xref: /linux/include/rdma/ib_verbs.h (revision bba2c3615bd6cfee7456d1130f2e6b01b3f4e9ba)
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 	IB_DEVICE_CC_DMA_BOUNCE = IB_UVERBS_DEVICE_CC_DMA_BOUNCE,
279 };
280 
281 enum ib_kernel_cap_flags {
282 	/*
283 	 * This device supports a per-device lkey or stag that can be
284 	 * used without performing a memory registration for the local
285 	 * memory.  Note that ULPs should never check this flag, but
286 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
287 	 * which will always contain a usable lkey.
288 	 */
289 	IBK_LOCAL_DMA_LKEY = 1 << 0,
290 	/* IB_QP_CREATE_INTEGRITY_EN is supported to implement T10-PI */
291 	IBK_INTEGRITY_HANDOVER = 1 << 1,
292 	/* IB_ACCESS_ON_DEMAND is supported during reg_user_mr() */
293 	IBK_ON_DEMAND_PAGING = 1 << 2,
294 	/* IB_MR_TYPE_SG_GAPS is supported */
295 	IBK_SG_GAPS_REG = 1 << 3,
296 	/* Driver supports RDMA_NLDEV_CMD_DELLINK */
297 	IBK_ALLOW_USER_UNREG = 1 << 4,
298 
299 	/* ipoib will use IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK */
300 	IBK_BLOCK_MULTICAST_LOOPBACK = 1 << 5,
301 	/* iopib will use IB_QP_CREATE_IPOIB_UD_LSO for its QPs */
302 	IBK_UD_TSO = 1 << 6,
303 	/* iopib will use the device ops:
304 	 *   get_vf_config
305 	 *   get_vf_guid
306 	 *   get_vf_stats
307 	 *   set_vf_guid
308 	 *   set_vf_link_state
309 	 */
310 	IBK_VIRTUAL_FUNCTION = 1 << 7,
311 	/* ipoib will use IB_QP_CREATE_NETDEV_USE for its QPs */
312 	IBK_RDMA_NETDEV_OPA = 1 << 8,
313 };
314 
315 enum ib_atomic_cap {
316 	IB_ATOMIC_NONE,
317 	IB_ATOMIC_HCA,
318 	IB_ATOMIC_GLOB
319 };
320 
321 enum ib_odp_general_cap_bits {
322 	IB_ODP_SUPPORT		= IB_UVERBS_ODP_SUPPORT,
323 	IB_ODP_SUPPORT_IMPLICIT = IB_UVERBS_ODP_SUPPORT_IMPLICIT,
324 };
325 
326 enum ib_odp_transport_cap_bits {
327 	IB_ODP_SUPPORT_SEND	= IB_UVERBS_ODP_SUPPORT_SEND,
328 	IB_ODP_SUPPORT_RECV	= IB_UVERBS_ODP_SUPPORT_RECV,
329 	IB_ODP_SUPPORT_WRITE	= IB_UVERBS_ODP_SUPPORT_WRITE,
330 	IB_ODP_SUPPORT_READ	= IB_UVERBS_ODP_SUPPORT_READ,
331 	IB_ODP_SUPPORT_ATOMIC	= IB_UVERBS_ODP_SUPPORT_ATOMIC,
332 	IB_ODP_SUPPORT_SRQ_RECV	= IB_UVERBS_ODP_SUPPORT_SRQ_RECV,
333 	IB_ODP_SUPPORT_FLUSH	= IB_UVERBS_ODP_SUPPORT_FLUSH,
334 	IB_ODP_SUPPORT_ATOMIC_WRITE	= IB_UVERBS_ODP_SUPPORT_ATOMIC_WRITE,
335 };
336 
337 struct ib_odp_caps {
338 	uint64_t general_caps;
339 	struct {
340 		uint32_t  rc_odp_caps;
341 		uint32_t  uc_odp_caps;
342 		uint32_t  ud_odp_caps;
343 		uint32_t  xrc_odp_caps;
344 	} per_transport_caps;
345 };
346 
347 struct ib_rss_caps {
348 	/* Corresponding bit will be set if qp type from
349 	 * 'enum ib_qp_type' is supported, e.g.
350 	 * supported_qpts |= 1 << IB_QPT_UD
351 	 */
352 	u32 supported_qpts;
353 	u32 max_rwq_indirection_tables;
354 	u32 max_rwq_indirection_table_size;
355 };
356 
357 enum ib_tm_cap_flags {
358 	/*  Support tag matching with rendezvous offload for RC transport */
359 	IB_TM_CAP_RNDV_RC = 1 << 0,
360 };
361 
362 struct ib_tm_caps {
363 	/* Max size of RNDV header */
364 	u32 max_rndv_hdr_size;
365 	/* Max number of entries in tag matching list */
366 	u32 max_num_tags;
367 	/* From enum ib_tm_cap_flags */
368 	u32 flags;
369 	/* Max number of outstanding list operations */
370 	u32 max_ops;
371 	/* Max number of SGE in tag matching entry */
372 	u32 max_sge;
373 };
374 
375 struct ib_cq_init_attr {
376 	unsigned int	cqe;
377 	u32		comp_vector;
378 	u32		flags;
379 };
380 
381 enum ib_cq_attr_mask {
382 	IB_CQ_MODERATE = 1 << 0,
383 };
384 
385 struct ib_cq_caps {
386 	u16     max_cq_moderation_count;
387 	u16     max_cq_moderation_period;
388 };
389 
390 struct ib_dm_mr_attr {
391 	u64		length;
392 	u64		offset;
393 	u32		access_flags;
394 };
395 
396 struct ib_dm_alloc_attr {
397 	u64	length;
398 	u32	alignment;
399 	u32	flags;
400 };
401 
402 struct ib_device_attr {
403 	u64			fw_ver;
404 	__be64			sys_image_guid;
405 	u64			max_mr_size;
406 	u64			page_size_cap;
407 	u32			vendor_id;
408 	u32			vendor_part_id;
409 	u32			hw_ver;
410 	int			max_qp;
411 	int			max_qp_wr;
412 	u64			device_cap_flags;
413 	u64			kernel_cap_flags;
414 	int			max_send_sge;
415 	int			max_recv_sge;
416 	int			max_sge_rd;
417 	int			max_cq;
418 	int			max_cqe;
419 	int			max_mr;
420 	int			max_pd;
421 	int			max_qp_rd_atom;
422 	int			max_ee_rd_atom;
423 	int			max_res_rd_atom;
424 	int			max_qp_init_rd_atom;
425 	int			max_ee_init_rd_atom;
426 	enum ib_atomic_cap	atomic_cap;
427 	enum ib_atomic_cap	masked_atomic_cap;
428 	int			max_ee;
429 	int			max_rdd;
430 	int			max_mw;
431 	int			max_raw_ipv6_qp;
432 	int			max_raw_ethy_qp;
433 	int			max_mcast_grp;
434 	int			max_mcast_qp_attach;
435 	int			max_total_mcast_qp_attach;
436 	int			max_ah;
437 	int			max_srq;
438 	int			max_srq_wr;
439 	int			max_srq_sge;
440 	unsigned int		max_fast_reg_page_list_len;
441 	unsigned int		max_pi_fast_reg_page_list_len;
442 	u16			max_pkeys;
443 	u8			local_ca_ack_delay;
444 	int			sig_prot_cap;
445 	int			sig_guard_cap;
446 	struct ib_odp_caps	odp_caps;
447 	uint64_t		timestamp_mask;
448 	uint64_t		hca_core_clock; /* in KHZ */
449 	struct ib_rss_caps	rss_caps;
450 	u32			max_wq_type_rq;
451 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
452 	struct ib_tm_caps	tm_caps;
453 	struct ib_cq_caps       cq_caps;
454 	u64			max_dm_size;
455 	/* Max entries for sgl for optimized performance per READ */
456 	u32			max_sgl_rd;
457 };
458 
459 enum ib_mtu {
460 	IB_MTU_256  = 1,
461 	IB_MTU_512  = 2,
462 	IB_MTU_1024 = 3,
463 	IB_MTU_2048 = 4,
464 	IB_MTU_4096 = 5
465 };
466 
467 enum opa_mtu {
468 	OPA_MTU_8192 = 6,
469 	OPA_MTU_10240 = 7
470 };
471 
472 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
473 {
474 	switch (mtu) {
475 	case IB_MTU_256:  return  256;
476 	case IB_MTU_512:  return  512;
477 	case IB_MTU_1024: return 1024;
478 	case IB_MTU_2048: return 2048;
479 	case IB_MTU_4096: return 4096;
480 	default: 	  return -1;
481 	}
482 }
483 
484 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
485 {
486 	if (mtu >= 4096)
487 		return IB_MTU_4096;
488 	else if (mtu >= 2048)
489 		return IB_MTU_2048;
490 	else if (mtu >= 1024)
491 		return IB_MTU_1024;
492 	else if (mtu >= 512)
493 		return IB_MTU_512;
494 	else
495 		return IB_MTU_256;
496 }
497 
498 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
499 {
500 	switch (mtu) {
501 	case OPA_MTU_8192:
502 		return 8192;
503 	case OPA_MTU_10240:
504 		return 10240;
505 	default:
506 		return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
507 	}
508 }
509 
510 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
511 {
512 	if (mtu >= 10240)
513 		return OPA_MTU_10240;
514 	else if (mtu >= 8192)
515 		return OPA_MTU_8192;
516 	else
517 		return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
518 }
519 
520 enum ib_port_state {
521 	IB_PORT_NOP		= 0,
522 	IB_PORT_DOWN		= 1,
523 	IB_PORT_INIT		= 2,
524 	IB_PORT_ARMED		= 3,
525 	IB_PORT_ACTIVE		= 4,
526 	IB_PORT_ACTIVE_DEFER	= 5
527 };
528 
529 static inline const char *__attribute_const__
530 ib_port_state_to_str(enum ib_port_state state)
531 {
532 	const char * const states[] = {
533 		[IB_PORT_NOP] = "NOP",
534 		[IB_PORT_DOWN] = "DOWN",
535 		[IB_PORT_INIT] = "INIT",
536 		[IB_PORT_ARMED] = "ARMED",
537 		[IB_PORT_ACTIVE] = "ACTIVE",
538 		[IB_PORT_ACTIVE_DEFER] = "ACTIVE_DEFER",
539 	};
540 
541 	if (state < ARRAY_SIZE(states))
542 		return states[state];
543 	return "UNKNOWN";
544 }
545 
546 enum ib_port_phys_state {
547 	IB_PORT_PHYS_STATE_SLEEP = 1,
548 	IB_PORT_PHYS_STATE_POLLING = 2,
549 	IB_PORT_PHYS_STATE_DISABLED = 3,
550 	IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
551 	IB_PORT_PHYS_STATE_LINK_UP = 5,
552 	IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
553 	IB_PORT_PHYS_STATE_PHY_TEST = 7,
554 };
555 
556 enum ib_port_width {
557 	IB_WIDTH_1X	= 1,
558 	IB_WIDTH_2X	= 16,
559 	IB_WIDTH_4X	= 2,
560 	IB_WIDTH_8X	= 4,
561 	IB_WIDTH_12X	= 8
562 };
563 
564 static inline int ib_width_enum_to_int(enum ib_port_width width)
565 {
566 	switch (width) {
567 	case IB_WIDTH_1X:  return  1;
568 	case IB_WIDTH_2X:  return  2;
569 	case IB_WIDTH_4X:  return  4;
570 	case IB_WIDTH_8X:  return  8;
571 	case IB_WIDTH_12X: return 12;
572 	default: 	  return -1;
573 	}
574 }
575 
576 enum ib_port_speed {
577 	IB_SPEED_SDR	= 1,
578 	IB_SPEED_DDR	= 2,
579 	IB_SPEED_QDR	= 4,
580 	IB_SPEED_FDR10	= 8,
581 	IB_SPEED_FDR	= 16,
582 	IB_SPEED_EDR	= 32,
583 	IB_SPEED_HDR	= 64,
584 	IB_SPEED_NDR	= 128,
585 	IB_SPEED_XDR	= 256,
586 };
587 
588 enum ib_stat_flag {
589 	IB_STAT_FLAG_OPTIONAL = 1 << 0,
590 };
591 
592 /**
593  * struct rdma_stat_desc - description of one rdma stat/counter
594  * @name: The name of the counter
595  * @flags: Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL
596  * @priv: Driver private information; Core code should not use
597  */
598 struct rdma_stat_desc {
599 	const char *name;
600 	unsigned int flags;
601 	const void *priv;
602 };
603 
604 /**
605  * struct rdma_hw_stats - collection of hardware stats and their management
606  * @lock: Mutex to protect parallel write access to lifespan and values
607  *    of counters, which are 64bits and not guaranteed to be written
608  *    atomicaly on 32bits systems.
609  * @timestamp: Used by the core code to track when the last update was
610  * @lifespan: Used by the core code to determine how old the counters
611  *   should be before being updated again.  Stored in jiffies, defaults
612  *   to 10 milliseconds, drivers can override the default be specifying
613  *   their own value during their allocation routine.
614  * @descs: Array of pointers to static descriptors used for the counters
615  *   in directory.
616  * @is_disabled: A bitmap to indicate each counter is currently disabled
617  *   or not.
618  * @num_counters: How many hardware counters there are.  If name is
619  *   shorter than this number, a kernel oops will result.  Driver authors
620  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
621  *   in their code to prevent this.
622  * @value: Array of u64 counters that are accessed by the sysfs code and
623  *   filled in by the drivers get_stats routine
624  */
625 struct rdma_hw_stats {
626 	struct mutex	lock; /* Protect lifespan and values[] */
627 	unsigned long	timestamp;
628 	unsigned long	lifespan;
629 	const struct rdma_stat_desc *descs;
630 	unsigned long	*is_disabled;
631 	int		num_counters;
632 	u64		value[] __counted_by(num_counters);
633 };
634 
635 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
636 
637 struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
638 	const struct rdma_stat_desc *descs, int num_counters,
639 	unsigned long lifespan);
640 
641 void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
642 
643 /* Define bits for the various functionality this port needs to be supported by
644  * the core.
