• 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/dma-mapping.h>
45 #include <linux/kref.h>
46 #include <linux/list.h>
47 #include <linux/rwsem.h>
48 #include <linux/workqueue.h>
49 #include <linux/irq_poll.h>
50 #include <uapi/linux/if_ether.h>
51 #include <net/ipv6.h>
52 #include <net/ip.h>
53 #include <linux/string.h>
54 #include <linux/slab.h>
55 #include <linux/netdevice.h>
56 #include <linux/refcount.h>
57 #include <linux/if_link.h>
58 #include <linux/atomic.h>
59 #include <linux/mmu_notifier.h>
60 #include <linux/uaccess.h>
61 #include <linux/cgroup_rdma.h>
62 #include <linux/irqflags.h>
63 #include <linux/preempt.h>
64 #include <linux/dim.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 #include <rdma/rdma_counter.h>
67 #include <rdma/restrack.h>
68 #include <rdma/signature.h>
69 #include <uapi/rdma/rdma_user_ioctl.h>
70 #include <uapi/rdma/ib_user_ioctl_verbs.h>
71 
72 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
73 
74 struct ib_umem_odp;
75 
76 extern struct workqueue_struct *ib_wq;
77 extern struct workqueue_struct *ib_comp_wq;
78 extern struct workqueue_struct *ib_comp_unbound_wq;
79 
80 __printf(3, 4) __cold
81 void ibdev_printk(const char *level, const struct ib_device *ibdev,
82 		  const char *format, ...);
83 __printf(2, 3) __cold
84 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
85 __printf(2, 3) __cold
86 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
87 __printf(2, 3) __cold
88 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
89 __printf(2, 3) __cold
90 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
91 __printf(2, 3) __cold
92 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
93 __printf(2, 3) __cold
94 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
95 __printf(2, 3) __cold
96 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
97 
98 #if defined(CONFIG_DYNAMIC_DEBUG) || \
99 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
100 #define ibdev_dbg(__dev, format, args...)                       \
101 	dynamic_ibdev_dbg(__dev, format, ##args)
102 #else
103 __printf(2, 3) __cold
104 static inline
ibdev_dbg(const struct ib_device * ibdev,const char * format,...)105 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
106 #endif
107 
108 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...)           \
109 do {                                                                    \
110 	static DEFINE_RATELIMIT_STATE(_rs,                              \
111 				      DEFAULT_RATELIMIT_INTERVAL,       \
112 				      DEFAULT_RATELIMIT_BURST);         \
113 	if (__ratelimit(&_rs))                                          \
114 		ibdev_level(ibdev, fmt, ##__VA_ARGS__);                 \
115 } while (0)
116 
117 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
118 	ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
119 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
120 	ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
121 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
122 	ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
123 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
124 	ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
125 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
126 	ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
127 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
128 	ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
129 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
130 	ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
131 
132 #if defined(CONFIG_DYNAMIC_DEBUG) || \
133 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
134 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
135 #define ibdev_dbg_ratelimited(ibdev, fmt, ...)                          \
136 do {                                                                    \
137 	static DEFINE_RATELIMIT_STATE(_rs,                              \
138 				      DEFAULT_RATELIMIT_INTERVAL,       \
139 				      DEFAULT_RATELIMIT_BURST);         \
140 	DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt);                 \
141 	if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs))      \
142 		__dynamic_ibdev_dbg(&descriptor, ibdev, fmt,            \
143 				    ##__VA_ARGS__);                     \
144 } while (0)
145 #else
146 __printf(2, 3) __cold
147 static inline
ibdev_dbg_ratelimited(const struct ib_device * ibdev,const char * format,...)148 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
149 #endif
150 
151 union ib_gid {
152 	u8	raw[16];
153 	struct {
154 		__be64	subnet_prefix;
155 		__be64	interface_id;
156 	} global;
157 };
158 
159 extern union ib_gid zgid;
160 
161 enum ib_gid_type {
162 	/* If link layer is Ethernet, this is RoCE V1 */
163 	IB_GID_TYPE_IB        = 0,
164 	IB_GID_TYPE_ROCE      = 0,
165 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
166 	IB_GID_TYPE_SIZE
167 };
168 
169 #define ROCE_V2_UDP_DPORT      4791
170 struct ib_gid_attr {
171 	struct net_device __rcu	*ndev;
172 	struct ib_device	*device;
173 	union ib_gid		gid;
174 	enum ib_gid_type	gid_type;
175 	u16			index;
176 	u8			port_num;
177 };
178 
179 enum {
180 	/* set the local administered indication */
181 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
182 };
183 
184 enum rdma_transport_type {
185 	RDMA_TRANSPORT_IB,
186 	RDMA_TRANSPORT_IWARP,
187 	RDMA_TRANSPORT_USNIC,
188 	RDMA_TRANSPORT_USNIC_UDP,
189 	RDMA_TRANSPORT_UNSPECIFIED,
190 };
191 
192 enum rdma_protocol_type {
193 	RDMA_PROTOCOL_IB,
194 	RDMA_PROTOCOL_IBOE,
195 	RDMA_PROTOCOL_IWARP,
196 	RDMA_PROTOCOL_USNIC_UDP
197 };
198 
199 __attribute_const__ enum rdma_transport_type
200 rdma_node_get_transport(unsigned int node_type);
201 
202 enum rdma_network_type {
203 	RDMA_NETWORK_IB,
204 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
205 	RDMA_NETWORK_IPV4,
206 	RDMA_NETWORK_IPV6
207 };
208 
ib_network_to_gid_type(enum rdma_network_type network_type)209 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
210 {
211 	if (network_type == RDMA_NETWORK_IPV4 ||
212 	    network_type == RDMA_NETWORK_IPV6)
213 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
214 
215 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
216 	return IB_GID_TYPE_IB;
217 }
218 
219 static inline enum rdma_network_type
rdma_gid_attr_network_type(const struct ib_gid_attr * attr)220 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
221 {
222 	if (attr->gid_type == IB_GID_TYPE_IB)
223 		return RDMA_NETWORK_IB;
224 
225 	if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
226 		return RDMA_NETWORK_IPV4;
227 	else
228 		return RDMA_NETWORK_IPV6;
229 }
230 
231 enum rdma_link_layer {
232 	IB_LINK_LAYER_UNSPECIFIED,
233 	IB_LINK_LAYER_INFINIBAND,
234 	IB_LINK_LAYER_ETHERNET,
235 };
236 
237 enum ib_device_cap_flags {
238 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
239 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
240 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
241 	IB_DEVICE_RAW_MULTI			= (1 << 3),
242 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
243 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
244 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
245 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
246 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
247 	/* Not in use, former INIT_TYPE		= (1 << 9),*/
248 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
249 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
250 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
251 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
252 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
253 
254 	/*
255 	 * This device supports a per-device lkey or stag that can be
256 	 * used without performing a memory registration for the local
257 	 * memory.  Note that ULPs should never check this flag, but
258 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
259 	 * which will always contain a usable lkey.
260 	 */
261 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
262 	/* Reserved, old SEND_W_INV		= (1 << 16),*/
263 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
264 	/*
265 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
266 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
267 	 * messages and can verify the validity of checksum for
268 	 * incoming messages.  Setting this flag implies that the
269 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
270 	 */
271 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
272 	IB_DEVICE_UD_TSO			= (1 << 19),
273 	IB_DEVICE_XRC				= (1 << 20),
274 
275 	/*
276 	 * This device supports the IB "base memory management extension",
277 	 * which includes support for fast registrations (IB_WR_REG_MR,
278 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
279 	 * also be set by any iWarp device which must support FRs to comply
280 	 * to the iWarp verbs spec.  iWarp devices also support the
281 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
282 	 * stag.
283 	 */
284 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
285 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
286 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
287 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
288 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
289 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
290 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
291 	/*
292 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
293 	 * support execution of WQEs that involve synchronization
294 	 * of I/O operations with single completion queue managed
295 	 * by hardware.
296 	 */
297 	IB_DEVICE_CROSS_CHANNEL			= (1 << 27),
298 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
299 	IB_DEVICE_INTEGRITY_HANDOVER		= (1 << 30),
300 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
301 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
302 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
303 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
304 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
305 	IB_DEVICE_RDMA_NETDEV_OPA_VNIC		= (1ULL << 35),
306 	/* The device supports padding incoming writes to cacheline. */
307 	IB_DEVICE_PCI_WRITE_END_PADDING		= (1ULL << 36),
308 	IB_DEVICE_ALLOW_USER_UNREG		= (1ULL << 37),
309 };
310 
311 enum ib_atomic_cap {
312 	IB_ATOMIC_NONE,
313 	IB_ATOMIC_HCA,
314 	IB_ATOMIC_GLOB
315 };
316 
317 enum ib_odp_general_cap_bits {
318 	IB_ODP_SUPPORT		= 1 << 0,
319 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
320 };
321 
322 enum ib_odp_transport_cap_bits {
323 	IB_ODP_SUPPORT_SEND	= 1 << 0,
324 	IB_ODP_SUPPORT_RECV	= 1 << 1,
325 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
326 	IB_ODP_SUPPORT_READ	= 1 << 3,
327 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
328 	IB_ODP_SUPPORT_SRQ_RECV	= 1 << 5,
329 };
330 
331 struct ib_odp_caps {
332 	uint64_t general_caps;
333 	struct {
334 		uint32_t  rc_odp_caps;
335 		uint32_t  uc_odp_caps;
336 		uint32_t  ud_odp_caps;
337 		uint32_t  xrc_odp_caps;
338 	} per_transport_caps;
339 };
340 
341 struct ib_rss_caps {
342 	/* Corresponding bit will be set if qp type from
343 	 * 'enum ib_qp_type' is supported, e.g.
344 	 * supported_qpts |= 1 << IB_QPT_UD
345 	 */
346 	u32 supported_qpts;
347 	u32 max_rwq_indirection_tables;
348 	u32 max_rwq_indirection_table_size;
349 };
350 
351 enum ib_tm_cap_flags {
352 	/*  Support tag matching with rendezvous offload for RC transport */
353 	IB_TM_CAP_RNDV_RC = 1 << 0,
354 };
355 
356 struct ib_tm_caps {
357 	/* Max size of RNDV header */
358 	u32 max_rndv_hdr_size;
359 	/* Max number of entries in tag matching list */
360 	u32 max_num_tags;
361 	/* From enum ib_tm_cap_flags */
362 	u32 flags;
363 	/* Max number of outstanding list operations */
364 	u32 max_ops;
365 	/* Max number of SGE in tag matching entry */
366 	u32 max_sge;
367 };
368 
369 struct ib_cq_init_attr {
370 	unsigned int	cqe;
371 	u32		comp_vector;
372 	u32		flags;
373 };
374 
375 enum ib_cq_attr_mask {
376 	IB_CQ_MODERATE = 1 << 0,
377 };
378 
379 struct ib_cq_caps {
380 	u16     max_cq_moderation_count;
381 	u16     max_cq_moderation_period;
382 };
383 
384 struct ib_dm_mr_attr {
385 	u64		length;
386 	u64		offset;
387 	u32		access_flags;
388 };
389 
390 struct ib_dm_alloc_attr {
391 	u64	length;
392 	u32	alignment;
393 	u32	flags;
394 };
395 
396 struct ib_device_attr {
397 	u64			fw_ver;
398 	__be64			sys_image_guid;
399 	u64			max_mr_size;
400 	u64			page_size_cap;
401 	u32			vendor_id;
402 	u32			vendor_part_id;
403 	u32			hw_ver;
404 	int			max_qp;
405 	int			max_qp_wr;
406 	u64			device_cap_flags;
407 	int			max_send_sge;
408 	int			max_recv_sge;
409 	int			max_sge_rd;
410 	int			max_cq;
411 	int			max_cqe;
412 	int			max_mr;
413 	int			max_pd;
414 	int			max_qp_rd_atom;
415 	int			max_ee_rd_atom;
416 	int			max_res_rd_atom;
417 	int			max_qp_init_rd_atom;
418 	int			max_ee_init_rd_atom;
419 	enum ib_atomic_cap	atomic_cap;
420 	enum ib_atomic_cap	masked_atomic_cap;
421 	int			max_ee;
422 	int			max_rdd;
423 	int			max_mw;
424 	int			max_raw_ipv6_qp;
425 	int			max_raw_ethy_qp;
426 	int			max_mcast_grp;
427 	int			max_mcast_qp_attach;
428 	int			max_total_mcast_qp_attach;
429 	int			max_ah;
430 	int			max_fmr;
431 	int			max_map_per_fmr;
432 	int			max_srq;
433 	int			max_srq_wr;
434 	int			max_srq_sge;
435 	unsigned int		max_fast_reg_page_list_len;
436 	unsigned int		max_pi_fast_reg_page_list_len;
437 	u16			max_pkeys;
438 	u8			local_ca_ack_delay;
439 	int			sig_prot_cap;
440 	int			sig_guard_cap;
441 	struct ib_odp_caps	odp_caps;
442 	uint64_t		timestamp_mask;
443 	uint64_t		hca_core_clock; /* in KHZ */
444 	struct ib_rss_caps	rss_caps;
445 	u32			max_wq_type_rq;
446 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
447 	struct ib_tm_caps	tm_caps;
448 	struct ib_cq_caps       cq_caps;
449 	u64			max_dm_size;
450 };
451 
452 enum ib_mtu {
453 	IB_MTU_256  = 1,
454 	IB_MTU_512  = 2,
455 	IB_MTU_1024 = 3,
456 	IB_MTU_2048 = 4,
457 	IB_MTU_4096 = 5
458 };
459 
ib_mtu_enum_to_int(enum ib_mtu mtu)460 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
461 {
462 	switch (mtu) {
463 	case IB_MTU_256:  return  256;
464 	case IB_MTU_512:  return  512;
465 	case IB_MTU_1024: return 1024;
466 	case IB_MTU_2048: return 2048;
467 	case IB_MTU_4096: return 4096;
468 	default: 	  return -1;
469 	}
470 }
471 
ib_mtu_int_to_enum(int mtu)472 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
473 {
474 	if (mtu >= 4096)
475 		return IB_MTU_4096;
476 	else if (mtu >= 2048)
477 		return IB_MTU_2048;
478 	else if (mtu >= 1024)
479 		return IB_MTU_1024;
480 	else if (mtu >= 512)
481 		return IB_MTU_512;
482 	else
483 		return IB_MTU_256;
484 }
485 
486 enum ib_port_state {
487 	IB_PORT_NOP		= 0,
488 	IB_PORT_DOWN		= 1,
489 	IB_PORT_INIT		= 2,
490 	IB_PORT_ARMED		= 3,
491 	IB_PORT_ACTIVE		= 4,
492 	IB_PORT_ACTIVE_DEFER	= 5
493 };
494 
495 enum ib_port_phys_state {
496 	IB_PORT_PHYS_STATE_SLEEP = 1,
497 	IB_PORT_PHYS_STATE_POLLING = 2,
498 	IB_PORT_PHYS_STATE_DISABLED = 3,
499 	IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
500 	IB_PORT_PHYS_STATE_LINK_UP = 5,
501 	IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
502 	IB_PORT_PHYS_STATE_PHY_TEST = 7,
503 };
504 
505 enum ib_port_width {
506 	IB_WIDTH_1X	= 1,
507 	IB_WIDTH_2X	= 16,
508 	IB_WIDTH_4X	= 2,
509 	IB_WIDTH_8X	= 4,
510 	IB_WIDTH_12X	= 8
511 };
512 
ib_width_enum_to_int(enum ib_port_width width)513 static inline int ib_width_enum_to_int(enum ib_port_width width)
514 {
515 	switch (width) {
516 	case IB_WIDTH_1X:  return  1;
517 	case IB_WIDTH_2X:  return  2;
518 	case IB_WIDTH_4X:  return  4;
519 	case IB_WIDTH_8X:  return  8;
520 	case IB_WIDTH_12X: return 12;
521 	default: 	  return -1;
522 	}
523 }
524 
525 enum ib_port_speed {
526 	IB_SPEED_SDR	= 1,
527 	IB_SPEED_DDR	= 2,
528 	IB_SPEED_QDR	= 4,
529 	IB_SPEED_FDR10	= 8,
530 	IB_SPEED_FDR	= 16,
531 	IB_SPEED_EDR	= 32,
532 	IB_SPEED_HDR	= 64
533 };
534 
535 /**
536  * struct rdma_hw_stats
537  * @lock - Mutex to protect parallel write access to lifespan and values
538  *    of counters, which are 64bits and not guaranteeed to be written
539  *    atomicaly on 32bits systems.
540  * @timestamp - Used by the core code to track when the last update was
541  * @lifespan - Used by the core code to determine how old the counters
542  *   should be before being updated again.  Stored in jiffies, defaults
543  *   to 10 milliseconds, drivers can override the default be specifying
544  *   their own value during their allocation routine.
545  * @name - Array of pointers to static names used for the counters in
546  *   directory.
547  * @num_counters - How many hardware counters there are.  If name is
548  *   shorter than this number, a kernel oops will result.  Driver authors
549  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
550  *   in their code to prevent this.
551  * @value - Array of u64 counters that are accessed by the sysfs code and
552  *   filled in by the drivers get_stats routine
553  */
554 struct rdma_hw_stats {
555 	struct mutex	lock; /* Protect lifespan and values[] */
556 	unsigned long	timestamp;
557 	unsigned long	lifespan;
558 	const char * const *names;
559 	int		num_counters;
560 	u64		value[];
561 };
562 
563 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
564 /**
565  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
566  *   for drivers.
567  * @names - Array of static const char *
568  * @num_counters - How many elements in array
569  * @lifespan - How many milliseconds between updates
570  */
rdma_alloc_hw_stats_struct(const char * const * names,int num_counters,unsigned long lifespan)571 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
572 		const char * const *names, int num_counters,
573 		unsigned long lifespan)
574 {
575 	struct rdma_hw_stats *stats;
576 
577 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
578 			GFP_KERNEL);
579 	if (!stats)
580 		return NULL;
581 	stats->names = names;
582 	stats->num_counters = num_counters;
583 	stats->lifespan = msecs_to_jiffies(lifespan);
584 
585 	return stats;
586 }
587 
588 
589 /* Define bits for the various functionality this port needs to be supported by
590  * the core.
