 			    struct ib_udata *udata);
3029 
3030 /**
3031  * ib_modify_qp - Modifies the attributes for the specified QP and then
3032  *   transitions the QP to the given state.
3033  * @qp: The QP to modify.
3034  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3035  *   the current values of selected QP attributes are returned.
3036  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3037  *   are being modified.
3038  */
3039 int ib_modify_qp(struct ib_qp *qp,
3040 		 struct ib_qp_attr *qp_attr,
3041 		 int qp_attr_mask);
3042 
3043 /**
3044  * ib_query_qp - Returns the attribute list and current values for the
3045  *   specified QP.
3046  * @qp: The QP to query.
3047  * @qp_attr: The attributes of the specified QP.
3048  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3049  * @qp_init_attr: Additional attributes of the selected QP.
3050  *
3051  * The qp_attr_mask may be used to limit the query to gathering only the
3052  * selected attributes.
3053  */
3054 int ib_query_qp(struct ib_qp *qp,
3055 		struct ib_qp_attr *qp_attr,
3056 		int qp_attr_mask,
3057 		struct ib_qp_init_attr *qp_init_attr);
3058 
3059 /**
3060  * ib_destroy_qp - Destroys the specified QP.
3061  * @qp: The QP to destroy.
3062  */
3063 int ib_destroy_qp(struct ib_qp *qp);
3064 
3065 /**
3066  * ib_open_qp - Obtain a reference to an existing sharable QP.
3067  * @xrcd - XRC domain
3068  * @qp_open_attr: Attributes identifying the QP to open.
3069  *
3070  * Returns a reference to a sharable QP.
3071  */
3072 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3073 			 struct ib_qp_open_attr *qp_open_attr);
3074 
3075 /**
3076  * ib_close_qp - Release an external reference to a QP.
3077  * @qp: The QP handle to release
3078  *
3079  * The opened QP handle is released by the caller.  The underlying
3080  * shared QP is not destroyed until all internal references are released.
3081  */
3082 int ib_close_qp(struct ib_qp *qp);
3083 
3084 /**
3085  * ib_post_send - Posts a list of work requests to the send queue of
3086  *   the specified QP.
3087  * @qp: The QP to post the work request on.
3088  * @send_wr: A list of work requests to post on the send queue.
3089  * @bad_send_wr: On an immediate failure, this parameter will reference
3090  *   the work request that failed to be posted on the QP.
3091  *
3092  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3093  * error is returned, the QP state shall not be affected,
3094  * ib_post_send() will return an immediate error after queueing any
3095  * earlier work requests in the list.
3096  */
ib_post_send(struct ib_qp * qp,struct ib_send_wr * send_wr,struct ib_send_wr ** bad_send_wr)3097 static inline int ib_post_send(struct ib_qp *qp,
3098 			       struct ib_send_wr *send_wr,
3099 			       struct ib_send_wr **bad_send_wr)
3100 {
3101 	return qp->device->post_send(qp, send_wr, bad_send_wr);
3102 }
3103 
3104 /**
3105  * ib_post_recv - Posts a list of work requests to the receive queue of
3106  *   the specified QP.
3107  * @qp: The QP to post the work request on.
3108  * @recv_wr: A list of work requests to post on the receive queue.
3109  * @bad_recv_wr: On an immediate failure, this parameter will reference
3110  *   the work request that failed to be posted on the QP.
3111  */
ib_post_recv(struct ib_qp * qp,struct ib_recv_wr * recv_wr,struct ib_recv_wr ** bad_recv_wr)3112 static inline int ib_post_recv(struct ib_qp *qp,
3113 			       struct ib_recv_wr *recv_wr,
3114 			       struct ib_recv_wr **bad_recv_wr)
3115 {
3116 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
3117 }
3118 
3119 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3120 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
3121 void ib_free_cq(struct ib_cq *cq);
3122 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3123 
3124 /**
3125  * ib_create_cq - Creates a CQ on the specified device.
3126  * @device: The device on which to create the CQ.
3127  * @comp_handler: A user-specified callback that is invoked when a
3128  *   completion event occurs on the CQ.
3129  * @event_handler: A user-specified callback that is invoked when an
3130  *   asynchronous event not associated with a completion occurs on the CQ.
3131  * @cq_context: Context associated with the CQ returned to the user via
3132  *   the associated completion and event handlers.