645  */
646 /* Management                           0x00000FFF */
647 #define RDMA_CORE_CAP_IB_MAD            0x00000001
648 #define RDMA_CORE_CAP_IB_SMI            0x00000002
649 #define RDMA_CORE_CAP_IB_CM             0x00000004
650 #define RDMA_CORE_CAP_IW_CM             0x00000008
651 #define RDMA_CORE_CAP_IB_SA             0x00000010
652 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
653 
654 /* Address format                       0x000FF000 */
655 #define RDMA_CORE_CAP_AF_IB             0x00001000
656 #define RDMA_CORE_CAP_ETH_AH            0x00002000
657 #define RDMA_CORE_CAP_OPA_AH            0x00004000
658 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
659 
660 /* Protocol                             0xFFF00000 */
661 #define RDMA_CORE_CAP_PROT_IB           0x00100000
662 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
663 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
664 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
665 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
666 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
667 
668 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
669 					| RDMA_CORE_CAP_PROT_ROCE     \
670 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
671 
672 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
673 					| RDMA_CORE_CAP_IB_MAD \
674 					| RDMA_CORE_CAP_IB_SMI \
675 					| RDMA_CORE_CAP_IB_CM  \
676 					| RDMA_CORE_CAP_IB_SA  \
677 					| RDMA_CORE_CAP_AF_IB)
678 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
679 					| RDMA_CORE_CAP_IB_MAD  \
680 					| RDMA_CORE_CAP_IB_CM   \
681 					| RDMA_CORE_CAP_AF_IB   \
682 					| RDMA_CORE_CAP_ETH_AH)
683 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
684 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
685 					| RDMA_CORE_CAP_IB_MAD  \
686 					| RDMA_CORE_CAP_IB_CM   \
687 					| RDMA_CORE_CAP_AF_IB   \
688 					| RDMA_CORE_CAP_ETH_AH)
689 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
690 					| RDMA_CORE_CAP_IW_CM)
691 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
692 					| RDMA_CORE_CAP_OPA_MAD)
693 
694 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
695 
696 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
697 
698 struct ib_port_attr {
699 	u64			subnet_prefix;
700 	enum ib_port_state	state;
701 	enum ib_mtu		max_mtu;
702 	enum ib_mtu		active_mtu;
703 	u32                     phys_mtu;
704 	int			gid_tbl_len;
705 	unsigned int		ip_gids:1;
706 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
707 	u32			port_cap_flags;
708 	u32			max_msg_sz;
709 	u32			bad_pkey_cntr;
710 	u32			qkey_viol_cntr;
711 	u16			pkey_tbl_len;
712 	u32			sm_lid;
713 	u32			lid;
714 	u8			lmc;
715 	u8			max_vl_num;
716 	u8			sm_sl;
717 	u8			subnet_timeout;
718 	u8			init_type_reply;
719 	u8			active_width;
720 	u16			active_speed;
721 	u8                      phys_state;
722 	u16			port_cap_flags2;
723 };
724 
725 enum ib_device_modify_flags {
726 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
727 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
728 };
729 
730 #define IB_DEVICE_NODE_DESC_MAX 64
731 
732 struct ib_device_modify {
733 	u64	sys_image_guid;
734 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
735 };
736 
737 enum ib_port_modify_flags {
738 	IB_PORT_SHUTDOWN		= 1,
739 	IB_PORT_INIT_TYPE		= (1<<2),
740 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
741 	IB_PORT_OPA_MASK_CHG		= (1<<4)
742 };
743 
744 struct ib_port_modify {
745 	u32	set_port_cap_mask;
746 	u32	clr_port_cap_mask;
747 	u8	init_type;
748 };
749 
750 enum ib_event_type {
751 	IB_EVENT_CQ_ERR,
752 	IB_EVENT_QP_FATAL,
753 	IB_EVENT_QP_REQ_ERR,
754 	IB_EVENT_QP_ACCESS_ERR,
755 	IB_EVENT_COMM_EST,
756 	IB_EVENT_SQ_DRAINED,
757 	IB_EVENT_PATH_MIG,
758 	IB_EVENT_PATH_MIG_ERR,
759 	IB_EVENT_DEVICE_FATAL,
760 	IB_EVENT_PORT_ACTIVE,
761 	IB_EVENT_PORT_ERR,
762 	IB_EVENT_LID_CHANGE,
763 	IB_EVENT_PKEY_CHANGE,
764 	IB_EVENT_SM_CHANGE,
765 	IB_EVENT_SRQ_ERR,
766 	IB_EVENT_SRQ_LIMIT_REACHED,
767 	IB_EVENT_QP_LAST_WQE_REACHED,
768 	IB_EVENT_CLIENT_REREGISTER,
769 	IB_EVENT_GID_CHANGE,
770 	IB_EVENT_WQ_FATAL,
771 	IB_EVENT_DEVICE_SPEED_CHANGE,
772 };
773 
774 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
775 
776 struct ib_event {
777 	struct ib_device	*device;
778 	union {
779 		struct ib_cq	*cq;
780 		struct ib_qp	*qp;
781 		struct ib_srq	*srq;
782 		struct ib_wq	*wq;
783 		u32		port_num;
784 	} element;
785 	enum ib_event_type	event;
786 };
787 
788 struct ib_event_handler {
789 	struct ib_device *device;
790 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
791 	struct list_head  list;
792 };
793 
794 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
795 	do {							\
796 		(_ptr)->device  = _device;			\
797 		(_ptr)->handler = _handler;			\
798 		INIT_LIST_HEAD(&(_ptr)->list);			\
799 	} while (0)
800 
801 struct ib_global_route {
802 	const struct ib_gid_attr *sgid_attr;
803 	union ib_gid	dgid;
804 	u32		flow_label;
805 	u8		sgid_index;
806 	u8		hop_limit;
807 	u8		traffic_class;
808 };
809 
810 struct ib_grh {
811 	__be32		version_tclass_flow;
812 	__be16		paylen;
813 	u8		next_hdr;
814 	u8		hop_limit;
815 	union ib_gid	sgid;
816 	union ib_gid	dgid;
817 };
818 
819 union rdma_network_hdr {
820 	struct ib_grh ibgrh;
821 	struct {
822 		/* The IB spec states that if it's IPv4, the header
823 		 * is located in the last 20 bytes of the header.
824 		 */
825 		u8		reserved[20];
826 		struct iphdr	roce4grh;
827 	};
828 };
829 
830 #define IB_QPN_MASK		0xFFFFFF
831 
832 enum {
833 	IB_MULTICAST_QPN = 0xffffff
834 };
835 
836 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
837 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
838 
839 enum ib_ah_flags {
840 	IB_AH_GRH	= 1
841 };
842 
843 enum ib_rate {
844 	IB_RATE_PORT_CURRENT = 0,
845 	IB_RATE_2_5_GBPS = 2,
846 	IB_RATE_5_GBPS   = 5,
847 	IB_RATE_10_GBPS  = 3,
848 	IB_RATE_20_GBPS  = 6,
849 	IB_RATE_30_GBPS  = 4,
850 	IB_RATE_40_GBPS  = 7,
851 	IB_RATE_60_GBPS  = 8,
852 	IB_RATE_80_GBPS  = 9,
853 	IB_RATE_120_GBPS = 10,
854 	IB_RATE_14_GBPS  = 11,
855 	IB_RATE_56_GBPS  = 12,
856 	IB_RATE_112_GBPS = 13,
857 	IB_RATE_168_GBPS = 14,
858 	IB_RATE_25_GBPS  = 15,
859 	IB_RATE_100_GBPS = 16,
860 	IB_RATE_200_GBPS = 17,
861 	IB_RATE_300_GBPS = 18,
862 	IB_RATE_28_GBPS  = 19,
863 	IB_RATE_50_GBPS  = 20,
864 	IB_RATE_400_GBPS = 21,
865 	IB_RATE_600_GBPS = 22,
866 	IB_RATE_800_GBPS = 23,
867 	IB_RATE_1600_GBPS = 25,
868 };
869 
870 /**
871  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
872  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
873  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
874  * @rate: rate to convert.
875  */
876 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
877 
878 /**
879  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
880  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
881  * @rate: rate to convert.
882  */
883 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
884 
885 struct ib_port_speed_info {
886 	const char *str;
887 	int rate;	/* in deci-Gb/sec (100 MBps units) */
888 };
889 
890 /**
891  * ib_port_attr_to_speed_info - Convert port attributes to speed information
892  * @attr: Port attributes containing active_speed and active_width
893  * @speed_info: Speed information to return
894  *
895  * Returns 0 on success, -EINVAL on error.
896  */
897 int ib_port_attr_to_speed_info(struct ib_port_attr *attr,
898 			       struct ib_port_speed_info *speed_info);
899 
900 /**
901  * enum ib_mr_type - memory region type
902  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
903  *                            normal registration
904  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
905  *                            register any arbitrary sg lists (without
906  *                            the normal mr constraints - see
907  *                            ib_map_mr_sg)
908  * @IB_MR_TYPE_DM:            memory region that is used for device
909  *                            memory registration
910  * @IB_MR_TYPE_USER:          memory region that is used for the user-space
911  *                            application
912  * @IB_MR_TYPE_DMA:           memory region that is used for DMA operations
913  *                            without address translations (VA=PA)
914  * @IB_MR_TYPE_INTEGRITY:     memory region that is used for
915  *                            data integrity operations
916  */
917 enum ib_mr_type {
918 	IB_MR_TYPE_MEM_REG,
919 	IB_MR_TYPE_SG_GAPS,
920 	IB_MR_TYPE_DM,
921 	IB_MR_TYPE_USER,
922 	IB_MR_TYPE_DMA,
923 	IB_MR_TYPE_INTEGRITY,
924 };
925 
926 enum ib_mr_status_check {
927 	IB_MR_CHECK_SIG_STATUS = 1,
928 };
929 
930 /**
931  * struct ib_mr_status - Memory region status container
932  *
933  * @fail_status: Bitmask of MR checks status. For each
934  *     failed check a corresponding status bit is set.
935  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
936  *     failure.
937  */
938 struct ib_mr_status {
939 	u32		    fail_status;
940 	struct ib_sig_err   sig_err;
941 };
942 
943 /**
944  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
945  * enum.
946  * @mult: multiple to convert.
947  */
948 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
949 
950 struct rdma_ah_init_attr {
951 	struct rdma_ah_attr *ah_attr;
952 	u32 flags;
953 	struct net_device *xmit_slave;
954 };
955 
956 enum rdma_ah_attr_type {
957 	RDMA_AH_ATTR_TYPE_UNDEFINED,
958 	RDMA_AH_ATTR_TYPE_IB,
959 	RDMA_AH_ATTR_TYPE_ROCE,
960 	RDMA_AH_ATTR_TYPE_OPA,
961 };
962 
963 struct ib_ah_attr {
964 	u16			dlid;
965 	u8			src_path_bits;
966 };
967 
968 struct roce_ah_attr {
969 	u8			dmac[ETH_ALEN];
970 };
971 
972 struct opa_ah_attr {
973 	u32			dlid;
974 	u8			src_path_bits;
975 	bool			make_grd;
976 };
977 
978 struct rdma_ah_attr {
979 	struct ib_global_route	grh;
980 	u8			sl;
981 	u8			static_rate;
982 	u32			port_num;
983 	u8			ah_flags;
984 	enum rdma_ah_attr_type type;
985 	union {
986 		struct ib_ah_attr ib;
987 		struct roce_ah_attr roce;
988 		struct opa_ah_attr opa;
989 	};
990 };
991 
992 enum ib_wc_status {
993 	IB_WC_SUCCESS,
994 	IB_WC_LOC_LEN_ERR,
995 	IB_WC_LOC_QP_OP_ERR,
996 	IB_WC_LOC_EEC_OP_ERR,
997 	IB_WC_LOC_PROT_ERR,
998 	IB_WC_WR_FLUSH_ERR,
999 	IB_WC_MW_BIND_ERR,
1000 	IB_WC_BAD_RESP_ERR,
1001 	IB_WC_LOC_ACCESS_ERR,
1002 	IB_WC_REM_INV_REQ_ERR,
1003 	IB_WC_REM_ACCESS_ERR,
1004 	IB_WC_REM_OP_ERR,
1005 	IB_WC_RETRY_EXC_ERR,
1006 	IB_WC_RNR_RETRY_EXC_ERR,
1007 	IB_WC_LOC_RDD_VIOL_ERR,
1008 	IB_WC_REM_INV_RD_REQ_ERR,
1009 	IB_WC_REM_ABORT_ERR,
1010 	IB_WC_INV_EECN_ERR,
1011 	IB_WC_INV_EEC_STATE_ERR,
1012 	IB_WC_FATAL_ERR,
1013 	IB_WC_RESP_TIMEOUT_ERR,
1014 	IB_WC_GENERAL_ERR
1015 };
1016 
1017 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
1018 
1019 enum ib_wc_opcode {
1020 	IB_WC_SEND = IB_UVERBS_WC_SEND,
1021 	IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
1022 	IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
1023 	IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
1024 	IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
1025 	IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
1026 	IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
1027 	IB_WC_LSO = IB_UVERBS_WC_TSO,
1028 	IB_WC_ATOMIC_WRITE = IB_UVERBS_WC_ATOMIC_WRITE,
1029 	IB_WC_REG_MR,
1030 	IB_WC_MASKED_COMP_SWAP,
1031 	IB_WC_MASKED_FETCH_ADD,
1032 	IB_WC_FLUSH = IB_UVERBS_WC_FLUSH,
1033 /*
1034  * Set value of IB_WC_RECV so consumers can test if a completion is a
1035  * receive by testing (opcode & IB_WC_RECV).
1036  */
1037 	IB_WC_RECV			= 1 << 7,
1038 	IB_WC_RECV_RDMA_WITH_IMM
1039 };
1040 
1041 enum ib_wc_flags {
1042 	IB_WC_GRH		= 1,
1043 	IB_WC_WITH_IMM		= (1<<1),
1044 	IB_WC_WITH_INVALIDATE	= (1<<2),
1045 	IB_WC_IP_CSUM_OK	= (1<<3),
1046 	IB_WC_WITH_SMAC		= (1<<4),
1047 	IB_WC_WITH_VLAN		= (1<<5),
1048 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
1049 };
1050 
1051 struct ib_wc {
1052 	union {
1053 		u64		wr_id;
1054 		struct ib_cqe	*wr_cqe;
1055 	};
1056 	enum ib_wc_status	status;
1057 	enum ib_wc_opcode	opcode;
1058 	u32			vendor_err;
1059 	u32			byte_len;
1060 	struct ib_qp	       *qp;
1061 	union {
1062 		__be32		imm_data;
1063 		u32		invalidate_rkey;
1064 	} ex;
1065 	u32			src_qp;
1066 	u32			slid;
1067 	int			wc_flags;
1068 	u16			pkey_index;
1069 	u8			sl;
1070 	u8			dlid_path_bits;
1071 	u32 port_num; /* valid only for DR SMPs on switches */
1072 	u8			smac[ETH_ALEN];
1073 	u16			vlan_id;
1074 	u8			network_hdr_type;
1075 };
1076 
1077 enum ib_cq_notify_flags {
1078 	IB_CQ_SOLICITED			= 1 << 0,
1079 	IB_CQ_NEXT_COMP			= 1 << 1,
1080 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1081 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
1082 };
1083 
1084 enum ib_srq_type {
1085 	IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1086 	IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1087 	IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1088 };
1089 
1090 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1091 {
1092 	return srq_type == IB_SRQT_XRC ||
1093 	       srq_type == IB_SRQT_TM;
1094 }
1095 
1096 enum ib_srq_attr_mask {
1097 	IB_SRQ_MAX_WR	= 1 << 0,
1098 	IB_SRQ_LIMIT	= 1 << 1,
1099 };
1100 
1101 struct ib_srq_attr {
1102 	u32	max_wr;
1103 	u32	max_sge;
1104 	u32	srq_limit;
1105 };
1106 
1107 struct ib_srq_init_attr {
1108 	void		      (*event_handler)(struct ib_event *, void *);
1109 	void		       *srq_context;
1110 	struct ib_srq_attr	attr;
1111 	enum ib_srq_type	srq_type;
1112 
1113 	struct {
1114 		struct ib_cq   *cq;
1115 		union {
1116 			struct {
1117 				struct ib_xrcd *xrcd;
1118 			} xrc;
1119 
1120 			struct {
1121 				u32		max_num_tags;
1122 			} tag_matching;
1123 		};
1124 	} ext;
1125 };
1126 
1127 struct ib_qp_cap {
1128 	u32	max_send_wr;
1129 	u32	max_recv_wr;
1130 	u32	max_send_sge;
1131 	u32	max_recv_sge;
1132 	u32	max_inline_data;
1133 
1134 	/*
1135 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1136 	 * ib_create_qp() will calculate the right amount of needed WRs
1137 	 * and MRs based on this.
1138 	 */
1139 	u32	max_rdma_ctxs;
1140 };
1141 
1142 enum ib_sig_type {
1143 	IB_SIGNAL_ALL_WR,
1144 	IB_SIGNAL_REQ_WR
1145 };
1146 
1147 enum ib_qp_type {
1148 	/*
1149 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1150 	 * here (and in that order) since the MAD layer uses them as
1151 	 * indices into a 2-entry table.
1152 	 */
1153 	IB_QPT_SMI,
1154 	IB_QPT_GSI,
1155 
1156 	IB_QPT_RC = IB_UVERBS_QPT_RC,
1157 	IB_QPT_UC = IB_UVERBS_QPT_UC,
1158 	IB_QPT_UD = IB_UVERBS_QPT_UD,
1159 	IB_QPT_RAW_IPV6,
1160 	IB_QPT_RAW_ETHERTYPE,
1161 	IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1162 	IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1163 	IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1164 	IB_QPT_MAX,
1165 	IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1166 	/* Reserve a range for qp types internal to the low level driver.
1167 	 * These qp types will not be visible at the IB core layer, so the
1168 	 * IB_QPT_MAX usages should not be affected in the core layer
1169 	 */
1170 	IB_QPT_RESERVED1 = 0x1000,
1171 	IB_QPT_RESERVED2,
1172 	IB_QPT_RESERVED3,
1173 	IB_QPT_RESERVED4,
1174 	IB_QPT_RESERVED5,
1175 	IB_QPT_RESERVED6,
1176 	IB_QPT_RESERVED7,
1177 	IB_QPT_RESERVED8,
1178 	IB_QPT_RESERVED9,
1179 	IB_QPT_RESERVED10,
1180 };
1181 
1182 enum ib_qp_create_flags {
1183 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1184 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	=
1185 		IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1186 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1187 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1188 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1189 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1190 	IB_QP_CREATE_INTEGRITY_EN		= 1 << 6,
1191 	IB_QP_CREATE_NETDEV_USE			= 1 << 7,
1192 	IB_QP_CREATE_SCATTER_FCS		=
1193 		IB_UVERBS_QP_CREATE_SCATTER_FCS,
1194 	IB_QP_CREATE_CVLAN_STRIPPING		=
1195 		IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1196 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1197 	IB_QP_CREATE_PCI_WRITE_END_PADDING	=
1198 		IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1199 	/* reserve bits 26-31 for low level drivers' internal use */
1200 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1201 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1202 };
1203 
1204 /*
1205  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1206  * callback to destroy the passed in QP.