591  */
592 /* Management                           0x00000FFF */
593 #define RDMA_CORE_CAP_IB_MAD            0x00000001
594 #define RDMA_CORE_CAP_IB_SMI            0x00000002
595 #define RDMA_CORE_CAP_IB_CM             0x00000004
596 #define RDMA_CORE_CAP_IW_CM             0x00000008
597 #define RDMA_CORE_CAP_IB_SA             0x00000010
598 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
599 
600 /* Address format                       0x000FF000 */
601 #define RDMA_CORE_CAP_AF_IB             0x00001000
602 #define RDMA_CORE_CAP_ETH_AH            0x00002000
603 #define RDMA_CORE_CAP_OPA_AH            0x00004000
604 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
605 
606 /* Protocol                             0xFFF00000 */
607 #define RDMA_CORE_CAP_PROT_IB           0x00100000
608 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
609 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
610 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
611 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
612 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
613 
614 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
615 					| RDMA_CORE_CAP_PROT_ROCE     \
616 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
617 
618 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
619 					| RDMA_CORE_CAP_IB_MAD \
620 					| RDMA_CORE_CAP_IB_SMI \
621 					| RDMA_CORE_CAP_IB_CM  \
622 					| RDMA_CORE_CAP_IB_SA  \
623 					| RDMA_CORE_CAP_AF_IB)
624 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
625 					| RDMA_CORE_CAP_IB_MAD  \
626 					| RDMA_CORE_CAP_IB_CM   \
627 					| RDMA_CORE_CAP_AF_IB   \
628 					| RDMA_CORE_CAP_ETH_AH)
629 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
630 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
631 					| RDMA_CORE_CAP_IB_MAD  \
632 					| RDMA_CORE_CAP_IB_CM   \
633 					| RDMA_CORE_CAP_AF_IB   \
634 					| RDMA_CORE_CAP_ETH_AH)
635 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
636 					| RDMA_CORE_CAP_IW_CM)
637 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
638 					| RDMA_CORE_CAP_OPA_MAD)
639 
640 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
641 
642 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
643 
644 struct ib_port_attr {
645 	u64			subnet_prefix;
646 	enum ib_port_state	state;
647 	enum ib_mtu		max_mtu;
648 	enum ib_mtu		active_mtu;
649 	int			gid_tbl_len;
650 	unsigned int		ip_gids:1;
651 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
652 	u32			port_cap_flags;
653 	u32			max_msg_sz;
654 	u32			bad_pkey_cntr;
655 	u32			qkey_viol_cntr;
656 	u16			pkey_tbl_len;
657 	u32			sm_lid;
658 	u32			lid;
659 	u8			lmc;
660 	u8			max_vl_num;
661 	u8			sm_sl;
662 	u8			subnet_timeout;
663 	u8			init_type_reply;
664 	u8			active_width;
665 	u8			active_speed;
666 	u8                      phys_state;
667 	u16			port_cap_flags2;
668 };
669 
670 enum ib_device_modify_flags {
671 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
672 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
673 };
674 
675 #define IB_DEVICE_NODE_DESC_MAX 64
676 
677 struct ib_device_modify {
678 	u64	sys_image_guid;
679 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
680 };
681 
682 enum ib_port_modify_flags {
683 	IB_PORT_SHUTDOWN		= 1,
684 	IB_PORT_INIT_TYPE		= (1<<2),
685 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
686 	IB_PORT_OPA_MASK_CHG		= (1<<4)
687 };
688 
689 struct ib_port_modify {
690 	u32	set_port_cap_mask;
691 	u32	clr_port_cap_mask;
692 	u8	init_type;
693 };
694 
695 enum ib_event_type {
696 	IB_EVENT_CQ_ERR,
697 	IB_EVENT_QP_FATAL,
698 	IB_EVENT_QP_REQ_ERR,
699 	IB_EVENT_QP_ACCESS_ERR,
700 	IB_EVENT_COMM_EST,
701 	IB_EVENT_SQ_DRAINED,
702 	IB_EVENT_PATH_MIG,
703 	IB_EVENT_PATH_MIG_ERR,
704 	IB_EVENT_DEVICE_FATAL,
705 	IB_EVENT_PORT_ACTIVE,
706 	IB_EVENT_PORT_ERR,
707 	IB_EVENT_LID_CHANGE,
708 	IB_EVENT_PKEY_CHANGE,
709 	IB_EVENT_SM_CHANGE,
710 	IB_EVENT_SRQ_ERR,
711 	IB_EVENT_SRQ_LIMIT_REACHED,
712 	IB_EVENT_QP_LAST_WQE_REACHED,
713 	IB_EVENT_CLIENT_REREGISTER,
714 	IB_EVENT_GID_CHANGE,
715 	IB_EVENT_WQ_FATAL,
716 };
717 
718 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
719 
720 struct ib_event {
721 	struct ib_device	*device;
722 	union {
723 		struct ib_cq	*cq;
724 		struct ib_qp	*qp;
725 		struct ib_srq	*srq;
726 		struct ib_wq	*wq;
727 		u8		port_num;
728 	} element;
729 	enum ib_event_type	event;
730 };
731 
732 struct ib_event_handler {
733 	struct ib_device *device;
734 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
735 	struct list_head  list;
736 };
737 
738 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
739 	do {							\
740 		(_ptr)->device  = _device;			\
741 		(_ptr)->handler = _handler;			\
742 		INIT_LIST_HEAD(&(_ptr)->list);			\
743 	} while (0)
744 
745 struct ib_global_route {
746 	const struct ib_gid_attr *sgid_attr;
747 	union ib_gid	dgid;
748 	u32		flow_label;
749 	u8		sgid_index;
750 	u8		hop_limit;
751 	u8		traffic_class;
752 };
753 
754 struct ib_grh {
755 	__be32		version_tclass_flow;
756 	__be16		paylen;
757 	u8		next_hdr;
758 	u8		hop_limit;
759 	union ib_gid	sgid;
760 	union ib_gid	dgid;
761 };
762 
763 union rdma_network_hdr {
764 	struct ib_grh ibgrh;
765 	struct {
766 		/* The IB spec states that if it's IPv4, the header
767 		 * is located in the last 20 bytes of the header.
768 		 */
769 		u8		reserved[20];
770 		struct iphdr	roce4grh;
771 	};
772 };
773 
774 #define IB_QPN_MASK		0xFFFFFF
775 
776 enum {
777 	IB_MULTICAST_QPN = 0xffffff
778 };
779 
780 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
781 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
782 
783 enum ib_ah_flags {
784 	IB_AH_GRH	= 1
785 };
786 
787 enum ib_rate {
788 	IB_RATE_PORT_CURRENT = 0,
789 	IB_RATE_2_5_GBPS = 2,
790 	IB_RATE_5_GBPS   = 5,
791 	IB_RATE_10_GBPS  = 3,
792 	IB_RATE_20_GBPS  = 6,
793 	IB_RATE_30_GBPS  = 4,
794 	IB_RATE_40_GBPS  = 7,
795 	IB_RATE_60_GBPS  = 8,
796 	IB_RATE_80_GBPS  = 9,
797 	IB_RATE_120_GBPS = 10,
798 	IB_RATE_14_GBPS  = 11,
799 	IB_RATE_56_GBPS  = 12,
800 	IB_RATE_112_GBPS = 13,
801 	IB_RATE_168_GBPS = 14,
802 	IB_RATE_25_GBPS  = 15,
803 	IB_RATE_100_GBPS = 16,
804 	IB_RATE_200_GBPS = 17,
805 	IB_RATE_300_GBPS = 18,
806 	IB_RATE_28_GBPS  = 19,
807 	IB_RATE_50_GBPS  = 20,
808 	IB_RATE_400_GBPS = 21,
809 	IB_RATE_600_GBPS = 22,
810 };
811 
812 /**
813  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
814  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
815  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
816  * @rate: rate to convert.
817  */
818 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
819 
820 /**
821  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
822  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
823  * @rate: rate to convert.
824  */
825 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
826 
827 
828 /**
829  * enum ib_mr_type - memory region type
830  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
831  *                            normal registration
832  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
833  *                            register any arbitrary sg lists (without
834  *                            the normal mr constraints - see
835  *                            ib_map_mr_sg)
836  * @IB_MR_TYPE_DM:            memory region that is used for device
837  *                            memory registration
838  * @IB_MR_TYPE_USER:          memory region that is used for the user-space
839  *                            application
840  * @IB_MR_TYPE_DMA:           memory region that is used for DMA operations
841  *                            without address translations (VA=PA)
842  * @IB_MR_TYPE_INTEGRITY:     memory region that is used for
843  *                            data integrity operations
844  */
845 enum ib_mr_type {
846 	IB_MR_TYPE_MEM_REG,
847 	IB_MR_TYPE_SG_GAPS,
848 	IB_MR_TYPE_DM,
849 	IB_MR_TYPE_USER,
850 	IB_MR_TYPE_DMA,
851 	IB_MR_TYPE_INTEGRITY,
852 };
853 
854 enum ib_mr_status_check {
855 	IB_MR_CHECK_SIG_STATUS = 1,
856 };
857 
858 /**
859  * struct ib_mr_status - Memory region status container
860  *
861  * @fail_status: Bitmask of MR checks status. For each
862  *     failed check a corresponding status bit is set.
863  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
864  *     failure.
865  */
866 struct ib_mr_status {
867 	u32		    fail_status;
868 	struct ib_sig_err   sig_err;
869 };
870 
871 /**
872  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
873  * enum.
874  * @mult: multiple to convert.
875  */
876 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
877 
878 enum rdma_ah_attr_type {
879 	RDMA_AH_ATTR_TYPE_UNDEFINED,
880 	RDMA_AH_ATTR_TYPE_IB,
881 	RDMA_AH_ATTR_TYPE_ROCE,
882 	RDMA_AH_ATTR_TYPE_OPA,
883 };
884 
885 struct ib_ah_attr {
886 	u16			dlid;
887 	u8			src_path_bits;
888 };
889 
890 struct roce_ah_attr {
891 	u8			dmac[ETH_ALEN];
892 };
893 
894 struct opa_ah_attr {
895 	u32			dlid;
896 	u8			src_path_bits;
897 	bool			make_grd;
898 };
899 
900 struct rdma_ah_attr {
901 	struct ib_global_route	grh;
902 	u8			sl;
903 	u8			static_rate;
904 	u8			port_num;
905 	u8			ah_flags;
906 	enum rdma_ah_attr_type type;
907 	union {
908 		struct ib_ah_attr ib;
909 		struct roce_ah_attr roce;
910 		struct opa_ah_attr opa;
911 	};
912 };
913 
914 enum ib_wc_status {
915 	IB_WC_SUCCESS,
916 	IB_WC_LOC_LEN_ERR,
917 	IB_WC_LOC_QP_OP_ERR,
918 	IB_WC_LOC_EEC_OP_ERR,
919 	IB_WC_LOC_PROT_ERR,
920 	IB_WC_WR_FLUSH_ERR,
921 	IB_WC_MW_BIND_ERR,
922 	IB_WC_BAD_RESP_ERR,
923 	IB_WC_LOC_ACCESS_ERR,
924 	IB_WC_REM_INV_REQ_ERR,
925 	IB_WC_REM_ACCESS_ERR,
926 	IB_WC_REM_OP_ERR,
927 	IB_WC_RETRY_EXC_ERR,
928 	IB_WC_RNR_RETRY_EXC_ERR,
929 	IB_WC_LOC_RDD_VIOL_ERR,
930 	IB_WC_REM_INV_RD_REQ_ERR,
931 	IB_WC_REM_ABORT_ERR,
932 	IB_WC_INV_EECN_ERR,
933 	IB_WC_INV_EEC_STATE_ERR,
934 	IB_WC_FATAL_ERR,
935 	IB_WC_RESP_TIMEOUT_ERR,
936 	IB_WC_GENERAL_ERR
937 };
938 
939 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
940 
941 enum ib_wc_opcode {
942 	IB_WC_SEND,
943 	IB_WC_RDMA_WRITE,
944 	IB_WC_RDMA_READ,
945 	IB_WC_COMP_SWAP,
946 	IB_WC_FETCH_ADD,
947 	IB_WC_LSO,
948 	IB_WC_LOCAL_INV,
949 	IB_WC_REG_MR,
950 	IB_WC_MASKED_COMP_SWAP,
951 	IB_WC_MASKED_FETCH_ADD,
952 /*
953  * Set value of IB_WC_RECV so consumers can test if a completion is a
954  * receive by testing (opcode & IB_WC_RECV).
955  */
956 	IB_WC_RECV			= 1 << 7,
957 	IB_WC_RECV_RDMA_WITH_IMM
958 };
959 
960 enum ib_wc_flags {
961 	IB_WC_GRH		= 1,
962 	IB_WC_WITH_IMM		= (1<<1),
963 	IB_WC_WITH_INVALIDATE	= (1<<2),
964 	IB_WC_IP_CSUM_OK	= (1<<3),
965 	IB_WC_WITH_SMAC		= (1<<4),
966 	IB_WC_WITH_VLAN		= (1<<5),
967 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
968 };
969 
970 struct ib_wc {
971 	union {
972 		u64		wr_id;
973 		struct ib_cqe	*wr_cqe;
974 	};
975 	enum ib_wc_status	status;
976 	enum ib_wc_opcode	opcode;
977 	u32			vendor_err;
978 	u32			byte_len;
979 	struct ib_qp	       *qp;
980 	union {
981 		__be32		imm_data;
982 		u32		invalidate_rkey;
983 	} ex;
984 	u32			src_qp;
985 	u32			slid;
986 	int			wc_flags;
987 	u16			pkey_index;
988 	u8			sl;
989 	u8			dlid_path_bits;
990 	u8			port_num;	/* valid only for DR SMPs on switches */
991 	u8			smac[ETH_ALEN];
992 	u16			vlan_id;
993 	u8			network_hdr_type;
994 };
995 
996 enum ib_cq_notify_flags {
997 	IB_CQ_SOLICITED			= 1 << 0,
998 	IB_CQ_NEXT_COMP			= 1 << 1,
999 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1000 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
1001 };
1002 
1003 enum ib_srq_type {
1004 	IB_SRQT_BASIC,
1005 	IB_SRQT_XRC,
1006 	IB_SRQT_TM,
1007 };
1008 
ib_srq_has_cq(enum ib_srq_type srq_type)1009 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1010 {
1011 	return srq_type == IB_SRQT_XRC ||
1012 	       srq_type == IB_SRQT_TM;
1013 }
1014 
1015 enum ib_srq_attr_mask {
1016 	IB_SRQ_MAX_WR	= 1 << 0,
1017 	IB_SRQ_LIMIT	= 1 << 1,
1018 };
1019 
1020 struct ib_srq_attr {
1021 	u32	max_wr;
1022 	u32	max_sge;
1023 	u32	srq_limit;
1024 };
1025 
1026 struct ib_srq_init_attr {
1027 	void		      (*event_handler)(struct ib_event *, void *);
1028 	void		       *srq_context;
1029 	struct ib_srq_attr	attr;
1030 	enum ib_srq_type	srq_type;
1031 
1032 	struct {
1033 		struct ib_cq   *cq;
1034 		union {
1035 			struct {
1036 				struct ib_xrcd *xrcd;
1037 			} xrc;
1038 
1039 			struct {
1040 				u32		max_num_tags;
1041 			} tag_matching;
1042 		};
1043 	} ext;
1044 };
1045 
1046 struct ib_qp_cap {
1047 	u32	max_send_wr;
1048 	u32	max_recv_wr;
1049 	u32	max_send_sge;
1050 	u32	max_recv_sge;
1051 	u32	max_inline_data;
1052 
1053 	/*
1054 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1055 	 * ib_create_qp() will calculate the right amount of neededed WRs
1056 	 * and MRs based on this.
1057 	 */
1058 	u32	max_rdma_ctxs;
1059 };
1060 
1061 enum ib_sig_type {
1062 	IB_SIGNAL_ALL_WR,
1063 	IB_SIGNAL_REQ_WR
1064 };
1065 
1066 enum ib_qp_type {
1067 	/*
1068 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1069 	 * here (and in that order) since the MAD layer uses them as
1070 	 * indices into a 2-entry table.
1071 	 */
1072 	IB_QPT_SMI,
1073 	IB_QPT_GSI,
1074 
1075 	IB_QPT_RC,
1076 	IB_QPT_UC,
1077 	IB_QPT_UD,
1078 	IB_QPT_RAW_IPV6,
1079 	IB_QPT_RAW_ETHERTYPE,
1080 	IB_QPT_RAW_PACKET = 8,
1081 	IB_QPT_XRC_INI = 9,
1082 	IB_QPT_XRC_TGT,
1083 	IB_QPT_MAX,
1084 	IB_QPT_DRIVER = 0xFF,
1085 	/* Reserve a range for qp types internal to the low level driver.
1086 	 * These qp types will not be visible at the IB core layer, so the
1087 	 * IB_QPT_MAX usages should not be affected in the core layer
1088 	 */
1089 	IB_QPT_RESERVED1 = 0x1000,
1090 	IB_QPT_RESERVED2,
1091 	IB_QPT_RESERVED3,
1092 	IB_QPT_RESERVED4,
1093 	IB_QPT_RESERVED5,
1094 	IB_QPT_RESERVED6,
1095 	IB_QPT_RESERVED7,
1096 	IB_QPT_RESERVED8,
1097 	IB_QPT_RESERVED9,
1098 	IB_QPT_RESERVED10,
1099 };
1100 
1101 enum ib_qp_create_flags {
1102 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1103 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1104 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1105 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1106 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1107 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1108 	IB_QP_CREATE_INTEGRITY_EN		= 1 << 6,
1109 	/* FREE					= 1 << 7, */
1110 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1111 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1112 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1113 	IB_QP_CREATE_PCI_WRITE_END_PADDING	= 1 << 11,
1114 	/* reserve bits 26-31 for low level drivers' internal use */
1115 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1116 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1117 };
1118 
1119 /*
1120  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1121  * callback to destroy the passed in QP.
1122  */
1123 
1124 struct ib_qp_init_attr {
1125 	/* Consumer's event_handler callback must not block */
1126 	void                  (*event_handler)(struct ib_event *, void *);
1127 
1128 	void		       *qp_context;
1129 	struct ib_cq	       *send_cq;
1130 	struct ib_cq	       *recv_cq;
1131 	struct ib_srq	       *srq;
1132 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1133 	struct ib_qp_cap	cap;
1134 	enum ib_sig_type	sq_sig_type;
1135 	enum ib_qp_type		qp_type;
1136 	u32			create_flags;
1137 
1138 	/*
1139 	 * Only needed for special QP types, or when using the RW API.
1140 	 */
1141 	u8			port_num;
1142 	struct ib_rwq_ind_table *rwq_ind_tbl;
1143 	u32			source_qpn;
1144 };
1145 
1146 struct ib_qp_open_attr {
1147 	void                  (*event_handler)(struct ib_event *, void *);
1148 	void		       *qp_context;
1149 	u32			qp_num;
1150 	enum ib_qp_type		qp_type;
1151 };
1152 
1153 enum ib_rnr_timeout {
1154 	IB_RNR_TIMER_655_36 =  0,
1155 	IB_RNR_TIMER_000_01 =  1,
1156 	IB_RNR_TIMER_000_02 =  2,
1157 	IB_RNR_TIMER_000_03 =  3,
1158 	IB_RNR_TIMER_000_04 =  4,
1159 	IB_RNR_TIMER_000_06 =  5,
1160 	IB_RNR_TIMER_000_08 =  6,
1161 	IB_RNR_TIMER_000_12 =  7,
1162 	IB_RNR_TIMER_000_16 =  8,
1163 	IB_RNR_TIMER_000_24 =  9,
1164 	IB_RNR_TIMER_000_32 = 10,
1165 	IB_RNR_TIMER_000_48 = 11,
1166 	IB_RNR_TIMER_000_64 = 12,
1167 	IB_RNR_TIMER_000_96 = 13,
1168 	IB_RNR_TIMER_001_28 = 14,
1169 	IB_RNR_TIMER_001_92 = 15,
1170 	IB_RNR_TIMER_002_56 = 16,
1171 	IB_RNR_TIMER_003_84 = 17,
1172 	IB_RNR_TIMER_005_12 = 18,
1173 	IB_RNR_TIMER_007_68 = 19,
1174 	IB_RNR_TIMER_010_24 = 20,
1175 	IB_RNR_TIMER_015_36 = 21,
1176 	IB_RNR_TIMER_020_48 = 22,
1177 	IB_RNR_TIMER_030_72 = 23,
1178 	IB_RNR_TIMER_040_96 = 24,
1179 	IB_RNR_TIMER_061_44 = 25,
1180 	IB_RNR_TIMER_081_92 = 26,
1181 	IB_RNR_TIMER_122_88 = 27,
1182 	IB_RNR_TIMER_163_84 = 28,
1183 	IB_RNR_TIMER_245_76 = 29,
1184 	IB_RNR_TIMER_327_68 = 30,
1185 	IB_RNR_TIMER_491_52 = 31
1186 };
1187 
1188 enum ib_qp_attr_mask {
1189 	IB_QP_STATE			= 1,
1190 	IB_QP_CUR_STATE			= (1<<1),
1191 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1192 	IB_QP_ACCESS_FLAGS		= (1<<3),
1193 	IB_QP_PKEY_INDEX		= (1<<4),
1194 	IB_QP_PORT			= (1<<5),
1195 	IB_QP_QKEY			= (1<<6),
1196 	IB_QP_AV			= (1<<7),
1197 	IB_QP_PATH_MTU			= (1<<8),
1198 	IB_QP_TIMEOUT			= (1<<9),
1199 	IB_QP_RETRY_CNT			= (1<<10),
1200 	IB_QP_RNR_RETRY			= (1<<11),
1201 	IB_QP_RQ_PSN			= (1<<12),
1202 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1203 	IB_QP_ALT_PATH			= (1<<14),
1204 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1205 	IB_QP_SQ_PSN			= (1<<16),
1206 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1207 	IB_QP_PATH_MIG_STATE		= (1<<18),
1208 	IB_QP_CAP			= (1<<19),
1209 	IB_QP_DEST_QPN			= (1<<20),
1210 	IB_QP_RESERVED1			= (1<<21),
1211 	IB_QP_RESERVED2			= (1<<22),
1212 	IB_QP_RESERVED3			= (1<<23),
1213 	IB_QP_RESERVED4			= (1<<24),
1214 	IB_QP_RATE_LIMIT		= (1<<25),
1215 };
1216 
1217 enum ib_qp_state {
1218 	IB_QPS_RESET,
1219 	IB_QPS_INIT,
1220 	IB_QPS_RTR,
1221 	IB_QPS_RTS,
1222 	IB_QPS_SQD,
1223 	IB_QPS_SQE,
1224 	IB_QPS_ERR
1225 };
1226 
1227 enum ib_mig_state {
1228 	IB_MIG_MIGRATED,
1229 	IB_MIG_REARM,
1230 	IB_MIG_ARMED
1231 };
1232 
1233 enum ib_mw_type {
1234 	IB_MW_TYPE_1 = 1,
1235 	IB_MW_TYPE_2 = 2
1236 };
1237 
1238 struct ib_qp_attr {
1239 	enum ib_qp_state	qp_state;
1240 	enum ib_qp_state	cur_qp_state;
1241 	enum ib_mtu		path_mtu;
1242 	enum ib_mig_state	path_mig_state;
1243 	u32			qkey;
1244 	u32			rq_psn;
1245 	u32			sq_psn;
1246 	u32			dest_qp_num;
1247 	int			qp_access_flags;
1248 	struct ib_qp_cap	cap;
1249 	struct rdma_ah_attr	ah_attr;
1250 	struct rdma_ah_attr	alt_ah_attr;
1251 	u16			pkey_index;
1252 	u16			alt_pkey_index;
1253 	u8			en_sqd_async_notify;
1254 	u8			sq_draining;
1255 	u8			max_rd_atomic;
1256 	u8			max_dest_rd_atomic;
1257 	u8			min_rnr_timer;
1258 	u8			port_num;
1259 	u8			timeout;
1260 	u8			retry_cnt;
1261 	u8			rnr_retry;
1262 	u8			alt_port_num;
1263 	u8			alt_timeout;
1264 	u32			rate_limit;
1265 };
1266 
1267 enum ib_wr_opcode {
1268 	/* These are shared with userspace */
1269 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1270 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1271 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1272 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1273 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1274 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1275 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1276 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1277 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1278 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1279 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1280 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1281 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1282 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1283 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1284 
1285 	/* These are kernel only and can not be issued by userspace */
1286 	IB_WR_REG_MR = 0x20,
1287 	IB_WR_REG_MR_INTEGRITY,
1288 
1289 	/* reserve values for low level drivers' internal use.