3133  * @cq_attr: The attributes the CQ should be created upon.
3134  *
3135  * Users can examine the cq structure to determine the actual CQ size.
3136  */
3137 struct ib_cq *ib_create_cq(struct ib_device *device,
3138 			   ib_comp_handler comp_handler,
3139 			   void (*event_handler)(struct ib_event *, void *),
3140 			   void *cq_context,
3141 			   const struct ib_cq_init_attr *cq_attr);
3142 
3143 /**
3144  * ib_resize_cq - Modifies the capacity of the CQ.
3145  * @cq: The CQ to resize.
3146  * @cqe: The minimum size of the CQ.
3147  *
3148  * Users can examine the cq structure to determine the actual CQ size.
3149  */
3150 int ib_resize_cq(struct ib_cq *cq, int cqe);
3151 
3152 /**
3153  * ib_modify_cq - Modifies moderation params of the CQ
3154  * @cq: The CQ to modify.
3155  * @cq_count: number of CQEs that will trigger an event
3156  * @cq_period: max period of time in usec before triggering an event
3157  *
3158  */
3159 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3160 
3161 /**
3162  * ib_destroy_cq - Destroys the specified CQ.
3163  * @cq: The CQ to destroy.
3164  */
3165 int ib_destroy_cq(struct ib_cq *cq);
3166 
3167 /**
3168  * ib_poll_cq - poll a CQ for completion(s)
3169  * @cq:the CQ being polled
3170  * @num_entries:maximum number of completions to return
3171  * @wc:array of at least @num_entries &struct ib_wc where completions
3172  *   will be returned
3173  *
3174  * Poll a CQ for (possibly multiple) completions.  If the return value
3175  * is < 0, an error occurred.  If the return value is >= 0, it is the
3176  * number of completions returned.  If the return value is
3177  * non-negative and < num_entries, then the CQ was emptied.
3178  */
ib_poll_cq(struct ib_cq * cq,int num_entries,struct ib_wc * wc)3179 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3180 			     struct ib_wc *wc)
3181 {
3182 	return cq->device->poll_cq(cq, num_entries, wc);
3183 }
3184 
3185 /**
3186  * ib_peek_cq - Returns the number of unreaped completions currently
3187  *   on the specified CQ.
3188  * @cq: The CQ to peek.
3189  * @wc_cnt: A minimum number of unreaped completions to check for.
3190  *
3191  * If the number of unreaped completions is greater than or equal to wc_cnt,
3192  * this function returns wc_cnt, otherwise, it returns the actual number of
3193  * unreaped completions.
3194  */
3195 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
3196 
3197 /**
3198  * ib_req_notify_cq - Request completion notification on a CQ.
3199  * @cq: The CQ to generate an event for.
3200  * @flags:
3201  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3202  *   to request an event on the next solicited event or next work
3203  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3204  *   may also be |ed in to request a hint about missed events, as
3205  *   described below.
3206  *
3207  * Return Value:
3208  *    < 0 means an error occurred while requesting notification
3209  *   == 0 means notification was requested successfully, and if
3210  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3211  *        were missed and it is safe to wait for another event.  In
3212  *        this case is it guaranteed that any work completions added
3213  *        to the CQ since the last CQ poll will trigger a completion
3214  *        notification event.
3215  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3216  *        in.  It means that the consumer must poll the CQ again to
3217  *        make sure it is empty to avoid missing an event because of a
3218  *        race between requesting notification and an entry being
3219  *        added to the CQ.  This return value means it is possible
3220  *        (but not guaranteed) that a work completion has been added
3221  *        to the CQ since the last poll without triggering a
3222  *        completion notification event.
3223  */
ib_req_notify_cq(struct ib_cq * cq,enum ib_cq_notify_flags flags)3224 static inline int ib_req_notify_cq(struct ib_cq *cq,
3225 				   enum ib_cq_notify_flags flags)
3226 {
3227 	return cq->device->req_notify_cq(cq, flags);
3228 }
3229 
3230 /**
3231  * ib_req_ncomp_notif - Request completion notification when there are
3232  *   at least the specified number of unreaped completions on the CQ.
3233  * @cq: The CQ to generate an event for.
3234  * @wc_cnt: The number of unreaped completions that should be on the
3235  *   CQ before an event is generated.