1207  */
1208 
1209 struct ib_qp_init_attr {
1210 	/* This callback occurs in workqueue context */
1211 	void                  (*event_handler)(struct ib_event *, void *);
1212 
1213 	void		       *qp_context;
1214 	struct ib_cq	       *send_cq;
1215 	struct ib_cq	       *recv_cq;
1216 	struct ib_srq	       *srq;
1217 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1218 	struct ib_qp_cap	cap;
1219 	enum ib_sig_type	sq_sig_type;
1220 	enum ib_qp_type		qp_type;
1221 	u32			create_flags;
1222 
1223 	/*
1224 	 * Only needed for special QP types, or when using the RW API.
1225 	 */
1226 	u32			port_num;
1227 	struct ib_rwq_ind_table *rwq_ind_tbl;
1228 	u32			source_qpn;
1229 };
1230 
1231 struct ib_qp_open_attr {
1232 	void                  (*event_handler)(struct ib_event *, void *);
1233 	void		       *qp_context;
1234 	u32			qp_num;
1235 	enum ib_qp_type		qp_type;
1236 };
1237 
1238 enum ib_rnr_timeout {
1239 	IB_RNR_TIMER_655_36 =  0,
1240 	IB_RNR_TIMER_000_01 =  1,
1241 	IB_RNR_TIMER_000_02 =  2,
1242 	IB_RNR_TIMER_000_03 =  3,
1243 	IB_RNR_TIMER_000_04 =  4,
1244 	IB_RNR_TIMER_000_06 =  5,
1245 	IB_RNR_TIMER_000_08 =  6,
1246 	IB_RNR_TIMER_000_12 =  7,
1247 	IB_RNR_TIMER_000_16 =  8,
1248 	IB_RNR_TIMER_000_24 =  9,
1249 	IB_RNR_TIMER_000_32 = 10,
1250 	IB_RNR_TIMER_000_48 = 11,
1251 	IB_RNR_TIMER_000_64 = 12,
1252 	IB_RNR_TIMER_000_96 = 13,
1253 	IB_RNR_TIMER_001_28 = 14,
1254 	IB_RNR_TIMER_001_92 = 15,
1255 	IB_RNR_TIMER_002_56 = 16,
1256 	IB_RNR_TIMER_003_84 = 17,
1257 	IB_RNR_TIMER_005_12 = 18,
1258 	IB_RNR_TIMER_007_68 = 19,
1259 	IB_RNR_TIMER_010_24 = 20,
1260 	IB_RNR_TIMER_015_36 = 21,
1261 	IB_RNR_TIMER_020_48 = 22,
1262 	IB_RNR_TIMER_030_72 = 23,
1263 	IB_RNR_TIMER_040_96 = 24,
1264 	IB_RNR_TIMER_061_44 = 25,
1265 	IB_RNR_TIMER_081_92 = 26,
1266 	IB_RNR_TIMER_122_88 = 27,
1267 	IB_RNR_TIMER_163_84 = 28,
1268 	IB_RNR_TIMER_245_76 = 29,
1269 	IB_RNR_TIMER_327_68 = 30,
1270 	IB_RNR_TIMER_491_52 = 31
1271 };
1272 
1273 enum ib_qp_attr_mask {
1274 	IB_QP_STATE			= 1,
1275 	IB_QP_CUR_STATE			= (1<<1),
1276 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1277 	IB_QP_ACCESS_FLAGS		= (1<<3),
1278 	IB_QP_PKEY_INDEX		= (1<<4),
1279 	IB_QP_PORT			= (1<<5),
1280 	IB_QP_QKEY			= (1<<6),
1281 	IB_QP_AV			= (1<<7),
1282 	IB_QP_PATH_MTU			= (1<<8),
1283 	IB_QP_TIMEOUT			= (1<<9),
1284 	IB_QP_RETRY_CNT			= (1<<10),
1285 	IB_QP_RNR_RETRY			= (1<<11),
1286 	IB_QP_RQ_PSN			= (1<<12),
1287 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1288 	IB_QP_ALT_PATH			= (1<<14),
1289 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1290 	IB_QP_SQ_PSN			= (1<<16),
1291 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1292 	IB_QP_PATH_MIG_STATE		= (1<<18),
1293 	IB_QP_CAP			= (1<<19),
1294 	IB_QP_DEST_QPN			= (1<<20),
1295 	IB_QP_RESERVED1			= (1<<21),
1296 	IB_QP_RESERVED2			= (1<<22),
1297 	IB_QP_RESERVED3			= (1<<23),
1298 	IB_QP_RESERVED4			= (1<<24),
1299 	IB_QP_RATE_LIMIT		= (1<<25),
1300 
1301 	IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1302 };
1303 
1304 enum ib_qp_state {
1305 	IB_QPS_RESET,
1306 	IB_QPS_INIT,
1307 	IB_QPS_RTR,
1308 	IB_QPS_RTS,
1309 	IB_QPS_SQD,
1310 	IB_QPS_SQE,
1311 	IB_QPS_ERR
1312 };
1313 
1314 enum ib_mig_state {
1315 	IB_MIG_MIGRATED,
1316 	IB_MIG_REARM,
1317 	IB_MIG_ARMED
1318 };
1319 
1320 enum ib_mw_type {
1321 	IB_MW_TYPE_1 = 1,
1322 	IB_MW_TYPE_2 = 2
1323 };
1324 
1325 struct ib_qp_attr {
1326 	enum ib_qp_state	qp_state;
1327 	enum ib_qp_state	cur_qp_state;
1328 	enum ib_mtu		path_mtu;
1329 	enum ib_mig_state	path_mig_state;
1330 	u32			qkey;
1331 	u32			rq_psn;
1332 	u32			sq_psn;
1333 	u32			dest_qp_num;
1334 	int			qp_access_flags;
1335 	struct ib_qp_cap	cap;
1336 	struct rdma_ah_attr	ah_attr;
1337 	struct rdma_ah_attr	alt_ah_attr;
1338 	u16			pkey_index;
1339 	u16			alt_pkey_index;
1340 	u8			en_sqd_async_notify;
1341 	u8			sq_draining;
1342 	u8			max_rd_atomic;
1343 	u8			max_dest_rd_atomic;
1344 	u8			min_rnr_timer;
1345 	u32			port_num;
1346 	u8			timeout;
1347 	u8			retry_cnt;
1348 	u8			rnr_retry;
1349 	u32			alt_port_num;
1350 	u8			alt_timeout;
1351 	u32			rate_limit;
1352 	struct net_device	*xmit_slave;
1353 };
1354 
1355 enum ib_wr_opcode {
1356 	/* These are shared with userspace */
1357 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1358 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1359 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1360 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1361 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1362 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1363 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1364 	IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1365 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1366 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1367 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1368 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1369 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1370 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1371 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1372 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1373 	IB_WR_FLUSH = IB_UVERBS_WR_FLUSH,
1374 	IB_WR_ATOMIC_WRITE = IB_UVERBS_WR_ATOMIC_WRITE,
1375 
1376 	/* These are kernel only and can not be issued by userspace */
1377 	IB_WR_REG_MR = 0x20,
1378 	IB_WR_REG_MR_INTEGRITY,
1379 
1380 	/* reserve values for low level drivers' internal use.
1381 	 * These values will not be used at all in the ib core layer.
1382 	 */
1383 	IB_WR_RESERVED1 = 0xf0,
1384 	IB_WR_RESERVED2,
1385 	IB_WR_RESERVED3,
1386 	IB_WR_RESERVED4,
1387 	IB_WR_RESERVED5,
1388 	IB_WR_RESERVED6,
1389 	IB_WR_RESERVED7,
1390 	IB_WR_RESERVED8,
1391 	IB_WR_RESERVED9,
1392 	IB_WR_RESERVED10,
1393 };
1394 
1395 enum ib_send_flags {
1396 	IB_SEND_FENCE		= 1,
1397 	IB_SEND_SIGNALED	= (1<<1),
1398 	IB_SEND_SOLICITED	= (1<<2),
1399 	IB_SEND_INLINE		= (1<<3),
1400 	IB_SEND_IP_CSUM		= (1<<4),
1401 
1402 	/* reserve bits 26-31 for low level drivers' internal use */
1403 	IB_SEND_RESERVED_START	= (1 << 26),
1404 	IB_SEND_RESERVED_END	= (1 << 31),
1405 };
1406 
1407 struct ib_sge {
1408 	u64	addr;
1409 	u32	length;
1410 	u32	lkey;
1411 };
1412 
1413 struct ib_cqe {
1414 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1415 };
1416 
1417 struct ib_send_wr {
1418 	struct ib_send_wr      *next;
1419 	union {
1420 		u64		wr_id;
1421 		struct ib_cqe	*wr_cqe;
1422 	};
1423 	struct ib_sge	       *sg_list;
1424 	int			num_sge;
1425 	enum ib_wr_opcode	opcode;
1426 	int			send_flags;
1427 	union {
1428 		__be32		imm_data;
1429 		u32		invalidate_rkey;
1430 	} ex;
1431 };
1432 
1433 struct ib_rdma_wr {
1434 	struct ib_send_wr	wr;
1435 	u64			remote_addr;
1436 	u32			rkey;
1437 };
1438 
1439 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1440 {
1441 	return container_of(wr, struct ib_rdma_wr, wr);
1442 }
1443 
1444 struct ib_atomic_wr {
1445 	struct ib_send_wr	wr;
1446 	u64			remote_addr;
1447 	u64			compare_add;
1448 	u64			swap;
1449 	u64			compare_add_mask;
1450 	u64			swap_mask;
1451 	u32			rkey;
1452 };
1453 
1454 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1455 {
1456 	return container_of(wr, struct ib_atomic_wr, wr);
1457 }
1458 
1459 struct ib_ud_wr {
1460 	struct ib_send_wr	wr;
1461 	struct ib_ah		*ah;
1462 	void			*header;
1463 	int			hlen;
1464 	int			mss;
1465 	u32			remote_qpn;
1466 	u32			remote_qkey;
1467 	u16			pkey_index; /* valid for GSI only */
1468 	u32			port_num; /* valid for DR SMPs on switch only */
1469 };
1470 
1471 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1472 {
1473 	return container_of(wr, struct ib_ud_wr, wr);
1474 }
1475 
1476 struct ib_reg_wr {
1477 	struct ib_send_wr	wr;
1478 	struct ib_mr		*mr;
1479 	u32			key;
1480 	int			access;
1481 };
1482 
1483 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1484 {
1485 	return container_of(wr, struct ib_reg_wr, wr);
1486 }
1487 
1488 struct ib_recv_wr {
1489 	struct ib_recv_wr      *next;
1490 	union {
1491 		u64		wr_id;
1492 		struct ib_cqe	*wr_cqe;
1493 	};
1494 	struct ib_sge	       *sg_list;
1495 	int			num_sge;
1496 };
1497 
1498 enum ib_access_flags {
1499 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1500 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1501 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1502 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1503 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1504 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1505 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1506 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1507 	IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1508 	IB_ACCESS_FLUSH_GLOBAL = IB_UVERBS_ACCESS_FLUSH_GLOBAL,
1509 	IB_ACCESS_FLUSH_PERSISTENT = IB_UVERBS_ACCESS_FLUSH_PERSISTENT,
1510 
1511 	IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1512 	IB_ACCESS_SUPPORTED =
1513 		((IB_ACCESS_FLUSH_PERSISTENT << 1) - 1) | IB_ACCESS_OPTIONAL,
1514 };
1515 
1516 /*
1517  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1518  * are hidden here instead of a uapi header!
1519  */
1520 enum ib_mr_rereg_flags {
1521 	IB_MR_REREG_TRANS	= 1,
1522 	IB_MR_REREG_PD		= (1<<1),
1523 	IB_MR_REREG_ACCESS	= (1<<2),
1524 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1525 };
1526 
1527 struct ib_umem;
1528 
1529 enum rdma_remove_reason {
1530 	/*
1531 	 * Userspace requested uobject deletion or initial try
1532 	 * to remove uobject via cleanup. Call could fail
1533 	 */
1534 	RDMA_REMOVE_DESTROY,
1535 	/* Context deletion. This call should delete the actual object itself */
1536 	RDMA_REMOVE_CLOSE,
1537 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1538 	RDMA_REMOVE_DRIVER_REMOVE,
1539 	/* uobj is being cleaned-up before being committed */
1540 	RDMA_REMOVE_ABORT,
1541 	/* The driver failed to destroy the uobject and is being disconnected */
1542 	RDMA_REMOVE_DRIVER_FAILURE,
1543 };
1544 
1545 struct ib_rdmacg_object {
1546 #ifdef CONFIG_CGROUP_RDMA
1547 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1548 #endif
1549 };
1550 
1551 struct ib_ucontext {
1552 	struct ib_device       *device;
1553 	struct ib_uverbs_file  *ufile;
1554 
1555 	struct ib_rdmacg_object	cg_obj;
1556 	u64 enabled_caps;
1557 	/*
1558 	 * Implementation details of the RDMA core, don't use in drivers:
1559 	 */
1560 	struct rdma_restrack_entry res;
1561 	struct xarray mmap_xa;
1562 };
1563 
1564 struct ib_uobject {
1565 	u64			user_handle;	/* handle given to us by userspace */
1566 	/* ufile & ucontext owning this object */
1567 	struct ib_uverbs_file  *ufile;
1568 	/* FIXME, save memory: ufile->context == context */
1569 	struct ib_ucontext     *context;	/* associated user context */
1570 	void		       *object;		/* containing object */
1571 	struct list_head	list;		/* link to context's list */
1572 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1573 	int			id;		/* index into kernel idr */
1574 	struct kref		ref;
1575 	atomic_t		usecnt;		/* protects exclusive access */
1576 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1577 
1578 	const struct uverbs_api_object *uapi_object;
1579 };
1580 
1581 /**
1582  * struct ib_udata - Driver request/response data from userspace
1583  * @inbuf: Pointer to request data from userspace
1584  * @outbuf: Pointer to response buffer in userspace
1585  * @inlen: Length of request data
1586  * @outlen: Length of response buffer
1587  *
1588  * struct ib_udata is used to hold the driver data request and response
1589  * structures defined in the uapi. They follow these rules for forwards and
1590  * backwards compatibility:
1591  *
1592  * 1) Userspace can provide a longer request so long as the trailing part the
1593  *    kernel doesn't understand is all zeros.
1594  *
1595  *    This provides a degree of safety if userspace wrongly tries to use a new
1596  *    feature the kernel does not understand with some non-zero value.
1597  *
1598  *    It allows a simpler rdma-core implementation because the library can
1599  *    simply always use the latest structs for the request, even if they are
1600  *    bigger. It simply has to avoid using the new members if they are not
1601  *    supported/required.
1602  *
1603  * 2) Userspace can provide a shorter request; the kernel will zero-pad it out
1604  *    to fill the storage. The newer kernel should understand that older
1605  *    userspace will provide 0 to new fields. The kernel has three options to
1606  *    enable new request fields:
1607  *
1608  *    - Input comp_mask that says the field is supported
1609  *    - Look for non-zero values
1610  *    - Check if the udata->inlen size covers the field
1611  *
1612  *    This also corrects any bugs related to not filling in request structures
1613  *    as the new helper always fully writes to the struct.
1614  *
1615  * 3) Userspace can provide a shorter or longer response struct. If shorter,
1616  *    the kernel reply is truncated. The kernel should be designed to not write
1617  *    to new reply fields unless userspace has affirmatively requested them.
1618  *
1619  *    If the user buffer is longer, the kernel will zero-fill it.