1290 	 * These values will not be used at all in the ib core layer.
1291 	 */
1292 	IB_WR_RESERVED1 = 0xf0,
1293 	IB_WR_RESERVED2,
1294 	IB_WR_RESERVED3,
1295 	IB_WR_RESERVED4,
1296 	IB_WR_RESERVED5,
1297 	IB_WR_RESERVED6,
1298 	IB_WR_RESERVED7,
1299 	IB_WR_RESERVED8,
1300 	IB_WR_RESERVED9,
1301 	IB_WR_RESERVED10,
1302 };
1303 
1304 enum ib_send_flags {
1305 	IB_SEND_FENCE		= 1,
1306 	IB_SEND_SIGNALED	= (1<<1),
1307 	IB_SEND_SOLICITED	= (1<<2),
1308 	IB_SEND_INLINE		= (1<<3),
1309 	IB_SEND_IP_CSUM		= (1<<4),
1310 
1311 	/* reserve bits 26-31 for low level drivers' internal use */
1312 	IB_SEND_RESERVED_START	= (1 << 26),
1313 	IB_SEND_RESERVED_END	= (1 << 31),
1314 };
1315 
1316 struct ib_sge {
1317 	u64	addr;
1318 	u32	length;
1319 	u32	lkey;
1320 };
1321 
1322 struct ib_cqe {
1323 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1324 };
1325 
1326 struct ib_send_wr {
1327 	struct ib_send_wr      *next;
1328 	union {
1329 		u64		wr_id;
1330 		struct ib_cqe	*wr_cqe;
1331 	};
1332 	struct ib_sge	       *sg_list;
1333 	int			num_sge;
1334 	enum ib_wr_opcode	opcode;
1335 	int			send_flags;
1336 	union {
1337 		__be32		imm_data;
1338 		u32		invalidate_rkey;
1339 	} ex;
1340 };
1341 
1342 struct ib_rdma_wr {
1343 	struct ib_send_wr	wr;
1344 	u64			remote_addr;
1345 	u32			rkey;
1346 };
1347 
rdma_wr(const struct ib_send_wr * wr)1348 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1349 {
1350 	return container_of(wr, struct ib_rdma_wr, wr);
1351 }
1352 
1353 struct ib_atomic_wr {
1354 	struct ib_send_wr	wr;
1355 	u64			remote_addr;
1356 	u64			compare_add;
1357 	u64			swap;
1358 	u64			compare_add_mask;
1359 	u64			swap_mask;
1360 	u32			rkey;
1361 };
1362 
atomic_wr(const struct ib_send_wr * wr)1363 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1364 {
1365 	return container_of(wr, struct ib_atomic_wr, wr);
1366 }
1367 
1368 struct ib_ud_wr {
1369 	struct ib_send_wr	wr;
1370 	struct ib_ah		*ah;
1371 	void			*header;
1372 	int			hlen;
1373 	int			mss;
1374 	u32			remote_qpn;
1375 	u32			remote_qkey;
1376 	u16			pkey_index; /* valid for GSI only */
1377 	u8			port_num;   /* valid for DR SMPs on switch only */
1378 };
1379 
ud_wr(const struct ib_send_wr * wr)1380 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1381 {
1382 	return container_of(wr, struct ib_ud_wr, wr);
1383 }
1384 
1385 struct ib_reg_wr {
1386 	struct ib_send_wr	wr;
1387 	struct ib_mr		*mr;
1388 	u32			key;
1389 	int			access;
1390 };
1391 
reg_wr(const struct ib_send_wr * wr)1392 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1393 {
1394 	return container_of(wr, struct ib_reg_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 = IB_UVERBS_ACCESS_LOCAL_WRITE,
1409 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1410 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1411 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1412 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1413 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1414 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1415 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1416 
1417 	IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1418 };
1419 
1420 /*
1421  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1422  * are hidden here instead of a uapi header!
1423  */
1424 enum ib_mr_rereg_flags {
1425 	IB_MR_REREG_TRANS	= 1,
1426 	IB_MR_REREG_PD		= (1<<1),
1427 	IB_MR_REREG_ACCESS	= (1<<2),
1428 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1429 };
1430 
1431 struct ib_fmr_attr {
1432 	int	max_pages;
1433 	int	max_maps;
1434 	u8	page_shift;
1435 };
1436 
1437 struct ib_umem;
1438 
1439 enum rdma_remove_reason {
1440 	/*
1441 	 * Userspace requested uobject deletion or initial try
1442 	 * to remove uobject via cleanup. Call could fail
1443 	 */
1444 	RDMA_REMOVE_DESTROY,
1445 	/* Context deletion. This call should delete the actual object itself */
1446 	RDMA_REMOVE_CLOSE,
1447 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1448 	RDMA_REMOVE_DRIVER_REMOVE,
1449 	/* uobj is being cleaned-up before being committed */
1450 	RDMA_REMOVE_ABORT,
1451 };
1452 
1453 struct ib_rdmacg_object {
1454 #ifdef CONFIG_CGROUP_RDMA
1455 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1456 #endif
1457 };
1458 
1459 struct ib_ucontext {
1460 	struct ib_device       *device;
1461 	struct ib_uverbs_file  *ufile;
1462 	/*
1463 	 * 'closing' can be read by the driver only during a destroy callback,
1464 	 * it is set when we are closing the file descriptor and indicates
1465 	 * that mm_sem may be locked.
1466 	 */
1467 	bool closing;
1468 
1469 	bool cleanup_retryable;
1470 
1471 	struct ib_rdmacg_object	cg_obj;
1472 	/*
1473 	 * Implementation details of the RDMA core, don't use in drivers:
1474 	 */
1475 	struct rdma_restrack_entry res;
1476 };
1477 
1478 struct ib_uobject {
1479 	u64			user_handle;	/* handle given to us by userspace */
1480 	/* ufile & ucontext owning this object */
1481 	struct ib_uverbs_file  *ufile;
1482 	/* FIXME, save memory: ufile->context == context */
1483 	struct ib_ucontext     *context;	/* associated user context */
1484 	void		       *object;		/* containing object */
1485 	struct list_head	list;		/* link to context's list */
1486 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1487 	int			id;		/* index into kernel idr */
1488 	struct kref		ref;
1489 	atomic_t		usecnt;		/* protects exclusive access */
1490 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1491 
1492 	const struct uverbs_api_object *uapi_object;
1493 };
1494 
1495 struct ib_udata {
1496 	const void __user *inbuf;
1497 	void __user *outbuf;
1498 	size_t       inlen;
1499 	size_t       outlen;
1500 };
1501 
1502 struct ib_pd {
1503 	u32			local_dma_lkey;
1504 	u32			flags;
1505 	struct ib_device       *device;
1506 	struct ib_uobject      *uobject;
1507 	atomic_t          	usecnt; /* count all resources */
1508 
1509 	u32			unsafe_global_rkey;
1510 
1511 	/*
1512 	 * Implementation details of the RDMA core, don't use in drivers:
1513 	 */
1514 	struct ib_mr	       *__internal_mr;
1515 	struct rdma_restrack_entry res;
1516 };
1517 
1518 struct ib_xrcd {
1519 	struct ib_device       *device;
1520 	atomic_t		usecnt; /* count all exposed resources */
1521 	struct inode	       *inode;
1522 
1523 	struct mutex		tgt_qp_mutex;
1524 	struct list_head	tgt_qp_list;
1525 };
1526 
1527 struct ib_ah {
1528 	struct ib_device	*device;
1529 	struct ib_pd		*pd;
1530 	struct ib_uobject	*uobject;
1531 	const struct ib_gid_attr *sgid_attr;
1532 	enum rdma_ah_attr_type	type;
1533 };
1534 
1535 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1536 
1537 enum ib_poll_context {
1538 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1539 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1540 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1541 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1542 };
1543 
1544 struct ib_cq {
1545 	struct ib_device       *device;
1546 	struct ib_uobject      *uobject;
1547 	ib_comp_handler   	comp_handler;
1548 	void                  (*event_handler)(struct ib_event *, void *);
1549 	void                   *cq_context;
1550 	int               	cqe;
1551 	atomic_t          	usecnt; /* count number of work queues */
1552 	enum ib_poll_context	poll_ctx;
1553 	struct ib_wc		*wc;
1554 	union {
1555 		struct irq_poll		iop;
1556 		struct work_struct	work;
1557 	};
1558 	struct workqueue_struct *comp_wq;
1559 	struct dim *dim;
1560 	/*
1561 	 * Implementation details of the RDMA core, don't use in drivers:
1562 	 */
1563 	struct rdma_restrack_entry res;
1564 };
1565 
1566 struct ib_srq {
1567 	struct ib_device       *device;
1568 	struct ib_pd	       *pd;
1569 	struct ib_uobject      *uobject;
1570 	void		      (*event_handler)(struct ib_event *, void *);
1571 	void		       *srq_context;
1572 	enum ib_srq_type	srq_type;
1573 	atomic_t		usecnt;
1574 
1575 	struct {
1576 		struct ib_cq   *cq;
1577 		union {
1578 			struct {
1579 				struct ib_xrcd *xrcd;
1580 				u32		srq_num;
1581 			} xrc;
1582 		};
1583 	} ext;
1584 };
1585 
1586 enum ib_raw_packet_caps {
1587 	/* Strip cvlan from incoming packet and report it in the matching work
1588 	 * completion is supported.
1589 	 */
1590 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1591 	/* Scatter FCS field of an incoming packet to host memory is supported.
1592 	 */
1593 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1594 	/* Checksum offloads are supported (for both send and receive). */
1595 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1596 	/* When a packet is received for an RQ with no receive WQEs, the
1597 	 * packet processing is delayed.
1598 	 */
1599 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1600 };
1601 
1602 enum ib_wq_type {
1603 	IB_WQT_RQ
1604 };
1605 
1606 enum ib_wq_state {
1607 	IB_WQS_RESET,
1608 	IB_WQS_RDY,
1609 	IB_WQS_ERR
1610 };
1611 
1612 struct ib_wq {
1613 	struct ib_device       *device;
1614 	struct ib_uobject      *uobject;
1615 	void		    *wq_context;
1616 	void		    (*event_handler)(struct ib_event *, void *);
1617 	struct ib_pd	       *pd;
1618 	struct ib_cq	       *cq;
1619 	u32		wq_num;
1620 	enum ib_wq_state       state;
1621 	enum ib_wq_type	wq_type;
1622 	atomic_t		usecnt;
1623 };
1624 
1625 enum ib_wq_flags {
1626 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1627 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1628 	IB_WQ_FLAGS_DELAY_DROP		= 1 << 2,
1629 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1630 };
1631 
1632 struct ib_wq_init_attr {
1633 	void		       *wq_context;
1634 	enum ib_wq_type	wq_type;
1635 	u32		max_wr;
1636 	u32		max_sge;
1637 	struct	ib_cq	       *cq;
1638 	void		    (*event_handler)(struct ib_event *, void *);
1639 	u32		create_flags; /* Use enum ib_wq_flags */
1640 };
1641 
1642 enum ib_wq_attr_mask {
1643 	IB_WQ_STATE		= 1 << 0,
1644 	IB_WQ_CUR_STATE		= 1 << 1,
1645 	IB_WQ_FLAGS		= 1 << 2,
1646 };
1647 
1648 struct ib_wq_attr {
1649 	enum	ib_wq_state	wq_state;
1650 	enum	ib_wq_state	curr_wq_state;
1651 	u32			flags; /* Use enum ib_wq_flags */
1652 	u32			flags_mask; /* Use enum ib_wq_flags */
1653 };
1654 
1655 struct ib_rwq_ind_table {
1656 	struct ib_device	*device;
1657 	struct ib_uobject      *uobject;
1658 	atomic_t		usecnt;
1659 	u32		ind_tbl_num;
1660 	u32		log_ind_tbl_size;
1661 	struct ib_wq	**ind_tbl;
1662 };
1663 
1664 struct ib_rwq_ind_table_init_attr {
1665 	u32		log_ind_tbl_size;
1666 	/* Each entry is a pointer to Receive Work Queue */
1667 	struct ib_wq	**ind_tbl;
1668 };
1669 
1670 enum port_pkey_state {
1671 	IB_PORT_PKEY_NOT_VALID = 0,
1672 	IB_PORT_PKEY_VALID = 1,
1673 	IB_PORT_PKEY_LISTED = 2,
1674 };
1675 
1676 struct ib_qp_security;
1677 
1678 struct ib_port_pkey {
1679 	enum port_pkey_state	state;
1680 	u16			pkey_index;
1681 	u8			port_num;
1682 	struct list_head	qp_list;
1683 	struct list_head	to_error_list;
1684 	struct ib_qp_security  *sec;
1685 };
1686 
1687 struct ib_ports_pkeys {
1688 	struct ib_port_pkey	main;
1689 	struct ib_port_pkey	alt;
1690 };
1691 
1692 struct ib_qp_security {
1693 	struct ib_qp	       *qp;
1694 	struct ib_device       *dev;
1695 	/* Hold this mutex when changing port and pkey settings. */
1696 	struct mutex		mutex;
1697 	struct ib_ports_pkeys  *ports_pkeys;
1698 	/* A list of all open shared QP handles.  Required to enforce security
1699 	 * properly for all users of a shared QP.
1700 	 */
1701 	struct list_head        shared_qp_list;
1702 	void                   *security;
1703 	bool			destroying;
1704 	atomic_t		error_list_count;
1705 	struct completion	error_complete;
1706 	int			error_comps_pending;
1707 };
1708 
1709 /*
1710  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1711  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1712  */
1713 struct ib_qp {
1714 	struct ib_device       *device;
1715 	struct ib_pd	       *pd;
1716 	struct ib_cq	       *send_cq;
1717 	struct ib_cq	       *recv_cq;
1718 	spinlock_t		mr_lock;
1719 	int			mrs_used;
1720 	struct list_head	rdma_mrs;
1721 	struct list_head	sig_mrs;
1722 	struct ib_srq	       *srq;
1723 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1724 	struct list_head	xrcd_list;
1725 
1726 	/* count times opened, mcast attaches, flow attaches */
1727 	atomic_t		usecnt;
1728 	struct list_head	open_list;
1729 	struct ib_qp           *real_qp;
1730 	struct ib_uobject      *uobject;
1731 	void                  (*event_handler)(struct ib_event *, void *);
1732 	void		       *qp_context;
1733 	/* sgid_attrs associated with the AV's */
1734 	const struct ib_gid_attr *av_sgid_attr;
1735 	const struct ib_gid_attr *alt_path_sgid_attr;
1736 	u32			qp_num;
1737 	u32			max_write_sge;
1738 	u32			max_read_sge;
1739 	enum ib_qp_type		qp_type;
1740 	struct ib_rwq_ind_table *rwq_ind_tbl;
1741 	struct ib_qp_security  *qp_sec;
1742 	u8			port;
1743 
1744 	bool			integrity_en;
1745 	/*
1746 	 * Implementation details of the RDMA core, don't use in drivers:
1747 	 */
1748 	struct rdma_restrack_entry     res;
1749 
1750 	/* The counter the qp is bind to */
1751 	struct rdma_counter    *counter;
1752 };
1753 
1754 struct ib_dm {
1755 	struct ib_device  *device;
1756 	u32		   length;
1757 	u32		   flags;
1758 	struct ib_uobject *uobject;
1759 	atomic_t	   usecnt;
1760 };
1761 
1762 struct ib_mr {
1763 	struct ib_device  *device;
1764 	struct ib_pd	  *pd;
1765 	u32		   lkey;
1766 	u32		   rkey;
1767 	u64		   iova;
1768 	u64		   length;
1769 	unsigned int	   page_size;
1770 	enum ib_mr_type	   type;
1771 	bool		   need_inval;
1772 	union {
1773 		struct ib_uobject	*uobject;	/* user */
1774 		struct list_head	qp_entry;	/* FR */
1775 	};
1776 
1777 	struct ib_dm      *dm;
1778 	struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1779 	/*
1780 	 * Implementation details of the RDMA core, don't use in drivers:
1781 	 */
1782 	struct rdma_restrack_entry res;
1783 };
1784 
1785 struct ib_mw {
1786 	struct ib_device	*device;
1787 	struct ib_pd		*pd;
1788 	struct ib_uobject	*uobject;
1789 	u32			rkey;
1790 	enum ib_mw_type         type;
1791 };
1792 
1793 struct ib_fmr {
1794 	struct ib_device	*device;
1795 	struct ib_pd		*pd;
1796 	struct list_head	list;
1797 	u32			lkey;
1798 	u32			rkey;
1799 };
1800 
1801 /* Supported steering options */
1802 enum ib_flow_attr_type {
1803 	/* steering according to rule specifications */
1804 	IB_FLOW_ATTR_NORMAL		= 0x0,
1805 	/* default unicast and multicast rule -
1806 	 * receive all Eth traffic which isn't steered to any QP
1807 	 */
1808 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1809 	/* default multicast rule -
1810 	 * receive all Eth multicast traffic which isn't steered to any QP
1811 	 */
1812 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1813 	/* sniffer rule - receive all port traffic */
1814 	IB_FLOW_ATTR_SNIFFER		= 0x3
1815 };
1816 
1817 /* Supported steering header types */
1818 enum ib_flow_spec_type {
1819 	/* L2 headers*/
1820 	IB_FLOW_SPEC_ETH		= 0x20,
1821 	IB_FLOW_SPEC_IB			= 0x22,
1822 	/* L3 header*/
1823 	IB_FLOW_SPEC_IPV4		= 0x30,
1824 	IB_FLOW_SPEC_IPV6		= 0x31,
1825 	IB_FLOW_SPEC_ESP                = 0x34,
1826 	/* L4 headers*/
1827 	IB_FLOW_SPEC_TCP		= 0x40,
1828 	IB_FLOW_SPEC_UDP		= 0x41,
1829 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1830 	IB_FLOW_SPEC_GRE		= 0x51,
1831 	IB_FLOW_SPEC_MPLS		= 0x60,
1832 	IB_FLOW_SPEC_INNER		= 0x100,
1833 	/* Actions */
1834 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1835 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1836 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1837 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1838 };
1839 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1840 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1841 
1842 /* Flow steering rule priority is set according to it's domain.