3236  */
ib_req_ncomp_notif(struct ib_cq * cq,int wc_cnt)3237 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3238 {
3239 	return cq->device->req_ncomp_notif ?
3240 		cq->device->req_ncomp_notif(cq, wc_cnt) :
3241 		-ENOSYS;
3242 }
3243 
3244 /**
3245  * ib_dma_mapping_error - check a DMA addr for error
3246  * @dev: The device for which the dma_addr was created
3247  * @dma_addr: The DMA address to check
3248  */
ib_dma_mapping_error(struct ib_device * dev,u64 dma_addr)3249 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3250 {
3251 	return dma_mapping_error(dev->dma_device, dma_addr);
3252 }
3253 
3254 /**
3255  * ib_dma_map_single - Map a kernel virtual address to DMA address
3256  * @dev: The device for which the dma_addr is to be created
3257  * @cpu_addr: The kernel virtual address
3258  * @size: The size of the region in bytes
3259  * @direction: The direction of the DMA
3260  */
ib_dma_map_single(struct ib_device * dev,void * cpu_addr,size_t size,enum dma_data_direction direction)3261 static inline u64 ib_dma_map_single(struct ib_device *dev,
3262 				    void *cpu_addr, size_t size,
3263 				    enum dma_data_direction direction)
3264 {
3265 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3266 }
3267 
3268 /**
3269  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3270  * @dev: The device for which the DMA address was created
3271  * @addr: The DMA address
3272  * @size: The size of the region in bytes
3273  * @direction: The direction of the DMA
3274  */
ib_dma_unmap_single(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction direction)3275 static inline void ib_dma_unmap_single(struct ib_device *dev,
3276 				       u64 addr, size_t size,
3277 				       enum dma_data_direction direction)
3278 {
3279 	dma_unmap_single(dev->dma_device, addr, size, direction);
3280 }
3281 
3282 /**
3283  * ib_dma_map_page - Map a physical page to DMA address
3284  * @dev: The device for which the dma_addr is to be created
3285  * @page: The page to be mapped
3286  * @offset: The offset within the page
3287  * @size: The size of the region in bytes
3288  * @direction: The direction of the DMA
3289  */
ib_dma_map_page(struct ib_device * dev,struct page * page,unsigned long offset,size_t size,enum dma_data_direction direction)3290 static inline u64 ib_dma_map_page(struct ib_device *dev,
3291 				  struct page *page,
3292 				  unsigned long offset,
3293 				  size_t size,
3294 					 enum dma_data_direction direction)
3295 {
3296 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3297 }
3298 
3299 /**
3300  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3301  * @dev: The device for which the DMA address was created
3302  * @addr: The DMA address
3303  * @size: The size of the region in bytes
3304  * @direction: The direction of the DMA
3305  */
ib_dma_unmap_page(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction direction)3306 static inline void ib_dma_unmap_page(struct ib_device *dev,
3307 				     u64 addr, size_t size,
3308 				     enum dma_data_direction direction)
3309 {
3310 	dma_unmap_page(dev->dma_device, addr, size, direction);
3311 }
3312 
3313 /**
3314  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3315  * @dev: The device for which the DMA addresses are to be created
3316  * @sg: The array of scatter/gather entries
3317  * @nents: The number of scatter/gather entries
3318  * @direction: The direction of the DMA
3319  */
ib_dma_map_sg(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction)3320 static inline int ib_dma_map_sg(struct ib_device *dev,
3321 				struct scatterlist *sg, int nents,
3322 				enum dma_data_direction direction)
3323 {
3324 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3325 }
3326 
3327 /**
3328  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3329  * @dev: The device for which the DMA addresses were created
3330  * @sg: The array of scatter/gather entries
3331  * @nents: The number of scatter/gather entries
3332  * @direction: The direction of the DMA
3333  */
ib_dma_unmap_sg(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction)3334 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3335 				   struct scatterlist *sg, int nents,
3336 				   enum dma_data_direction direction)
3337 {
3338 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
3339 }
3340 
ib_dma_map_sg_attrs(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction,unsigned long dma_attrs)3341 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3342 				      struct scatterlist *sg, int nents,
3343 				      enum dma_data_direction direction,
3344 				      unsigned long dma_attrs)
3345 {
3346 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3347 				dma_attrs);
3348 }
3349 
ib_dma_unmap_sg_attrs(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction,unsigned long dma_attrs)3350 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3351 					 struct scatterlist *sg, int nents,
3352 					 enum dma_data_direction direction,
3353 					 unsigned long dma_attrs)
3354 {
3355 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3356 }
3357 /**
3358  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3359  * @dev: The device for which the DMA addresses were created
3360  * @sg: The scatter/gather entry
3361  *
3362  * Note: this function is obsolete. To do: change all occurrences of
3363  * ib_sg_dma_address() into sg_dma_address().
