xref: /linux/include/rdma/ib_verbs.h (revision f30bc6f9b9cc492634a333be9c6aa9755ca1bf17)
1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
2 /*
3  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
4  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
5  * Copyright (c) 2004, 2020 Intel Corporation.  All rights reserved.
6  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
7  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
8  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
9  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
10  */
11 
12 #ifndef IB_VERBS_H
13 #define IB_VERBS_H
14 
15 #include <linux/ethtool.h>
16 #include <linux/types.h>
17 #include <linux/device.h>
18 #include <linux/bvec.h>
19 #include <linux/dma-mapping.h>
20 #include <linux/kref.h>
21 #include <linux/list.h>
22 #include <linux/rwsem.h>
23 #include <linux/workqueue.h>
24 #include <linux/irq_poll.h>
25 #include <uapi/linux/if_ether.h>
26 #include <net/ipv6.h>
27 #include <net/ip.h>
28 #include <linux/string.h>
29 #include <linux/slab.h>
30 #include <linux/netdevice.h>
31 #include <linux/refcount.h>
32 #include <linux/if_link.h>
33 #include <linux/atomic.h>
34 #include <linux/mmu_notifier.h>
35 #include <linux/uaccess.h>
36 #include <linux/cgroup_rdma.h>
37 #include <linux/irqflags.h>
38 #include <linux/preempt.h>
39 #include <linux/dim.h>
40 #include <uapi/rdma/ib_user_verbs.h>
41 #include <rdma/rdma_counter.h>
42 #include <rdma/restrack.h>
43 #include <rdma/signature.h>
44 #include <uapi/rdma/rdma_user_ioctl.h>
45 #include <uapi/rdma/ib_user_ioctl_verbs.h>
46 #include <linux/pci-tph.h>
47 #include <rdma/frmr_pools.h>
48 #include <linux/dma-buf.h>
49 
50 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
51 
52 struct ib_umem_odp;
53 struct ib_uqp_object;
54 struct ib_usrq_object;
55 struct ib_uwq_object;
56 struct rdma_cm_id;
57 struct ib_port;
58 struct hw_stats_device_data;
59 
60 extern struct workqueue_struct *ib_wq;
61 extern struct workqueue_struct *ib_comp_wq;
62 extern struct workqueue_struct *ib_comp_unbound_wq;
63 
64 struct ib_ucq_object;
65 
66 __printf(2, 3) __cold
67 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
68 __printf(2, 3) __cold
69 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
70 __printf(2, 3) __cold
71 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
72 __printf(2, 3) __cold
73 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
74 __printf(2, 3) __cold
75 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
76 __printf(2, 3) __cold
77 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
78 __printf(2, 3) __cold
79 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
80 
81 #if defined(CONFIG_DYNAMIC_DEBUG) || \
82 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
83 #define ibdev_dbg(__dev, format, args...)                       \
84 	dynamic_ibdev_dbg(__dev, format, ##args)
85 #else
86 __printf(2, 3) __cold
87 static inline
88 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
89 #endif
90 
91 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...)           \
92 do {                                                                    \
93 	static DEFINE_RATELIMIT_STATE(_rs,                              \
94 				      DEFAULT_RATELIMIT_INTERVAL,       \
95 				      DEFAULT_RATELIMIT_BURST);         \
96 	if (__ratelimit(&_rs))                                          \
97 		ibdev_level(ibdev, fmt, ##__VA_ARGS__);                 \
98 } while (0)
99 
100 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
101 	ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
102 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
103 	ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
104 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
105 	ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
106 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
107 	ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
108 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
109 	ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
110 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
111 	ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
112 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
113 	ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
114 
115 #if defined(CONFIG_DYNAMIC_DEBUG) || \
116 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
117 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
118 #define ibdev_dbg_ratelimited(ibdev, fmt, ...)                          \
119 do {                                                                    \
120 	static DEFINE_RATELIMIT_STATE(_rs,                              \
121 				      DEFAULT_RATELIMIT_INTERVAL,       \
122 				      DEFAULT_RATELIMIT_BURST);         \
123 	DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt);                 \
124 	if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs))      \
125 		__dynamic_ibdev_dbg(&descriptor, ibdev, fmt,            \
126 				    ##__VA_ARGS__);                     \
127 } while (0)
128 #else
129 __printf(2, 3) __cold
130 static inline
131 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
132 #endif
133 
134 union ib_gid {
135 	u8	raw[16];
136 	struct {
137 		__be64	subnet_prefix;
138 		__be64	interface_id;
139 	} global;
140 };
141 
142 extern union ib_gid zgid;
143 
144 enum ib_gid_type {
145 	IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
146 	IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
147 	IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
148 	IB_GID_TYPE_SIZE
149 };
150 
151 #define ROCE_V2_UDP_DPORT      4791
152 struct ib_gid_attr {
153 	struct net_device __rcu	*ndev;
154 	struct ib_device	*device;
155 	union ib_gid		gid;
156 	enum ib_gid_type	gid_type;
157 	u16			index;
158 	u32			port_num;
159 };
160 
161 enum {
162 	/* set the local administered indication */
163 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
164 };
165 
166 enum rdma_transport_type {
167 	RDMA_TRANSPORT_IB,
168 	RDMA_TRANSPORT_IWARP,
169 	RDMA_TRANSPORT_USNIC,
170 	RDMA_TRANSPORT_USNIC_UDP,
171 	RDMA_TRANSPORT_UNSPECIFIED,
172 };
173 
174 enum rdma_protocol_type {
175 	RDMA_PROTOCOL_IB,
176 	RDMA_PROTOCOL_IBOE,
177 	RDMA_PROTOCOL_IWARP,
178 	RDMA_PROTOCOL_USNIC_UDP
179 };
180 
181 __attribute_const__ enum rdma_transport_type
182 rdma_node_get_transport(unsigned int node_type);
183 
184 enum rdma_network_type {
185 	RDMA_NETWORK_IB,
186 	RDMA_NETWORK_ROCE_V1,
187 	RDMA_NETWORK_IPV4,
188 	RDMA_NETWORK_IPV6
189 };
190 
191 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
192 {
193 	if (network_type == RDMA_NETWORK_IPV4 ||
194 	    network_type == RDMA_NETWORK_IPV6)
195 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
196 	else if (network_type == RDMA_NETWORK_ROCE_V1)
197 		return IB_GID_TYPE_ROCE;
198 	else
199 		return IB_GID_TYPE_IB;
200 }
201 
202 static inline enum rdma_network_type
203 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
204 {
205 	if (attr->gid_type == IB_GID_TYPE_IB)
206 		return RDMA_NETWORK_IB;
207 
208 	if (attr->gid_type == IB_GID_TYPE_ROCE)
209 		return RDMA_NETWORK_ROCE_V1;
210 
211 	if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
212 		return RDMA_NETWORK_IPV4;
213 	else
214 		return RDMA_NETWORK_IPV6;
215 }
216 
217 enum rdma_link_layer {
218 	IB_LINK_LAYER_UNSPECIFIED,
219 	IB_LINK_LAYER_INFINIBAND,
220 	IB_LINK_LAYER_ETHERNET,
221 };
222 
223 enum ib_device_cap_flags {
224 	IB_DEVICE_RESIZE_MAX_WR = IB_UVERBS_DEVICE_RESIZE_MAX_WR,
225 	IB_DEVICE_BAD_PKEY_CNTR = IB_UVERBS_DEVICE_BAD_PKEY_CNTR,
226 	IB_DEVICE_BAD_QKEY_CNTR = IB_UVERBS_DEVICE_BAD_QKEY_CNTR,
227 	IB_DEVICE_RAW_MULTI = IB_UVERBS_DEVICE_RAW_MULTI,
228 	IB_DEVICE_AUTO_PATH_MIG = IB_UVERBS_DEVICE_AUTO_PATH_MIG,
229 	IB_DEVICE_CHANGE_PHY_PORT = IB_UVERBS_DEVICE_CHANGE_PHY_PORT,
230 	IB_DEVICE_UD_AV_PORT_ENFORCE = IB_UVERBS_DEVICE_UD_AV_PORT_ENFORCE,
231 	IB_DEVICE_CURR_QP_STATE_MOD = IB_UVERBS_DEVICE_CURR_QP_STATE_MOD,
232 	IB_DEVICE_SHUTDOWN_PORT = IB_UVERBS_DEVICE_SHUTDOWN_PORT,
233 	/* IB_DEVICE_INIT_TYPE = IB_UVERBS_DEVICE_INIT_TYPE, (not in use) */
234 	IB_DEVICE_PORT_ACTIVE_EVENT = IB_UVERBS_DEVICE_PORT_ACTIVE_EVENT,
235 	IB_DEVICE_SYS_IMAGE_GUID = IB_UVERBS_DEVICE_SYS_IMAGE_GUID,
236 	IB_DEVICE_RC_RNR_NAK_GEN = IB_UVERBS_DEVICE_RC_RNR_NAK_GEN,
237 	IB_DEVICE_SRQ_RESIZE = IB_UVERBS_DEVICE_SRQ_RESIZE,
238 	IB_DEVICE_N_NOTIFY_CQ = IB_UVERBS_DEVICE_N_NOTIFY_CQ,
239 
240 	/* Reserved, old SEND_W_INV = 1 << 16,*/
241 	IB_DEVICE_MEM_WINDOW = IB_UVERBS_DEVICE_MEM_WINDOW,
242 	/*
243 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
244 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
245 	 * messages and can verify the validity of checksum for
246 	 * incoming messages.  Setting this flag implies that the
247 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
248 	 */
249 	IB_DEVICE_UD_IP_CSUM = IB_UVERBS_DEVICE_UD_IP_CSUM,
250 	IB_DEVICE_XRC = IB_UVERBS_DEVICE_XRC,
251 
252 	/*
253 	 * This device supports the IB "base memory management extension",
254 	 * which includes support for fast registrations (IB_WR_REG_MR,
255 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
256 	 * also be set by any iWarp device which must support FRs to comply
257 	 * to the iWarp verbs spec.  iWarp devices also support the
258 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
259 	 * stag.
260 	 */
261 	IB_DEVICE_MEM_MGT_EXTENSIONS = IB_UVERBS_DEVICE_MEM_MGT_EXTENSIONS,
262 	IB_DEVICE_MEM_WINDOW_TYPE_2A = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2A,
263 	IB_DEVICE_MEM_WINDOW_TYPE_2B = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2B,
264 	IB_DEVICE_RC_IP_CSUM = IB_UVERBS_DEVICE_RC_IP_CSUM,
265 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
266 	IB_DEVICE_RAW_IP_CSUM = IB_UVERBS_DEVICE_RAW_IP_CSUM,
267 	IB_DEVICE_MANAGED_FLOW_STEERING =
268 		IB_UVERBS_DEVICE_MANAGED_FLOW_STEERING,
269 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
270 	IB_DEVICE_RAW_SCATTER_FCS = IB_UVERBS_DEVICE_RAW_SCATTER_FCS,
271 	/* The device supports padding incoming writes to cacheline. */
272 	IB_DEVICE_PCI_WRITE_END_PADDING =
273 		IB_UVERBS_DEVICE_PCI_WRITE_END_PADDING,
274 	/* Placement type attributes */
275 	IB_DEVICE_FLUSH_GLOBAL = IB_UVERBS_DEVICE_FLUSH_GLOBAL,
276 	IB_DEVICE_FLUSH_PERSISTENT = IB_UVERBS_DEVICE_FLUSH_PERSISTENT,
277 	IB_DEVICE_ATOMIC_WRITE = IB_UVERBS_DEVICE_ATOMIC_WRITE,
278 };
279 
280 enum ib_kernel_cap_flags {
281 	/*
282 	 * This device supports a per-device lkey or stag that can be
283 	 * used without performing a memory registration for the local
284 	 * memory.  Note that ULPs should never check this flag, but
285 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
286 	 * which will always contain a usable lkey.
287 	 */
288 	IBK_LOCAL_DMA_LKEY = 1 << 0,
289 	/* IB_QP_CREATE_INTEGRITY_EN is supported to implement T10-PI */
290 	IBK_INTEGRITY_HANDOVER = 1 << 1,
291 	/* IB_ACCESS_ON_DEMAND is supported during reg_user_mr() */
292 	IBK_ON_DEMAND_PAGING = 1 << 2,
293 	/* IB_MR_TYPE_SG_GAPS is supported */
294 	IBK_SG_GAPS_REG = 1 << 3,
295 	/* Driver supports RDMA_NLDEV_CMD_DELLINK */
296 	IBK_ALLOW_USER_UNREG = 1 << 4,
297 
298 	/* ipoib will use IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK */
299 	IBK_BLOCK_MULTICAST_LOOPBACK = 1 << 5,
300 	/* iopib will use IB_QP_CREATE_IPOIB_UD_LSO for its QPs */
301 	IBK_UD_TSO = 1 << 6,
302 	/* iopib will use the device ops:
303 	 *   get_vf_config
304 	 *   get_vf_guid
305 	 *   get_vf_stats
306 	 *   set_vf_guid
307 	 *   set_vf_link_state
308 	 */
309 	IBK_VIRTUAL_FUNCTION = 1 << 7,
310 	/* ipoib will use IB_QP_CREATE_NETDEV_USE for its QPs */
311 	IBK_RDMA_NETDEV_OPA = 1 << 8,
312 };
313 
314 enum ib_atomic_cap {
315 	IB_ATOMIC_NONE,
316 	IB_ATOMIC_HCA,
317 	IB_ATOMIC_GLOB
318 };
319 
320 enum ib_odp_general_cap_bits {
321 	IB_ODP_SUPPORT		= IB_UVERBS_ODP_SUPPORT,
322 	IB_ODP_SUPPORT_IMPLICIT = IB_UVERBS_ODP_SUPPORT_IMPLICIT,
323 };
324 
325 enum ib_odp_transport_cap_bits {
326 	IB_ODP_SUPPORT_SEND	= IB_UVERBS_ODP_SUPPORT_SEND,
327 	IB_ODP_SUPPORT_RECV	= IB_UVERBS_ODP_SUPPORT_RECV,
328 	IB_ODP_SUPPORT_WRITE	= IB_UVERBS_ODP_SUPPORT_WRITE,
329 	IB_ODP_SUPPORT_READ	= IB_UVERBS_ODP_SUPPORT_READ,
330 	IB_ODP_SUPPORT_ATOMIC	= IB_UVERBS_ODP_SUPPORT_ATOMIC,
331 	IB_ODP_SUPPORT_SRQ_RECV	= IB_UVERBS_ODP_SUPPORT_SRQ_RECV,
332 	IB_ODP_SUPPORT_FLUSH	= IB_UVERBS_ODP_SUPPORT_FLUSH,
333 	IB_ODP_SUPPORT_ATOMIC_WRITE	= IB_UVERBS_ODP_SUPPORT_ATOMIC_WRITE,
334 };
335 
336 struct ib_odp_caps {
337 	uint64_t general_caps;
338 	struct {
339 		uint32_t  rc_odp_caps;
340 		uint32_t  uc_odp_caps;
341 		uint32_t  ud_odp_caps;
342 		uint32_t  xrc_odp_caps;
343 	} per_transport_caps;
344 };
345 
346 struct ib_rss_caps {
347 	/* Corresponding bit will be set if qp type from
348 	 * 'enum ib_qp_type' is supported, e.g.
349 	 * supported_qpts |= 1 << IB_QPT_UD
350 	 */
351 	u32 supported_qpts;
352 	u32 max_rwq_indirection_tables;
353 	u32 max_rwq_indirection_table_size;
354 };
355 
356 enum ib_tm_cap_flags {
357 	/*  Support tag matching with rendezvous offload for RC transport */
358 	IB_TM_CAP_RNDV_RC = 1 << 0,
359 };
360 
361 struct ib_tm_caps {
362 	/* Max size of RNDV header */
363 	u32 max_rndv_hdr_size;
364 	/* Max number of entries in tag matching list */
365 	u32 max_num_tags;
366 	/* From enum ib_tm_cap_flags */
367 	u32 flags;
368 	/* Max number of outstanding list operations */
369 	u32 max_ops;
370 	/* Max number of SGE in tag matching entry */
371 	u32 max_sge;
372 };
373 
374 struct ib_cq_init_attr {
375 	unsigned int	cqe;
376 	u32		comp_vector;
377 	u32		flags;
378 };
379 
380 enum ib_cq_attr_mask {
381 	IB_CQ_MODERATE = 1 << 0,
382 };
383 
384 struct ib_cq_caps {
385 	u16     max_cq_moderation_count;
386 	u16     max_cq_moderation_period;
387 };
388 
389 struct ib_dm_mr_attr {
390 	u64		length;
391 	u64		offset;
392 	u32		access_flags;
393 };
394 
395 struct ib_dm_alloc_attr {
396 	u64	length;
397 	u32	alignment;
398 	u32	flags;
399 };
400 
401 struct ib_device_attr {
402 	u64			fw_ver;
403 	__be64			sys_image_guid;
404 	u64			max_mr_size;
405 	u64			page_size_cap;
406 	u32			vendor_id;
407 	u32			vendor_part_id;
408 	u32			hw_ver;
409 	int			max_qp;
410 	int			max_qp_wr;
411 	u64			device_cap_flags;
412 	u64			kernel_cap_flags;
413 	int			max_send_sge;
414 	int			max_recv_sge;
415 	int			max_sge_rd;
416 	int			max_cq;
417 	int			max_cqe;
418 	int			max_mr;
419 	int			max_pd;
420 	int			max_qp_rd_atom;
421 	int			max_ee_rd_atom;
422 	int			max_res_rd_atom;
423 	int			max_qp_init_rd_atom;
424 	int			max_ee_init_rd_atom;
425 	enum ib_atomic_cap	atomic_cap;
426 	enum ib_atomic_cap	masked_atomic_cap;
427 	int			max_ee;
428 	int			max_rdd;
429 	int			max_mw;
430 	int			max_raw_ipv6_qp;
431 	int			max_raw_ethy_qp;
432 	int			max_mcast_grp;
433 	int			max_mcast_qp_attach;
434 	int			max_total_mcast_qp_attach;
435 	int			max_ah;
436 	int			max_srq;
437 	int			max_srq_wr;
438 	int			max_srq_sge;
439 	unsigned int		max_fast_reg_page_list_len;
440 	unsigned int		max_pi_fast_reg_page_list_len;
441 	u16			max_pkeys;
442 	u8			local_ca_ack_delay;
443 	int			sig_prot_cap;
444 	int			sig_guard_cap;
445 	struct ib_odp_caps	odp_caps;
446 	uint64_t		timestamp_mask;
447 	uint64_t		hca_core_clock; /* in KHZ */
448 	struct ib_rss_caps	rss_caps;
449 	u32			max_wq_type_rq;
450 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
451 	struct ib_tm_caps	tm_caps;
452 	struct ib_cq_caps       cq_caps;
453 	u64			max_dm_size;
454 	/* Max entries for sgl for optimized performance per READ */
455 	u32			max_sgl_rd;
456 };
457 
458 enum ib_mtu {
459 	IB_MTU_256  = 1,
460 	IB_MTU_512  = 2,
461 	IB_MTU_1024 = 3,
462 	IB_MTU_2048 = 4,
463 	IB_MTU_4096 = 5
464 };
465 
466 enum opa_mtu {
467 	OPA_MTU_8192 = 6,
468 	OPA_MTU_10240 = 7
469 };
470 
471 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
472 {
473 	switch (mtu) {
474 	case IB_MTU_256:  return  256;
475 	case IB_MTU_512:  return  512;
476 	case IB_MTU_1024: return 1024;
477 	case IB_MTU_2048: return 2048;
478 	case IB_MTU_4096: return 4096;
479 	default: 	  return -1;
480 	}
481 }
482 
483 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
484 {
485 	if (mtu >= 4096)
486 		return IB_MTU_4096;
487 	else if (mtu >= 2048)
488 		return IB_MTU_2048;
489 	else if (mtu >= 1024)
490 		return IB_MTU_1024;
491 	else if (mtu >= 512)
492 		return IB_MTU_512;
493 	else
494 		return IB_MTU_256;
495 }
496 
497 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
498 {
499 	switch (mtu) {
500 	case OPA_MTU_8192:
501 		return 8192;
502 	case OPA_MTU_10240:
503 		return 10240;
504 	default:
505 		return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
506 	}
507 }
508 
509 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
510 {
511 	if (mtu >= 10240)
512 		return OPA_MTU_10240;
513 	else if (mtu >= 8192)
514 		return OPA_MTU_8192;
515 	else
516 		return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
517 }
518 
519 enum ib_port_state {
520 	IB_PORT_NOP		= 0,
521 	IB_PORT_DOWN		= 1,
522 	IB_PORT_INIT		= 2,
523 	IB_PORT_ARMED		= 3,
524 	IB_PORT_ACTIVE		= 4,
525 	IB_PORT_ACTIVE_DEFER	= 5
526 };
527 
528 static inline const char *__attribute_const__
529 ib_port_state_to_str(enum ib_port_state state)
530 {
531 	const char * const states[] = {
532 		[IB_PORT_NOP] = "NOP",
533 		[IB_PORT_DOWN] = "DOWN",
534 		[IB_PORT_INIT] = "INIT",
535 		[IB_PORT_ARMED] = "ARMED",
536 		[IB_PORT_ACTIVE] = "ACTIVE",
537 		[IB_PORT_ACTIVE_DEFER] = "ACTIVE_DEFER",
538 	};
539 
540 	if (state < ARRAY_SIZE(states))
541 		return states[state];
542 	return "UNKNOWN";
543 }
544 
545 enum ib_port_phys_state {
546 	IB_PORT_PHYS_STATE_SLEEP = 1,
547 	IB_PORT_PHYS_STATE_POLLING = 2,
548 	IB_PORT_PHYS_STATE_DISABLED = 3,
549 	IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
550 	IB_PORT_PHYS_STATE_LINK_UP = 5,
551 	IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
552 	IB_PORT_PHYS_STATE_PHY_TEST = 7,
553 };
554 
555 enum ib_port_width {
556 	IB_WIDTH_1X	= 1,
557 	IB_WIDTH_2X	= 16,
558 	IB_WIDTH_4X	= 2,
559 	IB_WIDTH_8X	= 4,
560 	IB_WIDTH_12X	= 8
561 };
562 
563 static inline int ib_width_enum_to_int(enum ib_port_width width)
564 {
565 	switch (width) {
566 	case IB_WIDTH_1X:  return  1;
567 	case IB_WIDTH_2X:  return  2;
568 	case IB_WIDTH_4X:  return  4;
569 	case IB_WIDTH_8X:  return  8;
570 	case IB_WIDTH_12X: return 12;
571 	default: 	  return -1;
572 	}
573 }
574 
575 enum ib_port_speed {
576 	IB_SPEED_SDR	= 1,
577 	IB_SPEED_DDR	= 2,
578 	IB_SPEED_QDR	= 4,
579 	IB_SPEED_FDR10	= 8,
580 	IB_SPEED_FDR	= 16,
581 	IB_SPEED_EDR	= 32,
582 	IB_SPEED_HDR	= 64,
583 	IB_SPEED_NDR	= 128,
584 	IB_SPEED_XDR	= 256,
585 };
586 
587 enum ib_stat_flag {
588 	IB_STAT_FLAG_OPTIONAL = 1 << 0,
589 };
590 
591 /**
592  * struct rdma_stat_desc - description of one rdma stat/counter
593  * @name: The name of the counter
594  * @flags: Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL
595  * @priv: Driver private information; Core code should not use
596  */
597 struct rdma_stat_desc {
598 	const char *name;
599 	unsigned int flags;
600 	const void *priv;
601 };
602 
603 /**
604  * struct rdma_hw_stats - collection of hardware stats and their management
605  * @lock: Mutex to protect parallel write access to lifespan and values
606  *    of counters, which are 64bits and not guaranteed to be written
607  *    atomicaly on 32bits systems.