1843  * Lower domain value means higher priority.
1844  */
1845 enum ib_flow_domain {
1846 	IB_FLOW_DOMAIN_USER,
1847 	IB_FLOW_DOMAIN_ETHTOOL,
1848 	IB_FLOW_DOMAIN_RFS,
1849 	IB_FLOW_DOMAIN_NIC,
1850 	IB_FLOW_DOMAIN_NUM /* Must be last */
1851 };
1852 
1853 enum ib_flow_flags {
1854 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1855 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1856 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1857 };
1858 
1859 struct ib_flow_eth_filter {
1860 	u8	dst_mac[6];
1861 	u8	src_mac[6];
1862 	__be16	ether_type;
1863 	__be16	vlan_tag;
1864 	/* Must be last */
1865 	u8	real_sz[0];
1866 };
1867 
1868 struct ib_flow_spec_eth {
1869 	u32			  type;
1870 	u16			  size;
1871 	struct ib_flow_eth_filter val;
1872 	struct ib_flow_eth_filter mask;
1873 };
1874 
1875 struct ib_flow_ib_filter {
1876 	__be16 dlid;
1877 	__u8   sl;
1878 	/* Must be last */
1879 	u8	real_sz[0];
1880 };
1881 
1882 struct ib_flow_spec_ib {
1883 	u32			 type;
1884 	u16			 size;
1885 	struct ib_flow_ib_filter val;
1886 	struct ib_flow_ib_filter mask;
1887 };
1888 
1889 /* IPv4 header flags */
1890 enum ib_ipv4_flags {
1891 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1892 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1893 				    last have this flag set */
1894 };
1895 
1896 struct ib_flow_ipv4_filter {
1897 	__be32	src_ip;
1898 	__be32	dst_ip;
1899 	u8	proto;
1900 	u8	tos;
1901 	u8	ttl;
1902 	u8	flags;
1903 	/* Must be last */
1904 	u8	real_sz[0];
1905 };
1906 
1907 struct ib_flow_spec_ipv4 {
1908 	u32			   type;
1909 	u16			   size;
1910 	struct ib_flow_ipv4_filter val;
1911 	struct ib_flow_ipv4_filter mask;
1912 };
1913 
1914 struct ib_flow_ipv6_filter {
1915 	u8	src_ip[16];
1916 	u8	dst_ip[16];
1917 	__be32	flow_label;
1918 	u8	next_hdr;
1919 	u8	traffic_class;
1920 	u8	hop_limit;
1921 	/* Must be last */
1922 	u8	real_sz[0];
1923 };
1924 
1925 struct ib_flow_spec_ipv6 {
1926 	u32			   type;
1927 	u16			   size;
1928 	struct ib_flow_ipv6_filter val;
1929 	struct ib_flow_ipv6_filter mask;
1930 };
1931 
1932 struct ib_flow_tcp_udp_filter {
1933 	__be16	dst_port;
1934 	__be16	src_port;
1935 	/* Must be last */
1936 	u8	real_sz[0];
1937 };
1938 
1939 struct ib_flow_spec_tcp_udp {
1940 	u32			      type;
1941 	u16			      size;
1942 	struct ib_flow_tcp_udp_filter val;
1943 	struct ib_flow_tcp_udp_filter mask;
1944 };
1945 
1946 struct ib_flow_tunnel_filter {
1947 	__be32	tunnel_id;
1948 	u8	real_sz[0];
1949 };
1950 
1951 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1952  * the tunnel_id from val has the vni value
1953  */
1954 struct ib_flow_spec_tunnel {
1955 	u32			      type;
1956 	u16			      size;
1957 	struct ib_flow_tunnel_filter  val;
1958 	struct ib_flow_tunnel_filter  mask;
1959 };
1960 
1961 struct ib_flow_esp_filter {
1962 	__be32	spi;
1963 	__be32  seq;
1964 	/* Must be last */
1965 	u8	real_sz[0];
1966 };
1967 
1968 struct ib_flow_spec_esp {
1969 	u32                           type;
1970 	u16			      size;
1971 	struct ib_flow_esp_filter     val;
1972 	struct ib_flow_esp_filter     mask;
1973 };
1974 
1975 struct ib_flow_gre_filter {
1976 	__be16 c_ks_res0_ver;
1977 	__be16 protocol;
1978 	__be32 key;
1979 	/* Must be last */
1980 	u8	real_sz[0];
1981 };
1982 
1983 struct ib_flow_spec_gre {
1984 	u32                           type;
1985 	u16			      size;
1986 	struct ib_flow_gre_filter     val;
1987 	struct ib_flow_gre_filter     mask;
1988 };
1989 
1990 struct ib_flow_mpls_filter {
1991 	__be32 tag;
1992 	/* Must be last */
1993 	u8	real_sz[0];
1994 };
1995 
1996 struct ib_flow_spec_mpls {
1997 	u32                           type;
1998 	u16			      size;
1999 	struct ib_flow_mpls_filter     val;
2000 	struct ib_flow_mpls_filter     mask;
2001 };
2002 
2003 struct ib_flow_spec_action_tag {
2004 	enum ib_flow_spec_type	      type;
2005 	u16			      size;
2006 	u32                           tag_id;
2007 };
2008 
2009 struct ib_flow_spec_action_drop {
2010 	enum ib_flow_spec_type	      type;
2011 	u16			      size;
2012 };
2013 
2014 struct ib_flow_spec_action_handle {
2015 	enum ib_flow_spec_type	      type;
2016 	u16			      size;
2017 	struct ib_flow_action	     *act;
2018 };
2019 
2020 enum ib_counters_description {
2021 	IB_COUNTER_PACKETS,
2022 	IB_COUNTER_BYTES,
2023 };
2024 
2025 struct ib_flow_spec_action_count {
2026 	enum ib_flow_spec_type type;
2027 	u16 size;
2028 	struct ib_counters *counters;
2029 };
2030 
2031 union ib_flow_spec {
2032 	struct {
2033 		u32			type;
2034 		u16			size;
2035 	};
2036 	struct ib_flow_spec_eth		eth;
2037 	struct ib_flow_spec_ib		ib;
2038 	struct ib_flow_spec_ipv4        ipv4;
2039 	struct ib_flow_spec_tcp_udp	tcp_udp;
2040 	struct ib_flow_spec_ipv6        ipv6;
2041 	struct ib_flow_spec_tunnel      tunnel;
2042 	struct ib_flow_spec_esp		esp;
2043 	struct ib_flow_spec_gre		gre;
2044 	struct ib_flow_spec_mpls	mpls;
2045 	struct ib_flow_spec_action_tag  flow_tag;
2046 	struct ib_flow_spec_action_drop drop;
2047 	struct ib_flow_spec_action_handle action;
2048 	struct ib_flow_spec_action_count flow_count;
2049 };
2050 
2051 struct ib_flow_attr {
2052 	enum ib_flow_attr_type type;
2053 	u16	     size;
2054 	u16	     priority;
2055 	u32	     flags;
2056 	u8	     num_of_specs;
2057 	u8	     port;
2058 	union ib_flow_spec flows[];
2059 };
2060 
2061 struct ib_flow {
2062 	struct ib_qp		*qp;
2063 	struct ib_device	*device;
2064 	struct ib_uobject	*uobject;
2065 };
2066 
2067 enum ib_flow_action_type {
2068 	IB_FLOW_ACTION_UNSPECIFIED,
2069 	IB_FLOW_ACTION_ESP = 1,
2070 };
2071 
2072 struct ib_flow_action_attrs_esp_keymats {
2073 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2074 	union {
2075 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2076 	} keymat;
2077 };
2078 
2079 struct ib_flow_action_attrs_esp_replays {
2080 	enum ib_uverbs_flow_action_esp_replay			protocol;
2081 	union {
2082 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2083 	} replay;
2084 };
2085 
2086 enum ib_flow_action_attrs_esp_flags {
2087 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2088 	 * This is done in order to share the same flags between user-space and
2089 	 * kernel and spare an unnecessary translation.
2090 	 */
2091 
2092 	/* Kernel flags */
2093 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2094 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2095 };
2096 
2097 struct ib_flow_spec_list {
2098 	struct ib_flow_spec_list	*next;
2099 	union ib_flow_spec		spec;
2100 };
2101 
2102 struct ib_flow_action_attrs_esp {
2103 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2104 	struct ib_flow_action_attrs_esp_replays		*replay;
2105 	struct ib_flow_spec_list			*encap;
2106 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2107 	 * Value of 0 is a valid value.
2108 	 */
2109 	u32						esn;
2110 	u32						spi;
2111 	u32						seq;
2112 	u32						tfc_pad;
2113 	/* Use enum ib_flow_action_attrs_esp_flags */
2114 	u64						flags;
2115 	u64						hard_limit_pkts;
2116 };
2117 
2118 struct ib_flow_action {
2119 	struct ib_device		*device;
2120 	struct ib_uobject		*uobject;
2121 	enum ib_flow_action_type	type;
2122 	atomic_t			usecnt;
2123 };
2124 
2125 struct ib_mad_hdr;
2126 struct ib_grh;
2127 
2128 enum ib_process_mad_flags {
2129 	IB_MAD_IGNORE_MKEY	= 1,
2130 	IB_MAD_IGNORE_BKEY	= 2,
2131 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2132 };
2133 
2134 enum ib_mad_result {
2135 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2136 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2137 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2138 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2139 };
2140 
2141 struct ib_port_cache {
2142 	u64		      subnet_prefix;
2143 	struct ib_pkey_cache  *pkey;
2144 	struct ib_gid_table   *gid;
2145 	u8                     lmc;
2146 	enum ib_port_state     port_state;
2147 };
2148 
2149 struct ib_cache {
2150 	rwlock_t                lock;
2151 };
2152 
2153 struct ib_port_immutable {
2154 	int                           pkey_tbl_len;
2155 	int                           gid_tbl_len;
2156 	u32                           core_cap_flags;
2157 	u32                           max_mad_size;
2158 };
2159 
2160 struct ib_port_data {
2161 	struct ib_device *ib_dev;
2162 
2163 	struct ib_port_immutable immutable;
2164 
2165 	spinlock_t pkey_list_lock;
2166 	struct list_head pkey_list;
2167 
2168 	struct ib_port_cache cache;
2169 
2170 	spinlock_t netdev_lock;
2171 	struct net_device __rcu *netdev;
2172 	struct hlist_node ndev_hash_link;
2173 	struct rdma_port_counter port_counter;
2174 	struct rdma_hw_stats *hw_stats;
2175 };
2176 
2177 /* rdma netdev type - specifies protocol type */
2178 enum rdma_netdev_t {
2179 	RDMA_NETDEV_OPA_VNIC,
2180 	RDMA_NETDEV_IPOIB,
2181 };
2182 
2183 /**
2184  * struct rdma_netdev - rdma netdev
2185  * For cases where netstack interfacing is required.
2186  */
2187 struct rdma_netdev {
2188 	void              *clnt_priv;
2189 	struct ib_device  *hca;
2190 	u8                 port_num;
2191 
2192 	/*
2193 	 * cleanup function must be specified.
2194 	 * FIXME: This is only used for OPA_VNIC and that usage should be
2195 	 * removed too.
2196 	 */
2197 	void (*free_rdma_netdev)(struct net_device *netdev);
2198 
2199 	/* control functions */
2200 	void (*set_id)(struct net_device *netdev, int id);
2201 	/* send packet */
2202 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2203 		    struct ib_ah *address, u32 dqpn);
2204 	/* multicast */
2205 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2206 			    union ib_gid *gid, u16 mlid,
2207 			    int set_qkey, u32 qkey);
2208 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2209 			    union ib_gid *gid, u16 mlid);
2210 };
2211 
2212 struct rdma_netdev_alloc_params {
2213 	size_t sizeof_priv;
2214 	unsigned int txqs;
2215 	unsigned int rxqs;
2216 	void *param;
2217 
2218 	int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2219 				      struct net_device *netdev, void *param);
2220 };
2221 
2222 struct ib_counters {
2223 	struct ib_device	*device;
2224 	struct ib_uobject	*uobject;
2225 	/* num of objects attached */
2226 	atomic_t	usecnt;
2227 };
2228 
2229 struct ib_counters_read_attr {
2230 	u64	*counters_buff;
2231 	u32	ncounters;
2232 	u32	flags; /* use enum ib_read_counters_flags */
2233 };
2234 
2235 struct uverbs_attr_bundle;
2236 struct iw_cm_id;
2237 struct iw_cm_conn_param;
2238 
2239 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member)                      \
2240 	.size_##ib_struct =                                                    \
2241 		(sizeof(struct drv_struct) +                                   \
2242 		 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) +      \
2243 		 BUILD_BUG_ON_ZERO(                                            \
2244 			 !__same_type(((struct drv_struct *)NULL)->member,     \
2245 				      struct ib_struct)))
2246 
2247 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp)                         \
2248 	((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2249 
2250 #define rdma_zalloc_drv_obj(ib_dev, ib_type)                                   \
2251 	rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2252 
2253 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2254 
2255 /**
2256  * struct ib_device_ops - InfiniBand device operations
2257  * This structure defines all the InfiniBand device operations, providers will
2258  * need to define the supported operations, otherwise they will be set to null.
2259  */
2260 struct ib_device_ops {
2261 	struct module *owner;
2262 	enum rdma_driver_id driver_id;
2263 	u32 uverbs_abi_ver;
2264 	unsigned int uverbs_no_driver_id_binding:1;
2265 
2266 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2267 			 const struct ib_send_wr **bad_send_wr);
2268 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2269 			 const struct ib_recv_wr **bad_recv_wr);
2270 	void (*drain_rq)(struct ib_qp *qp);
2271 	void (*drain_sq)(struct ib_qp *qp);
2272 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2273 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2274 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2275 	int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2276 	int (*post_srq_recv)(struct ib_srq *srq,
2277 			     const struct ib_recv_wr *recv_wr,
2278 			     const struct ib_recv_wr **bad_recv_wr);
2279 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2280 			   u8 port_num, const struct ib_wc *in_wc,
2281 			   const struct ib_grh *in_grh,
2282 			   const struct ib_mad_hdr *in_mad, size_t in_mad_size,
2283 			   struct ib_mad_hdr *out_mad, size_t *out_mad_size,
2284 			   u16 *out_mad_pkey_index);
2285 	int (*query_device)(struct ib_device *device,
2286 			    struct ib_device_attr *device_attr,
2287 			    struct ib_udata *udata);
2288 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2289 			     struct ib_device_modify *device_modify);
2290 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2291 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2292 						     int comp_vector);
2293 	int (*query_port)(struct ib_device *device, u8 port_num,
2294 			  struct ib_port_attr *port_attr);
2295 	int (*modify_port)(struct ib_device *device, u8 port_num,
2296 			   int port_modify_mask,
2297 			   struct ib_port_modify *port_modify);
2298 	/**
2299 	 * The following mandatory functions are used only at device
2300 	 * registration.  Keep functions such as these at the end of this
2301 	 * structure to avoid cache line misses when accessing struct ib_device
2302 	 * in fast paths.
2303 	 */
2304 	int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2305 				  struct ib_port_immutable *immutable);
2306 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2307 					       u8 port_num);
2308 	/**
2309 	 * When calling get_netdev, the HW vendor's driver should return the
2310 	 * net device of device @device at port @port_num or NULL if such
2311 	 * a net device doesn't exist. The vendor driver should call dev_hold
2312 	 * on this net device. The HW vendor's device driver must guarantee
2313 	 * that this function returns NULL before the net device has finished
2314 	 * NETDEV_UNREGISTER state.
2315 	 */
2316 	struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2317 	/**
2318 	 * rdma netdev operation
2319 	 *
2320 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2321 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2322 	 */
2323 	struct net_device *(*alloc_rdma_netdev)(
2324 		struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2325 		const char *name, unsigned char name_assign_type,
2326 		void (*setup)(struct net_device *));
2327 
2328 	int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2329 				      enum rdma_netdev_t type,
2330 				      struct rdma_netdev_alloc_params *params);
2331 	/**
2332 	 * query_gid should be return GID value for @device, when @port_num
2333 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2334 	 * is RoCE link layer.
2335 	 */
2336 	int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2337 			 union ib_gid *gid);
2338 	/**
2339 	 * When calling add_gid, the HW vendor's driver should add the gid
2340 	 * of device of port at gid index available at @attr. Meta-info of
2341 	 * that gid (for example, the network device related to this gid) is
2342 	 * available at @attr. @context allows the HW vendor driver to store
2343 	 * extra information together with a GID entry. The HW vendor driver may
2344 	 * allocate memory to contain this information and store it in @context
2345 	 * when a new GID entry is written to. Params are consistent until the
2346 	 * next call of add_gid or delete_gid. The function should return 0 on
2347 	 * success or error otherwise. The function could be called
2348 	 * concurrently for different ports. This function is only called when
2349 	 * roce_gid_table is used.
2350 	 */
2351 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2352 	/**
2353 	 * When calling del_gid, the HW vendor's driver should delete the
2354 	 * gid of device @device at gid index gid_index of port port_num
2355 	 * available in @attr.
2356 	 * Upon the deletion of a GID entry, the HW vendor must free any
2357 	 * allocated memory. The caller will clear @context afterwards.
2358 	 * This function is only called when roce_gid_table is used.