3364  */
ib_sg_dma_address(struct ib_device * dev,struct scatterlist * sg)3365 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3366 				    struct scatterlist *sg)
3367 {
3368 	return sg_dma_address(sg);
3369 }
3370 
3371 /**
3372  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3373  * @dev: The device for which the DMA addresses were created
3374  * @sg: The scatter/gather entry
3375  *
3376  * Note: this function is obsolete. To do: change all occurrences of
3377  * ib_sg_dma_len() into sg_dma_len().
3378  */
ib_sg_dma_len(struct ib_device * dev,struct scatterlist * sg)3379 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3380 					 struct scatterlist *sg)
3381 {
3382 	return sg_dma_len(sg);
3383 }
3384 
3385 /**
3386  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3387  * @dev: The device for which the DMA address was created
3388  * @addr: The DMA address
3389  * @size: The size of the region in bytes
3390  * @dir: The direction of the DMA
3391  */
ib_dma_sync_single_for_cpu(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction dir)3392 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3393 					      u64 addr,
3394 					      size_t size,
3395 					      enum dma_data_direction dir)
3396 {
3397 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3398 }
3399 
3400 /**
3401  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3402  * @dev: The device for which the DMA address was created
3403  * @addr: The DMA address
3404  * @size: The size of the region in bytes
3405  * @dir: The direction of the DMA
3406  */
ib_dma_sync_single_for_device(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction dir)3407 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3408 						 u64 addr,
3409 						 size_t size,
3410 						 enum dma_data_direction dir)
3411 {
3412 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3413 }
3414 
3415 /**
3416  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3417  * @dev: The device for which the DMA address is requested
3418  * @size: The size of the region to allocate in bytes
3419  * @dma_handle: A pointer for returning the DMA address of the region
3420  * @flag: memory allocator flags
3421  */
ib_dma_alloc_coherent(struct ib_device * dev,size_t size,dma_addr_t * dma_handle,gfp_t flag)3422 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3423 					   size_t size,
3424 					   dma_addr_t *dma_handle,
3425 					   gfp_t flag)
3426 {
3427 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3428 }
3429 
3430 /**
3431  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3432  * @dev: The device for which the DMA addresses were allocated
3433  * @size: The size of the region
3434  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3435  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3436  */
ib_dma_free_coherent(struct ib_device * dev,size_t size,void * cpu_addr,dma_addr_t dma_handle)3437 static inline void ib_dma_free_coherent(struct ib_device *dev,
3438 					size_t size, void *cpu_addr,
3439 					dma_addr_t dma_handle)
3440 {
3441 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3442 }
3443 
3444 /**
3445  * ib_dereg_mr - Deregisters a memory region and removes it from the
3446  *   HCA translation table.
3447  * @mr: The memory region to deregister.
3448  *
3449  * This function can fail, if the memory region has memory windows bound to it.
3450  */
3451 int ib_dereg_mr(struct ib_mr *mr);
3452 
3453 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3454 			  enum ib_mr_type mr_type,
3455 			  u32 max_num_sg);
3456 
3457 /**
3458  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3459  *   R_Key and L_Key.
3460  * @mr - struct ib_mr pointer to be updated.
3461  * @newkey - new key to be used.
3462  */
ib_update_fast_reg_key(struct ib_mr * mr,u8 newkey)3463 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3464 {
3465 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3466 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3467 }
3468 
3469 /**
3470  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3471  * for calculating a new rkey for type 2 memory windows.
3472  * @rkey - the rkey to increment.
3473  */
ib_inc_rkey(u32 rkey)3474 static inline u32 ib_inc_rkey(u32 rkey)
3475 {
3476 	const u32 mask = 0x000000ff;
3477 	return ((rkey + 1) & mask) | (rkey & ~mask);
3478 }
3479 
3480 /**
3481  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3482  * @pd: The protection domain associated with the unmapped region.