608  * @timestamp: Used by the core code to track when the last update was
609  * @lifespan: Used by the core code to determine how old the counters
610  *   should be before being updated again.  Stored in jiffies, defaults
611  *   to 10 milliseconds, drivers can override the default be specifying
612  *   their own value during their allocation routine.
613  * @descs: Array of pointers to static descriptors used for the counters
614  *   in directory.
615  * @is_disabled: A bitmap to indicate each counter is currently disabled
616  *   or not.
617  * @num_counters: How many hardware counters there are.  If name is
618  *   shorter than this number, a kernel oops will result.  Driver authors
619  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
620  *   in their code to prevent this.
621  * @value: Array of u64 counters that are accessed by the sysfs code and
622  *   filled in by the drivers get_stats routine
623  */
624 struct rdma_hw_stats {
625 	struct mutex	lock; /* Protect lifespan and values[] */
626 	unsigned long	timestamp;
627 	unsigned long	lifespan;
628 	const struct rdma_stat_desc *descs;
629 	unsigned long	*is_disabled;
630 	int		num_counters;
631 	u64		value[] __counted_by(num_counters);
632 };
633 
634 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
635 
636 struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
637 	const struct rdma_stat_desc *descs, int num_counters,
638 	unsigned long lifespan);
639 
640 void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
641 
642 /* Define bits for the various functionality this port needs to be supported by
643  * the core.
644  */
645 /* Management                           0x00000FFF */
646 #define RDMA_CORE_CAP_IB_MAD            0x00000001
647 #define RDMA_CORE_CAP_IB_SMI            0x00000002
648 #define RDMA_CORE_CAP_IB_CM             0x00000004
649 #define RDMA_CORE_CAP_IW_CM             0x00000008
650 #define RDMA_CORE_CAP_IB_SA             0x00000010
651 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
652 
653 /* Address format                       0x000FF000 */
654 #define RDMA_CORE_CAP_AF_IB             0x00001000
655 #define RDMA_CORE_CAP_ETH_AH            0x00002000
656 #define RDMA_CORE_CAP_OPA_AH            0x00004000
657 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
658 
659 /* Protocol                             0xFFF00000 */
660 #define RDMA_CORE_CAP_PROT_IB           0x00100000
661 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
662 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
663 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
664 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
665 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
666 
667 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
668 					| RDMA_CORE_CAP_PROT_ROCE     \
669 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
670 
671 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
672 					| RDMA_CORE_CAP_IB_MAD \
673 					| RDMA_CORE_CAP_IB_SMI \
674 					| RDMA_CORE_CAP_IB_CM  \
675 					| RDMA_CORE_CAP_IB_SA  \
676 					| RDMA_CORE_CAP_AF_IB)
677 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
678 					| RDMA_CORE_CAP_IB_MAD  \
679 					| RDMA_CORE_CAP_IB_CM   \
680 					| RDMA_CORE_CAP_AF_IB   \
681 					| RDMA_CORE_CAP_ETH_AH)
682 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
683 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
684 					| RDMA_CORE_CAP_IB_MAD  \
685 					| RDMA_CORE_CAP_IB_CM   \
686 					| RDMA_CORE_CAP_AF_IB   \
687 					| RDMA_CORE_CAP_ETH_AH)
688 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
689 					| RDMA_CORE_CAP_IW_CM)
690 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
691 					| RDMA_CORE_CAP_OPA_MAD)
692 
693 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
694 
695 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
696 
697 struct ib_port_attr {
698 	u64			subnet_prefix;
699 	enum ib_port_state	state;
700 	enum ib_mtu		max_mtu;
701 	enum ib_mtu		active_mtu;
702 	u32                     phys_mtu;
703 	int			gid_tbl_len;
704 	unsigned int		ip_gids:1;
705 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
706 	u32			port_cap_flags;
707 	u32			max_msg_sz;
708 	u32			bad_pkey_cntr;
709 	u32			qkey_viol_cntr;
710 	u16			pkey_tbl_len;
711 	u32			sm_lid;
712 	u32			lid;
713 	u8			lmc;
714 	u8			max_vl_num;
715 	u8			sm_sl;
716 	u8			subnet_timeout;
717 	u8			init_type_reply;
718 	u8			active_width;
719 	u16			active_speed;
720 	u8                      phys_state;
721 	u16			port_cap_flags2;
722 };
723 
724 enum ib_device_modify_flags {
725 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
726 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
727 };
728 
729 #define IB_DEVICE_NODE_DESC_MAX 64
730 
731 struct ib_device_modify {
732 	u64	sys_image_guid;
733 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
734 };
735 
736 enum ib_port_modify_flags {
737 	IB_PORT_SHUTDOWN		= 1,
738 	IB_PORT_INIT_TYPE		= (1<<2),
739 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
740 	IB_PORT_OPA_MASK_CHG		= (1<<4)
741 };
742 
743 struct ib_port_modify {
744 	u32	set_port_cap_mask;
745 	u32	clr_port_cap_mask;
746 	u8	init_type;
747 };
748 
749 enum ib_event_type {
750 	IB_EVENT_CQ_ERR,
751 	IB_EVENT_QP_FATAL,
752 	IB_EVENT_QP_REQ_ERR,
753 	IB_EVENT_QP_ACCESS_ERR,
754 	IB_EVENT_COMM_EST,
755 	IB_EVENT_SQ_DRAINED,
756 	IB_EVENT_PATH_MIG,
757 	IB_EVENT_PATH_MIG_ERR,
758 	IB_EVENT_DEVICE_FATAL,
759 	IB_EVENT_PORT_ACTIVE,
760 	IB_EVENT_PORT_ERR,
761 	IB_EVENT_LID_CHANGE,
762 	IB_EVENT_PKEY_CHANGE,
763 	IB_EVENT_SM_CHANGE,
764 	IB_EVENT_SRQ_ERR,
765 	IB_EVENT_SRQ_LIMIT_REACHED,
766 	IB_EVENT_QP_LAST_WQE_REACHED,
767 	IB_EVENT_CLIENT_REREGISTER,
768 	IB_EVENT_GID_CHANGE,
769 	IB_EVENT_WQ_FATAL,
770 	IB_EVENT_DEVICE_SPEED_CHANGE,
771 };
772 
773 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
774 
775 struct ib_event {
776 	struct ib_device	*device;
777 	union {
778 		struct ib_cq	*cq;
779 		struct ib_qp	*qp;
780 		struct ib_srq	*srq;
781 		struct ib_wq	*wq;
782 		u32		port_num;
783 	} element;
784 	enum ib_event_type	event;
785 };
786 
787 struct ib_event_handler {
788 	struct ib_device *device;
789 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
790 	struct list_head  list;
791 };
792 
793 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
794 	do {							\
795 		(_ptr)->device  = _device;			\
796 		(_ptr)->handler = _handler;			\
797 		INIT_LIST_HEAD(&(_ptr)->list);			\
798 	} while (0)
799 
800 struct ib_global_route {
801 	const struct ib_gid_attr *sgid_attr;
802 	union ib_gid	dgid;
803 	u32		flow_label;
804 	u8		sgid_index;
805 	u8		hop_limit;
806 	u8		traffic_class;
807 };
808 
809 struct ib_grh {
810 	__be32		version_tclass_flow;
811 	__be16		paylen;
812 	u8		next_hdr;
813 	u8		hop_limit;
814 	union ib_gid	sgid;
815 	union ib_gid	dgid;
816 };
817 
818 union rdma_network_hdr {
819 	struct ib_grh ibgrh;
820 	struct {
821 		/* The IB spec states that if it's IPv4, the header
822 		 * is located in the last 20 bytes of the header.
823 		 */
824 		u8		reserved[20];
825 		struct iphdr	roce4grh;
826 	};
827 };
828 
829 #define IB_QPN_MASK		0xFFFFFF
830 
831 enum {
832 	IB_MULTICAST_QPN = 0xffffff
833 };
834 
835 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
836 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
837 
838 enum ib_ah_flags {
839 	IB_AH_GRH	= 1
840 };
841 
842 enum ib_rate {
843 	IB_RATE_PORT_CURRENT = 0,
844 	IB_RATE_2_5_GBPS = 2,
845 	IB_RATE_5_GBPS   = 5,
846 	IB_RATE_10_GBPS  = 3,
847 	IB_RATE_20_GBPS  = 6,
848 	IB_RATE_30_GBPS  = 4,
849 	IB_RATE_40_GBPS  = 7,
850 	IB_RATE_60_GBPS  = 8,
851 	IB_RATE_80_GBPS  = 9,
852 	IB_RATE_120_GBPS = 10,
853 	IB_RATE_14_GBPS  = 11,
854 	IB_RATE_56_GBPS  = 12,
855 	IB_RATE_112_GBPS = 13,
856 	IB_RATE_168_GBPS = 14,
857 	IB_RATE_25_GBPS  = 15,
858 	IB_RATE_100_GBPS = 16,
859 	IB_RATE_200_GBPS = 17,
860 	IB_RATE_300_GBPS = 18,
861 	IB_RATE_28_GBPS  = 19,
862 	IB_RATE_50_GBPS  = 20,
863 	IB_RATE_400_GBPS = 21,
864 	IB_RATE_600_GBPS = 22,
865 	IB_RATE_800_GBPS = 23,
866 	IB_RATE_1600_GBPS = 25,
867 };
868 
869 /**
870  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
871  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
872  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
873  * @rate: rate to convert.
874  */
875 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
876 
877 /**
878  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
879  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
880  * @rate: rate to convert.
881  */
882 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
883 
884 struct ib_port_speed_info {
885 	const char *str;
886 	int rate;	/* in deci-Gb/sec (100 MBps units) */
887 };
888 
889 /**
890  * ib_port_attr_to_speed_info - Convert port attributes to speed information
891  * @attr: Port attributes containing active_speed and active_width
892  * @speed_info: Speed information to return
893  *
894  * Returns 0 on success, -EINVAL on error.
895  */
896 int ib_port_attr_to_speed_info(struct ib_port_attr *attr,
897 			       struct ib_port_speed_info *speed_info);
898 
899 /**
900  * enum ib_mr_type - memory region type
901  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
902  *                            normal registration
903  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
904  *                            register any arbitrary sg lists (without
905  *                            the normal mr constraints - see
906  *                            ib_map_mr_sg)
907  * @IB_MR_TYPE_DM:            memory region that is used for device
908  *                            memory registration
909  * @IB_MR_TYPE_USER:          memory region that is used for the user-space
910  *                            application
911  * @IB_MR_TYPE_DMA:           memory region that is used for DMA operations
912  *                            without address translations (VA=PA)
913  * @IB_MR_TYPE_INTEGRITY:     memory region that is used for
914  *                            data integrity operations
915  */
916 enum ib_mr_type {
917 	IB_MR_TYPE_MEM_REG,
918 	IB_MR_TYPE_SG_GAPS,
919 	IB_MR_TYPE_DM,
920 	IB_MR_TYPE_USER,
921 	IB_MR_TYPE_DMA,
922 	IB_MR_TYPE_INTEGRITY,
923 };
924 
925 enum ib_mr_status_check {
926 	IB_MR_CHECK_SIG_STATUS = 1,
927 };
928 
929 /**
930  * struct ib_mr_status - Memory region status container
931  *
932  * @fail_status: Bitmask of MR checks status. For each
933  *     failed check a corresponding status bit is set.
934  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
935  *     failure.
936  */
937 struct ib_mr_status {
938 	u32		    fail_status;
939 	struct ib_sig_err   sig_err;
940 };
941 
942 /**
943  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
944  * enum.
945  * @mult: multiple to convert.
946  */
947 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
948 
949 struct rdma_ah_init_attr {
950 	struct rdma_ah_attr *ah_attr;
951 	u32 flags;
952 	struct net_device *xmit_slave;
953 };
954 
955 enum rdma_ah_attr_type {
956 	RDMA_AH_ATTR_TYPE_UNDEFINED,
957 	RDMA_AH_ATTR_TYPE_IB,
958 	RDMA_AH_ATTR_TYPE_ROCE,
959 	RDMA_AH_ATTR_TYPE_OPA,
960 };
961 
962 struct ib_ah_attr {
963 	u16			dlid;
964 	u8			src_path_bits;
965 };
966 
967 struct roce_ah_attr {
968 	u8			dmac[ETH_ALEN];
969 };
970 
971 struct opa_ah_attr {
972 	u32			dlid;
973 	u8			src_path_bits;
974 	bool			make_grd;
975 };
976 
977 struct rdma_ah_attr {
978 	struct ib_global_route	grh;
979 	u8			sl;
980 	u8			static_rate;
981 	u32			port_num;
982 	u8			ah_flags;
983 	enum rdma_ah_attr_type type;
984 	union {
985 		struct ib_ah_attr ib;
986 		struct roce_ah_attr roce;
987 		struct opa_ah_attr opa;
988 	};
989 };
990 
991 enum ib_wc_status {
992 	IB_WC_SUCCESS,
993 	IB_WC_LOC_LEN_ERR,
994 	IB_WC_LOC_QP_OP_ERR,
995 	IB_WC_LOC_EEC_OP_ERR,
996 	IB_WC_LOC_PROT_ERR,
997 	IB_WC_WR_FLUSH_ERR,
998 	IB_WC_MW_BIND_ERR,
999 	IB_WC_BAD_RESP_ERR,
1000 	IB_WC_LOC_ACCESS_ERR,
1001 	IB_WC_REM_INV_REQ_ERR,
1002 	IB_WC_REM_ACCESS_ERR,
1003 	IB_WC_REM_OP_ERR,
1004 	IB_WC_RETRY_EXC_ERR,
1005 	IB_WC_RNR_RETRY_EXC_ERR,
1006 	IB_WC_LOC_RDD_VIOL_ERR,
1007 	IB_WC_REM_INV_RD_REQ_ERR,
1008 	IB_WC_REM_ABORT_ERR,
1009 	IB_WC_INV_EECN_ERR,
1010 	IB_WC_INV_EEC_STATE_ERR,
1011 	IB_WC_FATAL_ERR,
1012 	IB_WC_RESP_TIMEOUT_ERR,
1013 	IB_WC_GENERAL_ERR
1014 };
1015 
1016 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
1017 
1018 enum ib_wc_opcode {
1019 	IB_WC_SEND = IB_UVERBS_WC_SEND,
1020 	IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
1021 	IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
1022 	IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
1023 	IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
1024 	IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
1025 	IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
1026 	IB_WC_LSO = IB_UVERBS_WC_TSO,
1027 	IB_WC_ATOMIC_WRITE = IB_UVERBS_WC_ATOMIC_WRITE,
1028 	IB_WC_REG_MR,
1029 	IB_WC_MASKED_COMP_SWAP,
1030 	IB_WC_MASKED_FETCH_ADD,
1031 	IB_WC_FLUSH = IB_UVERBS_WC_FLUSH,
1032 /*
1033  * Set value of IB_WC_RECV so consumers can test if a completion is a
1034  * receive by testing (opcode & IB_WC_RECV).
1035  */
1036 	IB_WC_RECV			= 1 << 7,
1037 	IB_WC_RECV_RDMA_WITH_IMM
1038 };
1039 
1040 enum ib_wc_flags {
1041 	IB_WC_GRH		= 1,
1042 	IB_WC_WITH_IMM		= (1<<1),
1043 	IB_WC_WITH_INVALIDATE	= (1<<2),
1044 	IB_WC_IP_CSUM_OK	= (1<<3),
1045 	IB_WC_WITH_SMAC		= (1<<4),
1046 	IB_WC_WITH_VLAN		= (1<<5),
1047 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
1048 };
1049 
1050 struct ib_wc {
1051 	union {
1052 		u64		wr_id;
1053 		struct ib_cqe	*wr_cqe;
1054 	};
1055 	enum ib_wc_status	status;
1056 	enum ib_wc_opcode	opcode;
1057 	u32			vendor_err;
1058 	u32			byte_len;
1059 	struct ib_qp	       *qp;
1060 	union {
1061 		__be32		imm_data;
1062 		u32		invalidate_rkey;
1063 	} ex;
1064 	u32			src_qp;
1065 	u32			slid;
1066 	int			wc_flags;
1067 	u16			pkey_index;
1068 	u8			sl;
1069 	u8			dlid_path_bits;
1070 	u32 port_num; /* valid only for DR SMPs on switches */
1071 	u8			smac[ETH_ALEN];
1072 	u16			vlan_id;
1073 	u8			network_hdr_type;
1074 };
1075 
1076 enum ib_cq_notify_flags {
1077 	IB_CQ_SOLICITED			= 1 << 0,
1078 	IB_CQ_NEXT_COMP			= 1 << 1,
1079 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1080 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
1081 };
1082 
1083 enum ib_srq_type {
1084 	IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1085 	IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1086 	IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1087 };
1088 
1089 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1090 {
1091 	return srq_type == IB_SRQT_XRC ||
1092 	       srq_type == IB_SRQT_TM;
1093 }
1094 
1095 enum ib_srq_attr_mask {
1096 	IB_SRQ_MAX_WR	= 1 << 0,
1097 	IB_SRQ_LIMIT	= 1 << 1,
1098 };
1099 
1100 struct ib_srq_attr {
1101 	u32	max_wr;
1102 	u32	max_sge;
1103 	u32	srq_limit;
1104 };
1105 
1106 struct ib_srq_init_attr {
1107 	void		      (*event_handler)(struct ib_event *, void *);
1108 	void		       *srq_context;
1109 	struct ib_srq_attr	attr;
1110 	enum ib_srq_type	srq_type;
1111 
1112 	struct {
1113 		struct ib_cq   *cq;
1114 		union {
1115 			struct {
1116 				struct ib_xrcd *xrcd;
1117 			} xrc;
1118 
1119 			struct {
1120 				u32		max_num_tags;
1121 			} tag_matching;
1122 		};
1123 	} ext;
1124 };
1125 
1126 struct ib_qp_cap {
1127 	u32	max_send_wr;
1128 	u32	max_recv_wr;
1129 	u32	max_send_sge;
1130 	u32	max_recv_sge;
1131 	u32	max_inline_data;
1132 
1133 	/*
1134 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1135 	 * ib_create_qp() will calculate the right amount of needed WRs
1136 	 * and MRs based on this.
1137 	 */
1138 	u32	max_rdma_ctxs;
1139 };
1140 
1141 enum ib_sig_type {
1142 	IB_SIGNAL_ALL_WR,
1143 	IB_SIGNAL_REQ_WR
1144 };
1145 
1146 enum ib_qp_type {
1147 	/*
1148 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1149 	 * here (and in that order) since the MAD layer uses them as
1150 	 * indices into a 2-entry table.
1151 	 */
1152 	IB_QPT_SMI,
1153 	IB_QPT_GSI,
1154 
1155 	IB_QPT_RC = IB_UVERBS_QPT_RC,
1156 	IB_QPT_UC = IB_UVERBS_QPT_UC,
1157 	IB_QPT_UD = IB_UVERBS_QPT_UD,
1158 	IB_QPT_RAW_IPV6,
1159 	IB_QPT_RAW_ETHERTYPE,
1160 	IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1161 	IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1162 	IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1163 	IB_QPT_MAX,
1164 	IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1165 	/* Reserve a range for qp types internal to the low level driver.