2359 	 */
2360 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2361 	int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2362 			  u16 *pkey);
2363 	int (*alloc_ucontext)(struct ib_ucontext *context,
2364 			      struct ib_udata *udata);
2365 	void (*dealloc_ucontext)(struct ib_ucontext *context);
2366 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2367 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2368 	int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2369 	void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2370 	int (*create_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr,
2371 			 u32 flags, struct ib_udata *udata);
2372 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2373 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2374 	void (*destroy_ah)(struct ib_ah *ah, u32 flags);
2375 	int (*create_srq)(struct ib_srq *srq,
2376 			  struct ib_srq_init_attr *srq_init_attr,
2377 			  struct ib_udata *udata);
2378 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2379 			  enum ib_srq_attr_mask srq_attr_mask,
2380 			  struct ib_udata *udata);
2381 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2382 	void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2383 	struct ib_qp *(*create_qp)(struct ib_pd *pd,
2384 				   struct ib_qp_init_attr *qp_init_attr,
2385 				   struct ib_udata *udata);
2386 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2387 			 int qp_attr_mask, struct ib_udata *udata);
2388 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2389 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2390 	int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2391 	int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2392 			 struct ib_udata *udata);
2393 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2394 	void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2395 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2396 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2397 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2398 				     u64 virt_addr, int mr_access_flags,
2399 				     struct ib_udata *udata);
2400 	int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2401 			     u64 virt_addr, int mr_access_flags,
2402 			     struct ib_pd *pd, struct ib_udata *udata);
2403 	int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2404 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2405 				  u32 max_num_sg, struct ib_udata *udata);
2406 	struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2407 					    u32 max_num_data_sg,
2408 					    u32 max_num_meta_sg);
2409 	int (*advise_mr)(struct ib_pd *pd,
2410 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2411 			 struct ib_sge *sg_list, u32 num_sge,
2412 			 struct uverbs_attr_bundle *attrs);
2413 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2414 			 unsigned int *sg_offset);
2415 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2416 			       struct ib_mr_status *mr_status);
2417 	struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2418 				  struct ib_udata *udata);
2419 	int (*dealloc_mw)(struct ib_mw *mw);
2420 	struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2421 				    struct ib_fmr_attr *fmr_attr);
2422 	int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2423 			    u64 iova);
2424 	int (*unmap_fmr)(struct list_head *fmr_list);
2425 	int (*dealloc_fmr)(struct ib_fmr *fmr);
2426 	void (*invalidate_range)(struct ib_umem_odp *umem_odp,
2427 				 unsigned long start, unsigned long end);
2428 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2429 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2430 	struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2431 				      struct ib_udata *udata);
2432 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2433 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2434 				       struct ib_flow_attr *flow_attr,
2435 				       int domain, struct ib_udata *udata);
2436 	int (*destroy_flow)(struct ib_flow *flow_id);
2437 	struct ib_flow_action *(*create_flow_action_esp)(
2438 		struct ib_device *device,
2439 		const struct ib_flow_action_attrs_esp *attr,
2440 		struct uverbs_attr_bundle *attrs);
2441 	int (*destroy_flow_action)(struct ib_flow_action *action);
2442 	int (*modify_flow_action_esp)(
2443 		struct ib_flow_action *action,
2444 		const struct ib_flow_action_attrs_esp *attr,
2445 		struct uverbs_attr_bundle *attrs);
2446 	int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2447 				 int state);
2448 	int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2449 			     struct ifla_vf_info *ivf);
2450 	int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2451 			    struct ifla_vf_stats *stats);
2452 	int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2453 			   int type);
2454 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2455 				   struct ib_wq_init_attr *init_attr,
2456 				   struct ib_udata *udata);
2457 	void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2458 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2459 			 u32 wq_attr_mask, struct ib_udata *udata);
2460 	struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2461 		struct ib_device *device,
2462 		struct ib_rwq_ind_table_init_attr *init_attr,
2463 		struct ib_udata *udata);
2464 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2465 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2466 				  struct ib_ucontext *context,
2467 				  struct ib_dm_alloc_attr *attr,
2468 				  struct uverbs_attr_bundle *attrs);
2469 	int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2470 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2471 				   struct ib_dm_mr_attr *attr,
2472 				   struct uverbs_attr_bundle *attrs);
2473 	struct ib_counters *(*create_counters)(
2474 		struct ib_device *device, struct uverbs_attr_bundle *attrs);
2475 	int (*destroy_counters)(struct ib_counters *counters);
2476 	int (*read_counters)(struct ib_counters *counters,
2477 			     struct ib_counters_read_attr *counters_read_attr,
2478 			     struct uverbs_attr_bundle *attrs);
2479 	int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2480 			    int data_sg_nents, unsigned int *data_sg_offset,
2481 			    struct scatterlist *meta_sg, int meta_sg_nents,
2482 			    unsigned int *meta_sg_offset);
2483 
2484 	/**
2485 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2486 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2487 	 *   core when the device is removed.  A lifespan of -1 in the return
2488 	 *   struct tells the core to set a default lifespan.
2489 	 */
2490 	struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2491 						u8 port_num);
2492 	/**
2493 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2494 	 * @index - The index in the value array we wish to have updated, or
2495 	 *   num_counters if we want all stats updated
2496 	 * Return codes -
2497 	 *   < 0 - Error, no counters updated
2498 	 *   index - Updated the single counter pointed to by index
2499 	 *   num_counters - Updated all counters (will reset the timestamp
2500 	 *     and prevent further calls for lifespan milliseconds)
2501 	 * Drivers are allowed to update all counters in leiu of just the
2502 	 *   one given in index at their option
2503 	 */
2504 	int (*get_hw_stats)(struct ib_device *device,
2505 			    struct rdma_hw_stats *stats, u8 port, int index);
2506 	/*
2507 	 * This function is called once for each port when a ib device is
2508 	 * registered.
2509 	 */
2510 	int (*init_port)(struct ib_device *device, u8 port_num,
2511 			 struct kobject *port_sysfs);
2512 	/**
2513 	 * Allows rdma drivers to add their own restrack attributes.
2514 	 */
2515 	int (*fill_res_entry)(struct sk_buff *msg,
2516 			      struct rdma_restrack_entry *entry);
2517 
2518 	/* Device lifecycle callbacks */
2519 	/*
2520 	 * Called after the device becomes registered, before clients are
2521 	 * attached
2522 	 */
2523 	int (*enable_driver)(struct ib_device *dev);
2524 	/*
2525 	 * This is called as part of ib_dealloc_device().
2526 	 */
2527 	void (*dealloc_driver)(struct ib_device *dev);
2528 
2529 	/* iWarp CM callbacks */
2530 	void (*iw_add_ref)(struct ib_qp *qp);
2531 	void (*iw_rem_ref)(struct ib_qp *qp);
2532 	struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2533 	int (*iw_connect)(struct iw_cm_id *cm_id,
2534 			  struct iw_cm_conn_param *conn_param);
2535 	int (*iw_accept)(struct iw_cm_id *cm_id,
2536 			 struct iw_cm_conn_param *conn_param);
2537 	int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2538 			 u8 pdata_len);
2539 	int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2540 	int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2541 	/**
2542 	 * counter_bind_qp - Bind a QP to a counter.
2543 	 * @counter - The counter to be bound. If counter->id is zero then
2544 	 *   the driver needs to allocate a new counter and set counter->id
2545 	 */
2546 	int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2547 	/**
2548 	 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2549 	 *   counter and bind it onto the default one
2550 	 */
2551 	int (*counter_unbind_qp)(struct ib_qp *qp);
2552 	/**
2553 	 * counter_dealloc -De-allocate the hw counter
2554 	 */
2555 	int (*counter_dealloc)(struct rdma_counter *counter);
2556 	/**
2557 	 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2558 	 * the driver initialized data.
2559 	 */
2560 	struct rdma_hw_stats *(*counter_alloc_stats)(
2561 		struct rdma_counter *counter);
2562 	/**
2563 	 * counter_update_stats - Query the stats value of this counter
2564 	 */
2565 	int (*counter_update_stats)(struct rdma_counter *counter);
2566 
2567 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
2568 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
2569 	DECLARE_RDMA_OBJ_SIZE(ib_pd);
2570 	DECLARE_RDMA_OBJ_SIZE(ib_srq);
2571 	DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2572 };
2573 
2574 struct ib_core_device {
2575 	/* device must be the first element in structure until,
2576 	 * union of ib_core_device and device exists in ib_device.
2577 	 */
2578 	struct device dev;
2579 	possible_net_t rdma_net;
2580 	struct kobject *ports_kobj;
2581 	struct list_head port_list;
2582 	struct ib_device *owner; /* reach back to owner ib_device */
2583 };
2584 
2585 struct rdma_restrack_root;
2586 struct ib_device {
2587 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2588 	struct device                *dma_device;
2589 	struct ib_device_ops	     ops;
2590 	char                          name[IB_DEVICE_NAME_MAX];
2591 	struct rcu_head rcu_head;
2592 
2593 	struct list_head              event_handler_list;
2594 	/* Protects event_handler_list */
2595 	struct rw_semaphore event_handler_rwsem;
2596 
2597 	/* Protects QP's event_handler calls and open_qp list */
2598 	spinlock_t event_handler_lock;
2599 
2600 	struct rw_semaphore	      client_data_rwsem;
2601 	struct xarray                 client_data;
2602 	struct mutex                  unregistration_lock;
2603 
2604 	struct ib_cache               cache;
2605 	/**
2606 	 * port_data is indexed by port number
2607 	 */
2608 	struct ib_port_data *port_data;
2609 
2610 	int			      num_comp_vectors;
2611 
2612 	union {
2613 		struct device		dev;
2614 		struct ib_core_device	coredev;
2615 	};
2616 
2617 	/* First group for device attributes,
2618 	 * Second group for driver provided attributes (optional).
2619 	 * It is NULL terminated array.
2620 	 */
2621 	const struct attribute_group	*groups[3];
2622 
2623 	u64			     uverbs_cmd_mask;
2624 	u64			     uverbs_ex_cmd_mask;
2625 
2626 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2627 	__be64			     node_guid;
2628 	u32			     local_dma_lkey;
2629 	u16                          is_switch:1;
2630 	/* Indicates kernel verbs support, should not be used in drivers */
2631 	u16                          kverbs_provider:1;
2632 	/* CQ adaptive moderation (RDMA DIM) */
2633 	u16                          use_cq_dim:1;
2634 	u8                           node_type;
2635 	u8                           phys_port_cnt;
2636 	struct ib_device_attr        attrs;
2637 	struct attribute_group	     *hw_stats_ag;
2638 	struct rdma_hw_stats         *hw_stats;
2639 
2640 #ifdef CONFIG_CGROUP_RDMA
2641 	struct rdmacg_device         cg_device;
2642 #endif
2643 
2644 	u32                          index;
2645 	struct rdma_restrack_root *res;
2646 
2647 	const struct uapi_definition   *driver_def;
2648 
2649 	/*
2650 	 * Positive refcount indicates that the device is currently
2651 	 * registered and cannot be unregistered.
2652 	 */
2653 	refcount_t refcount;
2654 	struct completion unreg_completion;
2655 	struct work_struct unregistration_work;
2656 
2657 	const struct rdma_link_ops *link_ops;
2658 
2659 	/* Protects compat_devs xarray modifications */
2660 	struct mutex compat_devs_mutex;
2661 	/* Maintains compat devices for each net namespace */
2662 	struct xarray compat_devs;
2663 
2664 	/* Used by iWarp CM */
2665 	char iw_ifname[IFNAMSIZ];
2666 	u32 iw_driver_flags;
2667 };
2668 
2669 struct ib_client_nl_info;
2670 struct ib_client {
2671 	const char *name;
2672 	void (*add)   (struct ib_device *);
2673 	void (*remove)(struct ib_device *, void *client_data);
2674 	void (*rename)(struct ib_device *dev, void *client_data);
2675 	int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2676 			   struct ib_client_nl_info *res);
2677 	int (*get_global_nl_info)(struct ib_client_nl_info *res);
2678 
2679 	/* Returns the net_dev belonging to this ib_client and matching the
2680 	 * given parameters.
2681 	 * @dev:	 An RDMA device that the net_dev use for communication.
2682 	 * @port:	 A physical port number on the RDMA device.
2683 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2684 	 * @gid:	 A GID that the net_dev uses to communicate.
2685 	 * @addr:	 An IP address the net_dev is configured with.
2686 	 * @client_data: The device's client data set by ib_set_client_data().
2687 	 *
2688 	 * An ib_client that implements a net_dev on top of RDMA devices
2689 	 * (such as IP over IB) should implement this callback, allowing the
2690 	 * rdma_cm module to find the right net_dev for a given request.
2691 	 *
2692 	 * The caller is responsible for calling dev_put on the returned
2693 	 * netdev. */
2694 	struct net_device *(*get_net_dev_by_params)(
2695 			struct ib_device *dev,
2696 			u8 port,
2697 			u16 pkey,
2698 			const union ib_gid *gid,
2699 			const struct sockaddr *addr,
2700 			void *client_data);
2701 
2702 	refcount_t uses;
2703 	struct completion uses_zero;
2704 	u32 client_id;
2705 
2706 	/* kverbs are not required by the client */
2707 	u8 no_kverbs_req:1;
2708 };
2709 
2710 /*
2711  * IB block DMA iterator
2712  *
2713  * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2714  * to a HW supported page size.
2715  */
2716 struct ib_block_iter {
2717 	/* internal states */
2718 	struct scatterlist *__sg;	/* sg holding the current aligned block */
2719 	dma_addr_t __dma_addr;		/* unaligned DMA address of this block */
2720 	unsigned int __sg_nents;	/* number of SG entries */
2721 	unsigned int __sg_advance;	/* number of bytes to advance in sg in next step */
2722 	unsigned int __pg_bit;		/* alignment of current block */
2723 };
2724 
2725 struct ib_device *_ib_alloc_device(size_t size);
2726 #define ib_alloc_device(drv_struct, member)                                    \
2727 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2728 				      BUILD_BUG_ON_ZERO(offsetof(              \
2729 					      struct drv_struct, member))),    \
2730 		     struct drv_struct, member)
2731 
2732 void ib_dealloc_device(struct ib_device *device);
2733 
2734 void ib_get_device_fw_str(struct ib_device *device, char *str);
2735 
2736 int ib_register_device(struct ib_device *device, const char *name);
2737 void ib_unregister_device(struct ib_device *device);
2738 void ib_unregister_driver(enum rdma_driver_id driver_id);
2739 void ib_unregister_device_and_put(struct ib_device *device);
2740 void ib_unregister_device_queued(struct ib_device *ib_dev);
2741 
2742 int ib_register_client   (struct ib_client *client);
2743 void ib_unregister_client(struct ib_client *client);
2744 
2745 void __rdma_block_iter_start(struct ib_block_iter *biter,
2746 			     struct scatterlist *sglist,
2747 			     unsigned int nents,
2748 			     unsigned long pgsz);
2749 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2750 
2751 /**
2752  * rdma_block_iter_dma_address - get the aligned dma address of the current
2753  * block held by the block iterator.
2754  * @biter: block iterator holding the memory block
2755  */
2756 static inline dma_addr_t
rdma_block_iter_dma_address(struct ib_block_iter * biter)2757 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2758 {
2759 	return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2760 }
2761 
2762 /**
2763  * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2764  * @sglist: sglist to iterate over
2765  * @biter: block iterator holding the memory block
2766  * @nents: maximum number of sg entries to iterate over
2767  * @pgsz: best HW supported page size to use
2768  *
2769  * Callers may use rdma_block_iter_dma_address() to get each
2770  * blocks aligned DMA address.
2771  */
2772 #define rdma_for_each_block(sglist, biter, nents, pgsz)		\
2773 	for (__rdma_block_iter_start(biter, sglist, nents,	\
2774 				     pgsz);			\
2775 	     __rdma_block_iter_next(biter);)
2776 
2777 /**
2778  * ib_get_client_data - Get IB client context
2779  * @device:Device to get context for
2780  * @client:Client to get context for
2781  *
2782  * ib_get_client_data() returns the client context data set with
2783  * ib_set_client_data(). This can only be called while the client is
2784  * registered to the device, once the ib_client remove() callback returns this
2785  * cannot be called.
2786  */
ib_get_client_data(struct ib_device * device,struct ib_client * client)2787 static inline void *ib_get_client_data(struct ib_device *device,
2788 				       struct ib_client *client)
2789 {
2790 	return xa_load(&device->client_data, client->client_id);
2791 }
2792 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2793 			 void *data);
2794 void ib_set_device_ops(struct ib_device *device,
2795 		       const struct ib_device_ops *ops);
2796 
2797 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2798 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2799 		      unsigned long pfn, unsigned long size, pgprot_t prot);
2800 #else
rdma_user_mmap_io(struct ib_ucontext * ucontext,struct vm_area_struct * vma,unsigned long pfn,unsigned long size,pgprot_t prot)2801 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
2802 				    struct vm_area_struct *vma,
2803 				    unsigned long pfn, unsigned long size,
2804 				    pgprot_t prot)
2805 {
2806 	return -EINVAL;
2807 }
2808 #endif
2809 
ib_copy_from_udata(void * dest,struct ib_udata * udata,size_t len)2810 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2811 {
2812 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2813 }
2814 
ib_copy_to_udata(struct ib_udata * udata,void * src,size_t len)2815 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2816 {
2817 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2818 }
2819 
ib_is_buffer_cleared(const void __user * p,size_t len)2820 static inline bool ib_is_buffer_cleared(const void __user *p,
2821 					size_t len)
2822 {
2823 	bool ret;
2824 	u8 *buf;
2825 
2826 	if (len > USHRT_MAX)
2827 		return false;
2828 
2829 	buf = memdup_user(p, len);
2830 	if (IS_ERR(buf))
2831 		return false;
2832 
2833 	ret = !memchr_inv(buf, 0, len);
2834 	kfree(buf);
2835 	return ret;
2836 }
2837 
ib_is_udata_cleared(struct ib_udata * udata,size_t offset,size_t len)2838 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2839 				       size_t offset,
2840 				       size_t len)
2841 {
2842 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
2843 }
2844 
2845 /**
2846  * ib_is_destroy_retryable - Check whether the uobject destruction
2847  * is retryable.
2848  * @ret: The initial destruction return code
2849  * @why: remove reason
2850  * @uobj: The uobject that is destroyed
2851  *
2852  * This function is a helper function that IB layer and low-level drivers
2853  * can use to consider whether the destruction of the given uobject is
2854  * retry-able.
2855  * It checks the original return code, if it wasn't success the destruction
2856  * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2857  * the remove reason. (i.e. why).
2858  * Must be called with the object locked for destroy.
2859  */
ib_is_destroy_retryable(int ret,enum rdma_remove_reason why,struct ib_uobject * uobj)2860 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2861 					   struct ib_uobject *uobj)
2862 {
2863 	return ret && (why == RDMA_REMOVE_DESTROY ||
2864 		       uobj->context->cleanup_retryable);
2865 }
2866 
2867 /**
2868  * ib_destroy_usecnt - Called during destruction to check the usecnt
2869  * @usecnt: The usecnt atomic
2870  * @why: remove reason
2871  * @uobj: The uobject that is destroyed
2872  *
2873  * Non-zero usecnts will block destruction unless destruction was triggered by
2874  * a ucontext cleanup.
2875  */
ib_destroy_usecnt(atomic_t * usecnt,enum rdma_remove_reason why,struct ib_uobject * uobj)2876 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2877 				    enum rdma_remove_reason why,
2878 				    struct ib_uobject *uobj)
2879 {
2880 	if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2881 		return -EBUSY;
2882 	return 0;
2883 }
2884 
2885 /**
2886  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2887  * contains all required attributes and no attributes not allowed for
2888  * the given QP state transition.
2889  * @cur_state: Current QP state
2890  * @next_state: Next QP state
2891  * @type: QP type
2892  * @mask: Mask of supplied QP attributes
2893  *
2894  * This function is a helper function that a low-level driver's
2895  * modify_qp method can use to validate the consumer's input.  It
2896  * checks that cur_state and next_state are valid QP states, that a
2897  * transition from cur_state to next_state is allowed by the IB spec,
2898  * and that the attribute mask supplied is allowed for the transition.
2899  */
2900 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2901 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
2902 
2903 void ib_register_event_handler(struct ib_event_handler *event_handler);
2904 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2905 void ib_dispatch_event(const struct ib_event *event);
2906 
2907 int ib_query_port(struct ib_device *device,
2908 		  u8 port_num, struct ib_port_attr *port_attr);
2909 
2910 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2911 					       u8 port_num);
2912 
2913 /**
2914  * rdma_cap_ib_switch - Check if the device is IB switch
2915  * @device: Device to check
2916  *
2917  * Device driver is responsible for setting is_switch bit on
2918  * in ib_device structure at init time.
2919  *
2920  * Return: true if the device is IB switch.