3483  * @mr_access_flags: Specifies the memory access rights.
3484  * @fmr_attr: Attributes of the unmapped region.
3485  *
3486  * A fast memory region must be mapped before it can be used as part of
3487  * a work request.
3488  */
3489 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3490 			    int mr_access_flags,
3491 			    struct ib_fmr_attr *fmr_attr);
3492 
3493 /**
3494  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3495  * @fmr: The fast memory region to associate with the pages.
3496  * @page_list: An array of physical pages to map to the fast memory region.
3497  * @list_len: The number of pages in page_list.
3498  * @iova: The I/O virtual address to use with the mapped region.
3499  */
ib_map_phys_fmr(struct ib_fmr * fmr,u64 * page_list,int list_len,u64 iova)3500 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3501 				  u64 *page_list, int list_len,
3502 				  u64 iova)
3503 {
3504 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3505 }
3506 
3507 /**
3508  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3509  * @fmr_list: A linked list of fast memory regions to unmap.
3510  */
3511 int ib_unmap_fmr(struct list_head *fmr_list);
3512 
3513 /**
3514  * ib_dealloc_fmr - Deallocates a fast memory region.
3515  * @fmr: The fast memory region to deallocate.
3516  */
3517 int ib_dealloc_fmr(struct ib_fmr *fmr);
3518 
3519 /**
3520  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3521  * @qp: QP to attach to the multicast group.  The QP must be type
3522  *   IB_QPT_UD.
3523  * @gid: Multicast group GID.
3524  * @lid: Multicast group LID in host byte order.
3525  *
3526  * In order to send and receive multicast packets, subnet
3527  * administration must have created the multicast group and configured
3528  * the fabric appropriately.  The port associated with the specified
3529  * QP must also be a member of the multicast group.
3530  */
3531 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3532 
3533 /**
3534  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3535  * @qp: QP to detach from the multicast group.
3536  * @gid: Multicast group GID.
3537  * @lid: Multicast group LID in host byte order.
3538  */
3539 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3540 
3541 /**
3542  * ib_alloc_xrcd - Allocates an XRC domain.
3543  * @device: The device on which to allocate the XRC domain.
3544  */
3545 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3546 
3547 /**
3548  * ib_dealloc_xrcd - Deallocates an XRC domain.
3549  * @xrcd: The XRC domain to deallocate.
3550  */
3551 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3552 
3553 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3554 			       struct ib_flow_attr *flow_attr, int domain);
3555 int ib_destroy_flow(struct ib_flow *flow_id);
3556 
ib_check_mr_access(int flags)3557 static inline int ib_check_mr_access(int flags)
3558 {
3559 	/*
3560 	 * Local write permission is required if remote write or
3561 	 * remote atomic permission is also requested.
3562 	 */
3563 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3564 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3565 		return -EINVAL;
3566 
3567 	return 0;
3568 }
3569 
ib_access_writable(int access_flags)3570 static inline bool ib_access_writable(int access_flags)
3571 {
3572 	/*
3573 	 * We have writable memory backing the MR if any of the following
3574 	 * access flags are set.  "Local write" and "remote write" obviously
3575 	 * require write access.  "Remote atomic" can do things like fetch and
3576 	 * add, which will modify memory, and "MW bind" can change permissions
3577 	 * by binding a window.
3578 	 */
3579 	return access_flags &
3580 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
3581 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
3582 }
3583 
3584 /**
3585  * ib_check_mr_status: lightweight check of MR status.
3586  *     This routine may provide status checks on a selected
3587  *     ib_mr. first use is for signature status check.
3588  *
3589  * @mr: A memory region.
3590  * @check_mask: Bitmask of which checks to perform from
3591  *     ib_mr_status_check enumeration.
3592  * @mr_status: The container of relevant status checks.
3593  *     failed checks will be indicated in the status bitmask
3594  *     and the relevant info shall be in the error item.