1166 	 * These qp types will not be visible at the IB core layer, so the
1167 	 * IB_QPT_MAX usages should not be affected in the core layer
1168 	 */
1169 	IB_QPT_RESERVED1 = 0x1000,
1170 	IB_QPT_RESERVED2,
1171 	IB_QPT_RESERVED3,
1172 	IB_QPT_RESERVED4,
1173 	IB_QPT_RESERVED5,
1174 	IB_QPT_RESERVED6,
1175 	IB_QPT_RESERVED7,
1176 	IB_QPT_RESERVED8,
1177 	IB_QPT_RESERVED9,
1178 	IB_QPT_RESERVED10,
1179 };
1180 
1181 enum ib_qp_create_flags {
1182 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1183 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	=
1184 		IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1185 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1186 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1187 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1188 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1189 	IB_QP_CREATE_INTEGRITY_EN		= 1 << 6,
1190 	IB_QP_CREATE_NETDEV_USE			= 1 << 7,
1191 	IB_QP_CREATE_SCATTER_FCS		=
1192 		IB_UVERBS_QP_CREATE_SCATTER_FCS,
1193 	IB_QP_CREATE_CVLAN_STRIPPING		=
1194 		IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1195 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1196 	IB_QP_CREATE_PCI_WRITE_END_PADDING	=
1197 		IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1198 	/* reserve bits 26-31 for low level drivers' internal use */
1199 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1200 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1201 };
1202 
1203 /*
1204  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1205  * callback to destroy the passed in QP.
1206  */
1207 
1208 struct ib_qp_init_attr {
1209 	/* This callback occurs in workqueue context */
1210 	void                  (*event_handler)(struct ib_event *, void *);
1211 
1212 	void		       *qp_context;
1213 	struct ib_cq	       *send_cq;
1214 	struct ib_cq	       *recv_cq;
1215 	struct ib_srq	       *srq;
1216 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1217 	struct ib_qp_cap	cap;
1218 	enum ib_sig_type	sq_sig_type;
1219 	enum ib_qp_type		qp_type;
1220 	u32			create_flags;
1221 
1222 	/*
1223 	 * Only needed for special QP types, or when using the RW API.
1224 	 */
1225 	u32			port_num;
1226 	struct ib_rwq_ind_table *rwq_ind_tbl;
1227 	u32			source_qpn;
1228 };
1229 
1230 struct ib_qp_open_attr {
1231 	void                  (*event_handler)(struct ib_event *, void *);
1232 	void		       *qp_context;
1233 	u32			qp_num;
1234 	enum ib_qp_type		qp_type;
1235 };
1236 
1237 enum ib_rnr_timeout {
1238 	IB_RNR_TIMER_655_36 =  0,
1239 	IB_RNR_TIMER_000_01 =  1,
1240 	IB_RNR_TIMER_000_02 =  2,
1241 	IB_RNR_TIMER_000_03 =  3,
1242 	IB_RNR_TIMER_000_04 =  4,
1243 	IB_RNR_TIMER_000_06 =  5,
1244 	IB_RNR_TIMER_000_08 =  6,
1245 	IB_RNR_TIMER_000_12 =  7,
1246 	IB_RNR_TIMER_000_16 =  8,
1247 	IB_RNR_TIMER_000_24 =  9,
1248 	IB_RNR_TIMER_000_32 = 10,
1249 	IB_RNR_TIMER_000_48 = 11,
1250 	IB_RNR_TIMER_000_64 = 12,
1251 	IB_RNR_TIMER_000_96 = 13,
1252 	IB_RNR_TIMER_001_28 = 14,
1253 	IB_RNR_TIMER_001_92 = 15,
1254 	IB_RNR_TIMER_002_56 = 16,
1255 	IB_RNR_TIMER_003_84 = 17,
1256 	IB_RNR_TIMER_005_12 = 18,
1257 	IB_RNR_TIMER_007_68 = 19,
1258 	IB_RNR_TIMER_010_24 = 20,
1259 	IB_RNR_TIMER_015_36 = 21,
1260 	IB_RNR_TIMER_020_48 = 22,
1261 	IB_RNR_TIMER_030_72 = 23,
1262 	IB_RNR_TIMER_040_96 = 24,
1263 	IB_RNR_TIMER_061_44 = 25,
1264 	IB_RNR_TIMER_081_92 = 26,
1265 	IB_RNR_TIMER_122_88 = 27,
1266 	IB_RNR_TIMER_163_84 = 28,
1267 	IB_RNR_TIMER_245_76 = 29,
1268 	IB_RNR_TIMER_327_68 = 30,
1269 	IB_RNR_TIMER_491_52 = 31
1270 };
1271 
1272 enum ib_qp_attr_mask {
1273 	IB_QP_STATE			= 1,
1274 	IB_QP_CUR_STATE			= (1<<1),
1275 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1276 	IB_QP_ACCESS_FLAGS		= (1<<3),
1277 	IB_QP_PKEY_INDEX		= (1<<4),
1278 	IB_QP_PORT			= (1<<5),
1279 	IB_QP_QKEY			= (1<<6),
1280 	IB_QP_AV			= (1<<7),
1281 	IB_QP_PATH_MTU			= (1<<8),
1282 	IB_QP_TIMEOUT			= (1<<9),
1283 	IB_QP_RETRY_CNT			= (1<<10),
1284 	IB_QP_RNR_RETRY			= (1<<11),
1285 	IB_QP_RQ_PSN			= (1<<12),
1286 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1287 	IB_QP_ALT_PATH			= (1<<14),
1288 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1289 	IB_QP_SQ_PSN			= (1<<16),
1290 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1291 	IB_QP_PATH_MIG_STATE		= (1<<18),
1292 	IB_QP_CAP			= (1<<19),
1293 	IB_QP_DEST_QPN			= (1<<20),
1294 	IB_QP_RESERVED1			= (1<<21),
1295 	IB_QP_RESERVED2			= (1<<22),
1296 	IB_QP_RESERVED3			= (1<<23),
1297 	IB_QP_RESERVED4			= (1<<24),
1298 	IB_QP_RATE_LIMIT		= (1<<25),
1299 
1300 	IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1301 };
1302 
1303 enum ib_qp_state {
1304 	IB_QPS_RESET,
1305 	IB_QPS_INIT,
1306 	IB_QPS_RTR,
1307 	IB_QPS_RTS,
1308 	IB_QPS_SQD,
1309 	IB_QPS_SQE,
1310 	IB_QPS_ERR
1311 };
1312 
1313 enum ib_mig_state {
1314 	IB_MIG_MIGRATED,
1315 	IB_MIG_REARM,
1316 	IB_MIG_ARMED
1317 };
1318 
1319 enum ib_mw_type {
1320 	IB_MW_TYPE_1 = 1,
1321 	IB_MW_TYPE_2 = 2
1322 };
1323 
1324 struct ib_qp_attr {
1325 	enum ib_qp_state	qp_state;
1326 	enum ib_qp_state	cur_qp_state;
1327 	enum ib_mtu		path_mtu;
1328 	enum ib_mig_state	path_mig_state;
1329 	u32			qkey;
1330 	u32			rq_psn;
1331 	u32			sq_psn;
1332 	u32			dest_qp_num;
1333 	int			qp_access_flags;
1334 	struct ib_qp_cap	cap;
1335 	struct rdma_ah_attr	ah_attr;
1336 	struct rdma_ah_attr	alt_ah_attr;
1337 	u16			pkey_index;
1338 	u16			alt_pkey_index;
1339 	u8			en_sqd_async_notify;
1340 	u8			sq_draining;
1341 	u8			max_rd_atomic;
1342 	u8			max_dest_rd_atomic;
1343 	u8			min_rnr_timer;
1344 	u32			port_num;
1345 	u8			timeout;
1346 	u8			retry_cnt;
1347 	u8			rnr_retry;
1348 	u32			alt_port_num;
1349 	u8			alt_timeout;
1350 	u32			rate_limit;
1351 	struct net_device	*xmit_slave;
1352 };
1353 
1354 enum ib_wr_opcode {
1355 	/* These are shared with userspace */
1356 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1357 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1358 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1359 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1360 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1361 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1362 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1363 	IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1364 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1365 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1366 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1367 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1368 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1369 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1370 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1371 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1372 	IB_WR_FLUSH = IB_UVERBS_WR_FLUSH,
1373 	IB_WR_ATOMIC_WRITE = IB_UVERBS_WR_ATOMIC_WRITE,
1374 
1375 	/* These are kernel only and can not be issued by userspace */
1376 	IB_WR_REG_MR = 0x20,
1377 	IB_WR_REG_MR_INTEGRITY,
1378 
1379 	/* reserve values for low level drivers' internal use.
1380 	 * These values will not be used at all in the ib core layer.
1381 	 */
1382 	IB_WR_RESERVED1 = 0xf0,
1383 	IB_WR_RESERVED2,
1384 	IB_WR_RESERVED3,
1385 	IB_WR_RESERVED4,
1386 	IB_WR_RESERVED5,
1387 	IB_WR_RESERVED6,
1388 	IB_WR_RESERVED7,
1389 	IB_WR_RESERVED8,
1390 	IB_WR_RESERVED9,
1391 	IB_WR_RESERVED10,
1392 };
1393 
1394 enum ib_send_flags {
1395 	IB_SEND_FENCE		= 1,
1396 	IB_SEND_SIGNALED	= (1<<1),
1397 	IB_SEND_SOLICITED	= (1<<2),
1398 	IB_SEND_INLINE		= (1<<3),
1399 	IB_SEND_IP_CSUM		= (1<<4),
1400 
1401 	/* reserve bits 26-31 for low level drivers' internal use */
1402 	IB_SEND_RESERVED_START	= (1 << 26),
1403 	IB_SEND_RESERVED_END	= (1 << 31),
1404 };
1405 
1406 struct ib_sge {
1407 	u64	addr;
1408 	u32	length;
1409 	u32	lkey;
1410 };
1411 
1412 struct ib_cqe {
1413 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1414 };
1415 
1416 struct ib_send_wr {
1417 	struct ib_send_wr      *next;
1418 	union {
1419 		u64		wr_id;
1420 		struct ib_cqe	*wr_cqe;
1421 	};
1422 	struct ib_sge	       *sg_list;
1423 	int			num_sge;
1424 	enum ib_wr_opcode	opcode;
1425 	int			send_flags;
1426 	union {
1427 		__be32		imm_data;
1428 		u32		invalidate_rkey;
1429 	} ex;
1430 };
1431 
1432 struct ib_rdma_wr {
1433 	struct ib_send_wr	wr;
1434 	u64			remote_addr;
1435 	u32			rkey;
1436 };
1437 
1438 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1439 {
1440 	return container_of(wr, struct ib_rdma_wr, wr);
1441 }
1442 
1443 struct ib_atomic_wr {
1444 	struct ib_send_wr	wr;
1445 	u64			remote_addr;
1446 	u64			compare_add;
1447 	u64			swap;
1448 	u64			compare_add_mask;
1449 	u64			swap_mask;
1450 	u32			rkey;
1451 };
1452 
1453 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1454 {
1455 	return container_of(wr, struct ib_atomic_wr, wr);
1456 }
1457 
1458 struct ib_ud_wr {
1459 	struct ib_send_wr	wr;
1460 	struct ib_ah		*ah;
1461 	void			*header;
1462 	int			hlen;
1463 	int			mss;
1464 	u32			remote_qpn;
1465 	u32			remote_qkey;
1466 	u16			pkey_index; /* valid for GSI only */
1467 	u32			port_num; /* valid for DR SMPs on switch only */
1468 };
1469 
1470 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1471 {
1472 	return container_of(wr, struct ib_ud_wr, wr);
1473 }
1474 
1475 struct ib_reg_wr {
1476 	struct ib_send_wr	wr;
1477 	struct ib_mr		*mr;
1478 	u32			key;
1479 	int			access;
1480 };
1481 
1482 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1483 {
1484 	return container_of(wr, struct ib_reg_wr, wr);
1485 }
1486 
1487 struct ib_recv_wr {
1488 	struct ib_recv_wr      *next;
1489 	union {
1490 		u64		wr_id;
1491 		struct ib_cqe	*wr_cqe;
1492 	};
1493 	struct ib_sge	       *sg_list;
1494 	int			num_sge;
1495 };
1496 
1497 enum ib_access_flags {
1498 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1499 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1500 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1501 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1502 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1503 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1504 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1505 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1506 	IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1507 	IB_ACCESS_FLUSH_GLOBAL = IB_UVERBS_ACCESS_FLUSH_GLOBAL,
1508 	IB_ACCESS_FLUSH_PERSISTENT = IB_UVERBS_ACCESS_FLUSH_PERSISTENT,
1509 
1510 	IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1511 	IB_ACCESS_SUPPORTED =
1512 		((IB_ACCESS_FLUSH_PERSISTENT << 1) - 1) | IB_ACCESS_OPTIONAL,
1513 };
1514 
1515 /*
1516  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1517  * are hidden here instead of a uapi header!
1518  */
1519 enum ib_mr_rereg_flags {
1520 	IB_MR_REREG_TRANS	= 1,
1521 	IB_MR_REREG_PD		= (1<<1),
1522 	IB_MR_REREG_ACCESS	= (1<<2),
1523 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1524 };
1525 
1526 struct ib_umem;
1527 
1528 enum rdma_remove_reason {
1529 	/*
1530 	 * Userspace requested uobject deletion or initial try
1531 	 * to remove uobject via cleanup. Call could fail
1532 	 */
1533 	RDMA_REMOVE_DESTROY,
1534 	/* Context deletion. This call should delete the actual object itself */
1535 	RDMA_REMOVE_CLOSE,
1536 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1537 	RDMA_REMOVE_DRIVER_REMOVE,
1538 	/* uobj is being cleaned-up before being committed */
1539 	RDMA_REMOVE_ABORT,
1540 	/* The driver failed to destroy the uobject and is being disconnected */
1541 	RDMA_REMOVE_DRIVER_FAILURE,
1542 };
1543 
1544 struct ib_rdmacg_object {
1545 #ifdef CONFIG_CGROUP_RDMA
1546 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1547 #endif
1548 };
1549 
1550 struct ib_ucontext {
1551 	struct ib_device       *device;
1552 	struct ib_uverbs_file  *ufile;
1553 
1554 	struct ib_rdmacg_object	cg_obj;
1555 	u64 enabled_caps;
1556 	/*
1557 	 * Implementation details of the RDMA core, don't use in drivers:
1558 	 */
1559 	struct rdma_restrack_entry res;
1560 	struct xarray mmap_xa;
1561 };
1562 
1563 struct ib_uobject {
1564 	u64			user_handle;	/* handle given to us by userspace */
1565 	/* ufile & ucontext owning this object */
1566 	struct ib_uverbs_file  *ufile;
1567 	/* FIXME, save memory: ufile->context == context */
1568 	struct ib_ucontext     *context;	/* associated user context */
1569 	void		       *object;		/* containing object */
1570 	struct list_head	list;		/* link to context's list */
1571 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1572 	int			id;		/* index into kernel idr */
1573 	struct kref		ref;
1574 	atomic_t		usecnt;		/* protects exclusive access */
1575 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1576 
1577 	const struct uverbs_api_object *uapi_object;
1578 };
1579 
1580 struct ib_udata {
1581 	const void __user *inbuf;
1582 	void __user *outbuf;
1583 	size_t       inlen;
1584 	size_t       outlen;
1585 };
1586 
1587 struct ib_pd {
1588 	u32			local_dma_lkey;
1589 	u32			flags;
1590 	struct ib_device       *device;
1591 	struct ib_uobject      *uobject;
1592 	atomic_t          	usecnt; /* count all resources */
1593 
1594 	u32			unsafe_global_rkey;
1595 
1596 	/*
1597 	 * Implementation details of the RDMA core, don't use in drivers:
1598 	 */
1599 	struct ib_mr	       *__internal_mr;
1600 	struct rdma_restrack_entry res;
1601 };
1602 
1603 struct ib_xrcd {
1604 	struct ib_device       *device;
1605 	atomic_t		usecnt; /* count all exposed resources */
1606 	struct inode	       *inode;
1607 	struct rw_semaphore	tgt_qps_rwsem;
1608 	struct xarray		tgt_qps;
1609 };
1610 
1611 struct ib_ah {
1612 	struct ib_device	*device;
1613 	struct ib_pd		*pd;
1614 	struct ib_uobject	*uobject;
1615 	const struct ib_gid_attr *sgid_attr;
1616 	enum rdma_ah_attr_type	type;
1617 };
1618 
1619 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1620 
1621 enum ib_poll_context {
1622 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1623 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1624 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1625 	IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1626 
1627 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1628 };
1629 
1630 struct ib_cq {
1631 	struct ib_device       *device;
1632 	struct ib_ucq_object   *uobject;
1633 	ib_comp_handler   	comp_handler;
1634 	void                  (*event_handler)(struct ib_event *, void *);
1635 	void                   *cq_context;
1636 	int               	cqe;
1637 	unsigned int		cqe_used;
1638 	atomic_t          	usecnt; /* count number of work queues */
1639 	enum ib_poll_context	poll_ctx;
1640 	struct ib_wc		*wc;
1641 	struct list_head        pool_entry;
1642 	union {
1643 		struct irq_poll		iop;
1644 		struct work_struct	work;
1645 	};
1646 	struct workqueue_struct *comp_wq;
1647 	struct dim *dim;
1648 
1649 	/* updated only by trace points */
1650 	ktime_t timestamp;
1651 	u8 interrupt:1;
1652 	u8 shared:1;
1653 	unsigned int comp_vector;
1654 	struct ib_umem *umem;
1655 
1656 	/*
1657 	 * Implementation details of the RDMA core, don't use in drivers:
1658 	 */
1659 	struct rdma_restrack_entry res;
1660 };
1661 
1662 struct ib_srq {
1663 	struct ib_device       *device;
1664 	struct ib_pd	       *pd;
1665 	struct ib_usrq_object  *uobject;
1666 	void		      (*event_handler)(struct ib_event *, void *);
1667 	void		       *srq_context;
1668 	enum ib_srq_type	srq_type;
1669 	atomic_t		usecnt;
1670 
1671 	struct {
1672 		struct ib_cq   *cq;
1673 		union {
1674 			struct {
1675 				struct ib_xrcd *xrcd;
1676 				u32		srq_num;
1677 			} xrc;
1678 		};
1679 	} ext;
1680 
1681 	/*
1682 	 * Implementation details of the RDMA core, don't use in drivers:
1683 	 */
1684 	struct rdma_restrack_entry res;
1685 };
1686 
1687 enum ib_raw_packet_caps {
1688 	/*
1689 	 * Strip cvlan from incoming packet and report it in the matching work
1690 	 * completion is supported.
1691 	 */
1692 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING =
1693 		IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING,
1694 	/*
1695 	 * Scatter FCS field of an incoming packet to host memory is supported.
1696 	 */
1697 	IB_RAW_PACKET_CAP_SCATTER_FCS = IB_UVERBS_RAW_PACKET_CAP_SCATTER_FCS,
1698 	/* Checksum offloads are supported (for both send and receive). */
1699 	IB_RAW_PACKET_CAP_IP_CSUM = IB_UVERBS_RAW_PACKET_CAP_IP_CSUM,
1700 	/*
1701 	 * When a packet is received for an RQ with no receive WQEs, the
1702 	 * packet processing is delayed.
1703 	 */
1704 	IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP,
1705 };
1706 
1707 enum ib_wq_type {
1708 	IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1709 };
1710 
1711 enum ib_wq_state {
1712 	IB_WQS_RESET,
1713 	IB_WQS_RDY,
1714 	IB_WQS_ERR
1715 };
1716 
1717 struct ib_wq {
1718 	struct ib_device       *device;
1719 	struct ib_uwq_object   *uobject;
1720 	void		    *wq_context;
1721 	void		    (*event_handler)(struct ib_event *, void *);
1722 	struct ib_pd	       *pd;
1723 	struct ib_cq	       *cq;
1724 	u32		wq_num;
1725 	enum ib_wq_state       state;
1726 	enum ib_wq_type	wq_type;
1727 	atomic_t		usecnt;
1728 };
1729 
1730 enum ib_wq_flags {
1731 	IB_WQ_FLAGS_CVLAN_STRIPPING	= IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1732 	IB_WQ_FLAGS_SCATTER_FCS		= IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1733 	IB_WQ_FLAGS_DELAY_DROP		= IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1734 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1735 				IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1736 };
1737 
1738 struct ib_wq_init_attr {
1739 	void		       *wq_context;
1740 	enum ib_wq_type	wq_type;
1741 	u32		max_wr;
1742 	u32		max_sge;
1743 	struct	ib_cq	       *cq;
1744 	void		    (*event_handler)(struct ib_event *, void *);
1745 	u32		create_flags; /* Use enum ib_wq_flags */
1746 };
1747 
1748 enum ib_wq_attr_mask {
1749 	IB_WQ_STATE		= 1 << 0,
1750 	IB_WQ_CUR_STATE		= 1 << 1,
1751 	IB_WQ_FLAGS		= 1 << 2,
1752 };
1753 
1754 struct ib_wq_attr {
1755 	enum	ib_wq_state	wq_state;
1756 	enum	ib_wq_state	curr_wq_state;
1757 	u32			flags; /* Use enum ib_wq_flags */
1758 	u32			flags_mask; /* Use enum ib_wq_flags */
1759 };
1760 
1761 struct ib_rwq_ind_table {
1762 	struct ib_device	*device;
1763 	struct ib_uobject      *uobject;
1764 	atomic_t		usecnt;
1765 	u32		ind_tbl_num;
1766 	u32		log_ind_tbl_size;
1767 	struct ib_wq	**ind_tbl;
1768 };
1769 
1770 struct ib_rwq_ind_table_init_attr {
1771 	u32		log_ind_tbl_size;
1772 	/* Each entry is a pointer to Receive Work Queue */
1773 	struct ib_wq	**ind_tbl;
1774 };
1775 
1776 enum port_pkey_state {
1777 	IB_PORT_PKEY_NOT_VALID = 0,
1778 	IB_PORT_PKEY_VALID = 1,
1779 	IB_PORT_PKEY_LISTED = 2,
1780 };
1781 
1782 struct ib_qp_security;
1783 
1784 struct ib_port_pkey {
1785 	enum port_pkey_state	state;
1786 	u16			pkey_index;
1787 	u32			port_num;
1788 	struct list_head	qp_list;
1789 	struct list_head	to_error_list;
1790 	struct ib_qp_security  *sec;
1791 };
1792 
1793 struct ib_ports_pkeys {
1794 	struct ib_port_pkey	main;
1795 	struct ib_port_pkey	alt;
1796 };
1797 
1798 struct ib_qp_security {
1799 	struct ib_qp	       *qp;
1800 	struct ib_device       *dev;
1801 	/* Hold this mutex when changing port and pkey settings. */
1802 	struct mutex		mutex;
1803 	struct ib_ports_pkeys  *ports_pkeys;
1804 	/* A list of all open shared QP handles.  Required to enforce security
1805 	 * properly for all users of a shared QP.