2921  */
rdma_cap_ib_switch(const struct ib_device * device)2922 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2923 {
2924 	return device->is_switch;
2925 }
2926 
2927 /**
2928  * rdma_start_port - Return the first valid port number for the device
2929  * specified
2930  *
2931  * @device: Device to be checked
2932  *
2933  * Return start port number
2934  */
rdma_start_port(const struct ib_device * device)2935 static inline u8 rdma_start_port(const struct ib_device *device)
2936 {
2937 	return rdma_cap_ib_switch(device) ? 0 : 1;
2938 }
2939 
2940 /**
2941  * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2942  * @device - The struct ib_device * to iterate over
2943  * @iter - The unsigned int to store the port number
2944  */
2945 #define rdma_for_each_port(device, iter)                                       \
2946 	for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type(   \
2947 						     unsigned int, iter)));    \
2948 	     iter <= rdma_end_port(device); (iter)++)
2949 
2950 /**
2951  * rdma_end_port - Return the last valid port number for the device
2952  * specified
2953  *
2954  * @device: Device to be checked
2955  *
2956  * Return last port number
2957  */
rdma_end_port(const struct ib_device * device)2958 static inline u8 rdma_end_port(const struct ib_device *device)
2959 {
2960 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2961 }
2962 
rdma_is_port_valid(const struct ib_device * device,unsigned int port)2963 static inline int rdma_is_port_valid(const struct ib_device *device,
2964 				     unsigned int port)
2965 {
2966 	return (port >= rdma_start_port(device) &&
2967 		port <= rdma_end_port(device));
2968 }
2969 
rdma_is_grh_required(const struct ib_device * device,u8 port_num)2970 static inline bool rdma_is_grh_required(const struct ib_device *device,
2971 					u8 port_num)
2972 {
2973 	return device->port_data[port_num].immutable.core_cap_flags &
2974 	       RDMA_CORE_PORT_IB_GRH_REQUIRED;
2975 }
2976 
rdma_protocol_ib(const struct ib_device * device,u8 port_num)2977 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2978 {
2979 	return device->port_data[port_num].immutable.core_cap_flags &
2980 	       RDMA_CORE_CAP_PROT_IB;
2981 }
2982 
rdma_protocol_roce(const struct ib_device * device,u8 port_num)2983 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2984 {
2985 	return device->port_data[port_num].immutable.core_cap_flags &
2986 	       (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2987 }
2988 
rdma_protocol_roce_udp_encap(const struct ib_device * device,u8 port_num)2989 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2990 {
2991 	return device->port_data[port_num].immutable.core_cap_flags &
2992 	       RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2993 }
2994 
rdma_protocol_roce_eth_encap(const struct ib_device * device,u8 port_num)2995 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2996 {
2997 	return device->port_data[port_num].immutable.core_cap_flags &
2998 	       RDMA_CORE_CAP_PROT_ROCE;
2999 }
3000 
rdma_protocol_iwarp(const struct ib_device * device,u8 port_num)3001 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
3002 {
3003 	return device->port_data[port_num].immutable.core_cap_flags &
3004 	       RDMA_CORE_CAP_PROT_IWARP;
3005 }
3006 
rdma_ib_or_roce(const struct ib_device * device,u8 port_num)3007 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
3008 {
3009 	return rdma_protocol_ib(device, port_num) ||
3010 		rdma_protocol_roce(device, port_num);
3011 }
3012 
rdma_protocol_raw_packet(const struct ib_device * device,u8 port_num)3013 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
3014 {
3015 	return device->port_data[port_num].immutable.core_cap_flags &
3016 	       RDMA_CORE_CAP_PROT_RAW_PACKET;
3017 }
3018 
rdma_protocol_usnic(const struct ib_device * device,u8 port_num)3019 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
3020 {
3021 	return device->port_data[port_num].immutable.core_cap_flags &
3022 	       RDMA_CORE_CAP_PROT_USNIC;
3023 }
3024 
3025 /**
3026  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3027  * Management Datagrams.
3028  * @device: Device to check
3029  * @port_num: Port number to check
3030  *
3031  * Management Datagrams (MAD) are a required part of the InfiniBand
3032  * specification and are supported on all InfiniBand devices.  A slightly
3033  * extended version are also supported on OPA interfaces.
3034  *
3035  * Return: true if the port supports sending/receiving of MAD packets.
3036  */
rdma_cap_ib_mad(const struct ib_device * device,u8 port_num)3037 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
3038 {
3039 	return device->port_data[port_num].immutable.core_cap_flags &
3040 	       RDMA_CORE_CAP_IB_MAD;
3041 }
3042 
3043 /**
3044  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3045  * Management Datagrams.
3046  * @device: Device to check
3047  * @port_num: Port number to check
3048  *
3049  * Intel OmniPath devices extend and/or replace the InfiniBand Management
3050  * datagrams with their own versions.  These OPA MADs share many but not all of
3051  * the characteristics of InfiniBand MADs.
3052  *
3053  * OPA MADs differ in the following ways:
3054  *
3055  *    1) MADs are variable size up to 2K
3056  *       IBTA defined MADs remain fixed at 256 bytes
3057  *    2) OPA SMPs must carry valid PKeys
3058  *    3) OPA SMP packets are a different format
3059  *
3060  * Return: true if the port supports OPA MAD packet formats.
3061  */
rdma_cap_opa_mad(struct ib_device * device,u8 port_num)3062 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3063 {
3064 	return device->port_data[port_num].immutable.core_cap_flags &
3065 		RDMA_CORE_CAP_OPA_MAD;
3066 }
3067 
3068 /**
3069  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3070  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3071  * @device: Device to check
3072  * @port_num: Port number to check
3073  *
3074  * Each InfiniBand node is required to provide a Subnet Management Agent
3075  * that the subnet manager can access.  Prior to the fabric being fully
3076  * configured by the subnet manager, the SMA is accessed via a well known
3077  * interface called the Subnet Management Interface (SMI).  This interface
3078  * uses directed route packets to communicate with the SM to get around the
3079  * chicken and egg problem of the SM needing to know what's on the fabric
3080  * in order to configure the fabric, and needing to configure the fabric in
3081  * order to send packets to the devices on the fabric.  These directed
3082  * route packets do not need the fabric fully configured in order to reach
3083  * their destination.  The SMI is the only method allowed to send
3084  * directed route packets on an InfiniBand fabric.
3085  *
3086  * Return: true if the port provides an SMI.
3087  */
rdma_cap_ib_smi(const struct ib_device * device,u8 port_num)3088 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3089 {
3090 	return device->port_data[port_num].immutable.core_cap_flags &
3091 	       RDMA_CORE_CAP_IB_SMI;
3092 }
3093 
3094 /**
3095  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3096  * Communication Manager.
3097  * @device: Device to check
3098  * @port_num: Port number to check
3099  *
3100  * The InfiniBand Communication Manager is one of many pre-defined General
3101  * Service Agents (GSA) that are accessed via the General Service
3102  * Interface (GSI).  It's role is to facilitate establishment of connections
3103  * between nodes as well as other management related tasks for established
3104  * connections.
3105  *
3106  * Return: true if the port supports an IB CM (this does not guarantee that
3107  * a CM is actually running however).
3108  */
rdma_cap_ib_cm(const struct ib_device * device,u8 port_num)3109 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3110 {
3111 	return device->port_data[port_num].immutable.core_cap_flags &
3112 	       RDMA_CORE_CAP_IB_CM;
3113 }
3114 
3115 /**
3116  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3117  * Communication Manager.
3118  * @device: Device to check
3119  * @port_num: Port number to check
3120  *
3121  * Similar to above, but specific to iWARP connections which have a different
3122  * managment protocol than InfiniBand.
3123  *
3124  * Return: true if the port supports an iWARP CM (this does not guarantee that
3125  * a CM is actually running however).
3126  */
rdma_cap_iw_cm(const struct ib_device * device,u8 port_num)3127 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3128 {
3129 	return device->port_data[port_num].immutable.core_cap_flags &
3130 	       RDMA_CORE_CAP_IW_CM;
3131 }
3132 
3133 /**
3134  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3135  * Subnet Administration.
3136  * @device: Device to check
3137  * @port_num: Port number to check
3138  *
3139  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3140  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
3141  * fabrics, devices should resolve routes to other hosts by contacting the
3142  * SA to query the proper route.
3143  *
3144  * Return: true if the port should act as a client to the fabric Subnet
3145  * Administration interface.  This does not imply that the SA service is
3146  * running locally.
3147  */
rdma_cap_ib_sa(const struct ib_device * device,u8 port_num)3148 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3149 {
3150 	return device->port_data[port_num].immutable.core_cap_flags &
3151 	       RDMA_CORE_CAP_IB_SA;
3152 }
3153 
3154 /**
3155  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3156  * Multicast.
3157  * @device: Device to check
3158  * @port_num: Port number to check
3159  *
3160  * InfiniBand multicast registration is more complex than normal IPv4 or
3161  * IPv6 multicast registration.  Each Host Channel Adapter must register
3162  * with the Subnet Manager when it wishes to join a multicast group.  It
3163  * should do so only once regardless of how many queue pairs it subscribes
3164  * to this group.  And it should leave the group only after all queue pairs
3165  * attached to the group have been detached.
3166  *
3167  * Return: true if the port must undertake the additional adminstrative
3168  * overhead of registering/unregistering with the SM and tracking of the
3169  * total number of queue pairs attached to the multicast group.
3170  */
rdma_cap_ib_mcast(const struct ib_device * device,u8 port_num)3171 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3172 {
3173 	return rdma_cap_ib_sa(device, port_num);
3174 }
3175 
3176 /**
3177  * rdma_cap_af_ib - Check if the port of device has the capability
3178  * Native Infiniband Address.
3179  * @device: Device to check
3180  * @port_num: Port number to check
3181  *
3182  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3183  * GID.  RoCE uses a different mechanism, but still generates a GID via
3184  * a prescribed mechanism and port specific data.
3185  *
3186  * Return: true if the port uses a GID address to identify devices on the
3187  * network.
3188  */
rdma_cap_af_ib(const struct ib_device * device,u8 port_num)3189 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3190 {
3191 	return device->port_data[port_num].immutable.core_cap_flags &
3192 	       RDMA_CORE_CAP_AF_IB;
3193 }
3194 
3195 /**
3196  * rdma_cap_eth_ah - Check if the port of device has the capability
3197  * Ethernet Address Handle.
3198  * @device: Device to check
3199  * @port_num: Port number to check
3200  *
3201  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3202  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3203  * port.  Normally, packet headers are generated by the sending host
3204  * adapter, but when sending connectionless datagrams, we must manually
3205  * inject the proper headers for the fabric we are communicating over.
3206  *
3207  * Return: true if we are running as a RoCE port and must force the
3208  * addition of a Global Route Header built from our Ethernet Address
3209  * Handle into our header list for connectionless packets.
3210  */
rdma_cap_eth_ah(const struct ib_device * device,u8 port_num)3211 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3212 {
3213 	return device->port_data[port_num].immutable.core_cap_flags &
3214 	       RDMA_CORE_CAP_ETH_AH;
3215 }
3216 
3217 /**
3218  * rdma_cap_opa_ah - Check if the port of device supports
3219  * OPA Address handles
3220  * @device: Device to check
3221  * @port_num: Port number to check
3222  *
3223  * Return: true if we are running on an OPA device which supports
3224  * the extended OPA addressing.
3225  */
rdma_cap_opa_ah(struct ib_device * device,u8 port_num)3226 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3227 {
3228 	return (device->port_data[port_num].immutable.core_cap_flags &
3229 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3230 }
3231 
3232 /**
3233  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3234  *
3235  * @device: Device
3236  * @port_num: Port number
3237  *
3238  * This MAD size includes the MAD headers and MAD payload.  No other headers
3239  * are included.
3240  *
3241  * Return the max MAD size required by the Port.  Will return 0 if the port
3242  * does not support MADs
3243  */
rdma_max_mad_size(const struct ib_device * device,u8 port_num)3244 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3245 {
3246 	return device->port_data[port_num].immutable.max_mad_size;
3247 }
3248 
3249 /**
3250  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3251  * @device: Device to check
3252  * @port_num: Port number to check
3253  *
3254  * RoCE GID table mechanism manages the various GIDs for a device.
3255  *
3256  * NOTE: if allocating the port's GID table has failed, this call will still
3257  * return true, but any RoCE GID table API will fail.
3258  *
3259  * Return: true if the port uses RoCE GID table mechanism in order to manage
3260  * its GIDs.
3261  */
rdma_cap_roce_gid_table(const struct ib_device * device,u8 port_num)3262 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3263 					   u8 port_num)
3264 {
3265 	return rdma_protocol_roce(device, port_num) &&
3266 		device->ops.add_gid && device->ops.del_gid;
3267 }
3268 
3269 /*
3270  * Check if the device supports READ W/ INVALIDATE.
3271  */
rdma_cap_read_inv(struct ib_device * dev,u32 port_num)3272 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3273 {
3274 	/*
3275 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3276 	 * has support for it yet.
3277 	 */
3278 	return rdma_protocol_iwarp(dev, port_num);
3279 }
3280 
3281 /**
3282  * rdma_find_pg_bit - Find page bit given address and HW supported page sizes
3283  *
3284  * @addr: address
3285  * @pgsz_bitmap: bitmap of HW supported page sizes
3286  */
rdma_find_pg_bit(unsigned long addr,unsigned long pgsz_bitmap)3287 static inline unsigned int rdma_find_pg_bit(unsigned long addr,
3288 					    unsigned long pgsz_bitmap)
3289 {
3290 	unsigned long align;
3291 	unsigned long pgsz;
3292 
3293 	align = addr & -addr;
3294 
3295 	/* Find page bit such that addr is aligned to the highest supported
3296 	 * HW page size
3297 	 */
3298 	pgsz = pgsz_bitmap & ~(-align << 1);
3299 	if (!pgsz)
3300 		return __ffs(pgsz_bitmap);
3301 
3302 	return __fls(pgsz);
3303 }
3304 
3305 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3306 			 int state);
3307 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3308 		     struct ifla_vf_info *info);
3309 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3310 		    struct ifla_vf_stats *stats);
3311 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3312 		   int type);
3313 
3314 int ib_query_pkey(struct ib_device *device,
3315 		  u8 port_num, u16 index, u16 *pkey);
3316 
3317 int ib_modify_device(struct ib_device *device,
3318 		     int device_modify_mask,
3319 		     struct ib_device_modify *device_modify);
3320 
3321 int ib_modify_port(struct ib_device *device,
3322 		   u8 port_num, int port_modify_mask,
3323 		   struct ib_port_modify *port_modify);
3324 
3325 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3326 		u8 *port_num, u16 *index);
3327 
3328 int ib_find_pkey(struct ib_device *device,
3329 		 u8 port_num, u16 pkey, u16 *index);
3330 
3331 enum ib_pd_flags {
3332 	/*
3333 	 * Create a memory registration for all memory in the system and place
3334 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3335 	 * ULPs to avoid the overhead of dynamic MRs.
3336 	 *
3337 	 * This flag is generally considered unsafe and must only be used in
3338 	 * extremly trusted environments.  Every use of it will log a warning
3339 	 * in the kernel log.
3340 	 */
3341 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3342 };
3343 
3344 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3345 		const char *caller);
3346 
3347 #define ib_alloc_pd(device, flags) \
3348 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3349 
3350 /**
3351  * ib_dealloc_pd_user - Deallocate kernel/user PD
3352  * @pd: The protection domain
3353  * @udata: Valid user data or NULL for kernel objects
3354  */
3355 void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3356 
3357 /**
3358  * ib_dealloc_pd - Deallocate kernel PD
3359  * @pd: The protection domain
3360  *
3361  * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3362  */
ib_dealloc_pd(struct ib_pd * pd)3363 static inline void ib_dealloc_pd(struct ib_pd *pd)
3364 {
3365 	ib_dealloc_pd_user(pd, NULL);
3366 }
3367 
3368 enum rdma_create_ah_flags {
3369 	/* In a sleepable context */
3370 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3371 };
3372 
3373 /**
3374  * rdma_create_ah - Creates an address handle for the given address vector.
3375  * @pd: The protection domain associated with the address handle.
3376  * @ah_attr: The attributes of the address vector.
3377  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3378  *
3379  * The address handle is used to reference a local or global destination
3380  * in all UD QP post sends.
3381  */
3382 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3383 			     u32 flags);
3384 
3385 /**
3386  * rdma_create_user_ah - Creates an address handle for the given address vector.
3387  * It resolves destination mac address for ah attribute of RoCE type.
3388  * @pd: The protection domain associated with the address handle.
3389  * @ah_attr: The attributes of the address vector.
3390  * @udata: pointer to user's input output buffer information need by
3391  *         provider driver.
3392  *
3393  * It returns 0 on success and returns appropriate error code on error.
3394  * The address handle is used to reference a local or global destination
3395  * in all UD QP post sends.
3396  */
3397 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3398 				  struct rdma_ah_attr *ah_attr,
3399 				  struct ib_udata *udata);
3400 /**
3401  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3402  *   work completion.
3403  * @hdr: the L3 header to parse
3404  * @net_type: type of header to parse
3405  * @sgid: place to store source gid
3406  * @dgid: place to store destination gid
3407  */
3408 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3409 			      enum rdma_network_type net_type,
3410 			      union ib_gid *sgid, union ib_gid *dgid);
3411 
3412 /**
3413  * ib_get_rdma_header_version - Get the header version
3414  * @hdr: the L3 header to parse
3415  */
3416 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3417 
3418 /**
3419  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3420  *   work completion.
3421  * @device: Device on which the received message arrived.
3422  * @port_num: Port on which the received message arrived.
3423  * @wc: Work completion associated with the received message.
3424  * @grh: References the received global route header.  This parameter is
3425  *   ignored unless the work completion indicates that the GRH is valid.
3426  * @ah_attr: Returned attributes that can be used when creating an address
3427  *   handle for replying to the message.
3428  * When ib_init_ah_attr_from_wc() returns success,
3429  * (a) for IB link layer it optionally contains a reference to SGID attribute
3430  * when GRH is present for IB link layer.
3431  * (b) for RoCE link layer it contains a reference to SGID attribute.
3432  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3433  * attributes which are initialized using ib_init_ah_attr_from_wc().
3434  *
3435  */
3436 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3437 			    const struct ib_wc *wc, const struct ib_grh *grh,
3438 			    struct rdma_ah_attr *ah_attr);
3439 
3440 /**
3441  * ib_create_ah_from_wc - Creates an address handle associated with the
3442  *   sender of the specified work completion.
3443  * @pd: The protection domain associated with the address handle.
3444  * @wc: Work completion information associated with a received message.
3445  * @grh: References the received global route header.  This parameter is
3446  *   ignored unless the work completion indicates that the GRH is valid.
3447  * @port_num: The outbound port number to associate with the address.
3448  *
3449  * The address handle is used to reference a local or global destination
3450  * in all UD QP post sends.
3451  */
3452 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3453 				   const struct ib_grh *grh, u8 port_num);
3454 
3455 /**
3456  * rdma_modify_ah - Modifies the address vector associated with an address
3457  *   handle.
3458  * @ah: The address handle to modify.
3459  * @ah_attr: The new address vector attributes to associate with the
3460  *   address handle.
3461  */
3462 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3463 
3464 /**
3465  * rdma_query_ah - Queries the address vector associated with an address
3466  *   handle.
3467  * @ah: The address handle to query.
3468  * @ah_attr: The address vector attributes associated with the address
3469  *   handle.
3470  */
3471 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3472 
3473 enum rdma_destroy_ah_flags {
3474 	/* In a sleepable context */
3475 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3476 };
3477 
3478 /**
3479  * rdma_destroy_ah_user - Destroys an address handle.
3480  * @ah: The address handle to destroy.
3481  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3482  * @udata: Valid user data or NULL for kernel objects
3483  */
3484 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3485 
3486 /**
3487  * rdma_destroy_ah - Destroys an kernel address handle.
3488  * @ah: The address handle to destroy.
3489  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3490  *
3491  * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3492  */
rdma_destroy_ah(struct ib_ah * ah,u32 flags)3493 static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3494 {
3495 	return rdma_destroy_ah_user(ah, flags, NULL);
3496 }
3497 
3498 /**
3499  * ib_create_srq - Creates a SRQ associated with the specified protection
3500  *   domain.
3501  * @pd: The protection domain associated with the SRQ.
3502  * @srq_init_attr: A list of initial attributes required to create the
3503  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
3504  *   the actual capabilities of the created SRQ.
3505  *
3506  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3507  * requested size of the SRQ, and set to the actual values allocated
3508  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
3509  * will always be at least as large as the requested values.
3510  */
3511 struct ib_srq *ib_create_srq(struct ib_pd *pd,
3512 			     struct ib_srq_init_attr *srq_init_attr);
3513 
3514 /**
3515  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3516  * @srq: The SRQ to modify.
3517  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3518  *   the current values of selected SRQ attributes are returned.
3519  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3520  *   are being modified.
3521  *
3522  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3523  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3524  * the number of receives queued drops below the limit.
3525  */
3526 int ib_modify_srq(struct ib_srq *srq,
3527 		  struct ib_srq_attr *srq_attr,
3528 		  enum ib_srq_attr_mask srq_attr_mask);
3529 
3530 /**
3531  * ib_query_srq - Returns the attribute list and current values for the
3532  *   specified SRQ.