1620  *
1621  *    Userspace has three options to enable new response fields:
1622  *
1623  *    - Output comp_mask that says the field is supported
1624  *    - Look for non-zero values
1625  *    - Infer the output must be valid because the request contents demand it
1626  *      and old kernels will fail the request
1627  *
1628  * The following helper functions implement these semantics:
1629  *
1630  * ib_copy_validate_udata_in() - Checks the minimum length, and zero trailing::
1631  *
1632  *     struct driver_create_cq_req req;
1633  *     int err;
1634  *
1635  *     err = ib_copy_validate_udata_in(udata, req, end_member);
1636  *     if (err)
1637  *         return err;
1638  *
1639  * The third argument specifies the last member of the struct in the first
1640  * kernel version that introduced it, establishing the minimum required size.
1641  *
1642  * ib_copy_validate_udata_in_cm() - The above but also validate a
1643  * comp_mask member only has supported bits set::
1644  *
1645  *     err = ib_copy_validate_udata_in_cm(udata, req, first_version_last_member,
1646  *                                        DRIVER_CREATE_CQ_MASK_FEATURE_A |
1647  *                                        DRIVER_CREATE_CQ_MASK_FEATURE_B);
1648  *
1649  * ib_respond_udata() - Implements the response rules::
1650  *
1651  *     struct driver_create_cq_resp resp = {};
1652  *
1653  *     resp.some_field = value;
1654  *     return ib_respond_udata(udata, resp);
1655  *
1656  * ib_is_udata_in_empty() - Used instead of ib_copy_validate_udata_in() if the
1657  * driver does not have a request structure::
1658  *
1659  *     ret = ib_is_udata_in_empty(udata);
1660  *     if (ret)
1661  *         return ret;
1662  *
1663  * Similarly ib_respond_empty_udata() is used instead of ib_respond_udata() if
1664  * the driver does not have a response structure::
1665  *
1666  *    return ib_respond_empty_udata(udata);
1667  */
1668 struct ib_udata {
1669 	const void __user *inbuf;
1670 	void __user *outbuf;
1671 	size_t       inlen;
1672 	size_t       outlen;
1673 };
1674 
1675 struct ib_pd {
1676 	u32			local_dma_lkey;
1677 	u32			flags;
1678 	struct ib_device       *device;
1679 	struct ib_uobject      *uobject;
1680 	atomic_t          	usecnt; /* count all resources */
1681 
1682 	u32			unsafe_global_rkey;
1683 
1684 	/*
1685 	 * Implementation details of the RDMA core, don't use in drivers:
1686 	 */
1687 	struct ib_mr	       *__internal_mr;
1688 	struct rdma_restrack_entry res;
1689 };
1690 
1691 struct ib_xrcd {
1692 	struct ib_device       *device;
1693 	atomic_t		usecnt; /* count all exposed resources */
1694 	struct inode	       *inode;
1695 	struct rw_semaphore	tgt_qps_rwsem;
1696 	struct xarray		tgt_qps;
1697 };
1698 
1699 struct ib_ah {
1700 	struct ib_device	*device;
1701 	struct ib_pd		*pd;
1702 	struct ib_uobject	*uobject;
1703 	const struct ib_gid_attr *sgid_attr;
1704 	enum rdma_ah_attr_type	type;
1705 };
1706 
1707 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1708 
1709 enum ib_poll_context {
1710 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1711 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1712 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1713 	IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1714 
1715 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1716 };
1717 
1718 struct ib_cq {
1719 	struct ib_device       *device;
1720 	struct ib_ucq_object   *uobject;
1721 	ib_comp_handler   	comp_handler;
1722 	void                  (*event_handler)(struct ib_event *, void *);
1723 	void                   *cq_context;
1724 	int               	cqe;
1725 	unsigned int		cqe_used;
1726 	atomic_t          	usecnt; /* count number of work queues */
1727 	enum ib_poll_context	poll_ctx;
1728 	struct ib_wc		*wc;
1729 	struct list_head        pool_entry;
1730 	union {
1731 		struct irq_poll		iop;
1732 		struct work_struct	work;
1733 	};
1734 	struct workqueue_struct *comp_wq;
1735 	struct dim *dim;
1736 
1737 	/* updated only by trace points */
1738 	ktime_t timestamp;
1739 	u8 interrupt:1;
1740 	u8 shared:1;
1741 	unsigned int comp_vector;
1742 
1743 	/*
1744 	 * Implementation details of the RDMA core, don't use in drivers:
1745 	 */
1746 	struct rdma_restrack_entry res;
1747 };
1748 
1749 struct ib_srq {
1750 	struct ib_device       *device;
1751 	struct ib_pd	       *pd;
1752 	struct ib_usrq_object  *uobject;
1753 	void		      (*event_handler)(struct ib_event *, void *);
1754 	void		       *srq_context;
1755 	enum ib_srq_type	srq_type;
1756 	atomic_t		usecnt;
1757 
1758 	struct {
1759 		struct ib_cq   *cq;
1760 		union {
1761 			struct {
1762 				struct ib_xrcd *xrcd;
1763 				u32		srq_num;
1764 			} xrc;
1765 		};
1766 	} ext;
1767 
1768 	/*
1769 	 * Implementation details of the RDMA core, don't use in drivers:
1770 	 */
1771 	struct rdma_restrack_entry res;
1772 };
1773 
1774 enum ib_raw_packet_caps {
1775 	/*
1776 	 * Strip cvlan from incoming packet and report it in the matching work
1777 	 * completion is supported.
1778 	 */
1779 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING =
1780 		IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING,
1781 	/*
1782 	 * Scatter FCS field of an incoming packet to host memory is supported.
1783 	 */
1784 	IB_RAW_PACKET_CAP_SCATTER_FCS = IB_UVERBS_RAW_PACKET_CAP_SCATTER_FCS,
1785 	/* Checksum offloads are supported (for both send and receive). */
1786 	IB_RAW_PACKET_CAP_IP_CSUM = IB_UVERBS_RAW_PACKET_CAP_IP_CSUM,
1787 	/*
1788 	 * When a packet is received for an RQ with no receive WQEs, the
1789 	 * packet processing is delayed.
1790 	 */
1791 	IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP,
1792 };
1793 
1794 enum ib_wq_type {
1795 	IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1796 };
1797 
1798 enum ib_wq_state {
1799 	IB_WQS_RESET,
1800 	IB_WQS_RDY,
1801 	IB_WQS_ERR
1802 };
1803 
1804 struct ib_wq {
1805 	struct ib_device       *device;
1806 	struct ib_uwq_object   *uobject;
1807 	void		    *wq_context;
1808 	void		    (*event_handler)(struct ib_event *, void *);
1809 	struct ib_pd	       *pd;
1810 	struct ib_cq	       *cq;
1811 	u32		wq_num;
1812 	enum ib_wq_state       state;
1813 	enum ib_wq_type	wq_type;
1814 	atomic_t		usecnt;
1815 };
1816 
1817 enum ib_wq_flags {
1818 	IB_WQ_FLAGS_CVLAN_STRIPPING	= IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1819 	IB_WQ_FLAGS_SCATTER_FCS		= IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1820 	IB_WQ_FLAGS_DELAY_DROP		= IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1821 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1822 				IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1823 };
1824 
1825 struct ib_wq_init_attr {
1826 	void		       *wq_context;
1827 	enum ib_wq_type	wq_type;
1828 	u32		max_wr;
1829 	u32		max_sge;
1830 	struct	ib_cq	       *cq;
1831 	void		    (*event_handler)(struct ib_event *, void *);
1832 	u32		create_flags; /* Use enum ib_wq_flags */
1833 };
1834 
1835 enum ib_wq_attr_mask {
1836 	IB_WQ_STATE		= 1 << 0,
1837 	IB_WQ_CUR_STATE		= 1 << 1,
1838 	IB_WQ_FLAGS		= 1 << 2,
1839 };
1840 
1841 struct ib_wq_attr {
1842 	enum	ib_wq_state	wq_state;
1843 	enum	ib_wq_state	curr_wq_state;
1844 	u32			flags; /* Use enum ib_wq_flags */
1845 	u32			flags_mask; /* Use enum ib_wq_flags */
1846 };
1847 
1848 struct ib_rwq_ind_table {
1849 	struct ib_device	*device;
1850 	struct ib_uobject      *uobject;
1851 	atomic_t		usecnt;
1852 	u32		ind_tbl_num;
1853 	u32		log_ind_tbl_size;
1854 	struct ib_wq	**ind_tbl;
1855 };
1856 
1857 struct ib_rwq_ind_table_init_attr {
1858 	u32		log_ind_tbl_size;
1859 	/* Each entry is a pointer to Receive Work Queue */
1860 	struct ib_wq	**ind_tbl;
1861 };
1862 
1863 enum port_pkey_state {
1864 	IB_PORT_PKEY_NOT_VALID = 0,
1865 	IB_PORT_PKEY_VALID = 1,
1866 	IB_PORT_PKEY_LISTED = 2,
1867 };
1868 
1869 struct ib_qp_security;
1870 
1871 struct ib_port_pkey {
1872 	enum port_pkey_state	state;
1873 	u16			pkey_index;
1874 	u32			port_num;
1875 	struct list_head	qp_list;
1876 	struct list_head	to_error_list;
1877 	struct ib_qp_security  *sec;
1878 };
1879 
1880 struct ib_ports_pkeys {
1881 	struct ib_port_pkey	main;
1882 	struct ib_port_pkey	alt;
1883 };
1884 
1885 struct ib_qp_security {
1886 	struct ib_qp	       *qp;
1887 	struct ib_device       *dev;
1888 	/* Hold this mutex when changing port and pkey settings. */
1889 	struct mutex		mutex;
1890 	struct ib_ports_pkeys  *ports_pkeys;
1891 	/* A list of all open shared QP handles.  Required to enforce security
1892 	 * properly for all users of a shared QP.
1893 	 */
1894 	struct list_head        shared_qp_list;
1895 	void                   *security;
1896 	bool			destroying;
1897 	atomic_t		error_list_count;
1898 	struct completion	error_complete;
1899 	int			error_comps_pending;
1900 };
1901 
1902 /*
1903  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1904  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1905  */
1906 struct ib_qp {
1907 	struct ib_device       *device;
1908 	struct ib_pd	       *pd;
1909 	struct ib_cq	       *send_cq;
1910 	struct ib_cq	       *recv_cq;
1911 	spinlock_t		mr_lock;
1912 	int			mrs_used;
1913 	struct list_head	rdma_mrs;
1914 	struct list_head	sig_mrs;
1915 	struct ib_srq	       *srq;
1916 	struct completion	srq_completion;
1917 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1918 	struct list_head	xrcd_list;
1919 
1920 	/* count times opened, mcast attaches, flow attaches */
1921 	atomic_t		usecnt;
1922 	struct list_head	open_list;
1923 	struct ib_qp           *real_qp;
1924 	struct ib_uqp_object   *uobject;
1925 	void                  (*event_handler)(struct ib_event *, void *);
1926 	void                  (*registered_event_handler)(struct ib_event *, void *);
1927 	void		       *qp_context;
1928 	/* sgid_attrs associated with the AV's */
1929 	const struct ib_gid_attr *av_sgid_attr;
1930 	const struct ib_gid_attr *alt_path_sgid_attr;
1931 	u32			qp_num;
1932 	u32			max_write_sge;
1933 	u32			max_read_sge;
1934 	enum ib_qp_type		qp_type;
1935 	struct ib_rwq_ind_table *rwq_ind_tbl;
1936 	struct ib_qp_security  *qp_sec;
1937 	u32			port;
1938 
1939 	bool			integrity_en;
1940 	/*
1941 	 * Implementation details of the RDMA core, don't use in drivers:
1942 	 */
1943 	struct rdma_restrack_entry     res;
1944 
1945 	/* The counter the qp is bind to */
1946 	struct rdma_counter    *counter;
1947 };
1948 
1949 struct ib_dm {
1950 	struct ib_device  *device;
1951 	u32		   length;
1952 	u32		   flags;
1953 	struct ib_uobject *uobject;
1954 	atomic_t	   usecnt;
1955 };
1956 
1957 /* bit values to mark existence of ib_dmah fields */
1958 enum {
1959 	IB_DMAH_CPU_ID_EXISTS,
1960 	IB_DMAH_MEM_TYPE_EXISTS,
1961 	IB_DMAH_PH_EXISTS,
1962 };
1963 
1964 struct ib_dmah {
1965 	struct ib_device *device;
1966 	struct ib_uobject *uobject;
1967 	/*
1968 	 * Implementation details of the RDMA core, don't use in drivers:
1969 	 */
1970 	struct rdma_restrack_entry res;
1971 	u32 cpu_id;
1972 	enum tph_mem_type mem_type;
1973 	atomic_t usecnt;
1974 	u8 ph;
1975 	u8 valid_fields; /* use IB_DMAH_XXX_EXISTS */
1976 };
1977 
1978 struct ib_mr {
1979 	struct ib_device  *device;
1980 	/*
1981 	 * Due to IB_MR_REREG_PD pd is not a fixed pointer and can change. For a
1982 	 * user MR, this value should only be read from a system call that holds
1983 	 * the uobject lock, or the driver should disable in-place REREG_PD.