1806 	 */
1807 	struct list_head        shared_qp_list;
1808 	void                   *security;
1809 	bool			destroying;
1810 	atomic_t		error_list_count;
1811 	struct completion	error_complete;
1812 	int			error_comps_pending;
1813 };
1814 
1815 /*
1816  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1817  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1818  */
1819 struct ib_qp {
1820 	struct ib_device       *device;
1821 	struct ib_pd	       *pd;
1822 	struct ib_cq	       *send_cq;
1823 	struct ib_cq	       *recv_cq;
1824 	spinlock_t		mr_lock;
1825 	int			mrs_used;
1826 	struct list_head	rdma_mrs;
1827 	struct list_head	sig_mrs;
1828 	struct ib_srq	       *srq;
1829 	struct completion	srq_completion;
1830 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1831 	struct list_head	xrcd_list;
1832 
1833 	/* count times opened, mcast attaches, flow attaches */
1834 	atomic_t		usecnt;
1835 	struct list_head	open_list;
1836 	struct ib_qp           *real_qp;
1837 	struct ib_uqp_object   *uobject;
1838 	void                  (*event_handler)(struct ib_event *, void *);
1839 	void                  (*registered_event_handler)(struct ib_event *, void *);
1840 	void		       *qp_context;
1841 	/* sgid_attrs associated with the AV's */
1842 	const struct ib_gid_attr *av_sgid_attr;
1843 	const struct ib_gid_attr *alt_path_sgid_attr;
1844 	u32			qp_num;
1845 	u32			max_write_sge;
1846 	u32			max_read_sge;
1847 	enum ib_qp_type		qp_type;
1848 	struct ib_rwq_ind_table *rwq_ind_tbl;
1849 	struct ib_qp_security  *qp_sec;
1850 	u32			port;
1851 
1852 	bool			integrity_en;
1853 	/*
1854 	 * Implementation details of the RDMA core, don't use in drivers:
1855 	 */
1856 	struct rdma_restrack_entry     res;
1857 
1858 	/* The counter the qp is bind to */
1859 	struct rdma_counter    *counter;
1860 };
1861 
1862 struct ib_dm {
1863 	struct ib_device  *device;
1864 	u32		   length;
1865 	u32		   flags;
1866 	struct ib_uobject *uobject;
1867 	atomic_t	   usecnt;
1868 };
1869 
1870 /* bit values to mark existence of ib_dmah fields */
1871 enum {
1872 	IB_DMAH_CPU_ID_EXISTS,
1873 	IB_DMAH_MEM_TYPE_EXISTS,
1874 	IB_DMAH_PH_EXISTS,
1875 };
1876 
1877 struct ib_dmah {
1878 	struct ib_device *device;
1879 	struct ib_uobject *uobject;
1880 	/*
1881 	 * Implementation details of the RDMA core, don't use in drivers:
1882 	 */
1883 	struct rdma_restrack_entry res;
1884 	u32 cpu_id;
1885 	enum tph_mem_type mem_type;
1886 	atomic_t usecnt;
1887 	u8 ph;
1888 	u8 valid_fields; /* use IB_DMAH_XXX_EXISTS */
1889 };
1890 
1891 struct ib_mr {
1892 	struct ib_device  *device;
1893 	struct ib_pd	  *pd;
1894 	u32		   lkey;
1895 	u32		   rkey;
1896 	u64		   iova;
1897 	u64		   length;
1898 	unsigned int	   page_size;
1899 	enum ib_mr_type	   type;
1900 	bool		   need_inval;
1901 	union {
1902 		struct ib_uobject	*uobject;	/* user */
1903 		struct list_head	qp_entry;	/* FR */
1904 	};
1905 
1906 	struct ib_dm      *dm;
1907 	struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1908 	struct ib_dmah *dmah;
1909 	struct {
1910 		struct ib_frmr_pool *pool;
1911 		struct ib_frmr_key key;
1912 		u32 handle;
1913 	} frmr;
1914 	/*
1915 	 * Implementation details of the RDMA core, don't use in drivers:
1916 	 */
1917 	struct rdma_restrack_entry res;
1918 };
1919 
1920 struct ib_mw {
1921 	struct ib_device	*device;
1922 	struct ib_pd		*pd;
1923 	struct ib_uobject	*uobject;
1924 	u32			rkey;
1925 	enum ib_mw_type         type;
1926 };
1927 
1928 /* Supported steering options */
1929 enum ib_flow_attr_type {
1930 	/* steering according to rule specifications */
1931 	IB_FLOW_ATTR_NORMAL		= 0x0,
1932 	/* default unicast and multicast rule -
1933 	 * receive all Eth traffic which isn't steered to any QP
1934 	 */
1935 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1936 	/* default multicast rule -
1937 	 * receive all Eth multicast traffic which isn't steered to any QP
1938 	 */
1939 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1940 	/* sniffer rule - receive all port traffic */
1941 	IB_FLOW_ATTR_SNIFFER		= 0x3
1942 };
1943 
1944 /* Supported steering header types */
1945 enum ib_flow_spec_type {
1946 	/* L2 headers*/
1947 	IB_FLOW_SPEC_ETH		= 0x20,
1948 	IB_FLOW_SPEC_IB			= 0x22,
1949 	/* L3 header*/
1950 	IB_FLOW_SPEC_IPV4		= 0x30,
1951 	IB_FLOW_SPEC_IPV6		= 0x31,
1952 	IB_FLOW_SPEC_ESP                = 0x34,
1953 	/* L4 headers*/
1954 	IB_FLOW_SPEC_TCP		= 0x40,
1955 	IB_FLOW_SPEC_UDP		= 0x41,
1956 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1957 	IB_FLOW_SPEC_GRE		= 0x51,
1958 	IB_FLOW_SPEC_MPLS		= 0x60,
1959 	IB_FLOW_SPEC_INNER		= 0x100,
1960 	/* Actions */
1961 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1962 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1963 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1964 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1965 };
1966 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1967 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1968 
1969 enum ib_flow_flags {
1970 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1971 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1972 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1973 };
1974 
1975 struct ib_flow_eth_filter {
1976 	u8	dst_mac[6];
1977 	u8	src_mac[6];
1978 	__be16	ether_type;
1979 	__be16	vlan_tag;
1980 };
1981 
1982 struct ib_flow_spec_eth {
1983 	u32			  type;
1984 	u16			  size;
1985 	struct ib_flow_eth_filter val;
1986 	struct ib_flow_eth_filter mask;
1987 };
1988 
1989 struct ib_flow_ib_filter {
1990 	__be16 dlid;
1991 	__u8   sl;
1992 };
1993 
1994 struct ib_flow_spec_ib {
1995 	u32			 type;
1996 	u16			 size;
1997 	struct ib_flow_ib_filter val;
1998 	struct ib_flow_ib_filter mask;
1999 };
2000 
2001 /* IPv4 header flags */
2002 enum ib_ipv4_flags {
2003 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
2004 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
2005 				    last have this flag set */
2006 };
2007 
2008 struct ib_flow_ipv4_filter {
2009 	__be32	src_ip;
2010 	__be32	dst_ip;
2011 	u8	proto;
2012 	u8	tos;
2013 	u8	ttl;
2014 	u8	flags;
2015 };
2016 
2017 struct ib_flow_spec_ipv4 {
2018 	u32			   type;
2019 	u16			   size;
2020 	struct ib_flow_ipv4_filter val;
2021 	struct ib_flow_ipv4_filter mask;
2022 };
2023 
2024 struct ib_flow_ipv6_filter {
2025 	u8	src_ip[16];
2026 	u8	dst_ip[16];
2027 	__be32	flow_label;
2028 	u8	next_hdr;
2029 	u8	traffic_class;
2030 	u8	hop_limit;
2031 } __packed;
2032 
2033 struct ib_flow_spec_ipv6 {
2034 	u32			   type;
2035 	u16			   size;
2036 	struct ib_flow_ipv6_filter val;
2037 	struct ib_flow_ipv6_filter mask;
2038 };
2039 
2040 struct ib_flow_tcp_udp_filter {
2041 	__be16	dst_port;
2042 	__be16	src_port;
2043 };
2044 
2045 struct ib_flow_spec_tcp_udp {
2046 	u32			      type;
2047 	u16			      size;
2048 	struct ib_flow_tcp_udp_filter val;
2049 	struct ib_flow_tcp_udp_filter mask;
2050 };
2051 
2052 struct ib_flow_tunnel_filter {
2053 	__be32	tunnel_id;
2054 };
2055 
2056 /* ib_flow_spec_tunnel describes the Vxlan tunnel
2057  * the tunnel_id from val has the vni value
2058  */
2059 struct ib_flow_spec_tunnel {
2060 	u32			      type;
2061 	u16			      size;
2062 	struct ib_flow_tunnel_filter  val;
2063 	struct ib_flow_tunnel_filter  mask;
2064 };
2065 
2066 struct ib_flow_esp_filter {
2067 	__be32	spi;
2068 	__be32  seq;
2069 };
2070 
2071 struct ib_flow_spec_esp {
2072 	u32                           type;
2073 	u16			      size;
2074 	struct ib_flow_esp_filter     val;
2075 	struct ib_flow_esp_filter     mask;
2076 };
2077 
2078 struct ib_flow_gre_filter {
2079 	__be16 c_ks_res0_ver;
2080 	__be16 protocol;
2081 	__be32 key;
2082 };
2083 
2084 struct ib_flow_spec_gre {
2085 	u32                           type;
2086 	u16			      size;
2087 	struct ib_flow_gre_filter     val;
2088 	struct ib_flow_gre_filter     mask;
2089 };
2090 
2091 struct ib_flow_mpls_filter {
2092 	__be32 tag;
2093 };
2094 
2095 struct ib_flow_spec_mpls {
2096 	u32                           type;
2097 	u16			      size;
2098 	struct ib_flow_mpls_filter     val;
2099 	struct ib_flow_mpls_filter     mask;
2100 };
2101 
2102 struct ib_flow_spec_action_tag {
2103 	enum ib_flow_spec_type	      type;
2104 	u16			      size;
2105 	u32                           tag_id;
2106 };
2107 
2108 struct ib_flow_spec_action_drop {
2109 	enum ib_flow_spec_type	      type;
2110 	u16			      size;
2111 };
2112 
2113 struct ib_flow_spec_action_handle {
2114 	enum ib_flow_spec_type	      type;
2115 	u16			      size;
2116 	struct ib_flow_action	     *act;
2117 };
2118 
2119 enum ib_counters_description {
2120 	IB_COUNTER_PACKETS,
2121 	IB_COUNTER_BYTES,
2122 };
2123 
2124 struct ib_flow_spec_action_count {
2125 	enum ib_flow_spec_type type;
2126 	u16 size;
2127 	struct ib_counters *counters;
2128 };
2129 
2130 union ib_flow_spec {
2131 	struct {
2132 		u32			type;
2133 		u16			size;
2134 	};
2135 	struct ib_flow_spec_eth		eth;
2136 	struct ib_flow_spec_ib		ib;
2137 	struct ib_flow_spec_ipv4        ipv4;
2138 	struct ib_flow_spec_tcp_udp	tcp_udp;
2139 	struct ib_flow_spec_ipv6        ipv6;
2140 	struct ib_flow_spec_tunnel      tunnel;
2141 	struct ib_flow_spec_esp		esp;
2142 	struct ib_flow_spec_gre		gre;
2143 	struct ib_flow_spec_mpls	mpls;
2144 	struct ib_flow_spec_action_tag  flow_tag;
2145 	struct ib_flow_spec_action_drop drop;
2146 	struct ib_flow_spec_action_handle action;
2147 	struct ib_flow_spec_action_count flow_count;
2148 };
2149 
2150 struct ib_flow_attr {
2151 	enum ib_flow_attr_type type;
2152 	u16	     size;
2153 	u16	     priority;
2154 	u32	     flags;
2155 	u8	     num_of_specs;
2156 	u32	     port;
2157 	union ib_flow_spec flows[];
2158 };
2159 
2160 struct ib_flow {
2161 	struct ib_qp		*qp;
2162 	struct ib_device	*device;
2163 	struct ib_uobject	*uobject;
2164 };
2165 
2166 enum ib_flow_action_type {
2167 	IB_FLOW_ACTION_UNSPECIFIED,
2168 	IB_FLOW_ACTION_ESP = 1,
2169 };
2170 
2171 struct ib_flow_action_attrs_esp_keymats {
2172 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2173 	union {
2174 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2175 	} keymat;
2176 };
2177 
2178 struct ib_flow_action_attrs_esp_replays {
2179 	enum ib_uverbs_flow_action_esp_replay			protocol;
2180 	union {
2181 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2182 	} replay;
2183 };
2184 
2185 enum ib_flow_action_attrs_esp_flags {
2186 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2187 	 * This is done in order to share the same flags between user-space and
2188 	 * kernel and spare an unnecessary translation.
2189 	 */
2190 
2191 	/* Kernel flags */
2192 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2193 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2194 };
2195 
2196 struct ib_flow_spec_list {
2197 	struct ib_flow_spec_list	*next;
2198 	union ib_flow_spec		spec;
2199 };
2200 
2201 struct ib_flow_action_attrs_esp {
2202 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2203 	struct ib_flow_action_attrs_esp_replays		*replay;
2204 	struct ib_flow_spec_list			*encap;
2205 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2206 	 * Value of 0 is a valid value.
2207 	 */
2208 	u32						esn;
2209 	u32						spi;
2210 	u32						seq;
2211 	u32						tfc_pad;
2212 	/* Use enum ib_flow_action_attrs_esp_flags */
2213 	u64						flags;
2214 	u64						hard_limit_pkts;
2215 };
2216 
2217 struct ib_flow_action {
2218 	struct ib_device		*device;
2219 	struct ib_uobject		*uobject;
2220 	enum ib_flow_action_type	type;
2221 	atomic_t			usecnt;
2222 };
2223 
2224 struct ib_mad;
2225 
2226 enum ib_process_mad_flags {
2227 	IB_MAD_IGNORE_MKEY	= 1,
2228 	IB_MAD_IGNORE_BKEY	= 2,
2229 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2230 };
2231 
2232 enum ib_mad_result {
2233 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2234 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2235 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2236 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2237 };
2238 
2239 struct ib_port_cache {
2240 	u64		      subnet_prefix;
2241 	struct ib_pkey_cache  *pkey;
2242 	struct ib_gid_table   *gid;
2243 	u8                     lmc;
2244 	enum ib_port_state     port_state;
2245 	enum ib_port_state     last_port_state;
2246 };
2247 
2248 struct ib_port_immutable {
2249 	int                           pkey_tbl_len;
2250 	int                           gid_tbl_len;
2251 	u32                           core_cap_flags;
2252 	u32                           max_mad_size;
2253 };
2254 
2255 struct ib_port_data {
2256 	struct ib_device *ib_dev;
2257 
2258 	struct ib_port_immutable immutable;
2259 
2260 	spinlock_t pkey_list_lock;
2261 
2262 	spinlock_t netdev_lock;
2263 
2264 	struct list_head pkey_list;
2265 
2266 	struct ib_port_cache cache;
2267 
2268 	struct net_device __rcu *netdev;
2269 	netdevice_tracker netdev_tracker;
2270 	struct hlist_node ndev_hash_link;
2271 	struct rdma_port_counter port_counter;
2272 	struct ib_port *sysfs;
2273 };
2274 
2275 /* rdma netdev type - specifies protocol type */
2276 enum rdma_netdev_t {
2277 	RDMA_NETDEV_OPA_VNIC,
2278 	RDMA_NETDEV_IPOIB,
2279 };
2280 
2281 /**
2282  * struct rdma_netdev - rdma netdev
2283  * For cases where netstack interfacing is required.
2284  */
2285 struct rdma_netdev {
2286 	void              *clnt_priv;
2287 	struct ib_device  *hca;
2288 	u32		   port_num;
2289 	int                mtu;
2290 
2291 	/*
2292 	 * cleanup function must be specified.
2293 	 * FIXME: This is only used for OPA_VNIC and that usage should be
2294 	 * removed too.
2295 	 */
2296 	void (*free_rdma_netdev)(struct net_device *netdev);
2297 
2298 	/* control functions */
2299 	void (*set_id)(struct net_device *netdev, int id);
2300 	/* send packet */
2301 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2302 		    struct ib_ah *address, u32 dqpn);
2303 	/* multicast */
2304 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2305 			    union ib_gid *gid, u16 mlid,
2306 			    int set_qkey, u32 qkey);
2307 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2308 			    union ib_gid *gid, u16 mlid);
2309 	/* timeout */
2310 	void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2311 };
2312 
2313 struct rdma_netdev_alloc_params {
2314 	size_t sizeof_priv;
2315 	unsigned int txqs;
2316 	unsigned int rxqs;
2317 	void *param;
2318 
2319 	int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2320 				      struct net_device *netdev, void *param);
2321 };
2322 
2323 struct ib_odp_counters {
2324 	atomic64_t faults;
2325 	atomic64_t faults_handled;
2326 	atomic64_t invalidations;
2327 	atomic64_t invalidations_handled;
2328 	atomic64_t prefetch;
2329 };
2330 
2331 struct ib_counters {
2332 	struct ib_device	*device;
2333 	struct ib_uobject	*uobject;
2334 	/* num of objects attached */
2335 	atomic_t	usecnt;
2336 };
2337 
2338 struct ib_counters_read_attr {
2339 	u64	*counters_buff;
2340 	u32	ncounters;
2341 	u32	flags; /* use enum ib_read_counters_flags */
2342 };
2343 
2344 struct uverbs_attr_bundle;
2345 struct iw_cm_id;
2346 struct iw_cm_conn_param;
2347 
2348 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member)                      \
2349 	.size_##ib_struct =                                                    \
2350 		(sizeof(struct drv_struct) +                                   \
2351 		 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) +      \
2352 		 BUILD_BUG_ON_ZERO(                                            \
2353 			 !__same_type(((struct drv_struct *)NULL)->member,     \
2354 				      struct ib_struct)))
2355 
2356 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp)                          \
2357 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2358 					   gfp, false))
2359 
2360 #define rdma_zalloc_drv_obj_numa(ib_dev, ib_type)                              \
2361 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2362 					   GFP_KERNEL, true))
2363 
2364 #define rdma_zalloc_drv_obj(ib_dev, ib_type)                                   \
2365 	rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2366 
2367 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2368 
2369 struct rdma_user_mmap_entry {
2370 	struct kref ref;
2371 	struct ib_ucontext *ucontext;
2372 	unsigned long start_pgoff;
2373 	size_t npages;
2374 	bool driver_removed;
2375 	/* protects access to dmabufs */
2376 	struct mutex dmabufs_lock;
2377 	struct list_head dmabufs;
2378 };
2379 
2380 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2381 static inline u64
2382 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2383 {
2384 	return (u64)entry->start_pgoff << PAGE_SHIFT;
2385 }
2386 
2387 /**
2388  * struct ib_device_ops - InfiniBand device operations
2389  * This structure defines all the InfiniBand device operations, providers will
2390  * need to define the supported operations, otherwise they will be set to null.
2391  */
2392 struct ib_device_ops {
2393 	struct module *owner;
2394 	enum rdma_driver_id driver_id;
2395 	u32 uverbs_abi_ver;
2396 	unsigned int uverbs_no_driver_id_binding:1;
2397 
2398 	/*
2399 	 * NOTE: New drivers should not make use of device_group; instead new
2400 	 * device parameter should be exposed via netlink command. This
2401 	 * mechanism exists only for existing drivers.