3533  * @srq: The SRQ to query.
3534  * @srq_attr: The attributes of the specified SRQ.
3535  */
3536 int ib_query_srq(struct ib_srq *srq,
3537 		 struct ib_srq_attr *srq_attr);
3538 
3539 /**
3540  * ib_destroy_srq_user - Destroys the specified SRQ.
3541  * @srq: The SRQ to destroy.
3542  * @udata: Valid user data or NULL for kernel objects
3543  */
3544 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3545 
3546 /**
3547  * ib_destroy_srq - Destroys the specified kernel SRQ.
3548  * @srq: The SRQ to destroy.
3549  *
3550  * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3551  */
ib_destroy_srq(struct ib_srq * srq)3552 static inline int ib_destroy_srq(struct ib_srq *srq)
3553 {
3554 	return ib_destroy_srq_user(srq, NULL);
3555 }
3556 
3557 /**
3558  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3559  * @srq: The SRQ to post the work request on.
3560  * @recv_wr: A list of work requests to post on the receive queue.
3561  * @bad_recv_wr: On an immediate failure, this parameter will reference
3562  *   the work request that failed to be posted on the QP.
3563  */
ib_post_srq_recv(struct ib_srq * srq,const struct ib_recv_wr * recv_wr,const struct ib_recv_wr ** bad_recv_wr)3564 static inline int ib_post_srq_recv(struct ib_srq *srq,
3565 				   const struct ib_recv_wr *recv_wr,
3566 				   const struct ib_recv_wr **bad_recv_wr)
3567 {
3568 	const struct ib_recv_wr *dummy;
3569 
3570 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3571 					      bad_recv_wr ? : &dummy);
3572 }
3573 
3574 /**
3575  * ib_create_qp_user - Creates a QP associated with the specified protection
3576  *   domain.
3577  * @pd: The protection domain associated with the QP.
3578  * @qp_init_attr: A list of initial attributes required to create the
3579  *   QP.  If QP creation succeeds, then the attributes are updated to
3580  *   the actual capabilities of the created QP.
3581  * @udata: Valid user data or NULL for kernel objects
3582  */
3583 struct ib_qp *ib_create_qp_user(struct ib_pd *pd,
3584 				struct ib_qp_init_attr *qp_init_attr,
3585 				struct ib_udata *udata);
3586 
3587 /**
3588  * ib_create_qp - Creates a kernel QP associated with the specified protection
3589  *   domain.
3590  * @pd: The protection domain associated with the QP.
3591  * @qp_init_attr: A list of initial attributes required to create the
3592  *   QP.  If QP creation succeeds, then the attributes are updated to
3593  *   the actual capabilities of the created QP.
3594  * @udata: Valid user data or NULL for kernel objects
3595  *
3596  * NOTE: for user qp use ib_create_qp_user with valid udata!
3597  */
ib_create_qp(struct ib_pd * pd,struct ib_qp_init_attr * qp_init_attr)3598 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3599 					 struct ib_qp_init_attr *qp_init_attr)
3600 {
3601 	return ib_create_qp_user(pd, qp_init_attr, NULL);
3602 }
3603 
3604 /**
3605  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3606  * @qp: The QP to modify.
3607  * @attr: On input, specifies the QP attributes to modify.  On output,
3608  *   the current values of selected QP attributes are returned.
3609  * @attr_mask: A bit-mask used to specify which attributes of the QP
3610  *   are being modified.
3611  * @udata: pointer to user's input output buffer information
3612  *   are being modified.
3613  * It returns 0 on success and returns appropriate error code on error.
3614  */
3615 int ib_modify_qp_with_udata(struct ib_qp *qp,
3616 			    struct ib_qp_attr *attr,
3617 			    int attr_mask,
3618 			    struct ib_udata *udata);
3619 
3620 /**
3621  * ib_modify_qp - Modifies the attributes for the specified QP and then
3622  *   transitions the QP to the given state.
3623  * @qp: The QP to modify.
3624  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3625  *   the current values of selected QP attributes are returned.
3626  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3627  *   are being modified.
3628  */
3629 int ib_modify_qp(struct ib_qp *qp,
3630 		 struct ib_qp_attr *qp_attr,
3631 		 int qp_attr_mask);
3632 
3633 /**
3634  * ib_query_qp - Returns the attribute list and current values for the
3635  *   specified QP.
3636  * @qp: The QP to query.
3637  * @qp_attr: The attributes of the specified QP.
3638  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3639  * @qp_init_attr: Additional attributes of the selected QP.
3640  *
3641  * The qp_attr_mask may be used to limit the query to gathering only the
3642  * selected attributes.
3643  */
3644 int ib_query_qp(struct ib_qp *qp,
3645 		struct ib_qp_attr *qp_attr,
3646 		int qp_attr_mask,
3647 		struct ib_qp_init_attr *qp_init_attr);
3648 
3649 /**
3650  * ib_destroy_qp - Destroys the specified QP.
3651  * @qp: The QP to destroy.
3652  * @udata: Valid udata or NULL for kernel objects
3653  */
3654 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3655 
3656 /**
3657  * ib_destroy_qp - Destroys the specified kernel QP.
3658  * @qp: The QP to destroy.
3659  *
3660  * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3661  */
ib_destroy_qp(struct ib_qp * qp)3662 static inline int ib_destroy_qp(struct ib_qp *qp)
3663 {
3664 	return ib_destroy_qp_user(qp, NULL);
3665 }
3666 
3667 /**
3668  * ib_open_qp - Obtain a reference to an existing sharable QP.
3669  * @xrcd - XRC domain
3670  * @qp_open_attr: Attributes identifying the QP to open.
3671  *
3672  * Returns a reference to a sharable QP.
3673  */
3674 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3675 			 struct ib_qp_open_attr *qp_open_attr);
3676 
3677 /**
3678  * ib_close_qp - Release an external reference to a QP.
3679  * @qp: The QP handle to release
3680  *
3681  * The opened QP handle is released by the caller.  The underlying
3682  * shared QP is not destroyed until all internal references are released.
3683  */
3684 int ib_close_qp(struct ib_qp *qp);
3685 
3686 /**
3687  * ib_post_send - Posts a list of work requests to the send queue of
3688  *   the specified QP.
3689  * @qp: The QP to post the work request on.
3690  * @send_wr: A list of work requests to post on the send queue.
3691  * @bad_send_wr: On an immediate failure, this parameter will reference
3692  *   the work request that failed to be posted on the QP.
3693  *
3694  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3695  * error is returned, the QP state shall not be affected,
3696  * ib_post_send() will return an immediate error after queueing any
3697  * earlier work requests in the list.
3698  */
ib_post_send(struct ib_qp * qp,const struct ib_send_wr * send_wr,const struct ib_send_wr ** bad_send_wr)3699 static inline int ib_post_send(struct ib_qp *qp,
3700 			       const struct ib_send_wr *send_wr,
3701 			       const struct ib_send_wr **bad_send_wr)
3702 {
3703 	const struct ib_send_wr *dummy;
3704 
3705 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3706 }
3707 
3708 /**
3709  * ib_post_recv - Posts a list of work requests to the receive queue of
3710  *   the specified QP.
3711  * @qp: The QP to post the work request on.
3712  * @recv_wr: A list of work requests to post on the receive queue.
3713  * @bad_recv_wr: On an immediate failure, this parameter will reference
3714  *   the work request that failed to be posted on the QP.
3715  */
ib_post_recv(struct ib_qp * qp,const struct ib_recv_wr * recv_wr,const struct ib_recv_wr ** bad_recv_wr)3716 static inline int ib_post_recv(struct ib_qp *qp,
3717 			       const struct ib_recv_wr *recv_wr,
3718 			       const struct ib_recv_wr **bad_recv_wr)
3719 {
3720 	const struct ib_recv_wr *dummy;
3721 
3722 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3723 }
3724 
3725 struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
3726 				 int nr_cqe, int comp_vector,
3727 				 enum ib_poll_context poll_ctx,
3728 				 const char *caller, struct ib_udata *udata);
3729 
3730 /**
3731  * ib_alloc_cq_user: Allocate kernel/user CQ
3732  * @dev: The IB device
3733  * @private: Private data attached to the CQE
3734  * @nr_cqe: Number of CQEs in the CQ
3735  * @comp_vector: Completion vector used for the IRQs
3736  * @poll_ctx: Context used for polling the CQ
3737  * @udata: Valid user data or NULL for kernel objects
3738  */
ib_alloc_cq_user(struct ib_device * dev,void * private,int nr_cqe,int comp_vector,enum ib_poll_context poll_ctx,struct ib_udata * udata)3739 static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
3740 					     void *private, int nr_cqe,
3741 					     int comp_vector,
3742 					     enum ib_poll_context poll_ctx,
3743 					     struct ib_udata *udata)
3744 {
3745 	return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3746 				  KBUILD_MODNAME, udata);
3747 }
3748 
3749 /**
3750  * ib_alloc_cq: Allocate kernel CQ
3751  * @dev: The IB device
3752  * @private: Private data attached to the CQE
3753  * @nr_cqe: Number of CQEs in the CQ
3754  * @comp_vector: Completion vector used for the IRQs
3755  * @poll_ctx: Context used for polling the CQ
3756  *
3757  * NOTE: for user cq use ib_alloc_cq_user with valid udata!
3758  */
ib_alloc_cq(struct ib_device * dev,void * private,int nr_cqe,int comp_vector,enum ib_poll_context poll_ctx)3759 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3760 					int nr_cqe, int comp_vector,
3761 					enum ib_poll_context poll_ctx)
3762 {
3763 	return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3764 				NULL);
3765 }
3766 
3767 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3768 				int nr_cqe, enum ib_poll_context poll_ctx,
3769 				const char *caller);
3770 
3771 /**
3772  * ib_alloc_cq_any: Allocate kernel CQ
3773  * @dev: The IB device
3774  * @private: Private data attached to the CQE
3775  * @nr_cqe: Number of CQEs in the CQ
3776  * @poll_ctx: Context used for polling the CQ
3777  */
ib_alloc_cq_any(struct ib_device * dev,void * private,int nr_cqe,enum ib_poll_context poll_ctx)3778 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3779 					    void *private, int nr_cqe,
3780 					    enum ib_poll_context poll_ctx)
3781 {
3782 	return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3783 				 KBUILD_MODNAME);
3784 }
3785 
3786 /**
3787  * ib_free_cq_user - Free kernel/user CQ
3788  * @cq: The CQ to free
3789  * @udata: Valid user data or NULL for kernel objects
3790  */
3791 void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3792 
3793 /**
3794  * ib_free_cq - Free kernel CQ
3795  * @cq: The CQ to free
3796  *
3797  * NOTE: for user cq use ib_free_cq_user with valid udata!
3798  */
ib_free_cq(struct ib_cq * cq)3799 static inline void ib_free_cq(struct ib_cq *cq)
3800 {
3801 	ib_free_cq_user(cq, NULL);
3802 }
3803 
3804 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3805 
3806 /**
3807  * ib_create_cq - Creates a CQ on the specified device.
3808  * @device: The device on which to create the CQ.
3809  * @comp_handler: A user-specified callback that is invoked when a
3810  *   completion event occurs on the CQ.
3811  * @event_handler: A user-specified callback that is invoked when an
3812  *   asynchronous event not associated with a completion occurs on the CQ.
3813  * @cq_context: Context associated with the CQ returned to the user via
3814  *   the associated completion and event handlers.
3815  * @cq_attr: The attributes the CQ should be created upon.
3816  *
3817  * Users can examine the cq structure to determine the actual CQ size.
3818  */
3819 struct ib_cq *__ib_create_cq(struct ib_device *device,
3820 			     ib_comp_handler comp_handler,
3821 			     void (*event_handler)(struct ib_event *, void *),
3822 			     void *cq_context,
3823 			     const struct ib_cq_init_attr *cq_attr,
3824 			     const char *caller);
3825 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3826 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3827 
3828 /**
3829  * ib_resize_cq - Modifies the capacity of the CQ.
3830  * @cq: The CQ to resize.
3831  * @cqe: The minimum size of the CQ.
3832  *
3833  * Users can examine the cq structure to determine the actual CQ size.
3834  */
3835 int ib_resize_cq(struct ib_cq *cq, int cqe);
3836 
3837 /**
3838  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3839  * @cq: The CQ to modify.
3840  * @cq_count: number of CQEs that will trigger an event
3841  * @cq_period: max period of time in usec before triggering an event
3842  *
3843  */
3844 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3845 
3846 /**
3847  * ib_destroy_cq_user - Destroys the specified CQ.
3848  * @cq: The CQ to destroy.
3849  * @udata: Valid user data or NULL for kernel objects
3850  */
3851 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3852 
3853 /**
3854  * ib_destroy_cq - Destroys the specified kernel CQ.
3855  * @cq: The CQ to destroy.
3856  *
3857  * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3858  */
ib_destroy_cq(struct ib_cq * cq)3859 static inline void ib_destroy_cq(struct ib_cq *cq)
3860 {
3861 	ib_destroy_cq_user(cq, NULL);
3862 }
3863 
3864 /**
3865  * ib_poll_cq - poll a CQ for completion(s)
3866  * @cq:the CQ being polled
3867  * @num_entries:maximum number of completions to return
3868  * @wc:array of at least @num_entries &struct ib_wc where completions
3869  *   will be returned
3870  *
3871  * Poll a CQ for (possibly multiple) completions.  If the return value
3872  * is < 0, an error occurred.  If the return value is >= 0, it is the
3873  * number of completions returned.  If the return value is
3874  * non-negative and < num_entries, then the CQ was emptied.
3875  */
ib_poll_cq(struct ib_cq * cq,int num_entries,struct ib_wc * wc)3876 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3877 			     struct ib_wc *wc)
3878 {
3879 	return cq->device->ops.poll_cq(cq, num_entries, wc);
3880 }
3881 
3882 /**
3883  * ib_req_notify_cq - Request completion notification on a CQ.
3884  * @cq: The CQ to generate an event for.
3885  * @flags:
3886  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3887  *   to request an event on the next solicited event or next work
3888  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3889  *   may also be |ed in to request a hint about missed events, as
3890  *   described below.
3891  *
3892  * Return Value:
3893  *    < 0 means an error occurred while requesting notification
3894  *   == 0 means notification was requested successfully, and if
3895  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3896  *        were missed and it is safe to wait for another event.  In
3897  *        this case is it guaranteed that any work completions added
3898  *        to the CQ since the last CQ poll will trigger a completion
3899  *        notification event.
3900  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3901  *        in.  It means that the consumer must poll the CQ again to
3902  *        make sure it is empty to avoid missing an event because of a
3903  *        race between requesting notification and an entry being
3904  *        added to the CQ.  This return value means it is possible
3905  *        (but not guaranteed) that a work completion has been added
3906  *        to the CQ since the last poll without triggering a
3907  *        completion notification event.
3908  */
ib_req_notify_cq(struct ib_cq * cq,enum ib_cq_notify_flags flags)3909 static inline int ib_req_notify_cq(struct ib_cq *cq,
3910 				   enum ib_cq_notify_flags flags)
3911 {
3912 	return cq->device->ops.req_notify_cq(cq, flags);
3913 }
3914 
3915 /**
3916  * ib_req_ncomp_notif - Request completion notification when there are
3917  *   at least the specified number of unreaped completions on the CQ.
3918  * @cq: The CQ to generate an event for.
3919  * @wc_cnt: The number of unreaped completions that should be on the
3920  *   CQ before an event is generated.
3921  */
ib_req_ncomp_notif(struct ib_cq * cq,int wc_cnt)3922 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3923 {
3924 	return cq->device->ops.req_ncomp_notif ?
3925 		cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3926 		-ENOSYS;
3927 }
3928 
3929 /**
3930  * ib_dma_mapping_error - check a DMA addr for error
3931  * @dev: The device for which the dma_addr was created
3932  * @dma_addr: The DMA address to check
3933  */
ib_dma_mapping_error(struct ib_device * dev,u64 dma_addr)3934 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3935 {
3936 	return dma_mapping_error(dev->dma_device, dma_addr);
3937 }
3938 
3939 /**
3940  * ib_dma_map_single - Map a kernel virtual address to DMA address
3941  * @dev: The device for which the dma_addr is to be created
3942  * @cpu_addr: The kernel virtual address
3943  * @size: The size of the region in bytes
3944  * @direction: The direction of the DMA
3945  */
ib_dma_map_single(struct ib_device * dev,void * cpu_addr,size_t size,enum dma_data_direction direction)3946 static inline u64 ib_dma_map_single(struct ib_device *dev,
3947 				    void *cpu_addr, size_t size,
3948 				    enum dma_data_direction direction)
3949 {
3950 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3951 }
3952 
3953 /**
3954  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3955  * @dev: The device for which the DMA address was created
3956  * @addr: The DMA address
3957  * @size: The size of the region in bytes
3958  * @direction: The direction of the DMA
3959  */
ib_dma_unmap_single(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction direction)3960 static inline void ib_dma_unmap_single(struct ib_device *dev,
3961 				       u64 addr, size_t size,
3962 				       enum dma_data_direction direction)
3963 {
3964 	dma_unmap_single(dev->dma_device, addr, size, direction);
3965 }
3966 
3967 /**
3968  * ib_dma_map_page - Map a physical page to DMA address
3969  * @dev: The device for which the dma_addr is to be created
3970  * @page: The page to be mapped
3971  * @offset: The offset within the page
3972  * @size: The size of the region in bytes
3973  * @direction: The direction of the DMA
3974  */
ib_dma_map_page(struct ib_device * dev,struct page * page,unsigned long offset,size_t size,enum dma_data_direction direction)3975 static inline u64 ib_dma_map_page(struct ib_device *dev,
3976 				  struct page *page,
3977 				  unsigned long offset,
3978 				  size_t size,
3979 					 enum dma_data_direction direction)
3980 {
3981 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3982 }
3983 
3984 /**
3985  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3986  * @dev: The device for which the DMA address was created
3987  * @addr: The DMA address
3988  * @size: The size of the region in bytes
3989  * @direction: The direction of the DMA
3990  */
ib_dma_unmap_page(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction direction)3991 static inline void ib_dma_unmap_page(struct ib_device *dev,
3992 				     u64 addr, size_t size,
3993 				     enum dma_data_direction direction)
3994 {
3995 	dma_unmap_page(dev->dma_device, addr, size, direction);
3996 }
3997 
3998 /**
3999  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4000  * @dev: The device for which the DMA addresses are to be created
4001  * @sg: The array of scatter/gather entries
4002  * @nents: The number of scatter/gather entries
4003  * @direction: The direction of the DMA
4004  */
ib_dma_map_sg(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction)4005 static inline int ib_dma_map_sg(struct ib_device *dev,
4006 				struct scatterlist *sg, int nents,
4007 				enum dma_data_direction direction)
4008 {
4009 	return dma_map_sg(dev->dma_device, sg, nents, direction);
4010 }
4011 
4012 /**
4013  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4014  * @dev: The device for which the DMA addresses were created
4015  * @sg: The array of scatter/gather entries
4016  * @nents: The number of scatter/gather entries
4017  * @direction: The direction of the DMA
4018  */
ib_dma_unmap_sg(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction)4019 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4020 				   struct scatterlist *sg, int nents,
4021 				   enum dma_data_direction direction)
4022 {
4023 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
4024 }
4025 
ib_dma_map_sg_attrs(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction,unsigned long dma_attrs)4026 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4027 				      struct scatterlist *sg, int nents,
4028 				      enum dma_data_direction direction,
4029 				      unsigned long dma_attrs)
4030 {
4031 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4032 				dma_attrs);
4033 }
4034 
ib_dma_unmap_sg_attrs(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction,unsigned long dma_attrs)4035 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4036 					 struct scatterlist *sg, int nents,
4037 					 enum dma_data_direction direction,
4038 					 unsigned long dma_attrs)
4039 {
4040 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
4041 }
4042 
4043 /**
4044  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4045  * @dev: The device to query
4046  *
4047  * The returned value represents a size in bytes.