1984 	 */
1985 	struct ib_pd	  *pd;
1986 	u32		   lkey;
1987 	u32		   rkey;
1988 	u64		   iova;
1989 	u64		   length;
1990 	unsigned int	   page_size;
1991 	enum ib_mr_type	   type;
1992 	bool		   need_inval;
1993 	union {
1994 		struct ib_uobject	*uobject;	/* user */
1995 		struct list_head	qp_entry;	/* FR */
1996 	};
1997 
1998 	struct ib_dm      *dm;
1999 	struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
2000 	struct ib_dmah *dmah;
2001 	struct {
2002 		struct ib_frmr_pool *pool;
2003 		struct ib_frmr_key key;
2004 		u32 handle;
2005 	} frmr;
2006 	/*
2007 	 * Implementation details of the RDMA core, don't use in drivers:
2008 	 */
2009 	struct rdma_restrack_entry res;
2010 };
2011 
2012 struct ib_mw {
2013 	struct ib_device	*device;
2014 	struct ib_pd		*pd;
2015 	struct ib_uobject	*uobject;
2016 	u32			rkey;
2017 	enum ib_mw_type         type;
2018 };
2019 
2020 /* Supported steering options */
2021 enum ib_flow_attr_type {
2022 	/* steering according to rule specifications */
2023 	IB_FLOW_ATTR_NORMAL		= 0x0,
2024 	/* default unicast and multicast rule -
2025 	 * receive all Eth traffic which isn't steered to any QP
2026 	 */
2027 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
2028 	/* default multicast rule -
2029 	 * receive all Eth multicast traffic which isn't steered to any QP
2030 	 */
2031 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
2032 	/* sniffer rule - receive all port traffic */
2033 	IB_FLOW_ATTR_SNIFFER		= 0x3
2034 };
2035 
2036 /* Supported steering header types */
2037 enum ib_flow_spec_type {
2038 	/* L2 headers*/
2039 	IB_FLOW_SPEC_ETH		= 0x20,
2040 	IB_FLOW_SPEC_IB			= 0x22,
2041 	/* L3 header*/
2042 	IB_FLOW_SPEC_IPV4		= 0x30,
2043 	IB_FLOW_SPEC_IPV6		= 0x31,
2044 	IB_FLOW_SPEC_ESP                = 0x34,
2045 	/* L4 headers*/
2046 	IB_FLOW_SPEC_TCP		= 0x40,
2047 	IB_FLOW_SPEC_UDP		= 0x41,
2048 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
2049 	IB_FLOW_SPEC_GRE		= 0x51,
2050 	IB_FLOW_SPEC_MPLS		= 0x60,
2051 	IB_FLOW_SPEC_INNER		= 0x100,
2052 	/* Actions */
2053 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
2054 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
2055 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
2056 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
2057 };
2058 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
2059 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
2060 
2061 enum ib_flow_flags {
2062 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
2063 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
2064 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
2065 };
2066 
2067 struct ib_flow_eth_filter {
2068 	u8	dst_mac[6];
2069 	u8	src_mac[6];
2070 	__be16	ether_type;
2071 	__be16	vlan_tag;
2072 };
2073 
2074 struct ib_flow_spec_eth {
2075 	u32			  type;
2076 	u16			  size;
2077 	struct ib_flow_eth_filter val;
2078 	struct ib_flow_eth_filter mask;
2079 };
2080 
2081 struct ib_flow_ib_filter {
2082 	__be16 dlid;
2083 	__u8   sl;
2084 };
2085 
2086 struct ib_flow_spec_ib {
2087 	u32			 type;
2088 	u16			 size;
2089 	struct ib_flow_ib_filter val;
2090 	struct ib_flow_ib_filter mask;
2091 };
2092 
2093 /* IPv4 header flags */
2094 enum ib_ipv4_flags {
2095 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
2096 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
2097 				    last have this flag set */
2098 };
2099 
2100 struct ib_flow_ipv4_filter {
2101 	__be32	src_ip;
2102 	__be32	dst_ip;
2103 	u8	proto;
2104 	u8	tos;
2105 	u8	ttl;
2106 	u8	flags;
2107 };
2108 
2109 struct ib_flow_spec_ipv4 {
2110 	u32			   type;
2111 	u16			   size;
2112 	struct ib_flow_ipv4_filter val;
2113 	struct ib_flow_ipv4_filter mask;
2114 };
2115 
2116 struct ib_flow_ipv6_filter {
2117 	u8	src_ip[16];
2118 	u8	dst_ip[16];
2119 	__be32	flow_label;
2120 	u8	next_hdr;
2121 	u8	traffic_class;
2122 	u8	hop_limit;
2123 } __packed;
2124 
2125 struct ib_flow_spec_ipv6 {
2126 	u32			   type;
2127 	u16			   size;
2128 	struct ib_flow_ipv6_filter val;
2129 	struct ib_flow_ipv6_filter mask;
2130 };
2131 
2132 struct ib_flow_tcp_udp_filter {
2133 	__be16	dst_port;
2134 	__be16	src_port;
2135 };
2136 
2137 struct ib_flow_spec_tcp_udp {
2138 	u32			      type;
2139 	u16			      size;
2140 	struct ib_flow_tcp_udp_filter val;
2141 	struct ib_flow_tcp_udp_filter mask;
2142 };
2143 
2144 struct ib_flow_tunnel_filter {
2145 	__be32	tunnel_id;
2146 };
2147 
2148 /* ib_flow_spec_tunnel describes the Vxlan tunnel
2149  * the tunnel_id from val has the vni value
2150  */
2151 struct ib_flow_spec_tunnel {
2152 	u32			      type;
2153 	u16			      size;
2154 	struct ib_flow_tunnel_filter  val;
2155 	struct ib_flow_tunnel_filter  mask;
2156 };
2157 
2158 struct ib_flow_esp_filter {
2159 	__be32	spi;
2160 	__be32  seq;
2161 };
2162 
2163 struct ib_flow_spec_esp {
2164 	u32                           type;
2165 	u16			      size;
2166 	struct ib_flow_esp_filter     val;
2167 	struct ib_flow_esp_filter     mask;
2168 };
2169 
2170 struct ib_flow_gre_filter {
2171 	__be16 c_ks_res0_ver;
2172 	__be16 protocol;
2173 	__be32 key;
2174 };
2175 
2176 struct ib_flow_spec_gre {
2177 	u32                           type;
2178 	u16			      size;
2179 	struct ib_flow_gre_filter     val;
2180 	struct ib_flow_gre_filter     mask;
2181 };
2182 
2183 struct ib_flow_mpls_filter {
2184 	__be32 tag;
2185 };
2186 
2187 struct ib_flow_spec_mpls {
2188 	u32                           type;
2189 	u16			      size;
2190 	struct ib_flow_mpls_filter     val;
2191 	struct ib_flow_mpls_filter     mask;
2192 };
2193 
2194 struct ib_flow_spec_action_tag {
2195 	enum ib_flow_spec_type	      type;
2196 	u16			      size;
2197 	u32                           tag_id;
2198 };
2199 
2200 struct ib_flow_spec_action_drop {
2201 	enum ib_flow_spec_type	      type;
2202 	u16			      size;
2203 };
2204 
2205 struct ib_flow_spec_action_handle {
2206 	enum ib_flow_spec_type	      type;
2207 	u16			      size;
2208 	struct ib_flow_action	     *act;
2209 };
2210 
2211 enum ib_counters_description {
2212 	IB_COUNTER_PACKETS,
2213 	IB_COUNTER_BYTES,
2214 };
2215 
2216 struct ib_flow_spec_action_count {
2217 	enum ib_flow_spec_type type;
2218 	u16 size;
2219 	struct ib_counters *counters;
2220 };
2221 
2222 union ib_flow_spec {
2223 	struct {
2224 		u32			type;
2225 		u16			size;
2226 	};
2227 	struct ib_flow_spec_eth		eth;
2228 	struct ib_flow_spec_ib		ib;
2229 	struct ib_flow_spec_ipv4        ipv4;
2230 	struct ib_flow_spec_tcp_udp	tcp_udp;
2231 	struct ib_flow_spec_ipv6        ipv6;
2232 	struct ib_flow_spec_tunnel      tunnel;
2233 	struct ib_flow_spec_esp		esp;
2234 	struct ib_flow_spec_gre		gre;
2235 	struct ib_flow_spec_mpls	mpls;
2236 	struct ib_flow_spec_action_tag  flow_tag;
2237 	struct ib_flow_spec_action_drop drop;
2238 	struct ib_flow_spec_action_handle action;
2239 	struct ib_flow_spec_action_count flow_count;
2240 };
2241 
2242 struct ib_flow_attr {
2243 	enum ib_flow_attr_type type;
2244 	u16	     size;
2245 	u16	     priority;
2246 	u32	     flags;
2247 	u8	     num_of_specs;
2248 	u32	     port;
2249 	union ib_flow_spec flows[];
2250 };
2251 
2252 struct ib_flow {
2253 	struct ib_qp		*qp;
2254 	struct ib_device	*device;
2255 	struct ib_uobject	*uobject;
2256 };
2257 
2258 enum ib_flow_action_type {
2259 	IB_FLOW_ACTION_UNSPECIFIED,
2260 	IB_FLOW_ACTION_ESP = 1,
2261 };
2262 
2263 struct ib_flow_action_attrs_esp_keymats {
2264 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2265 	union {
2266 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2267 	} keymat;
2268 };
2269 
2270 struct ib_flow_action_attrs_esp_replays {
2271 	enum ib_uverbs_flow_action_esp_replay			protocol;
2272 	union {
2273 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2274 	} replay;
2275 };
2276 
2277 enum ib_flow_action_attrs_esp_flags {
2278 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2279 	 * This is done in order to share the same flags between user-space and
2280 	 * kernel and spare an unnecessary translation.
2281 	 */
2282 
2283 	/* Kernel flags */
2284 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2285 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2286 };
2287 
2288 struct ib_flow_spec_list {
2289 	struct ib_flow_spec_list	*next;
2290 	union ib_flow_spec		spec;
2291 };
2292 
2293 struct ib_flow_action_attrs_esp {
2294 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2295 	struct ib_flow_action_attrs_esp_replays		*replay;
2296 	struct ib_flow_spec_list			*encap;
2297 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2298 	 * Value of 0 is a valid value.
2299 	 */
2300 	u32						esn;
2301 	u32						spi;
2302 	u32						seq;
2303 	u32						tfc_pad;
2304 	/* Use enum ib_flow_action_attrs_esp_flags */
2305 	u64						flags;
2306 	u64						hard_limit_pkts;
2307 };
2308 
2309 struct ib_flow_action {
2310 	struct ib_device		*device;
2311 	struct ib_uobject		*uobject;
2312 	enum ib_flow_action_type	type;
2313 	atomic_t			usecnt;
2314 };
2315 
2316 struct ib_mad;
2317 
2318 enum ib_process_mad_flags {
2319 	IB_MAD_IGNORE_MKEY	= 1,
2320 	IB_MAD_IGNORE_BKEY	= 2,
2321 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2322 };
2323 
2324 enum ib_mad_result {
2325 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2326 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2327 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2328 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2329 };
2330 
2331 struct ib_port_cache {
2332 	u64		      subnet_prefix;
2333 	struct ib_pkey_cache  *pkey;
2334 	struct ib_gid_table   *gid;
2335 	u8                     lmc;
2336 	enum ib_port_state     port_state;
2337 	enum ib_port_state     last_port_state;
2338 };
2339 
2340 struct ib_port_immutable {
2341 	int                           pkey_tbl_len;
2342 	int                           gid_tbl_len;
2343 	u32                           core_cap_flags;
2344 	u32                           max_mad_size;
2345 };
2346 
2347 struct ib_port_data {
2348 	struct ib_device *ib_dev;
2349 
2350 	struct ib_port_immutable immutable;
2351 
2352 	spinlock_t pkey_list_lock;
2353 
2354 	spinlock_t netdev_lock;
2355 
2356 	struct list_head pkey_list;
2357 
2358 	struct ib_port_cache cache;
2359 
2360 	struct net_device __rcu *netdev;
2361 	netdevice_tracker netdev_tracker;
2362 	struct hlist_node ndev_hash_link;
2363 	struct rdma_port_counter port_counter;
2364 	struct ib_port *sysfs;
2365 };
2366 
2367 /* rdma netdev type - specifies protocol type */
2368 enum rdma_netdev_t {
2369 	RDMA_NETDEV_IPOIB,
2370 };
2371 
2372 /**
2373  * struct rdma_netdev - rdma netdev
2374  * For cases where netstack interfacing is required.
2375  */
2376 struct rdma_netdev {
2377 	void              *clnt_priv;
2378 	struct ib_device  *hca;
2379 	u32		   port_num;
2380 	int                mtu;
2381 
2382 	void (*free_rdma_netdev)(struct net_device *netdev);
2383 
2384 	/* control functions */
2385 	void (*set_id)(struct net_device *netdev, int id);
2386 	/* send packet */
2387 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2388 		    struct ib_ah *address, u32 dqpn);
2389 	/* multicast */
2390 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2391 			    union ib_gid *gid, u16 mlid,
2392 			    int set_qkey, u32 qkey);
2393 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2394 			    union ib_gid *gid, u16 mlid);
2395 	/* timeout */
2396 	void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2397 };
2398 
2399 struct rdma_netdev_alloc_params {
2400 	size_t sizeof_priv;
2401 	unsigned int txqs;
2402 	unsigned int rxqs;
2403 	void *param;
2404 
2405 	int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2406 				      struct net_device *netdev, void *param);
2407 };
2408 
2409 struct ib_odp_counters {
2410 	atomic64_t faults;
2411 	atomic64_t faults_handled;
2412 	atomic64_t invalidations;
2413 	atomic64_t invalidations_handled;
2414 	atomic64_t prefetch;
2415 };
2416 
2417 struct ib_counters {
2418 	struct ib_device	*device;
2419 	struct ib_uobject	*uobject;
2420 	/* num of objects attached */
2421 	atomic_t	usecnt;
2422 };
2423 
2424 struct ib_counters_read_attr {
2425 	u64	*counters_buff;
2426 	u32	ncounters;
2427 	u32	flags; /* use enum ib_read_counters_flags */
2428 };
2429 
2430 struct uverbs_attr_bundle;
2431 struct iw_cm_id;
2432 struct iw_cm_conn_param;
2433 
2434 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member)                      \
2435 	.size_##ib_struct =                                                    \
2436 		(sizeof(struct drv_struct) +                                   \
2437 		 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) +      \
2438 		 BUILD_BUG_ON_ZERO(                                            \
2439 			 !__same_type(((struct drv_struct *)NULL)->member,     \
2440 				      struct ib_struct)))
2441 
2442 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp)                          \
2443 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2444 					   gfp, false))
2445 
2446 #define rdma_zalloc_drv_obj_numa(ib_dev, ib_type)                              \
2447 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2448 					   GFP_KERNEL, true))
2449 
2450 #define rdma_zalloc_drv_obj(ib_dev, ib_type)                                   \
2451 	rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2452 
2453 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2454 
2455 struct rdma_user_mmap_entry {
2456 	struct kref ref;
2457 	struct ib_ucontext *ucontext;
2458 	unsigned long start_pgoff;
2459 	size_t npages;
2460 	bool driver_removed;
2461 	/* protects access to dmabufs */
2462 	struct mutex dmabufs_lock;
2463 	struct list_head dmabufs;
2464 };
2465 
2466 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2467 static inline u64
2468 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2469 {
2470 	return (u64)entry->start_pgoff << PAGE_SHIFT;
2471 }
2472 
2473 /**
2474  * struct ib_device_ops - InfiniBand device operations
2475  * This structure defines all the InfiniBand device operations, providers will
2476  * need to define the supported operations, otherwise they will be set to null.
2477  */
2478 struct ib_device_ops {
2479 	struct module *owner;
2480 	enum rdma_driver_id driver_id;
2481 	u32 uverbs_abi_ver;
2482 	unsigned int uverbs_no_driver_id_binding:1;
2483 	/*
2484 	 * Indicates the driver checks every op accepting a udata for the
2485 	 * correct size on input and always handles the output using the udata
2486 	 * helpers.
2487 	 */
2488 	unsigned int uverbs_robust_udata:1;
2489 
2490 	/*
2491 	 * NOTE: New drivers should not make use of device_group; instead new
2492 	 * device parameter should be exposed via netlink command. This
2493 	 * mechanism exists only for existing drivers.
2494 	 */
2495 	const struct attribute_group *device_group;
2496 	const struct attribute_group **port_groups;
2497 
2498 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2499 			 const struct ib_send_wr **bad_send_wr);
2500 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2501 			 const struct ib_recv_wr **bad_recv_wr);
2502 	void (*drain_rq)(struct ib_qp *qp);
2503 	void (*drain_sq)(struct ib_qp *qp);
2504 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2505 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2506 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2507 	int (*post_srq_recv)(struct ib_srq *srq,
2508 			     const struct ib_recv_wr *recv_wr,
2509 			     const struct ib_recv_wr **bad_recv_wr);
2510 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2511 			   u32 port_num, const struct ib_wc *in_wc,
2512 			   const struct ib_grh *in_grh,
2513 			   const struct ib_mad *in_mad, struct ib_mad *out_mad,
2514 			   size_t *out_mad_size, u16 *out_mad_pkey_index);
2515 	int (*query_device)(struct ib_device *device,
2516 			    struct ib_device_attr *device_attr,
2517 			    struct ib_udata *udata);
2518 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2519 			     struct ib_device_modify *device_modify);
2520 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2521 	int (*query_port)(struct ib_device *device, u32 port_num,
2522 			  struct ib_port_attr *port_attr);
2523 	int (*query_port_speed)(struct ib_device *device, u32 port_num,
2524 				u64 *speed);
2525 	int (*modify_port)(struct ib_device *device, u32 port_num,
2526 			   int port_modify_mask,
2527 			   struct ib_port_modify *port_modify);
2528 	/*
2529 	 * The following mandatory functions are used only at device
2530 	 * registration.  Keep functions such as these at the end of this
2531 	 * structure to avoid cache line misses when accessing struct ib_device
2532 	 * in fast paths.
2533 	 */
2534 	int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2535 				  struct ib_port_immutable *immutable);
2536 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2537 					       u32 port_num);
2538 	/*
2539 	 * When calling get_netdev, the HW vendor's driver should return the
2540 	 * net device of device @device at port @port_num or NULL if such
2541 	 * a net device doesn't exist. The vendor driver should call dev_hold
2542 	 * on this net device. The HW vendor's device driver must guarantee
2543 	 * that this function returns NULL before the net device has finished
2544 	 * NETDEV_UNREGISTER state.
2545 	 */
2546 	struct net_device *(*get_netdev)(struct ib_device *device,
2547 					 u32 port_num);
2548 	/*
2549 	 * rdma netdev operation
2550 	 *
2551 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2552 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2553 	 */
2554 	struct net_device *(*alloc_rdma_netdev)(
2555 		struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2556 		const char *name, unsigned char name_assign_type,
2557 		void (*setup)(struct net_device *));
2558 
2559 	int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2560 				      enum rdma_netdev_t type,
2561 				      struct rdma_netdev_alloc_params *params);
2562 	/*
2563 	 * query_gid should be return GID value for @device, when @port_num
2564 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2565 	 * is RoCE link layer.
2566 	 */
2567 	int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2568 			 union ib_gid *gid);
2569 	/*
2570 	 * When calling add_gid, the HW vendor's driver should add the gid
2571 	 * of device of port at gid index available at @attr. Meta-info of
2572 	 * that gid (for example, the network device related to this gid) is
2573 	 * available at @attr. @context allows the HW vendor driver to store
2574 	 * extra information together with a GID entry. The HW vendor driver may
2575 	 * allocate memory to contain this information and store it in @context
2576 	 * when a new GID entry is written to. Params are consistent until the
2577 	 * next call of add_gid or delete_gid. The function should return 0 on
2578 	 * success or error otherwise. The function could be called
2579 	 * concurrently for different ports. This function is only called when
2580 	 * roce_gid_table is used.