2402 	 */
2403 	const struct attribute_group *device_group;
2404 	const struct attribute_group **port_groups;
2405 
2406 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2407 			 const struct ib_send_wr **bad_send_wr);
2408 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2409 			 const struct ib_recv_wr **bad_recv_wr);
2410 	void (*drain_rq)(struct ib_qp *qp);
2411 	void (*drain_sq)(struct ib_qp *qp);
2412 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2413 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2414 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2415 	int (*post_srq_recv)(struct ib_srq *srq,
2416 			     const struct ib_recv_wr *recv_wr,
2417 			     const struct ib_recv_wr **bad_recv_wr);
2418 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2419 			   u32 port_num, const struct ib_wc *in_wc,
2420 			   const struct ib_grh *in_grh,
2421 			   const struct ib_mad *in_mad, struct ib_mad *out_mad,
2422 			   size_t *out_mad_size, u16 *out_mad_pkey_index);
2423 	int (*query_device)(struct ib_device *device,
2424 			    struct ib_device_attr *device_attr,
2425 			    struct ib_udata *udata);
2426 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2427 			     struct ib_device_modify *device_modify);
2428 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2429 	int (*query_port)(struct ib_device *device, u32 port_num,
2430 			  struct ib_port_attr *port_attr);
2431 	int (*query_port_speed)(struct ib_device *device, u32 port_num,
2432 				u64 *speed);
2433 	int (*modify_port)(struct ib_device *device, u32 port_num,
2434 			   int port_modify_mask,
2435 			   struct ib_port_modify *port_modify);
2436 	/*
2437 	 * The following mandatory functions are used only at device
2438 	 * registration.  Keep functions such as these at the end of this
2439 	 * structure to avoid cache line misses when accessing struct ib_device
2440 	 * in fast paths.
2441 	 */
2442 	int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2443 				  struct ib_port_immutable *immutable);
2444 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2445 					       u32 port_num);
2446 	/*
2447 	 * When calling get_netdev, the HW vendor's driver should return the
2448 	 * net device of device @device at port @port_num or NULL if such
2449 	 * a net device doesn't exist. The vendor driver should call dev_hold
2450 	 * on this net device. The HW vendor's device driver must guarantee
2451 	 * that this function returns NULL before the net device has finished
2452 	 * NETDEV_UNREGISTER state.
2453 	 */
2454 	struct net_device *(*get_netdev)(struct ib_device *device,
2455 					 u32 port_num);
2456 	/*
2457 	 * rdma netdev operation
2458 	 *
2459 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2460 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2461 	 */
2462 	struct net_device *(*alloc_rdma_netdev)(
2463 		struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2464 		const char *name, unsigned char name_assign_type,
2465 		void (*setup)(struct net_device *));
2466 
2467 	int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2468 				      enum rdma_netdev_t type,
2469 				      struct rdma_netdev_alloc_params *params);
2470 	/*
2471 	 * query_gid should be return GID value for @device, when @port_num
2472 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2473 	 * is RoCE link layer.
2474 	 */
2475 	int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2476 			 union ib_gid *gid);
2477 	/*
2478 	 * When calling add_gid, the HW vendor's driver should add the gid
2479 	 * of device of port at gid index available at @attr. Meta-info of
2480 	 * that gid (for example, the network device related to this gid) is
2481 	 * available at @attr. @context allows the HW vendor driver to store
2482 	 * extra information together with a GID entry. The HW vendor driver may
2483 	 * allocate memory to contain this information and store it in @context
2484 	 * when a new GID entry is written to. Params are consistent until the
2485 	 * next call of add_gid or delete_gid. The function should return 0 on
2486 	 * success or error otherwise. The function could be called
2487 	 * concurrently for different ports. This function is only called when
2488 	 * roce_gid_table is used.
2489 	 */
2490 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2491 	/*
2492 	 * When calling del_gid, the HW vendor's driver should delete the
2493 	 * gid of device @device at gid index gid_index of port port_num
2494 	 * available in @attr.
2495 	 * Upon the deletion of a GID entry, the HW vendor must free any
2496 	 * allocated memory. The caller will clear @context afterwards.
2497 	 * This function is only called when roce_gid_table is used.
2498 	 */
2499 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2500 	int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2501 			  u16 *pkey);
2502 	int (*alloc_ucontext)(struct ib_ucontext *context,
2503 			      struct ib_udata *udata);
2504 	void (*dealloc_ucontext)(struct ib_ucontext *context);
2505 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2506 	/*
2507 	 * This will be called once refcount of an entry in mmap_xa reaches
2508 	 * zero. The type of the memory that was mapped may differ between
2509 	 * entries and is opaque to the rdma_user_mmap interface.
2510 	 * Therefore needs to be implemented by the driver in mmap_free.
2511 	 */
2512 	void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2513 	int (*mmap_get_pfns)(struct rdma_user_mmap_entry *entry,
2514 			     struct phys_vec *phys_vec,
2515 			     struct p2pdma_provider **provider);
2516 	struct rdma_user_mmap_entry *(*pgoff_to_mmap_entry)(struct ib_ucontext *ucontext,
2517 							    off_t pg_off);
2518 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2519 	int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2520 	int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2521 	int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2522 			 struct ib_udata *udata);
2523 	int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2524 			      struct ib_udata *udata);
2525 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2526 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2527 	int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2528 	int (*create_srq)(struct ib_srq *srq,
2529 			  struct ib_srq_init_attr *srq_init_attr,
2530 			  struct ib_udata *udata);
2531 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2532 			  enum ib_srq_attr_mask srq_attr_mask,
2533 			  struct ib_udata *udata);
2534 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2535 	int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2536 	int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2537 			 struct ib_udata *udata);
2538 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2539 			 int qp_attr_mask, struct ib_udata *udata);
2540 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2541 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2542 	int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2543 	int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2544 			 struct uverbs_attr_bundle *attrs);
2545 	int (*create_user_cq)(struct ib_cq *cq,
2546 			      const struct ib_cq_init_attr *attr,
2547 			      struct uverbs_attr_bundle *attrs);
2548 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2549 	int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2550 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2551 	/*
2552 	 * pre_destroy_cq - Prevent a cq from generating any new work
2553 	 * completions, but not free any kernel resources
2554 	 */
2555 	int (*pre_destroy_cq)(struct ib_cq *cq);
2556 	/*
2557 	 * post_destroy_cq - Free all kernel resources
2558 	 */
2559 	void (*post_destroy_cq)(struct ib_cq *cq);
2560 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2561 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2562 				     u64 virt_addr, int mr_access_flags,
2563 				     struct ib_dmah *dmah,
2564 				     struct ib_udata *udata);
2565 	struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2566 					    u64 length, u64 virt_addr, int fd,
2567 					    int mr_access_flags,
2568 					    struct ib_dmah *dmah,
2569 					    struct uverbs_attr_bundle *attrs);
2570 	struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2571 				       u64 length, u64 virt_addr,
2572 				       int mr_access_flags, struct ib_pd *pd,
2573 				       struct ib_udata *udata);
2574 	int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2575 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2576 				  u32 max_num_sg);
2577 	struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2578 					    u32 max_num_data_sg,
2579 					    u32 max_num_meta_sg);
2580 	int (*advise_mr)(struct ib_pd *pd,
2581 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2582 			 struct ib_sge *sg_list, u32 num_sge,
2583 			 struct uverbs_attr_bundle *attrs);
2584 
2585 	/*
2586 	 * Kernel users should universally support relaxed ordering (RO), as
2587 	 * they are designed to read data only after observing the CQE and use
2588 	 * the DMA API correctly.
2589 	 *
2590 	 * Some drivers implicitly enable RO if platform supports it.
2591 	 */
2592 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2593 			 unsigned int *sg_offset);
2594 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2595 			       struct ib_mr_status *mr_status);
2596 	int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2597 	int (*dealloc_mw)(struct ib_mw *mw);
2598 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2599 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2600 	int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2601 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2602 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2603 				       struct ib_flow_attr *flow_attr,
2604 				       struct ib_udata *udata);
2605 	int (*destroy_flow)(struct ib_flow *flow_id);
2606 	int (*destroy_flow_action)(struct ib_flow_action *action);
2607 	int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2608 				 int state);
2609 	int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2610 			     struct ifla_vf_info *ivf);
2611 	int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2612 			    struct ifla_vf_stats *stats);
2613 	int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2614 			    struct ifla_vf_guid *node_guid,
2615 			    struct ifla_vf_guid *port_guid);
2616 	int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2617 			   int type);
2618 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2619 				   struct ib_wq_init_attr *init_attr,
2620 				   struct ib_udata *udata);
2621 	int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2622 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2623 			 u32 wq_attr_mask, struct ib_udata *udata);
2624 	int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2625 				    struct ib_rwq_ind_table_init_attr *init_attr,
2626 				    struct ib_udata *udata);
2627 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2628 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2629 				  struct ib_ucontext *context,
2630 				  struct ib_dm_alloc_attr *attr,
2631 				  struct uverbs_attr_bundle *attrs);
2632 	int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2633 	int (*alloc_dmah)(struct ib_dmah *ibdmah,
2634 			  struct uverbs_attr_bundle *attrs);
2635 	int (*dealloc_dmah)(struct ib_dmah *dmah, struct uverbs_attr_bundle *attrs);
2636 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2637 				   struct ib_dm_mr_attr *attr,
2638 				   struct uverbs_attr_bundle *attrs);
2639 	int (*create_counters)(struct ib_counters *counters,
2640 			       struct uverbs_attr_bundle *attrs);
2641 	int (*destroy_counters)(struct ib_counters *counters);
2642 	int (*read_counters)(struct ib_counters *counters,
2643 			     struct ib_counters_read_attr *counters_read_attr,
2644 			     struct uverbs_attr_bundle *attrs);
2645 	int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2646 			    int data_sg_nents, unsigned int *data_sg_offset,
2647 			    struct scatterlist *meta_sg, int meta_sg_nents,
2648 			    unsigned int *meta_sg_offset);
2649 
2650 	/*
2651 	 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2652 	 *   fill in the driver initialized data.  The struct is kfree()'ed by
2653 	 *   the sysfs core when the device is removed.  A lifespan of -1 in the
2654 	 *   return struct tells the core to set a default lifespan.
2655 	 */
2656 	struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2657 	struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2658 						     u32 port_num);
2659 	/*
2660 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2661 	 * @index - The index in the value array we wish to have updated, or
2662 	 *   num_counters if we want all stats updated
2663 	 * Return codes -
2664 	 *   < 0 - Error, no counters updated
2665 	 *   index - Updated the single counter pointed to by index
2666 	 *   num_counters - Updated all counters (will reset the timestamp
2667 	 *     and prevent further calls for lifespan milliseconds)
2668 	 * Drivers are allowed to update all counters in leiu of just the
2669 	 *   one given in index at their option
2670 	 */
2671 	int (*get_hw_stats)(struct ib_device *device,
2672 			    struct rdma_hw_stats *stats, u32 port, int index);
2673 
2674 	/*
2675 	 * modify_hw_stat - Modify the counter configuration
2676 	 * @enable: true/false when enable/disable a counter
2677 	 * Return codes - 0 on success or error code otherwise.
2678 	 */
2679 	int (*modify_hw_stat)(struct ib_device *device, u32 port,
2680 			      unsigned int counter_index, bool enable);
2681 	/*
2682 	 * Allows rdma drivers to add their own restrack attributes.
2683 	 */
2684 	int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2685 	int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2686 	int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2687 	int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2688 	int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2689 	int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2690 	int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2691 	int (*fill_res_srq_entry)(struct sk_buff *msg, struct ib_srq *ib_srq);
2692 	int (*fill_res_srq_entry_raw)(struct sk_buff *msg, struct ib_srq *ib_srq);
2693 
2694 	/* Device lifecycle callbacks */
2695 	/*
2696 	 * Called after the device becomes registered, before clients are
2697 	 * attached
2698 	 */
2699 	int (*enable_driver)(struct ib_device *dev);
2700 	/*
2701 	 * This is called as part of ib_dealloc_device().
2702 	 */
2703 	void (*dealloc_driver)(struct ib_device *dev);
2704 
2705 	/* iWarp CM callbacks */
2706 	void (*iw_add_ref)(struct ib_qp *qp);
2707 	void (*iw_rem_ref)(struct ib_qp *qp);
2708 	struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2709 	int (*iw_connect)(struct iw_cm_id *cm_id,
2710 			  struct iw_cm_conn_param *conn_param);
2711 	int (*iw_accept)(struct iw_cm_id *cm_id,
2712 			 struct iw_cm_conn_param *conn_param);
2713 	int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2714 			 u8 pdata_len);
2715 	int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2716 	int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2717 	/*
2718 	 * counter_bind_qp - Bind a QP to a counter.
2719 	 * @counter - The counter to be bound. If counter->id is zero then
2720 	 *   the driver needs to allocate a new counter and set counter->id
2721 	 */
2722 	int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp,
2723 			       u32 port);
2724 	/*
2725 	 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2726 	 *   counter and bind it onto the default one
2727 	 */
2728 	int (*counter_unbind_qp)(struct ib_qp *qp, u32 port);
2729 	/*
2730 	 * counter_dealloc -De-allocate the hw counter
2731 	 */
2732 	int (*counter_dealloc)(struct rdma_counter *counter);
2733 	/*
2734 	 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2735 	 * the driver initialized data.
2736 	 */
2737 	struct rdma_hw_stats *(*counter_alloc_stats)(
2738 		struct rdma_counter *counter);
2739 	/*
2740 	 * counter_update_stats - Query the stats value of this counter
2741 	 */
2742 	int (*counter_update_stats)(struct rdma_counter *counter);
2743 
2744 	/*
2745 	 * counter_init - Initialize the driver specific rdma counter struct.
2746 	 */
2747 	void (*counter_init)(struct rdma_counter *counter);
2748 
2749 	/*
2750 	 * Allows rdma drivers to add their own restrack attributes
2751 	 * dumped via 'rdma stat' iproute2 command.
2752 	 */
2753 	int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2754 
2755 	/* query driver for its ucontext properties */
2756 	int (*query_ucontext)(struct ib_ucontext *context,
2757 			      struct uverbs_attr_bundle *attrs);
2758 
2759 	/*
2760 	 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2761 	 * Everyone else relies on Linux memory management model.
2762 	 */
2763 	int (*get_numa_node)(struct ib_device *dev);
2764 
2765 	/*
2766 	 * add_sub_dev - Add a sub IB device
2767 	 */
2768 	struct ib_device *(*add_sub_dev)(struct ib_device *parent,
2769 					 enum rdma_nl_dev_type type,
2770 					 const char *name);
2771 
2772 	/*
2773 	 * del_sub_dev - Delete a sub IB device
2774 	 */
2775 	void (*del_sub_dev)(struct ib_device *sub_dev);
2776 
2777 	/*
2778 	 * ufile_cleanup - Attempt to cleanup ubojects HW resources inside
2779 	 * the ufile.
2780 	 */
2781 	void (*ufile_hw_cleanup)(struct ib_uverbs_file *ufile);
2782 
2783 	/*
2784 	 * report_port_event - Drivers need to implement this if they have
2785 	 * some private stuff to handle when link status changes.
2786 	 */
2787 	void (*report_port_event)(struct ib_device *ibdev,
2788 				  struct net_device *ndev, unsigned long event);
2789 
2790 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
2791 	DECLARE_RDMA_OBJ_SIZE(ib_counters);
2792 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
2793 	DECLARE_RDMA_OBJ_SIZE(ib_dmah);
2794 	DECLARE_RDMA_OBJ_SIZE(ib_mw);
2795 	DECLARE_RDMA_OBJ_SIZE(ib_pd);
2796 	DECLARE_RDMA_OBJ_SIZE(ib_qp);
2797 	DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2798 	DECLARE_RDMA_OBJ_SIZE(ib_srq);
2799 	DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2800 	DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2801 	DECLARE_RDMA_OBJ_SIZE(rdma_counter);
2802 };
2803 
2804 struct ib_core_device {
2805 	/* device must be the first element in structure until,
2806 	 * union of ib_core_device and device exists in ib_device.
2807 	 */
2808 	struct device dev;
2809 	possible_net_t rdma_net;
2810 	struct kobject *ports_kobj;
2811 	struct list_head port_list;
2812 	struct ib_device *owner; /* reach back to owner ib_device */
2813 };
2814 
2815 struct rdma_restrack_root;
2816 struct ib_device {
2817 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2818 	struct device                *dma_device;
2819 	struct ib_device_ops	     ops;
2820 	char                          name[IB_DEVICE_NAME_MAX];
2821 	struct rcu_head rcu_head;
2822 
2823 	struct list_head              event_handler_list;
2824 	/* Protects event_handler_list */
2825 	struct rw_semaphore event_handler_rwsem;
2826 
2827 	/* Protects QP's event_handler calls and open_qp list */
2828 	spinlock_t qp_open_list_lock;
2829 
2830 	struct rw_semaphore	      client_data_rwsem;
2831 	struct xarray                 client_data;
2832 	struct mutex                  unregistration_lock;
2833 
2834 	/* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2835 	rwlock_t cache_lock;
2836 	/**
2837 	 * port_data is indexed by port number
2838 	 */
2839 	struct ib_port_data *port_data;
2840 
2841 	int			      num_comp_vectors;
2842 
2843 	union {
2844 		struct device		dev;
2845 		struct ib_core_device	coredev;
2846 	};
2847 
2848 	/* First group is for device attributes,
2849 	 * Second group is for driver provided attributes (optional).
2850 	 * Third group is for the hw_stats
2851 	 * It is a NULL terminated array.
2852 	 */
2853 	const struct attribute_group	*groups[4];
2854 	u8				hw_stats_attr_index;
2855 
2856 	u64			     uverbs_cmd_mask;
2857 
2858 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2859 	__be64			     node_guid;
2860 	u32			     local_dma_lkey;
2861 	u16                          is_switch:1;
2862 	/* Indicates kernel verbs support, should not be used in drivers */
2863 	u16                          kverbs_provider:1;
2864 	/* CQ adaptive moderation (RDMA DIM) */
2865 	u16                          use_cq_dim:1;
2866 	u8                           node_type;
2867 	u32			     phys_port_cnt;
2868 	struct ib_device_attr        attrs;
2869 	struct hw_stats_device_data *hw_stats_data;
2870 
2871 #ifdef CONFIG_CGROUP_RDMA
2872 	struct rdmacg_device         cg_device;
2873 #endif
2874 
2875 	u32                          index;
2876 
2877 	spinlock_t                   cq_pools_lock;
2878 	struct list_head             cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2879 
2880 	struct rdma_restrack_root *res;
2881 
2882 	const struct uapi_definition   *driver_def;
2883 
2884 	/*
2885 	 * Positive refcount indicates that the device is currently
2886 	 * registered and cannot be unregistered.
2887 	 */
2888 	refcount_t refcount;
2889 	struct completion unreg_completion;
2890 	struct work_struct unregistration_work;
2891 
2892 	const struct rdma_link_ops *link_ops;
2893 
2894 	/* Protects compat_devs xarray modifications */
2895 	struct mutex compat_devs_mutex;
2896 	/* Maintains compat devices for each net namespace */
2897 	struct xarray compat_devs;
2898 
2899 	/* Used by iWarp CM */
2900 	char iw_ifname[IFNAMSIZ];
2901 	u32 iw_driver_flags;
2902 	u32 lag_flags;
2903 
2904 	/* A parent device has a list of sub-devices */
2905 	struct mutex subdev_lock;
2906 	struct list_head subdev_list_head;
2907 
2908 	/* A sub device has a type and a parent */
2909 	enum rdma_nl_dev_type type;
2910 	struct ib_device *parent;
2911 	struct list_head subdev_list;
2912 
2913 	enum rdma_nl_name_assign_type name_assign_type;
2914 
2915 	struct ib_frmr_pools *frmr_pools;
2916 };
2917 
2918 static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
2919 				    gfp_t gfp, bool is_numa_aware)
2920 {
2921 	if (is_numa_aware && dev->ops.get_numa_node)
2922 		return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
2923 
2924 	return kzalloc(size, gfp);
2925 }
2926 
2927 struct ib_client_nl_info;
2928 struct ib_client {
2929 	const char *name;
2930 	int (*add)(struct ib_device *ibdev);
2931 	void (*remove)(struct ib_device *, void *client_data);
2932 	void (*rename)(struct ib_device *dev, void *client_data);
2933 	int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2934 			   struct ib_client_nl_info *res);
2935 	int (*get_global_nl_info)(struct ib_client_nl_info *res);
2936 
2937 	/* Returns the net_dev belonging to this ib_client and matching the
2938 	 * given parameters.
2939 	 * @dev:	 An RDMA device that the net_dev use for communication.
2940 	 * @port:	 A physical port number on the RDMA device.
2941 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2942 	 * @gid:	 A GID that the net_dev uses to communicate.
2943 	 * @addr:	 An IP address the net_dev is configured with.
2944 	 * @client_data: The device's client data set by ib_set_client_data().
2945 	 *
2946 	 * An ib_client that implements a net_dev on top of RDMA devices
2947 	 * (such as IP over IB) should implement this callback, allowing the
2948 	 * rdma_cm module to find the right net_dev for a given request.