4048  */
ib_dma_max_seg_size(struct ib_device * dev)4049 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4050 {
4051 	return dma_get_max_seg_size(dev->dma_device);
4052 }
4053 
4054 /**
4055  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4056  * @dev: The device for which the DMA address was created
4057  * @addr: The DMA address
4058  * @size: The size of the region in bytes
4059  * @dir: The direction of the DMA
4060  */
ib_dma_sync_single_for_cpu(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction dir)4061 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4062 					      u64 addr,
4063 					      size_t size,
4064 					      enum dma_data_direction dir)
4065 {
4066 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4067 }
4068 
4069 /**
4070  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4071  * @dev: The device for which the DMA address was created
4072  * @addr: The DMA address
4073  * @size: The size of the region in bytes
4074  * @dir: The direction of the DMA
4075  */
ib_dma_sync_single_for_device(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction dir)4076 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4077 						 u64 addr,
4078 						 size_t size,
4079 						 enum dma_data_direction dir)
4080 {
4081 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4082 }
4083 
4084 /**
4085  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4086  * @dev: The device for which the DMA address is requested
4087  * @size: The size of the region to allocate in bytes
4088  * @dma_handle: A pointer for returning the DMA address of the region
4089  * @flag: memory allocator flags
4090  */
ib_dma_alloc_coherent(struct ib_device * dev,size_t size,dma_addr_t * dma_handle,gfp_t flag)4091 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4092 					   size_t size,
4093 					   dma_addr_t *dma_handle,
4094 					   gfp_t flag)
4095 {
4096 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4097 }
4098 
4099 /**
4100  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4101  * @dev: The device for which the DMA addresses were allocated
4102  * @size: The size of the region
4103  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4104  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4105  */
ib_dma_free_coherent(struct ib_device * dev,size_t size,void * cpu_addr,dma_addr_t dma_handle)4106 static inline void ib_dma_free_coherent(struct ib_device *dev,
4107 					size_t size, void *cpu_addr,
4108 					dma_addr_t dma_handle)
4109 {
4110 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4111 }
4112 
4113 /**
4114  * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4115  *   HCA translation table.
4116  * @mr: The memory region to deregister.
4117  * @udata: Valid user data or NULL for kernel object
4118  *
4119  * This function can fail, if the memory region has memory windows bound to it.
4120  */
4121 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4122 
4123 /**
4124  * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4125  *   HCA translation table.
4126  * @mr: The memory region to deregister.
4127  *
4128  * This function can fail, if the memory region has memory windows bound to it.
4129  *
4130  * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4131  */
ib_dereg_mr(struct ib_mr * mr)4132 static inline int ib_dereg_mr(struct ib_mr *mr)
4133 {
4134 	return ib_dereg_mr_user(mr, NULL);
4135 }
4136 
4137 struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type,
4138 			       u32 max_num_sg, struct ib_udata *udata);
4139 
ib_alloc_mr(struct ib_pd * pd,enum ib_mr_type mr_type,u32 max_num_sg)4140 static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
4141 					enum ib_mr_type mr_type, u32 max_num_sg)
4142 {
4143 	return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL);
4144 }
4145 
4146 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4147 				    u32 max_num_data_sg,
4148 				    u32 max_num_meta_sg);
4149 
4150 /**
4151  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4152  *   R_Key and L_Key.
4153  * @mr - struct ib_mr pointer to be updated.
4154  * @newkey - new key to be used.
4155  */
ib_update_fast_reg_key(struct ib_mr * mr,u8 newkey)4156 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4157 {
4158 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4159 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4160 }
4161 
4162 /**
4163  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4164  * for calculating a new rkey for type 2 memory windows.
4165  * @rkey - the rkey to increment.
4166  */
ib_inc_rkey(u32 rkey)4167 static inline u32 ib_inc_rkey(u32 rkey)
4168 {
4169 	const u32 mask = 0x000000ff;
4170 	return ((rkey + 1) & mask) | (rkey & ~mask);
4171 }
4172 
4173 /**
4174  * ib_alloc_fmr - Allocates a unmapped fast memory region.
4175  * @pd: The protection domain associated with the unmapped region.
4176  * @mr_access_flags: Specifies the memory access rights.
4177  * @fmr_attr: Attributes of the unmapped region.
4178  *
4179  * A fast memory region must be mapped before it can be used as part of
4180  * a work request.
4181  */
4182 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
4183 			    int mr_access_flags,
4184 			    struct ib_fmr_attr *fmr_attr);
4185 
4186 /**
4187  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
4188  * @fmr: The fast memory region to associate with the pages.
4189  * @page_list: An array of physical pages to map to the fast memory region.
4190  * @list_len: The number of pages in page_list.
4191  * @iova: The I/O virtual address to use with the mapped region.
4192  */
ib_map_phys_fmr(struct ib_fmr * fmr,u64 * page_list,int list_len,u64 iova)4193 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
4194 				  u64 *page_list, int list_len,
4195 				  u64 iova)
4196 {
4197 	return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
4198 }
4199 
4200 /**
4201  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
4202  * @fmr_list: A linked list of fast memory regions to unmap.
4203  */
4204 int ib_unmap_fmr(struct list_head *fmr_list);
4205 
4206 /**
4207  * ib_dealloc_fmr - Deallocates a fast memory region.
4208  * @fmr: The fast memory region to deallocate.
4209  */
4210 int ib_dealloc_fmr(struct ib_fmr *fmr);
4211 
4212 /**
4213  * ib_attach_mcast - Attaches the specified QP to a multicast group.
4214  * @qp: QP to attach to the multicast group.  The QP must be type
4215  *   IB_QPT_UD.
4216  * @gid: Multicast group GID.
4217  * @lid: Multicast group LID in host byte order.
4218  *
4219  * In order to send and receive multicast packets, subnet
4220  * administration must have created the multicast group and configured
4221  * the fabric appropriately.  The port associated with the specified
4222  * QP must also be a member of the multicast group.
4223  */
4224 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4225 
4226 /**
4227  * ib_detach_mcast - Detaches the specified QP from a multicast group.
4228  * @qp: QP to detach from the multicast group.
4229  * @gid: Multicast group GID.
4230  * @lid: Multicast group LID in host byte order.
4231  */
4232 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4233 
4234 /**
4235  * ib_alloc_xrcd - Allocates an XRC domain.
4236  * @device: The device on which to allocate the XRC domain.
4237  * @caller: Module name for kernel consumers
4238  */
4239 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
4240 #define ib_alloc_xrcd(device) \
4241 	__ib_alloc_xrcd((device), KBUILD_MODNAME)
4242 
4243 /**
4244  * ib_dealloc_xrcd - Deallocates an XRC domain.
4245  * @xrcd: The XRC domain to deallocate.
4246  * @udata: Valid user data or NULL for kernel object
4247  */
4248 int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
4249 
ib_check_mr_access(int flags)4250 static inline int ib_check_mr_access(int flags)
4251 {
4252 	/*
4253 	 * Local write permission is required if remote write or
4254 	 * remote atomic permission is also requested.
4255 	 */
4256 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4257 	    !(flags & IB_ACCESS_LOCAL_WRITE))
4258 		return -EINVAL;
4259 
4260 	if (flags & ~IB_ACCESS_SUPPORTED)
4261 		return -EINVAL;
4262 
4263 	return 0;
4264 }
4265 
ib_access_writable(int access_flags)4266 static inline bool ib_access_writable(int access_flags)
4267 {
4268 	/*
4269 	 * We have writable memory backing the MR if any of the following
4270 	 * access flags are set.  "Local write" and "remote write" obviously
4271 	 * require write access.  "Remote atomic" can do things like fetch and
4272 	 * add, which will modify memory, and "MW bind" can change permissions
4273 	 * by binding a window.
4274 	 */
4275 	return access_flags &
4276 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
4277 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4278 }
4279 
4280 /**
4281  * ib_check_mr_status: lightweight check of MR status.
4282  *     This routine may provide status checks on a selected
4283  *     ib_mr. first use is for signature status check.
4284  *
4285  * @mr: A memory region.
4286  * @check_mask: Bitmask of which checks to perform from
4287  *     ib_mr_status_check enumeration.
4288  * @mr_status: The container of relevant status checks.
4289  *     failed checks will be indicated in the status bitmask
4290  *     and the relevant info shall be in the error item.
4291  */
4292 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4293 		       struct ib_mr_status *mr_status);
4294 
4295 /**
4296  * ib_device_try_get: Hold a registration lock
4297  * device: The device to lock
4298  *
4299  * A device under an active registration lock cannot become unregistered. It
4300  * is only possible to obtain a registration lock on a device that is fully
4301  * registered, otherwise this function returns false.
4302  *
4303  * The registration lock is only necessary for actions which require the
4304  * device to still be registered. Uses that only require the device pointer to
4305  * be valid should use get_device(&ibdev->dev) to hold the memory.
4306  *
4307  */
ib_device_try_get(struct ib_device * dev)4308 static inline bool ib_device_try_get(struct ib_device *dev)
4309 {
4310 	return refcount_inc_not_zero(&dev->refcount);
4311 }
4312 
4313 void ib_device_put(struct ib_device *device);
4314 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4315 					  enum rdma_driver_id driver_id);
4316 struct ib_device *ib_device_get_by_name(const char *name,
4317 					enum rdma_driver_id driver_id);
4318 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4319 					    u16 pkey, const union ib_gid *gid,
4320 					    const struct sockaddr *addr);
4321 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4322 			 unsigned int port);
4323 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4324 
4325 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4326 			   struct ib_wq_init_attr *init_attr);
4327 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
4328 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4329 		 u32 wq_attr_mask);
4330 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4331 						 struct ib_rwq_ind_table_init_attr*
4332 						 wq_ind_table_init_attr);
4333 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4334 
4335 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4336 		 unsigned int *sg_offset, unsigned int page_size);
4337 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4338 		    int data_sg_nents, unsigned int *data_sg_offset,
4339 		    struct scatterlist *meta_sg, int meta_sg_nents,
4340 		    unsigned int *meta_sg_offset, unsigned int page_size);
4341 
4342 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)4343 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4344 		  unsigned int *sg_offset, unsigned int page_size)
4345 {
4346 	int n;
4347 
4348 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4349 	mr->iova = 0;
4350 
4351 	return n;
4352 }
4353 
4354 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4355 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4356 
4357 void ib_drain_rq(struct ib_qp *qp);
4358 void ib_drain_sq(struct ib_qp *qp);
4359 void ib_drain_qp(struct ib_qp *qp);
4360 
4361 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4362 
rdma_ah_retrieve_dmac(struct rdma_ah_attr * attr)4363 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4364 {
4365 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4366 		return attr->roce.dmac;
4367 	return NULL;
4368 }
4369 
rdma_ah_set_dlid(struct rdma_ah_attr * attr,u32 dlid)4370 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4371 {
4372 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4373 		attr->ib.dlid = (u16)dlid;
4374 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4375 		attr->opa.dlid = dlid;
4376 }
4377 
rdma_ah_get_dlid(const struct rdma_ah_attr * attr)4378 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4379 {
4380 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4381 		return attr->ib.dlid;
4382 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4383 		return attr->opa.dlid;
4384 	return 0;
4385 }
4386 
rdma_ah_set_sl(struct rdma_ah_attr * attr,u8 sl)4387 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4388 {
4389 	attr->sl = sl;
4390 }
4391 
rdma_ah_get_sl(const struct rdma_ah_attr * attr)4392 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4393 {
4394 	return attr->sl;
4395 }
4396 
rdma_ah_set_path_bits(struct rdma_ah_attr * attr,u8 src_path_bits)4397 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4398 					 u8 src_path_bits)
4399 {
4400 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4401 		attr->ib.src_path_bits = src_path_bits;
4402 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4403 		attr->opa.src_path_bits = src_path_bits;
4404 }
4405 
rdma_ah_get_path_bits(const struct rdma_ah_attr * attr)4406 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4407 {
4408 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4409 		return attr->ib.src_path_bits;
4410 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4411 		return attr->opa.src_path_bits;
4412 	return 0;
4413 }
4414 
rdma_ah_set_make_grd(struct rdma_ah_attr * attr,bool make_grd)4415 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4416 					bool make_grd)
4417 {
4418 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4419 		attr->opa.make_grd = make_grd;
4420 }
4421 
rdma_ah_get_make_grd(const struct rdma_ah_attr * attr)4422 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4423 {
4424 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4425 		return attr->opa.make_grd;
4426 	return false;
4427 }
4428 
rdma_ah_set_port_num(struct rdma_ah_attr * attr,u8 port_num)4429 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4430 {
4431 	attr->port_num = port_num;
4432 }
4433 
rdma_ah_get_port_num(const struct rdma_ah_attr * attr)4434 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4435 {
4436 	return attr->port_num;
4437 }
4438 
rdma_ah_set_static_rate(struct rdma_ah_attr * attr,u8 static_rate)4439 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4440 					   u8 static_rate)
4441 {
4442 	attr->static_rate = static_rate;
4443 }
4444 
rdma_ah_get_static_rate(const struct rdma_ah_attr * attr)4445 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4446 {
4447 	return attr->static_rate;
4448 }
4449 
rdma_ah_set_ah_flags(struct rdma_ah_attr * attr,enum ib_ah_flags flag)4450 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4451 					enum ib_ah_flags flag)
4452 {
4453 	attr->ah_flags = flag;
4454 }
4455 
4456 static inline enum ib_ah_flags
rdma_ah_get_ah_flags(const struct rdma_ah_attr * attr)4457 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4458 {
4459 	return attr->ah_flags;
4460 }
4461 
4462 static inline const struct ib_global_route
rdma_ah_read_grh(const struct rdma_ah_attr * attr)4463 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4464 {
4465 	return &attr->grh;
4466 }
4467 
4468 /*To retrieve and modify the grh */
4469 static inline struct ib_global_route
rdma_ah_retrieve_grh(struct rdma_ah_attr * attr)4470 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4471 {
4472 	return &attr->grh;
4473 }
4474 
rdma_ah_set_dgid_raw(struct rdma_ah_attr * attr,void * dgid)4475 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4476 {
4477 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4478 
4479 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4480 }
4481 
rdma_ah_set_subnet_prefix(struct rdma_ah_attr * attr,__be64 prefix)4482 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4483 					     __be64 prefix)
4484 {
4485 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4486 
4487 	grh->dgid.global.subnet_prefix = prefix;
4488 }
4489 
rdma_ah_set_interface_id(struct rdma_ah_attr * attr,__be64 if_id)4490 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4491 					    __be64 if_id)
4492 {
4493 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4494 
4495 	grh->dgid.global.interface_id = if_id;
4496 }
4497 
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)4498 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4499 				   union ib_gid *dgid, u32 flow_label,
4500 				   u8 sgid_index, u8 hop_limit,
4501 				   u8 traffic_class)
4502 {
4503 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4504 
4505 	attr->ah_flags = IB_AH_GRH;
4506 	if (dgid)
4507 		grh->dgid = *dgid;
4508 	grh->flow_label = flow_label;
4509 	grh->sgid_index = sgid_index;
4510 	grh->hop_limit = hop_limit;
4511 	grh->traffic_class = traffic_class;
4512 	grh->sgid_attr = NULL;
4513 }
4514 
4515 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4516 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4517 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4518 			     const struct ib_gid_attr *sgid_attr);
4519 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4520 		       const struct rdma_ah_attr *src);
4521 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4522 			  const struct rdma_ah_attr *new);
4523 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4524 
4525 /**
4526  * rdma_ah_find_type - Return address handle type.
4527  *
4528  * @dev: Device to be checked
4529  * @port_num: Port number
4530  */
rdma_ah_find_type(struct ib_device * dev,u8 port_num)4531 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4532 						       u8 port_num)
4533 {
4534 	if (rdma_protocol_roce(dev, port_num))
4535 		return RDMA_AH_ATTR_TYPE_ROCE;
4536 	if (rdma_protocol_ib(dev, port_num)) {
4537 		if (rdma_cap_opa_ah(dev, port_num))
4538 			return RDMA_AH_ATTR_TYPE_OPA;
4539 		return RDMA_AH_ATTR_TYPE_IB;
4540 	}
4541 
4542 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4543 }
4544 
4545 /**
4546  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4547  *     In the current implementation the only way to get
4548  *     get the 32bit lid is from other sources for OPA.
4549  *     For IB, lids will always be 16bits so cast the
4550  *     value accordingly.
4551  *
4552  * @lid: A 32bit LID
4553  */
ib_lid_cpu16(u32 lid)4554 static inline u16 ib_lid_cpu16(u32 lid)
4555 {
4556 	WARN_ON_ONCE(lid & 0xFFFF0000);
4557 	return (u16)lid;
4558 }
4559 
4560 /**
4561  * ib_lid_be16 - Return lid in 16bit BE encoding.
4562  *
4563  * @lid: A 32bit LID
4564  */
ib_lid_be16(u32 lid)4565 static inline __be16 ib_lid_be16(u32 lid)
4566 {
4567 	WARN_ON_ONCE(lid & 0xFFFF0000);
4568 	return cpu_to_be16((u16)lid);
4569 }
4570 
4571 /**
4572  * ib_get_vector_affinity - Get the affinity mappings of a given completion
4573  *   vector
4574  * @device:         the rdma device
4575  * @comp_vector:    index of completion vector
4576  *
4577  * Returns NULL on failure, otherwise a corresponding cpu map of the
4578  * completion vector (returns all-cpus map if the device driver doesn't
4579  * implement get_vector_affinity).
4580  */
4581 static inline const struct cpumask *
ib_get_vector_affinity(struct ib_device * device,int comp_vector)4582 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4583 {
4584 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4585 	    !device->ops.get_vector_affinity)
4586 		return NULL;
4587 
4588 	return device->ops.get_vector_affinity(device, comp_vector);
4589 
4590 }
4591 
4592 /**
4593  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4594  * and add their gids, as needed, to the relevant RoCE devices.
4595  *
4596  * @device:         the rdma device
4597  */
4598 void rdma_roce_rescan_device(struct ib_device *ibdev);
4599 
4600 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4601 
4602 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4603 
4604 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4605 				     enum rdma_netdev_t type, const char *name,
4606 				     unsigned char name_assign_type,
4607 				     void (*setup)(struct net_device *));
4608 
4609 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4610 		     enum rdma_netdev_t type, const char *name,
4611 		     unsigned char name_assign_type,
4612 		     void (*setup)(struct net_device *),
4613 		     struct net_device *netdev);
4614 
4615 /**
4616  * rdma_set_device_sysfs_group - Set device attributes group to have
4617  *				 driver specific sysfs entries at
4618  *				 for infiniband class.
4619  *
4620  * @device:	device pointer for which attributes to be created
4621  * @group:	Pointer to group which should be added when device
4622  *		is registered with sysfs.
4623  * rdma_set_device_sysfs_group() allows existing drivers to expose one
4624  * group per device to have sysfs attributes.
4625  *
4626  * NOTE: New drivers should not make use of this API; instead new device
4627  * parameter should be exposed via netlink command. This API and mechanism
4628  * exist only for existing drivers.
4629  */
4630 static inline void
rdma_set_device_sysfs_group(struct ib_device * dev,const struct attribute_group * group)4631 rdma_set_device_sysfs_group(struct ib_device *dev,
4632 			    const struct attribute_group *group)
4633 {
4634 	dev->groups[1] = group;
4635 }
4636 
4637 /**
4638  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4639  *
4640  * @device:	device pointer for which ib_device pointer to retrieve
4641  *
4642  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4643  *
4644  */
rdma_device_to_ibdev(struct device * device)4645 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4646 {
4647 	struct ib_core_device *coredev =
4648 		container_of(device, struct ib_core_device, dev);
4649 
4650 	return coredev->owner;
4651 }
4652 
4653 /**
4654  * rdma_device_to_drv_device - Helper macro to reach back to driver's
4655  *			       ib_device holder structure from device pointer.
4656  *
4657  * NOTE: New drivers should not make use of this API; This API is only for
4658  * existing drivers who have exposed sysfs entries using
4659  * rdma_set_device_sysfs_group().
4660  */
4661 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
4662 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4663 
4664 bool rdma_dev_access_netns(const struct ib_device *device,
4665 			   const struct net *net);
4666 #endif /* IB_VERBS_H */
4667