2581 	 */
2582 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2583 	/*
2584 	 * When calling del_gid, the HW vendor's driver should delete the
2585 	 * gid of device @device at gid index gid_index of port port_num
2586 	 * available in @attr.
2587 	 * Upon the deletion of a GID entry, the HW vendor must free any
2588 	 * allocated memory. The caller will clear @context afterwards.
2589 	 * This function is only called when roce_gid_table is used.
2590 	 */
2591 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2592 	int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2593 			  u16 *pkey);
2594 	int (*alloc_ucontext)(struct ib_ucontext *context,
2595 			      struct ib_udata *udata);
2596 	void (*dealloc_ucontext)(struct ib_ucontext *context);
2597 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2598 	/*
2599 	 * This will be called once refcount of an entry in mmap_xa reaches
2600 	 * zero. The type of the memory that was mapped may differ between
2601 	 * entries and is opaque to the rdma_user_mmap interface.
2602 	 * Therefore needs to be implemented by the driver in mmap_free.
2603 	 */
2604 	void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2605 	int (*mmap_get_pfns)(struct rdma_user_mmap_entry *entry,
2606 			     struct phys_vec *phys_vec,
2607 			     struct p2pdma_provider **provider);
2608 	struct rdma_user_mmap_entry *(*pgoff_to_mmap_entry)(struct ib_ucontext *ucontext,
2609 							    off_t pg_off);
2610 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2611 	int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2612 	int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2613 	int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2614 			 struct ib_udata *udata);
2615 	int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2616 			      struct ib_udata *udata);
2617 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2618 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2619 	int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2620 	int (*create_srq)(struct ib_srq *srq,
2621 			  struct ib_srq_init_attr *srq_init_attr,
2622 			  struct ib_udata *udata);
2623 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2624 			  enum ib_srq_attr_mask srq_attr_mask,
2625 			  struct ib_udata *udata);
2626 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2627 	int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2628 	int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2629 			 struct ib_udata *udata);
2630 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2631 			 int qp_attr_mask, struct ib_udata *udata);
2632 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2633 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2634 	int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2635 	int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2636 			 struct uverbs_attr_bundle *attrs);
2637 	int (*create_user_cq)(struct ib_cq *cq,
2638 			      const struct ib_cq_init_attr *attr,
2639 			      struct uverbs_attr_bundle *attrs);
2640 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2641 	int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2642 	int (*resize_user_cq)(struct ib_cq *cq, unsigned int cqe,
2643 			      struct ib_udata *udata);
2644 	/*
2645 	 * pre_destroy_cq - Prevent a cq from generating any new work
2646 	 * completions, but not free any kernel resources
2647 	 */
2648 	int (*pre_destroy_cq)(struct ib_cq *cq);
2649 	/*
2650 	 * post_destroy_cq - Free all kernel resources
2651 	 */
2652 	void (*post_destroy_cq)(struct ib_cq *cq);
2653 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2654 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2655 				     u64 virt_addr, int mr_access_flags,
2656 				     struct ib_dmah *dmah,
2657 				     struct ib_udata *udata);
2658 	struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2659 					    u64 length, u64 virt_addr, int fd,
2660 					    int mr_access_flags,
2661 					    struct ib_dmah *dmah,
2662 					    struct uverbs_attr_bundle *attrs);
2663 	struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2664 				       u64 length, u64 virt_addr,
2665 				       int mr_access_flags, struct ib_pd *pd,
2666 				       struct ib_udata *udata);
2667 	int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2668 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2669 				  u32 max_num_sg);
2670 	struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2671 					    u32 max_num_data_sg,
2672 					    u32 max_num_meta_sg);
2673 	int (*advise_mr)(struct ib_pd *pd,
2674 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2675 			 struct ib_sge *sg_list, u32 num_sge,
2676 			 struct uverbs_attr_bundle *attrs);
2677 
2678 	/*
2679 	 * Kernel users should universally support relaxed ordering (RO), as
2680 	 * they are designed to read data only after observing the CQE and use
2681 	 * the DMA API correctly.
2682 	 *
2683 	 * Some drivers implicitly enable RO if platform supports it.
2684 	 */
2685 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2686 			 unsigned int *sg_offset);
2687 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2688 			       struct ib_mr_status *mr_status);
2689 	int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2690 	int (*dealloc_mw)(struct ib_mw *mw);
2691 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2692 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2693 	int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2694 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2695 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2696 				       struct ib_flow_attr *flow_attr,
2697 				       struct ib_udata *udata);
2698 	int (*destroy_flow)(struct ib_flow *flow_id);
2699 	int (*destroy_flow_action)(struct ib_flow_action *action);
2700 	int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2701 				 int state);
2702 	int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2703 			     struct ifla_vf_info *ivf);
2704 	int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2705 			    struct ifla_vf_stats *stats);
2706 	int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2707 			    struct ifla_vf_guid *node_guid,
2708 			    struct ifla_vf_guid *port_guid);
2709 	int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2710 			   int type);
2711 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2712 				   struct ib_wq_init_attr *init_attr,
2713 				   struct ib_udata *udata);
2714 	int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2715 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2716 			 u32 wq_attr_mask, struct ib_udata *udata);
2717 	int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2718 				    struct ib_rwq_ind_table_init_attr *init_attr,
2719 				    struct ib_udata *udata);
2720 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2721 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2722 				  struct ib_ucontext *context,
2723 				  struct ib_dm_alloc_attr *attr,
2724 				  struct uverbs_attr_bundle *attrs);
2725 	int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2726 	int (*alloc_dmah)(struct ib_dmah *ibdmah,
2727 			  struct uverbs_attr_bundle *attrs);
2728 	int (*dealloc_dmah)(struct ib_dmah *dmah, struct uverbs_attr_bundle *attrs);
2729 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2730 				   struct ib_dm_mr_attr *attr,
2731 				   struct uverbs_attr_bundle *attrs);
2732 	int (*create_counters)(struct ib_counters *counters,
2733 			       struct uverbs_attr_bundle *attrs);
2734 	int (*destroy_counters)(struct ib_counters *counters);
2735 	int (*read_counters)(struct ib_counters *counters,
2736 			     struct ib_counters_read_attr *counters_read_attr,
2737 			     struct uverbs_attr_bundle *attrs);
2738 	int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2739 			    int data_sg_nents, unsigned int *data_sg_offset,
2740 			    struct scatterlist *meta_sg, int meta_sg_nents,
2741 			    unsigned int *meta_sg_offset);
2742 
2743 	/*
2744 	 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2745 	 *   fill in the driver initialized data.  The struct is kfree()'ed by
2746 	 *   the sysfs core when the device is removed.  A lifespan of -1 in the
2747 	 *   return struct tells the core to set a default lifespan.
2748 	 */
2749 	struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2750 	struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2751 						     u32 port_num);
2752 	/*
2753 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2754 	 * @index - The index in the value array we wish to have updated, or
2755 	 *   num_counters if we want all stats updated
2756 	 * Return codes -
2757 	 *   < 0 - Error, no counters updated
2758 	 *   index - Updated the single counter pointed to by index
2759 	 *   num_counters - Updated all counters (will reset the timestamp
2760 	 *     and prevent further calls for lifespan milliseconds)
2761 	 * Drivers are allowed to update all counters in leiu of just the
2762 	 *   one given in index at their option
2763 	 */
2764 	int (*get_hw_stats)(struct ib_device *device,
2765 			    struct rdma_hw_stats *stats, u32 port, int index);
2766 
2767 	/*
2768 	 * modify_hw_stat - Modify the counter configuration
2769 	 * @enable: true/false when enable/disable a counter
2770 	 * Return codes - 0 on success or error code otherwise.
2771 	 */
2772 	int (*modify_hw_stat)(struct ib_device *device, u32 port,
2773 			      unsigned int counter_index, bool enable);
2774 	/*
2775 	 * Allows rdma drivers to add their own restrack attributes.
2776 	 */
2777 	int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2778 	int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2779 	int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2780 	int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2781 	int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2782 	int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2783 	int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2784 	int (*fill_res_srq_entry)(struct sk_buff *msg, struct ib_srq *ib_srq);
2785 	int (*fill_res_srq_entry_raw)(struct sk_buff *msg, struct ib_srq *ib_srq);
2786 
2787 	/* Device lifecycle callbacks */
2788 	/*
2789 	 * Called after the device becomes registered, before clients are
2790 	 * attached
2791 	 */
2792 	int (*enable_driver)(struct ib_device *dev);
2793 	/*
2794 	 * This is called as part of ib_dealloc_device().
2795 	 */
2796 	void (*dealloc_driver)(struct ib_device *dev);
2797 
2798 	/* iWarp CM callbacks */
2799 	void (*iw_add_ref)(struct ib_qp *qp);
2800 	void (*iw_rem_ref)(struct ib_qp *qp);
2801 	struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2802 	int (*iw_connect)(struct iw_cm_id *cm_id,
2803 			  struct iw_cm_conn_param *conn_param);
2804 	int (*iw_accept)(struct iw_cm_id *cm_id,
2805 			 struct iw_cm_conn_param *conn_param);
2806 	int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2807 			 u8 pdata_len);
2808 	int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2809 	int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2810 	/*
2811 	 * counter_bind_qp - Bind a QP to a counter.
2812 	 * @counter - The counter to be bound. If counter->id is zero then
2813 	 *   the driver needs to allocate a new counter and set counter->id
2814 	 */
2815 	int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp,
2816 			       u32 port);
2817 	/*
2818 	 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2819 	 *   counter and bind it onto the default one
2820 	 */
2821 	int (*counter_unbind_qp)(struct ib_qp *qp, u32 port);
2822 	/*
2823 	 * counter_dealloc -De-allocate the hw counter
2824 	 */
2825 	int (*counter_dealloc)(struct rdma_counter *counter);
2826 	/*
2827 	 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2828 	 * the driver initialized data.
2829 	 */
2830 	struct rdma_hw_stats *(*counter_alloc_stats)(
2831 		struct rdma_counter *counter);
2832 	/*
2833 	 * counter_update_stats - Query the stats value of this counter
2834 	 */
2835 	int (*counter_update_stats)(struct rdma_counter *counter);
2836 
2837 	/*
2838 	 * counter_init - Initialize the driver specific rdma counter struct.
2839 	 */
2840 	void (*counter_init)(struct rdma_counter *counter);
2841 
2842 	/*
2843 	 * Allows rdma drivers to add their own restrack attributes
2844 	 * dumped via 'rdma stat' iproute2 command.
2845 	 */
2846 	int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2847 
2848 	/* query driver for its ucontext properties */
2849 	int (*query_ucontext)(struct ib_ucontext *context,
2850 			      struct uverbs_attr_bundle *attrs);
2851 
2852 	/*
2853 	 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2854 	 * Everyone else relies on Linux memory management model.
2855 	 */
2856 	int (*get_numa_node)(struct ib_device *dev);
2857 
2858 	/*
2859 	 * add_sub_dev - Add a sub IB device
2860 	 */
2861 	struct ib_device *(*add_sub_dev)(struct ib_device *parent,
2862 					 enum rdma_nl_dev_type type,
2863 					 const char *name);
2864 
2865 	/*
2866 	 * del_sub_dev - Delete a sub IB device
2867 	 */
2868 	void (*del_sub_dev)(struct ib_device *sub_dev);
2869 
2870 	/*
2871 	 * ufile_cleanup - Attempt to cleanup ubojects HW resources inside
2872 	 * the ufile.
2873 	 */
2874 	void (*ufile_hw_cleanup)(struct ib_uverbs_file *ufile);
2875 
2876 	/*
2877 	 * report_port_event - Drivers need to implement this if they have
2878 	 * some private stuff to handle when link status changes.
2879 	 */
2880 	void (*report_port_event)(struct ib_device *ibdev,
2881 				  struct net_device *ndev, unsigned long event);
2882 
2883 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
2884 	DECLARE_RDMA_OBJ_SIZE(ib_counters);
2885 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
2886 	DECLARE_RDMA_OBJ_SIZE(ib_dmah);
2887 	DECLARE_RDMA_OBJ_SIZE(ib_mw);
2888 	DECLARE_RDMA_OBJ_SIZE(ib_pd);
2889 	DECLARE_RDMA_OBJ_SIZE(ib_qp);
2890 	DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2891 	DECLARE_RDMA_OBJ_SIZE(ib_srq);
2892 	DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2893 	DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2894 	DECLARE_RDMA_OBJ_SIZE(rdma_counter);
2895 };
2896 
2897 struct ib_core_device {
2898 	/* device must be the first element in structure until,
2899 	 * union of ib_core_device and device exists in ib_device.
2900 	 */
2901 	struct device dev;
2902 	possible_net_t rdma_net;
2903 	struct kobject *ports_kobj;
2904 	struct list_head port_list;
2905 	struct ib_device *owner; /* reach back to owner ib_device */
2906 };
2907 
2908 struct rdma_restrack_root;
2909 struct ib_device {
2910 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2911 	struct device                *dma_device;
2912 	struct ib_device_ops	     ops;
2913 	char                          name[IB_DEVICE_NAME_MAX];
2914 	struct rcu_head rcu_head;
2915 
2916 	struct list_head              event_handler_list;
2917 	/* Protects event_handler_list */
2918 	struct rw_semaphore event_handler_rwsem;
2919 
2920 	/* Protects QP's event_handler calls and open_qp list */
2921 	spinlock_t qp_open_list_lock;
2922 
2923 	struct rw_semaphore	      client_data_rwsem;
2924 	struct xarray                 client_data;
2925 	struct mutex                  unregistration_lock;
2926 
2927 	/* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2928 	rwlock_t cache_lock;
2929 	/**
2930 	 * port_data is indexed by port number
2931 	 */
2932 	struct ib_port_data *port_data;
2933 
2934 	int			      num_comp_vectors;
2935 
2936 	union {
2937 		struct device		dev;
2938 		struct ib_core_device	coredev;
2939 	};
2940 
2941 	/* First group is for device attributes,
2942 	 * Second group is for driver provided attributes (optional).
2943 	 * Third group is for the hw_stats
2944 	 * It is a NULL terminated array.
2945 	 */
2946 	const struct attribute_group	*groups[4];
2947 	u8				hw_stats_attr_index;
2948 
2949 	u64			     uverbs_cmd_mask;
2950 
2951 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2952 	__be64			     node_guid;
2953 	u32			     local_dma_lkey;
2954 	u16                          is_switch:1;
2955 	/* Indicates kernel verbs support, should not be used in drivers */
2956 	u16                          kverbs_provider:1;
2957 	/* CQ adaptive moderation (RDMA DIM) */
2958 	u16                          use_cq_dim:1;
2959 	/* CoCo guest with DMA bounce buffering required */
2960 	u16                          cc_dma_bounce:1;
2961 	u8                           node_type;
2962 	u32			     phys_port_cnt;
2963 	struct ib_device_attr        attrs;
2964 	struct hw_stats_device_data *hw_stats_data;
2965 
2966 #ifdef CONFIG_CGROUP_RDMA
2967 	struct rdmacg_device         cg_device;
2968 #endif
2969 
2970 	u32                          index;
2971 
2972 	spinlock_t                   cq_pools_lock;
2973 	struct list_head             cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2974 
2975 	struct rdma_restrack_root *res;
2976 
2977 	const struct uapi_definition   *driver_def;
2978 
2979 	/*
2980 	 * Positive refcount indicates that the device is currently
2981 	 * registered and cannot be unregistered.