2949 	 *
2950 	 * The caller is responsible for calling dev_put on the returned
2951 	 * netdev. */
2952 	struct net_device *(*get_net_dev_by_params)(
2953 			struct ib_device *dev,
2954 			u32 port,
2955 			u16 pkey,
2956 			const union ib_gid *gid,
2957 			const struct sockaddr *addr,
2958 			void *client_data);
2959 
2960 	refcount_t uses;
2961 	struct completion uses_zero;
2962 	u32 client_id;
2963 
2964 	/* kverbs are not required by the client */
2965 	u8 no_kverbs_req:1;
2966 };
2967 
2968 struct ib_device *_ib_alloc_device(size_t size, struct net *net);
2969 #define ib_alloc_device(drv_struct, member)                                    \
2970 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2971 				      BUILD_BUG_ON_ZERO(offsetof(              \
2972 					      struct drv_struct, member)),     \
2973 				      &init_net),			       \
2974 		     struct drv_struct, member)
2975 
2976 #define ib_alloc_device_with_net(drv_struct, member, net)		       \
2977 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2978 				      BUILD_BUG_ON_ZERO(offsetof(              \
2979 					struct drv_struct, member)), net),     \
2980 		     struct drv_struct, member)
2981 
2982 void ib_dealloc_device(struct ib_device *device);
2983 
2984 void ib_get_device_fw_str(struct ib_device *device, char *str);
2985 
2986 int ib_register_device(struct ib_device *device, const char *name,
2987 		       struct device *dma_device);
2988 void ib_unregister_device(struct ib_device *device);
2989 void ib_unregister_driver(enum rdma_driver_id driver_id);
2990 void ib_unregister_device_and_put(struct ib_device *device);
2991 void ib_unregister_device_queued(struct ib_device *ib_dev);
2992 
2993 int ib_register_client   (struct ib_client *client);
2994 void ib_unregister_client(struct ib_client *client);
2995 
2996 /**
2997  * ib_get_client_data - Get IB client context
2998  * @device:Device to get context for
2999  * @client:Client to get context for
3000  *
3001  * ib_get_client_data() returns the client context data set with
3002  * ib_set_client_data(). This can only be called while the client is
3003  * registered to the device, once the ib_client remove() callback returns this
3004  * cannot be called.
3005  */
3006 static inline void *ib_get_client_data(struct ib_device *device,
3007 				       struct ib_client *client)
3008 {
3009 	return xa_load(&device->client_data, client->client_id);
3010 }
3011 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
3012 			 void *data);
3013 void ib_set_device_ops(struct ib_device *device,
3014 		       const struct ib_device_ops *ops);
3015 
3016 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
3017 		      unsigned long pfn, unsigned long size, pgprot_t prot,
3018 		      struct rdma_user_mmap_entry *entry);
3019 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
3020 				struct rdma_user_mmap_entry *entry,
3021 				size_t length);
3022 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
3023 				      struct rdma_user_mmap_entry *entry,
3024 				      size_t length, u32 min_pgoff,
3025 				      u32 max_pgoff);
3026 
3027 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
3028 void rdma_user_mmap_disassociate(struct ib_device *device);
3029 #else
3030 static inline void rdma_user_mmap_disassociate(struct ib_device *device)
3031 {
3032 }
3033 #endif
3034 
3035 static inline int
3036 rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
3037 				  struct rdma_user_mmap_entry *entry,
3038 				  size_t length, u32 pgoff)
3039 {
3040 	return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff,
3041 						 pgoff);
3042 }
3043 
3044 struct rdma_user_mmap_entry *
3045 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
3046 			       unsigned long pgoff);
3047 struct rdma_user_mmap_entry *
3048 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
3049 			 struct vm_area_struct *vma);
3050 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
3051 
3052 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
3053 
3054 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
3055 {
3056 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
3057 }
3058 
3059 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
3060 {
3061 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
3062 }
3063 
3064 static inline bool ib_is_buffer_cleared(const void __user *p,
3065 					size_t len)
3066 {
3067 	bool ret;
3068 	u8 *buf;
3069 
3070 	if (len > USHRT_MAX)
3071 		return false;
3072 
3073 	buf = memdup_user(p, len);
3074 	if (IS_ERR(buf))
3075 		return false;
3076 
3077 	ret = !memchr_inv(buf, 0, len);
3078 	kfree(buf);
3079 	return ret;
3080 }
3081 
3082 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
3083 				       size_t offset,
3084 				       size_t len)
3085 {
3086 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
3087 }
3088 
3089 /**
3090  * ib_modify_qp_is_ok - Check that the supplied attribute mask
3091  * contains all required attributes and no attributes not allowed for
3092  * the given QP state transition.
3093  * @cur_state: Current QP state
3094  * @next_state: Next QP state
3095  * @type: QP type
3096  * @mask: Mask of supplied QP attributes
3097  *
3098  * This function is a helper function that a low-level driver's
3099  * modify_qp method can use to validate the consumer's input.  It
3100  * checks that cur_state and next_state are valid QP states, that a
3101  * transition from cur_state to next_state is allowed by the IB spec,
3102  * and that the attribute mask supplied is allowed for the transition.
3103  */
3104 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
3105 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
3106 
3107 void ib_register_event_handler(struct ib_event_handler *event_handler);
3108 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
3109 void ib_dispatch_event(const struct ib_event *event);
3110 
3111 int ib_query_port(struct ib_device *device,
3112 		  u32 port_num, struct ib_port_attr *port_attr);
3113 
3114 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3115 					       u32 port_num);
3116 
3117 /**
3118  * rdma_cap_ib_switch - Check if the device is IB switch
3119  * @device: Device to check
3120  *
3121  * Device driver is responsible for setting is_switch bit on
3122  * in ib_device structure at init time.
3123  *
3124  * Return: true if the device is IB switch.
3125  */
3126 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3127 {
3128 	return device->is_switch;
3129 }
3130 
3131 /**
3132  * rdma_start_port - Return the first valid port number for the device
3133  * specified
3134  *
3135  * @device: Device to be checked
3136  *
3137  * Return start port number
3138  */
3139 static inline u32 rdma_start_port(const struct ib_device *device)
3140 {
3141 	return rdma_cap_ib_switch(device) ? 0 : 1;
3142 }
3143 
3144 /**
3145  * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3146  * @device: The struct ib_device * to iterate over
3147  * @iter: The unsigned int to store the port number
3148  */
3149 #define rdma_for_each_port(device, iter)                                       \
3150 	for (iter = rdma_start_port(device +				       \
3151 				    BUILD_BUG_ON_ZERO(!__same_type(u32,	       \
3152 								   iter)));    \
3153 	     iter <= rdma_end_port(device); iter++)
3154 
3155 /**
3156  * rdma_end_port - Return the last valid port number for the device
3157  * specified
3158  *
3159  * @device: Device to be checked
3160  *
3161  * Return last port number
3162  */
3163 static inline u32 rdma_end_port(const struct ib_device *device)
3164 {
3165 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3166 }
3167 
3168 static inline int rdma_is_port_valid(const struct ib_device *device,
3169 				     unsigned int port)
3170 {
3171 	return (port >= rdma_start_port(device) &&
3172 		port <= rdma_end_port(device));
3173 }
3174 
3175 static inline bool rdma_is_grh_required(const struct ib_device *device,
3176 					u32 port_num)
3177 {
3178 	return device->port_data[port_num].immutable.core_cap_flags &
3179 	       RDMA_CORE_PORT_IB_GRH_REQUIRED;
3180 }
3181 
3182 static inline bool rdma_protocol_ib(const struct ib_device *device,
3183 				    u32 port_num)
3184 {
3185 	return device->port_data[port_num].immutable.core_cap_flags &
3186 	       RDMA_CORE_CAP_PROT_IB;
3187 }
3188 
3189 static inline bool rdma_protocol_roce(const struct ib_device *device,
3190 				      u32 port_num)
3191 {
3192 	return device->port_data[port_num].immutable.core_cap_flags &
3193 	       (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3194 }
3195 
3196 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3197 						u32 port_num)
3198 {
3199 	return device->port_data[port_num].immutable.core_cap_flags &
3200 	       RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3201 }
3202 
3203 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3204 						u32 port_num)
3205 {
3206 	return device->port_data[port_num].immutable.core_cap_flags &
3207 	       RDMA_CORE_CAP_PROT_ROCE;
3208 }
3209 
3210 static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3211 				       u32 port_num)
3212 {
3213 	return device->port_data[port_num].immutable.core_cap_flags &
3214 	       RDMA_CORE_CAP_PROT_IWARP;
3215 }
3216 
3217 static inline bool rdma_ib_or_roce(const struct ib_device *device,
3218 				   u32 port_num)
3219 {
3220 	return rdma_protocol_ib(device, port_num) ||
3221 		rdma_protocol_roce(device, port_num);
3222 }
3223 
3224 static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3225 					    u32 port_num)
3226 {
3227 	return device->port_data[port_num].immutable.core_cap_flags &
3228 	       RDMA_CORE_CAP_PROT_RAW_PACKET;
3229 }
3230 
3231 static inline bool rdma_protocol_usnic(const struct ib_device *device,
3232 				       u32 port_num)
3233 {
3234 	return device->port_data[port_num].immutable.core_cap_flags &
3235 	       RDMA_CORE_CAP_PROT_USNIC;
3236 }
3237 
3238 /**
3239  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3240  * Management Datagrams.
3241  * @device: Device to check
3242  * @port_num: Port number to check
3243  *
3244  * Management Datagrams (MAD) are a required part of the InfiniBand
3245  * specification and are supported on all InfiniBand devices.  A slightly
3246  * extended version are also supported on OPA interfaces.
3247  *
3248  * Return: true if the port supports sending/receiving of MAD packets.
3249  */
3250 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3251 {
3252 	return device->port_data[port_num].immutable.core_cap_flags &
3253 	       RDMA_CORE_CAP_IB_MAD;
3254 }
3255 
3256 /**
3257  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3258  * Management Datagrams.
3259  * @device: Device to check
3260  * @port_num: Port number to check
3261  *
3262  * Intel OmniPath devices extend and/or replace the InfiniBand Management
3263  * datagrams with their own versions.  These OPA MADs share many but not all of
3264  * the characteristics of InfiniBand MADs.
3265  *
3266  * OPA MADs differ in the following ways:
3267  *
3268  *    1) MADs are variable size up to 2K
3269  *       IBTA defined MADs remain fixed at 256 bytes
3270  *    2) OPA SMPs must carry valid PKeys
3271  *    3) OPA SMP packets are a different format
3272  *
3273  * Return: true if the port supports OPA MAD packet formats.
3274  */
3275 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3276 {
3277 	return device->port_data[port_num].immutable.core_cap_flags &
3278 		RDMA_CORE_CAP_OPA_MAD;
3279 }
3280 
3281 /**
3282  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3283  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3284  * @device: Device to check
3285  * @port_num: Port number to check
3286  *
3287  * Each InfiniBand node is required to provide a Subnet Management Agent
3288  * that the subnet manager can access.  Prior to the fabric being fully
3289  * configured by the subnet manager, the SMA is accessed via a well known
3290  * interface called the Subnet Management Interface (SMI).  This interface
3291  * uses directed route packets to communicate with the SM to get around the
3292  * chicken and egg problem of the SM needing to know what's on the fabric
3293  * in order to configure the fabric, and needing to configure the fabric in
3294  * order to send packets to the devices on the fabric.  These directed
3295  * route packets do not need the fabric fully configured in order to reach
3296  * their destination.  The SMI is the only method allowed to send
3297  * directed route packets on an InfiniBand fabric.
3298  *
3299  * Return: true if the port provides an SMI.
3300  */
3301 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3302 {
3303 	return device->port_data[port_num].immutable.core_cap_flags &
3304 	       RDMA_CORE_CAP_IB_SMI;
3305 }
3306 
3307 /**
3308  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3309  * Communication Manager.
3310  * @device: Device to check
3311  * @port_num: Port number to check
3312  *
3313  * The InfiniBand Communication Manager is one of many pre-defined General
3314  * Service Agents (GSA) that are accessed via the General Service
3315  * Interface (GSI).  It's role is to facilitate establishment of connections
3316  * between nodes as well as other management related tasks for established
3317  * connections.
3318  *
3319  * Return: true if the port supports an IB CM (this does not guarantee that
3320  * a CM is actually running however).
3321  */
3322 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3323 {
3324 	return device->port_data[port_num].immutable.core_cap_flags &
3325 	       RDMA_CORE_CAP_IB_CM;
3326 }
3327 
3328 /**
3329  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3330  * Communication Manager.
3331  * @device: Device to check
3332  * @port_num: Port number to check
3333  *
3334  * Similar to above, but specific to iWARP connections which have a different
3335  * managment protocol than InfiniBand.
3336  *
3337  * Return: true if the port supports an iWARP CM (this does not guarantee that
3338  * a CM is actually running however).
3339  */
3340 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3341 {
3342 	return device->port_data[port_num].immutable.core_cap_flags &
3343 	       RDMA_CORE_CAP_IW_CM;
3344 }
3345 
3346 /**
3347  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3348  * Subnet Administration.
3349  * @device: Device to check
3350  * @port_num: Port number to check
3351  *
3352  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3353  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
3354  * fabrics, devices should resolve routes to other hosts by contacting the
3355  * SA to query the proper route.
3356  *
3357  * Return: true if the port should act as a client to the fabric Subnet
3358  * Administration interface.  This does not imply that the SA service is
3359  * running locally.
3360  */
3361 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3362 {
3363 	return device->port_data[port_num].immutable.core_cap_flags &
3364 	       RDMA_CORE_CAP_IB_SA;
3365 }
3366 
3367 /**
3368  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3369  * Multicast.
3370  * @device: Device to check
3371  * @port_num: Port number to check
3372  *
3373  * InfiniBand multicast registration is more complex than normal IPv4 or
3374  * IPv6 multicast registration.  Each Host Channel Adapter must register
3375  * with the Subnet Manager when it wishes to join a multicast group.  It
3376  * should do so only once regardless of how many queue pairs it subscribes
3377  * to this group.  And it should leave the group only after all queue pairs
3378  * attached to the group have been detached.
3379  *
3380  * Return: true if the port must undertake the additional adminstrative
3381  * overhead of registering/unregistering with the SM and tracking of the
3382  * total number of queue pairs attached to the multicast group.
3383  */
3384 static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3385 				     u32 port_num)
3386 {
3387 	return rdma_cap_ib_sa(device, port_num);
3388 }
3389 
3390 /**
3391  * rdma_cap_af_ib - Check if the port of device has the capability
3392  * Native Infiniband Address.
3393  * @device: Device to check
3394  * @port_num: Port number to check
3395  *
3396  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3397  * GID.  RoCE uses a different mechanism, but still generates a GID via
3398  * a prescribed mechanism and port specific data.
3399  *
3400  * Return: true if the port uses a GID address to identify devices on the
3401  * network.
3402  */
3403 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3404 {
3405 	return device->port_data[port_num].immutable.core_cap_flags &
3406 	       RDMA_CORE_CAP_AF_IB;
3407 }
3408 
3409 /**
3410  * rdma_cap_eth_ah - Check if the port of device has the capability
3411  * Ethernet Address Handle.
3412  * @device: Device to check
3413  * @port_num: Port number to check
3414  *
3415  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3416  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3417  * port.  Normally, packet headers are generated by the sending host
3418  * adapter, but when sending connectionless datagrams, we must manually
3419  * inject the proper headers for the fabric we are communicating over.
3420  *
3421  * Return: true if we are running as a RoCE port and must force the
3422  * addition of a Global Route Header built from our Ethernet Address
3423  * Handle into our header list for connectionless packets.
3424  */
3425 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3426 {
3427 	return device->port_data[port_num].immutable.core_cap_flags &
3428 	       RDMA_CORE_CAP_ETH_AH;
3429 }
3430 
3431 /**
3432  * rdma_cap_opa_ah - Check if the port of device supports
3433  * OPA Address handles
3434  * @device: Device to check
3435  * @port_num: Port number to check
3436  *
3437  * Return: true if we are running on an OPA device which supports
3438  * the extended OPA addressing.
3439  */
3440 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3441 {
3442 	return (device->port_data[port_num].immutable.core_cap_flags &
3443 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3444 }
3445 
3446 /**
3447  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3448  *
3449  * @device: Device
3450  * @port_num: Port number
3451  *
3452  * This MAD size includes the MAD headers and MAD payload.  No other headers
3453  * are included.
3454  *
3455  * Return the max MAD size required by the Port.  Will return 0 if the port
3456  * does not support MADs
3457  */
3458 static inline size_t rdma_max_mad_size(const struct ib_device *device,
3459 				       u32 port_num)
3460 {
3461 	return device->port_data[port_num].immutable.max_mad_size;
3462 }
3463 
3464 /**
3465  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3466  * @device: Device to check
3467  * @port_num: Port number to check
3468  *
3469  * RoCE GID table mechanism manages the various GIDs for a device.
3470  *
3471  * NOTE: if allocating the port's GID table has failed, this call will still
3472  * return true, but any RoCE GID table API will fail.
3473  *
3474  * Return: true if the port uses RoCE GID table mechanism in order to manage
3475  * its GIDs.
3476  */
3477 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3478 					   u32 port_num)
3479 {
3480 	return rdma_protocol_roce(device, port_num) &&
3481 		device->ops.add_gid && device->ops.del_gid;
3482 }
3483 
3484 /*
3485  * Check if the device supports READ W/ INVALIDATE.
3486  */
3487 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3488 {
3489 	/*
3490 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3491 	 * has support for it yet.
3492 	 */
3493 	return rdma_protocol_iwarp(dev, port_num);
3494 }
3495 
3496 /**
3497  * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3498  * @device: Device
3499  * @port_num: 1 based Port number
3500  *
3501  * Return true if port is an Intel OPA port , false if not
3502  */
3503 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3504 					  u32 port_num)
3505 {
3506 	return (device->port_data[port_num].immutable.core_cap_flags &
3507 		RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3508 }
3509 
3510 /**
3511  * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3512  * @device: Device
3513  * @port: Port number
3514  * @mtu: enum value of MTU
3515  *
3516  * Return the MTU size supported by the port as an integer value. Will return
3517  * -1 if enum value of mtu is not supported.
3518  */
3519 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3520 				       int mtu)
3521 {
3522 	if (rdma_core_cap_opa_port(device, port))
3523 		return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3524 	else
3525 		return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3526 }
3527 
3528 /**
3529  * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3530  * @device: Device
3531  * @port: Port number
3532  * @attr: port attribute
3533  *
3534  * Return the MTU size supported by the port as an integer value.
3535  */
3536 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3537 				     struct ib_port_attr *attr)
3538 {
3539 	if (rdma_core_cap_opa_port(device, port))
3540 		return attr->phys_mtu;
3541 	else
3542 		return ib_mtu_enum_to_int(attr->max_mtu);
3543 }
3544 
3545 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3546 			 int state);
3547 int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3548 		     struct ifla_vf_info *info);
3549 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3550 		    struct ifla_vf_stats *stats);
3551 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3552 		    struct ifla_vf_guid *node_guid,
3553 		    struct ifla_vf_guid *port_guid);
3554 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3555 		   int type);
3556 
3557 int ib_query_pkey(struct ib_device *device,
3558 		  u32 port_num, u16 index, u16 *pkey);
3559 
3560 int ib_modify_device(struct ib_device *device,
3561 		     int device_modify_mask,
3562 		     struct ib_device_modify *device_modify);
3563 
3564 int ib_modify_port(struct ib_device *device,
3565 		   u32 port_num, int port_modify_mask,
3566 		   struct ib_port_modify *port_modify);
3567 
3568 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3569 		u32 *port_num, u16 *index);
3570 
3571 int ib_find_pkey(struct ib_device *device,
3572 		 u32 port_num, u16 pkey, u16 *index);
3573 
3574 enum ib_pd_flags {
3575 	/*
3576 	 * Create a memory registration for all memory in the system and place
3577 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3578 	 * ULPs to avoid the overhead of dynamic MRs.
3579 	 *
3580 	 * This flag is generally considered unsafe and must only be used in
3581 	 * extremly trusted environments.  Every use of it will log a warning
3582 	 * in the kernel log.
3583 	 */
3584 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3585 };
3586 
3587 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3588 		const char *caller);
3589 
3590 /**
3591  * ib_alloc_pd - Allocates an unused protection domain.
3592  * @device: The device on which to allocate the protection domain.
3593  * @flags: protection domain flags
3594  *
3595  * A protection domain object provides an association between QPs, shared
3596  * receive queues, address handles, memory regions, and memory windows.
3597  *
3598  * Every PD has a local_dma_lkey which can be used as the lkey value for local
3599  * memory operations.
3600  */
3601 #define ib_alloc_pd(device, flags) \
3602 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3603 
3604 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3605 
3606 /**
3607  * ib_dealloc_pd - Deallocate kernel PD
3608  * @pd: The protection domain
3609  *
3610  * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3611  */
3612 static inline void ib_dealloc_pd(struct ib_pd *pd)
3613 {
3614 	int ret = ib_dealloc_pd_user(pd, NULL);
3615 
3616 	WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3617 }
3618 
3619 enum rdma_create_ah_flags {
3620 	/* In a sleepable context */
3621 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3622 };
3623 
3624 /**
3625  * rdma_create_ah - Creates an address handle for the given address vector.
3626  * @pd: The protection domain associated with the address handle.
3627  * @ah_attr: The attributes of the address vector.
3628  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3629  *
3630  * The address handle is used to reference a local or global destination
3631  * in all UD QP post sends.
3632  */
3633 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3634 			     u32 flags);
3635 
3636 /**
3637  * rdma_create_user_ah - Creates an address handle for the given address vector.
3638  * It resolves destination mac address for ah attribute of RoCE type.
3639  * @pd: The protection domain associated with the address handle.
3640  * @ah_attr: The attributes of the address vector.
3641  * @udata: pointer to user's input output buffer information need by
3642  *         provider driver.
3643  *
3644  * It returns 0 on success and returns appropriate error code on error.
3645  * The address handle is used to reference a local or global destination
3646  * in all UD QP post sends.
3647  */
3648 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3649 				  struct rdma_ah_attr *ah_attr,
3650 				  struct ib_udata *udata);
3651 /**
3652  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3653  *   work completion.