3595  */
3596 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3597 		       struct ib_mr_status *mr_status);
3598 
3599 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3600 					    u16 pkey, const union ib_gid *gid,
3601 					    const struct sockaddr *addr);
3602 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3603 			   struct ib_wq_init_attr *init_attr);
3604 int ib_destroy_wq(struct ib_wq *wq);
3605 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3606 		 u32 wq_attr_mask);
3607 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3608 						 struct ib_rwq_ind_table_init_attr*
3609 						 wq_ind_table_init_attr);
3610 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3611 
3612 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3613 		 unsigned int *sg_offset, unsigned int page_size);
3614 
3615 static inline int
ib_map_mr_sg_zbva(struct ib_mr * mr,struct scatterlist * sg,int sg_nents,unsigned int * sg_offset,unsigned int page_size)3616 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3617 		  unsigned int *sg_offset, unsigned int page_size)
3618 {
3619 	int n;
3620 
3621 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3622 	mr->iova = 0;
3623 
3624 	return n;
3625 }
3626 
3627 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3628 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3629 
3630 void ib_drain_rq(struct ib_qp *qp);
3631 void ib_drain_sq(struct ib_qp *qp);
3632 void ib_drain_qp(struct ib_qp *qp);
3633 
3634 int ib_resolve_eth_dmac(struct ib_device *device,
3635 			struct rdma_ah_attr *ah_attr);
3636 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3637 
rdma_ah_retrieve_dmac(struct rdma_ah_attr * attr)3638 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3639 {
3640 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3641 		return attr->roce.dmac;
3642 	return NULL;
3643 }
3644 
rdma_ah_set_dlid(struct rdma_ah_attr * attr,u32 dlid)3645 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3646 {
3647 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3648 		attr->ib.dlid = (u16)dlid;
3649 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3650 		attr->opa.dlid = dlid;
3651 }
3652 
rdma_ah_get_dlid(const struct rdma_ah_attr * attr)3653 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3654 {
3655 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3656 		return attr->ib.dlid;
3657 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3658 		return attr->opa.dlid;
3659 	return 0;
3660 }
3661 
rdma_ah_set_sl(struct rdma_ah_attr * attr,u8 sl)3662 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3663 {
3664 	attr->sl = sl;
3665 }
3666 
rdma_ah_get_sl(const struct rdma_ah_attr * attr)3667 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3668 {
3669 	return attr->sl;
3670 }
3671 
rdma_ah_set_path_bits(struct rdma_ah_attr * attr,u8 src_path_bits)3672 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3673 					 u8 src_path_bits)
3674 {
3675 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3676 		attr->ib.src_path_bits = src_path_bits;
3677 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3678 		attr->opa.src_path_bits = src_path_bits;
3679 }
3680 
rdma_ah_get_path_bits(const struct rdma_ah_attr * attr)3681 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3682 {
3683 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3684 		return attr->ib.src_path_bits;
3685 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3686 		return attr->opa.src_path_bits;
3687 	return 0;
3688 }
3689 
rdma_ah_set_make_grd(struct rdma_ah_attr * attr,bool make_grd)3690 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
3691 					bool make_grd)
3692 {
3693 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3694 		attr->opa.make_grd = make_grd;
3695 }
3696 
rdma_ah_get_make_grd(const struct rdma_ah_attr * attr)3697 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
3698 {
3699 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3700 		return attr->opa.make_grd;
3701 	return false;
3702 }
3703 
rdma_ah_set_port_num(struct rdma_ah_attr * attr,u8 port_num)3704 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3705 {
3706 	attr->port_num = port_num;
3707 }
3708 
rdma_ah_get_port_num(const struct rdma_ah_attr * attr)3709 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3710 {
3711 	return attr->port_num;
3712 }
3713 
rdma_ah_set_static_rate(struct rdma_ah_attr * attr,u8 static_rate)3714 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3715 					   u8 static_rate)
3716 {
3717 	attr->static_rate = static_rate;
3718 }
3719 