2982 	 */
2983 	refcount_t refcount;
2984 	struct completion unreg_completion;
2985 	struct work_struct unregistration_work;
2986 
2987 	const struct rdma_link_ops *link_ops;
2988 
2989 	/* Protects compat_devs xarray modifications */
2990 	struct mutex compat_devs_mutex;
2991 	/* Maintains compat devices for each net namespace */
2992 	struct xarray compat_devs;
2993 
2994 	/* Used by iWarp CM */
2995 	char iw_ifname[IFNAMSIZ];
2996 	u32 iw_driver_flags;
2997 	u32 lag_flags;
2998 
2999 	/* A parent device has a list of sub-devices */
3000 	struct mutex subdev_lock;
3001 	struct list_head subdev_list_head;
3002 
3003 	/* A sub device has a type and a parent */
3004 	enum rdma_nl_dev_type type;
3005 	struct ib_device *parent;
3006 	struct list_head subdev_list;
3007 
3008 	enum rdma_nl_name_assign_type name_assign_type;
3009 
3010 	struct ib_frmr_pools *frmr_pools;
3011 };
3012 
3013 static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
3014 				    gfp_t gfp, bool is_numa_aware)
3015 {
3016 	if (is_numa_aware && dev->ops.get_numa_node)
3017 		return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
3018 
3019 	return kzalloc(size, gfp);
3020 }
3021 
3022 struct ib_client_nl_info;
3023 struct ib_client {
3024 	const char *name;
3025 	int (*add)(struct ib_device *ibdev);
3026 	void (*remove)(struct ib_device *, void *client_data);
3027 	void (*rename)(struct ib_device *dev, void *client_data);
3028 	int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
3029 			   struct ib_client_nl_info *res);
3030 	int (*get_global_nl_info)(struct ib_client_nl_info *res);
3031 
3032 	/* Returns the net_dev belonging to this ib_client and matching the
3033 	 * given parameters.
3034 	 * @dev:	 An RDMA device that the net_dev use for communication.
3035 	 * @port:	 A physical port number on the RDMA device.
3036 	 * @pkey:	 P_Key that the net_dev uses if applicable.
3037 	 * @gid:	 A GID that the net_dev uses to communicate.
3038 	 * @addr:	 An IP address the net_dev is configured with.
3039 	 * @client_data: The device's client data set by ib_set_client_data().
3040 	 *
3041 	 * An ib_client that implements a net_dev on top of RDMA devices
3042 	 * (such as IP over IB) should implement this callback, allowing the
3043 	 * rdma_cm module to find the right net_dev for a given request.
3044 	 *
3045 	 * The caller is responsible for calling dev_put on the returned
3046 	 * netdev. */
3047 	struct net_device *(*get_net_dev_by_params)(
3048 			struct ib_device *dev,
3049 			u32 port,
3050 			u16 pkey,
3051 			const union ib_gid *gid,
3052 			const struct sockaddr *addr,
3053 			void *client_data);
3054 
3055 	refcount_t uses;
3056 	struct completion uses_zero;
3057 	u32 client_id;
3058 
3059 	/* kverbs are not required by the client */
3060 	u8 no_kverbs_req:1;
3061 };
3062 
3063 struct ib_device *_ib_alloc_device(size_t size, struct net *net);
3064 #define ib_alloc_device(drv_struct, member)                                    \
3065 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
3066 				      BUILD_BUG_ON_ZERO(offsetof(              \
3067 					      struct drv_struct, member)),     \
3068 				      &init_net),			       \
3069 		     struct drv_struct, member)
3070 
3071 #define ib_alloc_device_with_net(drv_struct, member, net)		       \
3072 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
3073 				      BUILD_BUG_ON_ZERO(offsetof(              \
3074 					struct drv_struct, member)), net),     \
3075 		     struct drv_struct, member)
3076 
3077 void ib_dealloc_device(struct ib_device *device);
3078 
3079 void ib_get_device_fw_str(struct ib_device *device, char *str);
3080 
3081 int ib_register_device(struct ib_device *device, const char *name,
3082 		       struct device *dma_device);
3083 void ib_unregister_device(struct ib_device *device);
3084 void ib_unregister_driver(enum rdma_driver_id driver_id);
3085 void ib_unregister_device_and_put(struct ib_device *device);
3086 void ib_unregister_device_queued(struct ib_device *ib_dev);
3087 
3088 int ib_register_client   (struct ib_client *client);
3089 void ib_unregister_client(struct ib_client *client);
3090 
3091 /**
3092  * ib_get_client_data - Get IB client context
3093  * @device:Device to get context for
3094  * @client:Client to get context for
3095  *
3096  * ib_get_client_data() returns the client context data set with
3097  * ib_set_client_data(). This can only be called while the client is
3098  * registered to the device, once the ib_client remove() callback returns this
3099  * cannot be called.
3100  */
3101 static inline void *ib_get_client_data(struct ib_device *device,
3102 				       struct ib_client *client)
3103 {
3104 	return xa_load(&device->client_data, client->client_id);
3105 }
3106 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
3107 			 void *data);
3108 void ib_set_device_ops(struct ib_device *device,
3109 		       const struct ib_device_ops *ops);
3110 
3111 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
3112 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
3113 		      unsigned long pfn, unsigned long size, pgprot_t prot,
3114 		      struct rdma_user_mmap_entry *entry);
3115 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
3116 				struct rdma_user_mmap_entry *entry,
3117 				size_t length);
3118 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
3119 				      struct rdma_user_mmap_entry *entry,
3120 				      size_t length, u32 min_pgoff,
3121 				      u32 max_pgoff);
3122 
3123 void rdma_user_mmap_disassociate(struct ib_device *device);
3124 
3125 static inline int
3126 rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
3127 				  struct rdma_user_mmap_entry *entry,
3128 				  size_t length, u32 pgoff)
3129 {
3130 	return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff,
3131 						 pgoff);
3132 }
3133 
3134 struct rdma_user_mmap_entry *
3135 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
3136 			       unsigned long pgoff);
3137 struct rdma_user_mmap_entry *
3138 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
3139 			 struct vm_area_struct *vma);
3140 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
3141 
3142 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
3143 #else
3144 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext,
3145 				    struct vm_area_struct *vma,
3146 				    unsigned long pfn, unsigned long size,
3147 				    pgprot_t prot,
3148 				    struct rdma_user_mmap_entry *entry)
3149 {
3150 	return -EINVAL;
3151 }
3152 
3153 static inline int
3154 rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
3155 			    struct rdma_user_mmap_entry *entry, size_t length)
3156 {
3157 	return -EINVAL;
3158 }
3159 
3160 static inline int
3161 rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
3162 				  struct rdma_user_mmap_entry *entry,
3163 				  size_t length, u32 min_pgoff, u32 max_pgoff)
3164 {
3165 	return -EINVAL;
3166 }
3167 
3168 static inline void rdma_user_mmap_disassociate(struct ib_device *device)
3169 {
3170 }
3171 
3172 static inline int
3173 rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
3174 				  struct rdma_user_mmap_entry *entry,
3175 				  size_t length, u32 pgoff)
3176 {
3177 	return -EINVAL;
3178 }
3179 
3180 static inline struct rdma_user_mmap_entry *
3181 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
3182 			       unsigned long pgoff)
3183 {
3184 	return NULL;
3185 }
3186 
3187 static inline struct rdma_user_mmap_entry *
3188 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
3189 			 struct vm_area_struct *vma)
3190 {
3191 	return NULL;
3192 }
3193 
3194 static inline void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry)
3195 {
3196 }
3197 
3198 static inline void
3199 rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry)
3200 {
3201 }
3202 #endif
3203 
3204 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
3205 {
3206 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
3207 }
3208 
3209 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
3210 {
3211 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
3212 }
3213 
3214 static inline bool ib_is_buffer_cleared(const void __user *p,
3215 					size_t len)
3216 {
3217 	bool ret;
3218 	u8 *buf;
3219 
3220 	if (len > USHRT_MAX)
3221 		return false;
3222 
3223 	buf = memdup_user(p, len);
3224 	if (IS_ERR(buf))
3225 		return false;
3226 
3227 	ret = !memchr_inv(buf, 0, len);
3228 	kfree(buf);
3229 	return ret;
3230 }
3231 
3232 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
3233 				       size_t offset,
3234 				       size_t len)
3235 {
3236 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
3237 }
3238 
3239 /**
3240  * ib_modify_qp_is_ok - Check that the supplied attribute mask
3241  * contains all required attributes and no attributes not allowed for
3242  * the given QP state transition.
3243  * @cur_state: Current QP state
3244  * @next_state: Next QP state
3245  * @type: QP type
3246  * @mask: Mask of supplied QP attributes
3247  *
3248  * This function is a helper function that a low-level driver's
3249  * modify_qp method can use to validate the consumer's input.  It
3250  * checks that cur_state and next_state are valid QP states, that a
3251  * transition from cur_state to next_state is allowed by the IB spec,
3252  * and that the attribute mask supplied is allowed for the transition.
3253  */
3254 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
3255 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
3256 
3257 void ib_register_event_handler(struct ib_event_handler *event_handler);
3258 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
3259 void ib_dispatch_event(const struct ib_event *event);
3260 
3261 int ib_query_port(struct ib_device *device,
3262 		  u32 port_num, struct ib_port_attr *port_attr);
3263 
3264 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3265 					       u32 port_num);
3266 
3267 /**
3268  * rdma_cap_ib_switch - Check if the device is IB switch
3269  * @device: Device to check
3270  *
3271  * Device driver is responsible for setting is_switch bit on
3272  * in ib_device structure at init time.
3273  *
3274  * Return: true if the device is IB switch.
3275  */
3276 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3277 {
3278 	return device->is_switch;
3279 }
3280 
3281 /**
3282  * rdma_start_port - Return the first valid port number for the device
3283  * specified
3284  *
3285  * @device: Device to be checked
3286  *
3287  * Return start port number
3288  */
3289 static inline u32 rdma_start_port(const struct ib_device *device)
3290 {
3291 	return rdma_cap_ib_switch(device) ? 0 : 1;
3292 }
3293 
3294 /**
3295  * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3296  * @device: The struct ib_device * to iterate over
3297  * @iter: The unsigned int to store the port number
3298  */
3299 #define rdma_for_each_port(device, iter)                                       \
3300 	for (iter = rdma_start_port(device +				       \
3301 				    BUILD_BUG_ON_ZERO(!__same_type(u32,	       \
3302 								   iter)));    \
3303 	     iter <= rdma_end_port(device); iter++)
3304 
3305 /**
3306  * rdma_end_port - Return the last valid port number for the device
3307  * specified
3308  *
3309  * @device: Device to be checked
3310  *
3311  * Return last port number
3312  */
3313 static inline u32 rdma_end_port(const struct ib_device *device)
3314 {
3315 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3316 }
3317 
3318 static inline int rdma_is_port_valid(const struct ib_device *device,
3319 				     unsigned int port)
3320 {
3321 	return (port >= rdma_start_port(device) &&
3322 		port <= rdma_end_port(device));
3323 }
3324 
3325 static inline bool rdma_is_grh_required(const struct ib_device *device,
3326 					u32 port_num)
3327 {
3328 	return device->port_data[port_num].immutable.core_cap_flags &
3329 	       RDMA_CORE_PORT_IB_GRH_REQUIRED;
3330 }
3331 
3332 static inline bool rdma_protocol_ib(const struct ib_device *device,
3333 				    u32 port_num)
3334 {
3335 	return device->port_data[port_num].immutable.core_cap_flags &
3336 	       RDMA_CORE_CAP_PROT_IB;
3337 }
3338 
3339 static inline bool rdma_protocol_roce(const struct ib_device *device,
3340 				      u32 port_num)
3341 {
3342 	return device->port_data[port_num].immutable.core_cap_flags &
3343 	       (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3344 }
3345 
3346 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3347 						u32 port_num)
3348 {
3349 	return device->port_data[port_num].immutable.core_cap_flags &
3350 	       RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3351 }
3352 
3353 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3354 						u32 port_num)
3355 {
3356 	return device->port_data[port_num].immutable.core_cap_flags &
3357 	       RDMA_CORE_CAP_PROT_ROCE;
3358 }
3359 
3360 static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3361 				       u32 port_num)
3362 {
3363 	return device->port_data[port_num].immutable.core_cap_flags &
3364 	       RDMA_CORE_CAP_PROT_IWARP;
3365 }
3366 
3367 static inline bool rdma_ib_or_roce(const struct ib_device *device,
3368 				   u32 port_num)
3369 {
3370 	return rdma_protocol_ib(device, port_num) ||
3371 		rdma_protocol_roce(device, port_num);
3372 }
3373 
3374 static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3375 					    u32 port_num)
3376 {
3377 	return device->port_data[port_num].immutable.core_cap_flags &
3378 	       RDMA_CORE_CAP_PROT_RAW_PACKET;
3379 }
3380 
3381 static inline bool rdma_protocol_usnic(const struct ib_device *device,
3382 				       u32 port_num)
3383 {
3384 	return device->port_data[port_num].immutable.core_cap_flags &
3385 	       RDMA_CORE_CAP_PROT_USNIC;
3386 }
3387 
3388 /**
3389  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3390  * Management Datagrams.
3391  * @device: Device to check
3392  * @port_num: Port number to check
3393  *
3394  * Management Datagrams (MAD) are a required part of the InfiniBand
3395  * specification and are supported on all InfiniBand devices.  A slightly
3396  * extended version are also supported on OPA interfaces.
3397  *
3398  * Return: true if the port supports sending/receiving of MAD packets.
3399  */
3400 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3401 {
3402 	return device->port_data[port_num].immutable.core_cap_flags &
3403 	       RDMA_CORE_CAP_IB_MAD;
3404 }
3405 
3406 /**
3407  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3408  * Management Datagrams.
3409  * @device: Device to check
3410  * @port_num: Port number to check
3411  *
3412  * Intel OmniPath devices extend and/or replace the InfiniBand Management
3413  * datagrams with their own versions.  These OPA MADs share many but not all of
3414  * the characteristics of InfiniBand MADs.
3415  *
3416  * OPA MADs differ in the following ways:
3417  *
3418  *    1) MADs are variable size up to 2K
3419  *       IBTA defined MADs remain fixed at 256 bytes
3420  *    2) OPA SMPs must carry valid PKeys
3421  *    3) OPA SMP packets are a different format
3422  *
3423  * Return: true if the port supports OPA MAD packet formats.
3424  */
3425 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3426 {
3427 	return device->port_data[port_num].immutable.core_cap_flags &
3428 		RDMA_CORE_CAP_OPA_MAD;
3429 }
3430 
3431 /**
3432  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3433  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3434  * @device: Device to check
3435  * @port_num: Port number to check
3436  *
3437  * Each InfiniBand node is required to provide a Subnet Management Agent
3438  * that the subnet manager can access.  Prior to the fabric being fully
3439  * configured by the subnet manager, the SMA is accessed via a well known
3440  * interface called the Subnet Management Interface (SMI).  This interface
3441  * uses directed route packets to communicate with the SM to get around the
3442  * chicken and egg problem of the SM needing to know what's on the fabric
3443  * in order to configure the fabric, and needing to configure the fabric in
3444  * order to send packets to the devices on the fabric.  These directed
3445  * route packets do not need the fabric fully configured in order to reach
3446  * their destination.  The SMI is the only method allowed to send
3447  * directed route packets on an InfiniBand fabric.
3448  *
3449  * Return: true if the port provides an SMI.
3450  */
3451 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3452 {
3453 	return device->port_data[port_num].immutable.core_cap_flags &
3454 	       RDMA_CORE_CAP_IB_SMI;
3455 }
3456 
3457 /**
3458  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3459  * Communication Manager.
3460  * @device: Device to check
3461  * @port_num: Port number to check
3462  *
3463  * The InfiniBand Communication Manager is one of many pre-defined General
3464  * Service Agents (GSA) that are accessed via the General Service
3465  * Interface (GSI).  It's role is to facilitate establishment of connections
3466  * between nodes as well as other management related tasks for established
3467  * connections.
3468  *
3469  * Return: true if the port supports an IB CM (this does not guarantee that
3470  * a CM is actually running however).
3471  */
3472 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3473 {
3474 	return device->port_data[port_num].immutable.core_cap_flags &
3475 	       RDMA_CORE_CAP_IB_CM;
3476 }
3477 
3478 /**
3479  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3480  * Communication Manager.
3481  * @device: Device to check
3482  * @port_num: Port number to check
3483  *
3484  * Similar to above, but specific to iWARP connections which have a different
3485  * managment protocol than InfiniBand.
3486  *
3487  * Return: true if the port supports an iWARP CM (this does not guarantee that
3488  * a CM is actually running however).
3489  */
3490 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3491 {
3492 	return device->port_data[port_num].immutable.core_cap_flags &
3493 	       RDMA_CORE_CAP_IW_CM;
3494 }
3495 
3496 /**
3497  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3498  * Subnet Administration.
3499  * @device: Device to check
3500  * @port_num: Port number to check
3501  *
3502  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3503  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
3504  * fabrics, devices should resolve routes to other hosts by contacting the
3505  * SA to query the proper route.