3654  * @hdr: the L3 header to parse
3655  * @net_type: type of header to parse
3656  * @sgid: place to store source gid
3657  * @dgid: place to store destination gid
3658  */
3659 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3660 			      enum rdma_network_type net_type,
3661 			      union ib_gid *sgid, union ib_gid *dgid);
3662 
3663 /**
3664  * ib_get_rdma_header_version - Get the header version
3665  * @hdr: the L3 header to parse
3666  */
3667 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3668 
3669 /**
3670  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3671  *   work completion.
3672  * @device: Device on which the received message arrived.
3673  * @port_num: Port on which the received message arrived.
3674  * @wc: Work completion associated with the received message.
3675  * @grh: References the received global route header.  This parameter is
3676  *   ignored unless the work completion indicates that the GRH is valid.
3677  * @ah_attr: Returned attributes that can be used when creating an address
3678  *   handle for replying to the message.
3679  * When ib_init_ah_attr_from_wc() returns success,
3680  * (a) for IB link layer it optionally contains a reference to SGID attribute
3681  * when GRH is present for IB link layer.
3682  * (b) for RoCE link layer it contains a reference to SGID attribute.
3683  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3684  * attributes which are initialized using ib_init_ah_attr_from_wc().
3685  *
3686  */
3687 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3688 			    const struct ib_wc *wc, const struct ib_grh *grh,
3689 			    struct rdma_ah_attr *ah_attr);
3690 
3691 /**
3692  * ib_create_ah_from_wc - Creates an address handle associated with the
3693  *   sender of the specified work completion.
3694  * @pd: The protection domain associated with the address handle.
3695  * @wc: Work completion information associated with a received message.
3696  * @grh: References the received global route header.  This parameter is
3697  *   ignored unless the work completion indicates that the GRH is valid.
3698  * @port_num: The outbound port number to associate with the address.
3699  *
3700  * The address handle is used to reference a local or global destination
3701  * in all UD QP post sends.
3702  */
3703 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3704 				   const struct ib_grh *grh, u32 port_num);
3705 
3706 /**
3707  * rdma_modify_ah - Modifies the address vector associated with an address
3708  *   handle.
3709  * @ah: The address handle to modify.
3710  * @ah_attr: The new address vector attributes to associate with the
3711  *   address handle.
3712  */
3713 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3714 
3715 /**
3716  * rdma_query_ah - Queries the address vector associated with an address
3717  *   handle.
3718  * @ah: The address handle to query.
3719  * @ah_attr: The address vector attributes associated with the address
3720  *   handle.
3721  */
3722 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3723 
3724 enum rdma_destroy_ah_flags {
3725 	/* In a sleepable context */
3726 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3727 };
3728 
3729 /**
3730  * rdma_destroy_ah_user - Destroys an address handle.
3731  * @ah: The address handle to destroy.
3732  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3733  * @udata: Valid user data or NULL for kernel objects
3734  */
3735 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3736 
3737 /**
3738  * rdma_destroy_ah - Destroys an kernel address handle.
3739  * @ah: The address handle to destroy.
3740  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3741  *
3742  * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3743  */
3744 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3745 {
3746 	int ret = rdma_destroy_ah_user(ah, flags, NULL);
3747 
3748 	WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3749 }
3750 
3751 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3752 				  struct ib_srq_init_attr *srq_init_attr,
3753 				  struct ib_usrq_object *uobject,
3754 				  struct ib_udata *udata);
3755 static inline struct ib_srq *
3756 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3757 {
3758 	if (!pd->device->ops.create_srq)
3759 		return ERR_PTR(-EOPNOTSUPP);
3760 
3761 	return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3762 }
3763 
3764 /**
3765  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3766  * @srq: The SRQ to modify.
3767  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3768  *   the current values of selected SRQ attributes are returned.
3769  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3770  *   are being modified.
3771  *
3772  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3773  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3774  * the number of receives queued drops below the limit.
3775  */
3776 int ib_modify_srq(struct ib_srq *srq,
3777 		  struct ib_srq_attr *srq_attr,
3778 		  enum ib_srq_attr_mask srq_attr_mask);
3779 
3780 /**
3781  * ib_query_srq - Returns the attribute list and current values for the
3782  *   specified SRQ.
3783  * @srq: The SRQ to query.
3784  * @srq_attr: The attributes of the specified SRQ.
3785  */
3786 int ib_query_srq(struct ib_srq *srq,
3787 		 struct ib_srq_attr *srq_attr);
3788 
3789 /**
3790  * ib_destroy_srq_user - Destroys the specified SRQ.
3791  * @srq: The SRQ to destroy.
3792  * @udata: Valid user data or NULL for kernel objects
3793  */
3794 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3795 
3796 /**
3797  * ib_destroy_srq - Destroys the specified kernel SRQ.
3798  * @srq: The SRQ to destroy.
3799  *
3800  * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3801  */
3802 static inline void ib_destroy_srq(struct ib_srq *srq)
3803 {
3804 	int ret = ib_destroy_srq_user(srq, NULL);
3805 
3806 	WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3807 }
3808 
3809 /**
3810  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3811  * @srq: The SRQ to post the work request on.
3812  * @recv_wr: A list of work requests to post on the receive queue.
3813  * @bad_recv_wr: On an immediate failure, this parameter will reference
3814  *   the work request that failed to be posted on the QP.
3815  */
3816 static inline int ib_post_srq_recv(struct ib_srq *srq,
3817 				   const struct ib_recv_wr *recv_wr,
3818 				   const struct ib_recv_wr **bad_recv_wr)
3819 {
3820 	const struct ib_recv_wr *dummy;
3821 
3822 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3823 					      bad_recv_wr ? : &dummy);
3824 }
3825 
3826 struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3827 				  struct ib_qp_init_attr *qp_init_attr,
3828 				  const char *caller);
3829 /**
3830  * ib_create_qp - Creates a kernel QP associated with the specific protection
3831  * domain.
3832  * @pd: The protection domain associated with the QP.
3833  * @init_attr: A list of initial attributes required to create the
3834  *   QP.  If QP creation succeeds, then the attributes are updated to
3835  *   the actual capabilities of the created QP.
3836  */
3837 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3838 					 struct ib_qp_init_attr *init_attr)
3839 {
3840 	return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
3841 }
3842 
3843 /**
3844  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3845  * @qp: The QP to modify.
3846  * @attr: On input, specifies the QP attributes to modify.  On output,
3847  *   the current values of selected QP attributes are returned.
3848  * @attr_mask: A bit-mask used to specify which attributes of the QP
3849  *   are being modified.
3850  * @udata: pointer to user's input output buffer information
3851  *   are being modified.
3852  * It returns 0 on success and returns appropriate error code on error.
3853  */
3854 int ib_modify_qp_with_udata(struct ib_qp *qp,
3855 			    struct ib_qp_attr *attr,
3856 			    int attr_mask,
3857 			    struct ib_udata *udata);
3858 
3859 /**
3860  * ib_modify_qp - Modifies the attributes for the specified QP and then
3861  *   transitions the QP to the given state.
3862  * @qp: The QP to modify.
3863  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3864  *   the current values of selected QP attributes are returned.
3865  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3866  *   are being modified.
3867  */
3868 int ib_modify_qp(struct ib_qp *qp,
3869 		 struct ib_qp_attr *qp_attr,
3870 		 int qp_attr_mask);
3871 
3872 /**
3873  * ib_query_qp - Returns the attribute list and current values for the
3874  *   specified QP.
3875  * @qp: The QP to query.
3876  * @qp_attr: The attributes of the specified QP.
3877  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3878  * @qp_init_attr: Additional attributes of the selected QP.
3879  *
3880  * The qp_attr_mask may be used to limit the query to gathering only the
3881  * selected attributes.
3882  */
3883 int ib_query_qp(struct ib_qp *qp,
3884 		struct ib_qp_attr *qp_attr,
3885 		int qp_attr_mask,
3886 		struct ib_qp_init_attr *qp_init_attr);
3887 
3888 /**
3889  * ib_destroy_qp - Destroys the specified QP.
3890  * @qp: The QP to destroy.
3891  * @udata: Valid udata or NULL for kernel objects
3892  */
3893 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3894 
3895 /**
3896  * ib_destroy_qp - Destroys the specified kernel QP.
3897  * @qp: The QP to destroy.
3898  *
3899  * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3900  */
3901 static inline int ib_destroy_qp(struct ib_qp *qp)
3902 {
3903 	return ib_destroy_qp_user(qp, NULL);
3904 }
3905 
3906 /**
3907  * ib_open_qp - Obtain a reference to an existing sharable QP.
3908  * @xrcd: XRC domain
3909  * @qp_open_attr: Attributes identifying the QP to open.
3910  *
3911  * Returns a reference to a sharable QP.
3912  */
3913 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3914 			 struct ib_qp_open_attr *qp_open_attr);
3915 
3916 /**
3917  * ib_close_qp - Release an external reference to a QP.
3918  * @qp: The QP handle to release
3919  *
3920  * The opened QP handle is released by the caller.  The underlying
3921  * shared QP is not destroyed until all internal references are released.
3922  */
3923 int ib_close_qp(struct ib_qp *qp);
3924 
3925 /**
3926  * ib_post_send - Posts a list of work requests to the send queue of
3927  *   the specified QP.
3928  * @qp: The QP to post the work request on.
3929  * @send_wr: A list of work requests to post on the send queue.
3930  * @bad_send_wr: On an immediate failure, this parameter will reference
3931  *   the work request that failed to be posted on the QP.
3932  *
3933  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3934  * error is returned, the QP state shall not be affected,
3935  * ib_post_send() will return an immediate error after queueing any
3936  * earlier work requests in the list.
3937  */
3938 static inline int ib_post_send(struct ib_qp *qp,
3939 			       const struct ib_send_wr *send_wr,
3940 			       const struct ib_send_wr **bad_send_wr)
3941 {
3942 	const struct ib_send_wr *dummy;
3943 
3944 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3945 }
3946 
3947 /**
3948  * ib_post_recv - Posts a list of work requests to the receive queue of
3949  *   the specified QP.
3950  * @qp: The QP to post the work request on.
3951  * @recv_wr: A list of work requests to post on the receive queue.
3952  * @bad_recv_wr: On an immediate failure, this parameter will reference
3953  *   the work request that failed to be posted on the QP.
3954  */
3955 static inline int ib_post_recv(struct ib_qp *qp,
3956 			       const struct ib_recv_wr *recv_wr,
3957 			       const struct ib_recv_wr **bad_recv_wr)
3958 {
3959 	const struct ib_recv_wr *dummy;
3960 
3961 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3962 }
3963 
3964 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3965 			    int comp_vector, enum ib_poll_context poll_ctx,
3966 			    const char *caller);
3967 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3968 					int nr_cqe, int comp_vector,
3969 					enum ib_poll_context poll_ctx)
3970 {
3971 	return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3972 			     KBUILD_MODNAME);
3973 }
3974 
3975 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3976 				int nr_cqe, enum ib_poll_context poll_ctx,
3977 				const char *caller);
3978 
3979 /**
3980  * ib_alloc_cq_any: Allocate kernel CQ
3981  * @dev: The IB device
3982  * @private: Private data attached to the CQE
3983  * @nr_cqe: Number of CQEs in the CQ
3984  * @poll_ctx: Context used for polling the CQ
3985  */
3986 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3987 					    void *private, int nr_cqe,
3988 					    enum ib_poll_context poll_ctx)
3989 {
3990 	return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3991 				 KBUILD_MODNAME);
3992 }
3993 
3994 void ib_free_cq(struct ib_cq *cq);
3995 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3996 
3997 /**
3998  * ib_create_cq - Creates a CQ on the specified device.
3999  * @device: The device on which to create the CQ.
4000  * @comp_handler: A user-specified callback that is invoked when a
4001  *   completion event occurs on the CQ.
4002  * @event_handler: A user-specified callback that is invoked when an
4003  *   asynchronous event not associated with a completion occurs on the CQ.
4004  * @cq_context: Context associated with the CQ returned to the user via
4005  *   the associated completion and event handlers.
4006  * @cq_attr: The attributes the CQ should be created upon.
4007  *
4008  * Users can examine the cq structure to determine the actual CQ size.
4009  */
4010 struct ib_cq *__ib_create_cq(struct ib_device *device,
4011 			     ib_comp_handler comp_handler,
4012 			     void (*event_handler)(struct ib_event *, void *),
4013 			     void *cq_context,
4014 			     const struct ib_cq_init_attr *cq_attr,
4015 			     const char *caller);
4016 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
4017 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
4018 
4019 /**
4020  * ib_resize_cq - Modifies the capacity of the CQ.
4021  * @cq: The CQ to resize.
4022  * @cqe: The minimum size of the CQ.
4023  *
4024  * Users can examine the cq structure to determine the actual CQ size.
4025  */
4026 int ib_resize_cq(struct ib_cq *cq, int cqe);
4027 
4028 /**
4029  * rdma_set_cq_moderation - Modifies moderation params of the CQ
4030  * @cq: The CQ to modify.
4031  * @cq_count: number of CQEs that will trigger an event
4032  * @cq_period: max period of time in usec before triggering an event
4033  *
4034  */
4035 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
4036 
4037 /**
4038  * ib_destroy_cq_user - Destroys the specified CQ.
4039  * @cq: The CQ to destroy.
4040  * @udata: Valid user data or NULL for kernel objects
4041  */
4042 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
4043 
4044 /**
4045  * ib_destroy_cq - Destroys the specified kernel CQ.
4046  * @cq: The CQ to destroy.
4047  *
4048  * NOTE: for user cq use ib_destroy_cq_user with valid udata!
4049  */
4050 static inline void ib_destroy_cq(struct ib_cq *cq)
4051 {
4052 	int ret = ib_destroy_cq_user(cq, NULL);
4053 
4054 	WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
4055 }
4056 
4057 /**
4058  * ib_poll_cq - poll a CQ for completion(s)
4059  * @cq:the CQ being polled
4060  * @num_entries:maximum number of completions to return
4061  * @wc:array of at least @num_entries &struct ib_wc where completions
4062  *   will be returned
4063  *
4064  * Poll a CQ for (possibly multiple) completions.  If the return value
4065  * is < 0, an error occurred.  If the return value is >= 0, it is the
4066  * number of completions returned.  If the return value is
4067  * non-negative and < num_entries, then the CQ was emptied.
4068  */
4069 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
4070 			     struct ib_wc *wc)
4071 {
4072 	return cq->device->ops.poll_cq(cq, num_entries, wc);
4073 }
4074 
4075 /**
4076  * ib_req_notify_cq - Request completion notification on a CQ.
4077  * @cq: The CQ to generate an event for.
4078  * @flags:
4079  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
4080  *   to request an event on the next solicited event or next work
4081  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
4082  *   may also be |ed in to request a hint about missed events, as
4083  *   described below.
4084  *
4085  * Return Value:
4086  *    < 0 means an error occurred while requesting notification
4087  *   == 0 means notification was requested successfully, and if
4088  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
4089  *        were missed and it is safe to wait for another event.  In
4090  *        this case is it guaranteed that any work completions added
4091  *        to the CQ since the last CQ poll will trigger a completion
4092  *        notification event.
4093  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
4094  *        in.  It means that the consumer must poll the CQ again to
4095  *        make sure it is empty to avoid missing an event because of a
4096  *        race between requesting notification and an entry being
4097  *        added to the CQ.  This return value means it is possible
4098  *        (but not guaranteed) that a work completion has been added
4099  *        to the CQ since the last poll without triggering a
4100  *        completion notification event.
4101  */
4102 static inline int ib_req_notify_cq(struct ib_cq *cq,
4103 				   enum ib_cq_notify_flags flags)
4104 {
4105 	return cq->device->ops.req_notify_cq(cq, flags);
4106 }
4107 
4108 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4109 			     int comp_vector_hint,
4110 			     enum ib_poll_context poll_ctx);
4111 
4112 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4113 
4114 /*
4115  * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
4116  * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
4117  * address into the dma address.
4118  */
4119 static inline bool ib_uses_virt_dma(struct ib_device *dev)
4120 {
4121 	return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
4122 }
4123 
4124 /*
4125  * Check if a IB device's underlying DMA mapping supports P2PDMA transfers.
4126  */
4127 static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev)
4128 {
4129 	if (ib_uses_virt_dma(dev))
4130 		return false;
4131 
4132 	return dma_pci_p2pdma_supported(dev->dma_device);
4133 }
4134 
4135 /**
4136  * ib_virt_dma_to_ptr - Convert a dma_addr to a kernel pointer
4137  * @dma_addr: The DMA address
4138  *
4139  * Used by ib_uses_virt_dma() devices to get back to the kernel pointer after
4140  * going through the dma_addr marshalling.
4141  */
4142 static inline void *ib_virt_dma_to_ptr(u64 dma_addr)
4143 {
4144 	/* virt_dma mode maps the kvs's directly into the dma addr */
4145 	return (void *)(uintptr_t)dma_addr;
4146 }
4147 
4148 /**
4149  * ib_virt_dma_to_page - Convert a dma_addr to a struct page
4150  * @dma_addr: The DMA address
4151  *
4152  * Used by ib_uses_virt_dma() device to get back to the struct page after going
4153  * through the dma_addr marshalling.
4154  */
4155 static inline struct page *ib_virt_dma_to_page(u64 dma_addr)
4156 {
4157 	return virt_to_page(ib_virt_dma_to_ptr(dma_addr));
4158 }
4159 
4160 /**
4161  * ib_dma_mapping_error - check a DMA addr for error
4162  * @dev: The device for which the dma_addr was created
4163  * @dma_addr: The DMA address to check
4164  */
4165 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4166 {
4167 	if (ib_uses_virt_dma(dev))
4168 		return 0;
4169 	return dma_mapping_error(dev->dma_device, dma_addr);
4170 }
4171 
4172 /**
4173  * ib_dma_map_single - Map a kernel virtual address to DMA address
4174  * @dev: The device for which the dma_addr is to be created
4175  * @cpu_addr: The kernel virtual address
4176  * @size: The size of the region in bytes
4177  * @direction: The direction of the DMA
4178  */
4179 static inline u64 ib_dma_map_single(struct ib_device *dev,
4180 				    void *cpu_addr, size_t size,
4181 				    enum dma_data_direction direction)
4182 {
4183 	if (ib_uses_virt_dma(dev))
4184 		return (uintptr_t)cpu_addr;
4185 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4186 }
4187 
4188 /**
4189  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4190  * @dev: The device for which the DMA address was created
4191  * @addr: The DMA address
4192  * @size: The size of the region in bytes
4193  * @direction: The direction of the DMA
4194  */
4195 static inline void ib_dma_unmap_single(struct ib_device *dev,
4196 				       u64 addr, size_t size,
4197 				       enum dma_data_direction direction)
4198 {
4199 	if (!ib_uses_virt_dma(dev))
4200 		dma_unmap_single(dev->dma_device, addr, size, direction);
4201 }
4202 
4203 /**
4204  * ib_dma_map_page - Map a physical page to DMA address
4205  * @dev: The device for which the dma_addr is to be created
4206  * @page: The page to be mapped
4207  * @offset: The offset within the page
4208  * @size: The size of the region in bytes
4209  * @direction: The direction of the DMA
4210  */
4211 static inline u64 ib_dma_map_page(struct ib_device *dev,
4212 				  struct page *page,
4213 				  unsigned long offset,
4214 				  size_t size,
4215 					 enum dma_data_direction direction)
4216 {
4217 	if (ib_uses_virt_dma(dev))
4218 		return (uintptr_t)(page_address(page) + offset);
4219 	return dma_map_page(dev->dma_device, page, offset, size, direction);
4220 }
4221 
4222 /**
4223  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4224  * @dev: The device for which the DMA address was created
4225  * @addr: The DMA address
4226  * @size: The size of the region in bytes
4227  * @direction: The direction of the DMA
4228  */
4229 static inline void ib_dma_unmap_page(struct ib_device *dev,
4230 				     u64 addr, size_t size,
4231 				     enum dma_data_direction direction)
4232 {
4233 	if (!ib_uses_virt_dma(dev))
4234 		dma_unmap_page(dev->dma_device, addr, size, direction);
4235 }
4236 
4237 /**
4238  * ib_dma_map_bvec - Map a bio_vec to DMA address
4239  * @dev: The device for which the dma_addr is to be created
4240  * @bvec: The bio_vec to map
4241  * @direction: The direction of the DMA
4242  *
4243  * Returns a DMA address for the bio_vec. The caller must check the
4244  * result with ib_dma_mapping_error() before use; a failed mapping
4245  * must not be passed to ib_dma_unmap_bvec().
4246  *
4247  * For software RDMA devices (rxe, siw), returns a virtual address
4248  * and no actual DMA mapping occurs.
4249  */
4250 static inline u64 ib_dma_map_bvec(struct ib_device *dev,
4251 				  struct bio_vec *bvec,
4252 				  enum dma_data_direction direction)
4253 {
4254 	if (ib_uses_virt_dma(dev))
4255 		return (uintptr_t)bvec_virt(bvec);
4256 	return dma_map_phys(dev->dma_device, bvec_phys(bvec),
4257 			    bvec->bv_len, direction, 0);
4258 }
4259 
4260 /**
4261  * ib_dma_unmap_bvec - Unmap a bio_vec DMA mapping
4262  * @dev: The device for which the DMA address was created
4263  * @addr: The DMA address returned by ib_dma_map_bvec()
4264  * @size: The size of the region in bytes
4265  * @direction: The direction of the DMA
4266  *
4267  * Releases a DMA mapping created by ib_dma_map_bvec(). For software
4268  * RDMA devices this is a no-op since no actual mapping occurred.