rdma_ah_get_static_rate(const struct rdma_ah_attr * attr)3720 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3721 {
3722 	return attr->static_rate;
3723 }
3724 
rdma_ah_set_ah_flags(struct rdma_ah_attr * attr,enum ib_ah_flags flag)3725 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3726 					enum ib_ah_flags flag)
3727 {
3728 	attr->ah_flags = flag;
3729 }
3730 
3731 static inline enum ib_ah_flags
rdma_ah_get_ah_flags(const struct rdma_ah_attr * attr)3732 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3733 {
3734 	return attr->ah_flags;
3735 }
3736 
3737 static inline const struct ib_global_route
rdma_ah_read_grh(const struct rdma_ah_attr * attr)3738 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
3739 {
3740 	return &attr->grh;
3741 }
3742 
3743 /*To retrieve and modify the grh */
3744 static inline struct ib_global_route
rdma_ah_retrieve_grh(struct rdma_ah_attr * attr)3745 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
3746 {
3747 	return &attr->grh;
3748 }
3749 
rdma_ah_set_dgid_raw(struct rdma_ah_attr * attr,void * dgid)3750 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
3751 {
3752 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3753 
3754 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
3755 }
3756 
rdma_ah_set_subnet_prefix(struct rdma_ah_attr * attr,__be64 prefix)3757 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
3758 					     __be64 prefix)
3759 {
3760 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3761 
3762 	grh->dgid.global.subnet_prefix = prefix;
3763 }
3764 
rdma_ah_set_interface_id(struct rdma_ah_attr * attr,__be64 if_id)3765 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
3766 					    __be64 if_id)
3767 {
3768 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3769 
3770 	grh->dgid.global.interface_id = if_id;
3771 }
3772 
rdma_ah_set_grh(struct rdma_ah_attr * attr,union ib_gid * dgid,u32 flow_label,u8 sgid_index,u8 hop_limit,u8 traffic_class)3773 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
3774 				   union ib_gid *dgid, u32 flow_label,
3775 				   u8 sgid_index, u8 hop_limit,
3776 				   u8 traffic_class)
3777 {
3778 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3779 
3780 	attr->ah_flags = IB_AH_GRH;
3781 	if (dgid)
3782 		grh->dgid = *dgid;
3783 	grh->flow_label = flow_label;
3784 	grh->sgid_index = sgid_index;
3785 	grh->hop_limit = hop_limit;
3786 	grh->traffic_class = traffic_class;
3787 }
3788 
3789 /**
3790  * rdma_ah_find_type - Return address handle type.
3791  *
3792  * @dev: Device to be checked
3793  * @port_num: Port number
3794  */
rdma_ah_find_type(struct ib_device * dev,u8 port_num)3795 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
3796 						       u8 port_num)
3797 {
3798 	if (rdma_protocol_roce(dev, port_num))
3799 		return RDMA_AH_ATTR_TYPE_ROCE;
3800 	if (rdma_protocol_ib(dev, port_num)) {
3801 		if (rdma_cap_opa_ah(dev, port_num))
3802 			return RDMA_AH_ATTR_TYPE_OPA;
3803 		return RDMA_AH_ATTR_TYPE_IB;
3804 	}
3805 
3806 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
3807 }
3808 
3809 /**
3810  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
3811  *     In the current implementation the only way to get
3812  *     get the 32bit lid is from other sources for OPA.
3813  *     For IB, lids will always be 16bits so cast the
3814  *     value accordingly.
3815  *
3816  * @lid: A 32bit LID
3817  */
ib_lid_cpu16(u32 lid)3818 static inline u16 ib_lid_cpu16(u32 lid)
3819 {
3820 	WARN_ON_ONCE(lid & 0xFFFF0000);
3821 	return (u16)lid;
3822 }
3823 
3824 /**
3825  * ib_lid_be16 - Return lid in 16bit BE encoding.
3826  *
3827  * @lid: A 32bit LID
3828  */
ib_lid_be16(u32 lid)3829 static inline __be16 ib_lid_be16(u32 lid)
3830 {
3831 	WARN_ON_ONCE(lid & 0xFFFF0000);
3832 	return cpu_to_be16((u16)lid);
3833 }
3834 
3835 /**
3836  * ib_get_vector_affinity - Get the affinity mappings of a given completion
3837  *   vector
3838  * @device:         the rdma device
3839  * @comp_vector:    index of completion vector
3840  *
3841  * Returns NULL on failure, otherwise a corresponding cpu map of the
3842  * completion vector (returns all-cpus map if the device driver doesn't
3843  * implement get_vector_affinity).
3844  */
3845 static inline const struct cpumask *
ib_get_vector_affinity(struct ib_device * device,int comp_vector)3846 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
3847 {
3848 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
3849 	    !device->get_vector_affinity)
3850 		return NULL;
3851 
3852 	return device->get_vector_affinity(device, comp_vector);
3853 
3854 }
3855 
3856 #endif /* IB_VERBS_H */
3857