4269  */
4270 static inline void ib_dma_unmap_bvec(struct ib_device *dev,
4271 				     u64 addr, size_t size,
4272 				     enum dma_data_direction direction)
4273 {
4274 	if (!ib_uses_virt_dma(dev))
4275 		dma_unmap_phys(dev->dma_device, addr, size, direction, 0);
4276 }
4277 
4278 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4279 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4280 				      struct scatterlist *sg, int nents,
4281 				      enum dma_data_direction direction,
4282 				      unsigned long dma_attrs)
4283 {
4284 	if (ib_uses_virt_dma(dev))
4285 		return ib_dma_virt_map_sg(dev, sg, nents);
4286 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4287 				dma_attrs);
4288 }
4289 
4290 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4291 					 struct scatterlist *sg, int nents,
4292 					 enum dma_data_direction direction,
4293 					 unsigned long dma_attrs)
4294 {
4295 	if (!ib_uses_virt_dma(dev))
4296 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4297 				   dma_attrs);
4298 }
4299 
4300 /**
4301  * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4302  * @dev: The device for which the DMA addresses are to be created
4303  * @sgt: The sg_table object describing the buffer
4304  * @direction: The direction of the DMA
4305  * @dma_attrs: Optional DMA attributes for the map operation
4306  */
4307 static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4308 					   struct sg_table *sgt,
4309 					   enum dma_data_direction direction,
4310 					   unsigned long dma_attrs)
4311 {
4312 	int nents;
4313 
4314 	if (ib_uses_virt_dma(dev)) {
4315 		nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4316 		if (!nents)
4317 			return -EIO;
4318 		sgt->nents = nents;
4319 		return 0;
4320 	}
4321 	return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4322 }
4323 
4324 static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4325 					      struct sg_table *sgt,
4326 					      enum dma_data_direction direction,
4327 					      unsigned long dma_attrs)
4328 {
4329 	if (!ib_uses_virt_dma(dev))
4330 		dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4331 }
4332 
4333 /**
4334  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4335  * @dev: The device for which the DMA addresses are to be created
4336  * @sg: The array of scatter/gather entries
4337  * @nents: The number of scatter/gather entries
4338  * @direction: The direction of the DMA
4339  */
4340 static inline int ib_dma_map_sg(struct ib_device *dev,
4341 				struct scatterlist *sg, int nents,
4342 				enum dma_data_direction direction)
4343 {
4344 	return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4345 }
4346 
4347 /**
4348  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4349  * @dev: The device for which the DMA addresses were created
4350  * @sg: The array of scatter/gather entries
4351  * @nents: The number of scatter/gather entries
4352  * @direction: The direction of the DMA
4353  */
4354 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4355 				   struct scatterlist *sg, int nents,
4356 				   enum dma_data_direction direction)
4357 {
4358 	ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4359 }
4360 
4361 /**
4362  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4363  * @dev: The device to query
4364  *
4365  * The returned value represents a size in bytes.
4366  */
4367 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4368 {
4369 	if (ib_uses_virt_dma(dev))
4370 		return UINT_MAX;
4371 	return dma_get_max_seg_size(dev->dma_device);
4372 }
4373 
4374 /**
4375  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4376  * @dev: The device for which the DMA address was created
4377  * @addr: The DMA address
4378  * @size: The size of the region in bytes
4379  * @dir: The direction of the DMA
4380  */
4381 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4382 					      u64 addr,
4383 					      size_t size,
4384 					      enum dma_data_direction dir)
4385 {
4386 	if (!ib_uses_virt_dma(dev))
4387 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4388 }
4389 
4390 /**
4391  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4392  * @dev: The device for which the DMA address was created
4393  * @addr: The DMA address
4394  * @size: The size of the region in bytes
4395  * @dir: The direction of the DMA
4396  */
4397 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4398 						 u64 addr,
4399 						 size_t size,
4400 						 enum dma_data_direction dir)
4401 {
4402 	if (!ib_uses_virt_dma(dev))
4403 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4404 }
4405 
4406 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4407  * space. This function should be called when 'current' is the owning MM.
4408  */
4409 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4410 			     u64 virt_addr, int mr_access_flags);
4411 
4412 /* ib_advise_mr -  give an advice about an address range in a memory region */
4413 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4414 		 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4415 /**
4416  * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4417  *   HCA translation table.
4418  * @mr: The memory region to deregister.
4419  * @udata: Valid user data or NULL for kernel object
4420  *
4421  * This function can fail, if the memory region has memory windows bound to it.
4422  */
4423 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4424 
4425 /**
4426  * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4427  *   HCA translation table.
4428  * @mr: The memory region to deregister.
4429  *
4430  * This function can fail, if the memory region has memory windows bound to it.
4431  *
4432  * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4433  */
4434 static inline int ib_dereg_mr(struct ib_mr *mr)
4435 {
4436 	return ib_dereg_mr_user(mr, NULL);
4437 }
4438 
4439 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4440 			  u32 max_num_sg);
4441 
4442 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4443 				    u32 max_num_data_sg,
4444 				    u32 max_num_meta_sg);
4445 
4446 /**
4447  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4448  *   R_Key and L_Key.
4449  * @mr: struct ib_mr pointer to be updated.
4450  * @newkey: new key to be used.
4451  */
4452 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4453 {
4454 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4455 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4456 }
4457 
4458 /**
4459  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4460  * for calculating a new rkey for type 2 memory windows.
4461  * @rkey: the rkey to increment.
4462  */
4463 static inline u32 ib_inc_rkey(u32 rkey)
4464 {
4465 	const u32 mask = 0x000000ff;
4466 	return ((rkey + 1) & mask) | (rkey & ~mask);
4467 }
4468 
4469 /**
4470  * ib_attach_mcast - Attaches the specified QP to a multicast group.
4471  * @qp: QP to attach to the multicast group.  The QP must be type
4472  *   IB_QPT_UD.
4473  * @gid: Multicast group GID.
4474  * @lid: Multicast group LID in host byte order.
4475  *
4476  * In order to send and receive multicast packets, subnet
4477  * administration must have created the multicast group and configured
4478  * the fabric appropriately.  The port associated with the specified
4479  * QP must also be a member of the multicast group.
4480  */
4481 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4482 
4483 /**
4484  * ib_detach_mcast - Detaches the specified QP from a multicast group.
4485  * @qp: QP to detach from the multicast group.
4486  * @gid: Multicast group GID.
4487  * @lid: Multicast group LID in host byte order.
4488  */
4489 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4490 
4491 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4492 				   struct inode *inode, struct ib_udata *udata);
4493 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4494 
4495 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4496 				     unsigned int flags)
4497 {
4498 	u64 device_cap = ib_dev->attrs.device_cap_flags;
4499 
4500 	/*
4501 	 * Local write permission is required if remote write or
4502 	 * remote atomic permission is also requested.
4503 	 */
4504 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4505 	    !(flags & IB_ACCESS_LOCAL_WRITE))
4506 		return -EINVAL;
4507 
4508 	if (flags & ~IB_ACCESS_SUPPORTED)
4509 		return -EINVAL;
4510 
4511 	if (flags & IB_ACCESS_ON_DEMAND &&
4512 	    !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING))
4513 		return -EOPNOTSUPP;
4514 
4515 	if ((flags & IB_ACCESS_FLUSH_GLOBAL &&
4516 	    !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) ||
4517 	    (flags & IB_ACCESS_FLUSH_PERSISTENT &&
4518 	    !(device_cap & IB_DEVICE_FLUSH_PERSISTENT)))
4519 		return -EOPNOTSUPP;
4520 
4521 	return 0;
4522 }
4523 
4524 static inline bool ib_access_writable(int access_flags)
4525 {
4526 	/*
4527 	 * We have writable memory backing the MR if any of the following
4528 	 * access flags are set.  "Local write" and "remote write" obviously
4529 	 * require write access.  "Remote atomic" can do things like fetch and
4530 	 * add, which will modify memory, and "MW bind" can change permissions
4531 	 * by binding a window.
4532 	 */
4533 	return access_flags &
4534 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
4535 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4536 }
4537 
4538 /**
4539  * ib_check_mr_status: lightweight check of MR status.
4540  *     This routine may provide status checks on a selected
4541  *     ib_mr. first use is for signature status check.
4542  *
4543  * @mr: A memory region.
4544  * @check_mask: Bitmask of which checks to perform from
4545  *     ib_mr_status_check enumeration.
4546  * @mr_status: The container of relevant status checks.
4547  *     failed checks will be indicated in the status bitmask
4548  *     and the relevant info shall be in the error item.
4549  */
4550 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4551 		       struct ib_mr_status *mr_status);
4552 
4553 /**
4554  * ib_device_try_get: Hold a registration lock
4555  * @dev: The device to lock
4556  *
4557  * A device under an active registration lock cannot become unregistered. It
4558  * is only possible to obtain a registration lock on a device that is fully
4559  * registered, otherwise this function returns false.
4560  *
4561  * The registration lock is only necessary for actions which require the
4562  * device to still be registered. Uses that only require the device pointer to
4563  * be valid should use get_device(&ibdev->dev) to hold the memory.
4564  *
4565  */
4566 static inline bool ib_device_try_get(struct ib_device *dev)
4567 {
4568 	return refcount_inc_not_zero(&dev->refcount);
4569 }
4570 
4571 void ib_device_put(struct ib_device *device);
4572 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4573 					  enum rdma_driver_id driver_id);
4574 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4575 					    u16 pkey, const union ib_gid *gid,
4576 					    const struct sockaddr *addr);
4577 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4578 			 unsigned int port);
4579 struct net_device *ib_device_get_netdev(struct ib_device *ib_dev,
4580 					u32 port);
4581 int ib_query_netdev_port(struct ib_device *ibdev, struct net_device *ndev,
4582 			 u32 *port);
4583 
4584 static inline enum ib_port_state ib_get_curr_port_state(struct net_device *net_dev)
4585 {
4586 	return (netif_running(net_dev) && netif_carrier_ok(net_dev)) ?
4587 		IB_PORT_ACTIVE : IB_PORT_DOWN;
4588 }
4589 
4590 void ib_dispatch_port_state_event(struct ib_device *ibdev,
4591 				  struct net_device *ndev);
4592 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4593 			   struct ib_wq_init_attr *init_attr);
4594 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4595 
4596 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4597 		 unsigned int *sg_offset, unsigned int page_size);
4598 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4599 		    int data_sg_nents, unsigned int *data_sg_offset,
4600 		    struct scatterlist *meta_sg, int meta_sg_nents,
4601 		    unsigned int *meta_sg_offset, unsigned int page_size);
4602 
4603 static inline int
4604 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4605 		  unsigned int *sg_offset, unsigned int page_size)
4606 {
4607 	int n;
4608 
4609 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4610 	mr->iova = 0;
4611 
4612 	return n;
4613 }
4614 
4615 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4616 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4617 
4618 void ib_drain_rq(struct ib_qp *qp);
4619 void ib_drain_sq(struct ib_qp *qp);
4620 void ib_drain_qp(struct ib_qp *qp);
4621 
4622 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4623 		     u8 *width);
4624 
4625 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4626 {
4627 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4628 		return attr->roce.dmac;
4629 	return NULL;
4630 }
4631 
4632 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4633 {
4634 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4635 		attr->ib.dlid = (u16)dlid;
4636 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4637 		attr->opa.dlid = dlid;
4638 }
4639 
4640 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4641 {
4642 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4643 		return attr->ib.dlid;
4644 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4645 		return attr->opa.dlid;
4646 	return 0;
4647 }
4648 
4649 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4650 {
4651 	attr->sl = sl;
4652 }
4653 
4654 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4655 {
4656 	return attr->sl;
4657 }
4658 
4659 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4660 					 u8 src_path_bits)
4661 {
4662 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4663 		attr->ib.src_path_bits = src_path_bits;
4664 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4665 		attr->opa.src_path_bits = src_path_bits;
4666 }
4667 
4668 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4669 {
4670 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4671 		return attr->ib.src_path_bits;
4672 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4673 		return attr->opa.src_path_bits;
4674 	return 0;
4675 }
4676 
4677 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4678 					bool make_grd)
4679 {
4680 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4681 		attr->opa.make_grd = make_grd;
4682 }
4683 
4684 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4685 {
4686 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4687 		return attr->opa.make_grd;
4688 	return false;
4689 }
4690 
4691 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4692 {
4693 	attr->port_num = port_num;
4694 }
4695 
4696 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4697 {
4698 	return attr->port_num;
4699 }
4700 
4701 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4702 					   u8 static_rate)
4703 {
4704 	attr->static_rate = static_rate;
4705 }
4706 
4707 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4708 {
4709 	return attr->static_rate;
4710 }
4711 
4712 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4713 					enum ib_ah_flags flag)
4714 {
4715 	attr->ah_flags = flag;
4716 }
4717 
4718 static inline enum ib_ah_flags
4719 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4720 {
4721 	return attr->ah_flags;
4722 }
4723 
4724 static inline const struct ib_global_route
4725 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4726 {
4727 	return &attr->grh;
4728 }
4729 
4730 /*To retrieve and modify the grh */
4731 static inline struct ib_global_route
4732 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4733 {
4734 	return &attr->grh;
4735 }
4736 
4737 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4738 {
4739 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4740 
4741 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4742 }
4743 
4744 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4745 					     __be64 prefix)
4746 {
4747 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4748 
4749 	grh->dgid.global.subnet_prefix = prefix;
4750 }
4751 
4752 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4753 					    __be64 if_id)
4754 {
4755 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4756 
4757 	grh->dgid.global.interface_id = if_id;
4758 }
4759 
4760 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4761 				   union ib_gid *dgid, u32 flow_label,
4762 				   u8 sgid_index, u8 hop_limit,
4763 				   u8 traffic_class)
4764 {
4765 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4766 
4767 	attr->ah_flags = IB_AH_GRH;
4768 	if (dgid)
4769 		grh->dgid = *dgid;
4770 	grh->flow_label = flow_label;
4771 	grh->sgid_index = sgid_index;
4772 	grh->hop_limit = hop_limit;
4773 	grh->traffic_class = traffic_class;
4774 	grh->sgid_attr = NULL;
4775 }
4776 
4777 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4778 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4779 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4780 			     const struct ib_gid_attr *sgid_attr);
4781 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4782 		       const struct rdma_ah_attr *src);
4783 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4784 			  const struct rdma_ah_attr *new);
4785 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4786 
4787 /**
4788  * rdma_ah_find_type - Return address handle type.
4789  *
4790  * @dev: Device to be checked
4791  * @port_num: Port number
4792  */
4793 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4794 						       u32 port_num)
4795 {
4796 	if (rdma_protocol_roce(dev, port_num))
4797 		return RDMA_AH_ATTR_TYPE_ROCE;
4798 	if (rdma_protocol_ib(dev, port_num)) {
4799 		if (rdma_cap_opa_ah(dev, port_num))
4800 			return RDMA_AH_ATTR_TYPE_OPA;
4801 		return RDMA_AH_ATTR_TYPE_IB;
4802 	}
4803 	if (dev->type == RDMA_DEVICE_TYPE_SMI)
4804 		return RDMA_AH_ATTR_TYPE_IB;
4805 
4806 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4807 }
4808 
4809 /**
4810  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4811  *     In the current implementation the only way to
4812  *     get the 32bit lid is from other sources for OPA.
4813  *     For IB, lids will always be 16bits so cast the
4814  *     value accordingly.
4815  *
4816  * @lid: A 32bit LID
4817  */
4818 static inline u16 ib_lid_cpu16(u32 lid)
4819 {
4820 	WARN_ON_ONCE(lid & 0xFFFF0000);
4821 	return (u16)lid;
4822 }
4823 
4824 /**
4825  * ib_lid_be16 - Return lid in 16bit BE encoding.
4826  *
4827  * @lid: A 32bit LID
4828  */
4829 static inline __be16 ib_lid_be16(u32 lid)
4830 {
4831 	WARN_ON_ONCE(lid & 0xFFFF0000);
4832 	return cpu_to_be16((u16)lid);
4833 }
4834 
4835 /**
4836  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4837  * and add their gids, as needed, to the relevant RoCE devices.
4838  *
4839  * @ibdev:         the rdma device
4840  */
4841 void rdma_roce_rescan_device(struct ib_device *ibdev);
4842 void rdma_roce_rescan_port(struct ib_device *ib_dev, u32 port);
4843 void roce_del_all_netdev_gids(struct ib_device *ib_dev,
4844 			      u32 port, struct net_device *ndev);
4845 
4846 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4847 
4848 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
4849 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4850 bool rdma_uattrs_has_raw_cap(const struct uverbs_attr_bundle *attrs);
4851 #else
4852 static inline int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs)
4853 {
4854 	return 0;
4855 }
4856 static inline bool
4857 rdma_uattrs_has_raw_cap(const struct uverbs_attr_bundle *attrs)
4858 {
4859 	return false;
4860 }
4861 #endif
4862 
4863 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4864 				     enum rdma_netdev_t type, const char *name,
4865 				     unsigned char name_assign_type,
4866 				     void (*setup)(struct net_device *));
4867 
4868 int rdma_init_netdev(struct ib_device *device, u32 port_num,
4869 		     enum rdma_netdev_t type, const char *name,
4870 		     unsigned char name_assign_type,
4871 		     void (*setup)(struct net_device *),
4872 		     struct net_device *netdev);
4873 
4874 /**
4875  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4876  *
4877  * @device:	device pointer for which ib_device pointer to retrieve
4878  *
4879  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4880  *
4881  */
4882 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4883 {
4884 	struct ib_core_device *coredev =
4885 		container_of(device, struct ib_core_device, dev);
4886 
4887 	return coredev->owner;
4888 }
4889 
4890 /**
4891  * ibdev_to_node - return the NUMA node for a given ib_device
4892  * @ibdev:	device to get the NUMA node for.
4893  */
4894 static inline int ibdev_to_node(struct ib_device *ibdev)
4895 {
4896 	struct device *parent = ibdev->dev.parent;
4897 
4898 	if (!parent)
4899 		return NUMA_NO_NODE;
4900 	return dev_to_node(parent);
4901 }
4902 
4903 /**
4904  * rdma_device_to_drv_device - Helper macro to reach back to driver's
4905  *			       ib_device holder structure from device pointer.
4906  *
4907  * NOTE: New drivers should not make use of this API; This API is only for
4908  * existing drivers who have exposed sysfs entries using
4909  * ops->device_group.
4910  */
4911 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
4912 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4913 
4914 bool rdma_dev_access_netns(const struct ib_device *device,
4915 			   const struct net *net);
4916 
4917 bool rdma_dev_has_raw_cap(const struct ib_device *dev);
4918 static inline struct net *rdma_dev_net(struct ib_device *device)
4919 {
4920 	return read_pnet(&device->coredev.rdma_net);
4921 }
4922 
4923 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4924 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4925 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4926 
4927 /**
4928  * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4929  *                               on the flow_label
4930  * @fl: flow_label value
4931  *
4932  * This function will convert the 20 bit flow_label input to a valid RoCE v2
4933  * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4934  * convention.
4935  */
4936 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4937 {
4938 	u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4939 
4940 	fl_low ^= fl_high >> 14;
4941 	return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4942 }
4943 
4944 /**
4945  * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4946  *                        local and remote qpn values
4947  *
4948  * This function folded the multiplication results of two qpns, 24 bit each,
4949  * fields, and converts it to a 20 bit results.
4950  *
4951  * This function will create symmetric flow_label value based on the local
4952  * and remote qpn values. this will allow both the requester and responder
4953  * to calculate the same flow_label for a given connection.
4954  *
4955  * This helper function should be used by driver in case the upper layer
4956  * provide a zero flow_label value. This is to improve entropy of RDMA
4957  * traffic in the network.
4958  */
4959 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4960 {
4961 	u64 v = (u64)lqpn * rqpn;
4962 
4963 	v ^= v >> 20;
4964 	v ^= v >> 40;
4965 
4966 	return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4967 }
4968 
4969 /**
4970  * rdma_get_udp_sport - Calculate and set UDP source port based on the flow
4971  *                      label. If flow label is not defined in GRH then
4972  *                      calculate it based on lqpn/rqpn.
4973  *
4974  * @fl:                 flow label from GRH
4975  * @lqpn:               local qp number
4976  * @rqpn:               remote qp number
4977  */
4978 static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
4979 {
4980 	if (!fl)
4981 		fl = rdma_calc_flow_label(lqpn, rqpn);
4982 
4983 	return rdma_flow_label_to_udp_sport(fl);
4984 }
4985 
4986 const struct ib_port_immutable*
4987 ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4988 
4989 /** ib_add_sub_device - Add a sub IB device on an existing one
4990  *
4991  * @parent: The IB device that needs to add a sub device
4992  * @type: The type of the new sub device
4993  * @name: The name of the new sub device
4994  *
4995  *
4996  * Return 0 on success, an error code otherwise
4997  */
4998 int ib_add_sub_device(struct ib_device *parent,
4999 		      enum rdma_nl_dev_type type,
5000 		      const char *name);
5001 
5002 
5003 /** ib_del_sub_device_and_put - Delect an IB sub device while holding a 'get'
5004  *
5005  * @sub: The sub device that is going to be deleted
5006  *
5007  * Return 0 on success, an error code otherwise
5008  */
5009 int ib_del_sub_device_and_put(struct ib_device *sub);
5010 
5011 static inline void ib_mark_name_assigned_by_user(struct ib_device *ibdev)
5012 {
5013 	ibdev->name_assign_type = RDMA_NAME_ASSIGN_TYPE_USER;
5014 }
5015 
5016 #endif /* IB_VERBS_H */
5017