3506  *
3507  * Return: true if the port should act as a client to the fabric Subnet
3508  * Administration interface.  This does not imply that the SA service is
3509  * running locally.
3510  */
3511 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3512 {
3513 	return device->port_data[port_num].immutable.core_cap_flags &
3514 	       RDMA_CORE_CAP_IB_SA;
3515 }
3516 
3517 /**
3518  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3519  * Multicast.
3520  * @device: Device to check
3521  * @port_num: Port number to check
3522  *
3523  * InfiniBand multicast registration is more complex than normal IPv4 or
3524  * IPv6 multicast registration.  Each Host Channel Adapter must register
3525  * with the Subnet Manager when it wishes to join a multicast group.  It
3526  * should do so only once regardless of how many queue pairs it subscribes
3527  * to this group.  And it should leave the group only after all queue pairs
3528  * attached to the group have been detached.
3529  *
3530  * Return: true if the port must undertake the additional adminstrative
3531  * overhead of registering/unregistering with the SM and tracking of the
3532  * total number of queue pairs attached to the multicast group.
3533  */
3534 static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3535 				     u32 port_num)
3536 {
3537 	return rdma_cap_ib_sa(device, port_num);
3538 }
3539 
3540 /**
3541  * rdma_cap_af_ib - Check if the port of device has the capability
3542  * Native Infiniband Address.
3543  * @device: Device to check
3544  * @port_num: Port number to check
3545  *
3546  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3547  * GID.  RoCE uses a different mechanism, but still generates a GID via
3548  * a prescribed mechanism and port specific data.
3549  *
3550  * Return: true if the port uses a GID address to identify devices on the
3551  * network.
3552  */
3553 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3554 {
3555 	return device->port_data[port_num].immutable.core_cap_flags &
3556 	       RDMA_CORE_CAP_AF_IB;
3557 }
3558 
3559 /**
3560  * rdma_cap_eth_ah - Check if the port of device has the capability
3561  * Ethernet Address Handle.
3562  * @device: Device to check
3563  * @port_num: Port number to check
3564  *
3565  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3566  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3567  * port.  Normally, packet headers are generated by the sending host
3568  * adapter, but when sending connectionless datagrams, we must manually
3569  * inject the proper headers for the fabric we are communicating over.
3570  *
3571  * Return: true if we are running as a RoCE port and must force the
3572  * addition of a Global Route Header built from our Ethernet Address
3573  * Handle into our header list for connectionless packets.
3574  */
3575 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3576 {
3577 	return device->port_data[port_num].immutable.core_cap_flags &
3578 	       RDMA_CORE_CAP_ETH_AH;
3579 }
3580 
3581 /**
3582  * rdma_cap_opa_ah - Check if the port of device supports
3583  * OPA Address handles
3584  * @device: Device to check
3585  * @port_num: Port number to check
3586  *
3587  * Return: true if we are running on an OPA device which supports
3588  * the extended OPA addressing.
3589  */
3590 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3591 {
3592 	return (device->port_data[port_num].immutable.core_cap_flags &
3593 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3594 }
3595 
3596 /**
3597  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3598  *
3599  * @device: Device
3600  * @port_num: Port number
3601  *
3602  * This MAD size includes the MAD headers and MAD payload.  No other headers
3603  * are included.
3604  *
3605  * Return the max MAD size required by the Port.  Will return 0 if the port
3606  * does not support MADs
3607  */
3608 static inline size_t rdma_max_mad_size(const struct ib_device *device,
3609 				       u32 port_num)
3610 {
3611 	return device->port_data[port_num].immutable.max_mad_size;
3612 }
3613 
3614 /**
3615  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3616  * @device: Device to check
3617  * @port_num: Port number to check
3618  *
3619  * RoCE GID table mechanism manages the various GIDs for a device.
3620  *
3621  * NOTE: if allocating the port's GID table has failed, this call will still
3622  * return true, but any RoCE GID table API will fail.
3623  *
3624  * Return: true if the port uses RoCE GID table mechanism in order to manage
3625  * its GIDs.
3626  */
3627 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3628 					   u32 port_num)
3629 {
3630 	return rdma_protocol_roce(device, port_num) &&
3631 		device->ops.add_gid && device->ops.del_gid;
3632 }
3633 
3634 /*
3635  * Check if the device supports READ W/ INVALIDATE.
3636  */
3637 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3638 {
3639 	/*
3640 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3641 	 * has support for it yet.
3642 	 */
3643 	return rdma_protocol_iwarp(dev, port_num);
3644 }
3645 
3646 /**
3647  * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3648  * @device: Device
3649  * @port_num: 1 based Port number
3650  *
3651  * Return true if port is an Intel OPA port , false if not
3652  */
3653 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3654 					  u32 port_num)
3655 {
3656 	return (device->port_data[port_num].immutable.core_cap_flags &
3657 		RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3658 }
3659 
3660 /**
3661  * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3662  * @device: Device
3663  * @port: Port number
3664  * @mtu: enum value of MTU
3665  *
3666  * Return the MTU size supported by the port as an integer value. Will return
3667  * -1 if enum value of mtu is not supported.
3668  */
3669 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3670 				       int mtu)
3671 {
3672 	if (rdma_core_cap_opa_port(device, port))
3673 		return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3674 	else
3675 		return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3676 }
3677 
3678 /**
3679  * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3680  * @device: Device
3681  * @port: Port number
3682  * @attr: port attribute
3683  *
3684  * Return the MTU size supported by the port as an integer value.
3685  */
3686 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3687 				     struct ib_port_attr *attr)
3688 {
3689 	if (rdma_core_cap_opa_port(device, port))
3690 		return attr->phys_mtu;
3691 	else
3692 		return ib_mtu_enum_to_int(attr->max_mtu);
3693 }
3694 
3695 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3696 			 int state);
3697 int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3698 		     struct ifla_vf_info *info);
3699 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3700 		    struct ifla_vf_stats *stats);
3701 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3702 		    struct ifla_vf_guid *node_guid,
3703 		    struct ifla_vf_guid *port_guid);
3704 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3705 		   int type);
3706 
3707 int ib_query_pkey(struct ib_device *device,
3708 		  u32 port_num, u16 index, u16 *pkey);
3709 
3710 int ib_modify_device(struct ib_device *device,
3711 		     int device_modify_mask,
3712 		     struct ib_device_modify *device_modify);
3713 
3714 int ib_modify_port(struct ib_device *device,
3715 		   u32 port_num, int port_modify_mask,
3716 		   struct ib_port_modify *port_modify);
3717 
3718 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3719 		u32 *port_num, u16 *index);
3720 
3721 int ib_find_pkey(struct ib_device *device,
3722 		 u32 port_num, u16 pkey, u16 *index);
3723 
3724 enum ib_pd_flags {
3725 	/*
3726 	 * Create a memory registration for all memory in the system and place
3727 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3728 	 * ULPs to avoid the overhead of dynamic MRs.
3729 	 *
3730 	 * This flag is generally considered unsafe and must only be used in
3731 	 * extremly trusted environments.  Every use of it will log a warning
3732 	 * in the kernel log.
3733 	 */
3734 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3735 };
3736 
3737 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3738 		const char *caller);
3739 
3740 /**
3741  * ib_alloc_pd - Allocates an unused protection domain.
3742  * @device: The device on which to allocate the protection domain.
3743  * @flags: protection domain flags
3744  *
3745  * A protection domain object provides an association between QPs, shared
3746  * receive queues, address handles, memory regions, and memory windows.
3747  *
3748  * Every PD has a local_dma_lkey which can be used as the lkey value for local
3749  * memory operations.
3750  */
3751 #define ib_alloc_pd(device, flags) \
3752 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3753 
3754 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3755 
3756 /**
3757  * ib_dealloc_pd - Deallocate kernel PD
3758  * @pd: The protection domain
3759  *
3760  * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3761  */
3762 static inline void ib_dealloc_pd(struct ib_pd *pd)
3763 {
3764 	int ret = ib_dealloc_pd_user(pd, NULL);
3765 
3766 	WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3767 }
3768 
3769 enum rdma_create_ah_flags {
3770 	/* In a sleepable context */
3771 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3772 };
3773 
3774 /**
3775  * rdma_create_ah - Creates an address handle for the given address vector.
3776  * @pd: The protection domain associated with the address handle.
3777  * @ah_attr: The attributes of the address vector.
3778  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3779  *
3780  * The address handle is used to reference a local or global destination
3781  * in all UD QP post sends.
3782  */
3783 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3784 			     u32 flags);
3785 
3786 /**
3787  * rdma_create_user_ah - Creates an address handle for the given address vector.
3788  * It resolves destination mac address for ah attribute of RoCE type.
3789  * @pd: The protection domain associated with the address handle.
3790  * @ah_attr: The attributes of the address vector.
3791  * @udata: pointer to user's input output buffer information need by
3792  *         provider driver.
3793  *
3794  * It returns 0 on success and returns appropriate error code on error.
3795  * The address handle is used to reference a local or global destination
3796  * in all UD QP post sends.
3797  */
3798 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3799 				  struct rdma_ah_attr *ah_attr,
3800 				  struct ib_udata *udata);
3801 /**
3802  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3803  *   work completion.
3804  * @hdr: the L3 header to parse
3805  * @net_type: type of header to parse
3806  * @sgid: place to store source gid
3807  * @dgid: place to store destination gid
3808  */
3809 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3810 			      enum rdma_network_type net_type,
3811 			      union ib_gid *sgid, union ib_gid *dgid);
3812 
3813 /**
3814  * ib_get_rdma_header_version - Get the header version
3815  * @hdr: the L3 header to parse
3816  */
3817 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3818 
3819 /**
3820  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3821  *   work completion.
3822  * @device: Device on which the received message arrived.
3823  * @port_num: Port on which the received message arrived.
3824  * @wc: Work completion associated with the received message.
3825  * @grh: References the received global route header.  This parameter is
3826  *   ignored unless the work completion indicates that the GRH is valid.
3827  * @ah_attr: Returned attributes that can be used when creating an address
3828  *   handle for replying to the message.
3829  * When ib_init_ah_attr_from_wc() returns success,
3830  * (a) for IB link layer it optionally contains a reference to SGID attribute
3831  * when GRH is present for IB link layer.
3832  * (b) for RoCE link layer it contains a reference to SGID attribute.
3833  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3834  * attributes which are initialized using ib_init_ah_attr_from_wc().
3835  *
3836  */
3837 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3838 			    const struct ib_wc *wc, const struct ib_grh *grh,
3839 			    struct rdma_ah_attr *ah_attr);
3840 
3841 /**
3842  * ib_create_ah_from_wc - Creates an address handle associated with the
3843  *   sender of the specified work completion.
3844  * @pd: The protection domain associated with the address handle.
3845  * @wc: Work completion information associated with a received message.
3846  * @grh: References the received global route header.  This parameter is
3847  *   ignored unless the work completion indicates that the GRH is valid.
3848  * @port_num: The outbound port number to associate with the address.
3849  *
3850  * The address handle is used to reference a local or global destination
3851  * in all UD QP post sends.
3852  */
3853 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3854 				   const struct ib_grh *grh, u32 port_num);
3855 
3856 /**
3857  * rdma_modify_ah - Modifies the address vector associated with an address
3858  *   handle.
3859  * @ah: The address handle to modify.
3860  * @ah_attr: The new address vector attributes to associate with the
3861  *   address handle.
3862  */
3863 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3864 
3865 /**
3866  * rdma_query_ah - Queries the address vector associated with an address
3867  *   handle.
3868  * @ah: The address handle to query.
3869  * @ah_attr: The address vector attributes associated with the address
3870  *   handle.
3871  */
3872 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3873 
3874 enum rdma_destroy_ah_flags {
3875 	/* In a sleepable context */
3876 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3877 };
3878 
3879 /**
3880  * rdma_destroy_ah_user - Destroys an address handle.
3881  * @ah: The address handle to destroy.
3882  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3883  * @udata: Valid user data or NULL for kernel objects
3884  */
3885 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3886 
3887 /**
3888  * rdma_destroy_ah - Destroys an kernel address handle.
3889  * @ah: The address handle to destroy.
3890  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3891  *
3892  * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3893  */
3894 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3895 {
3896 	int ret = rdma_destroy_ah_user(ah, flags, NULL);
3897 
3898 	WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3899 }
3900 
3901 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3902 				  struct ib_srq_init_attr *srq_init_attr,
3903 				  struct ib_usrq_object *uobject,
3904 				  struct ib_udata *udata);
3905 static inline struct ib_srq *
3906 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3907 {
3908 	if (!pd->device->ops.create_srq)
3909 		return ERR_PTR(-EOPNOTSUPP);
3910 
3911 	return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3912 }
3913 
3914 /**
3915  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3916  * @srq: The SRQ to modify.
3917  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3918  *   the current values of selected SRQ attributes are returned.
3919  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3920  *   are being modified.
3921  *
3922  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3923  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3924  * the number of receives queued drops below the limit.
3925  */
3926 int ib_modify_srq(struct ib_srq *srq,
3927 		  struct ib_srq_attr *srq_attr,
3928 		  enum ib_srq_attr_mask srq_attr_mask);
3929 
3930 /**
3931  * ib_query_srq - Returns the attribute list and current values for the
3932  *   specified SRQ.
3933  * @srq: The SRQ to query.
3934  * @srq_attr: The attributes of the specified SRQ.
3935  */
3936 int ib_query_srq(struct ib_srq *srq,
3937 		 struct ib_srq_attr *srq_attr);
3938 
3939 /**
3940  * ib_destroy_srq_user - Destroys the specified SRQ.
3941  * @srq: The SRQ to destroy.
3942  * @udata: Valid user data or NULL for kernel objects
3943  */
3944 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3945 
3946 /**
3947  * ib_destroy_srq - Destroys the specified kernel SRQ.
3948  * @srq: The SRQ to destroy.
3949  *
3950  * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3951  */
3952 static inline void ib_destroy_srq(struct ib_srq *srq)
3953 {
3954 	int ret = ib_destroy_srq_user(srq, NULL);
3955 
3956 	WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3957 }
3958 
3959 /**
3960  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3961  * @srq: The SRQ to post the work request on.
3962  * @recv_wr: A list of work requests to post on the receive queue.
3963  * @bad_recv_wr: On an immediate failure, this parameter will reference
3964  *   the work request that failed to be posted on the QP.
3965  */
3966 static inline int ib_post_srq_recv(struct ib_srq *srq,
3967 				   const struct ib_recv_wr *recv_wr,
3968 				   const struct ib_recv_wr **bad_recv_wr)
3969 {
3970 	const struct ib_recv_wr *dummy;
3971 
3972 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3973 					      bad_recv_wr ? : &dummy);
3974 }
3975 
3976 struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3977 				  struct ib_qp_init_attr *qp_init_attr,
3978 				  const char *caller);
3979 /**
3980  * ib_create_qp - Creates a kernel QP associated with the specific protection
3981  * domain.
3982  * @pd: The protection domain associated with the QP.
3983  * @init_attr: A list of initial attributes required to create the
3984  *   QP.  If QP creation succeeds, then the attributes are updated to
3985  *   the actual capabilities of the created QP.
3986  */
3987 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3988 					 struct ib_qp_init_attr *init_attr)
3989 {
3990 	return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
3991 }
3992 
3993 /**
3994  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3995  * @qp: The QP to modify.
3996  * @attr: On input, specifies the QP attributes to modify.  On output,
3997  *   the current values of selected QP attributes are returned.
3998  * @attr_mask: A bit-mask used to specify which attributes of the QP
3999  *   are being modified.
4000  * @udata: pointer to user's input output buffer information
4001  *   are being modified.
4002  * It returns 0 on success and returns appropriate error code on error.
4003  */
4004 int ib_modify_qp_with_udata(struct ib_qp *qp,
4005 			    struct ib_qp_attr *attr,
4006 			    int attr_mask,
4007 			    struct ib_udata *udata);
4008 
4009 /**
4010  * ib_modify_qp - Modifies the attributes for the specified QP and then
4011  *   transitions the QP to the given state.
4012  * @qp: The QP to modify.
4013  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
4014  *   the current values of selected QP attributes are returned.
4015  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
4016  *   are being modified.
4017  */
4018 int ib_modify_qp(struct ib_qp *qp,
4019 		 struct ib_qp_attr *qp_attr,
4020 		 int qp_attr_mask);
4021 
4022 /**
4023  * ib_query_qp - Returns the attribute list and current values for the
4024  *   specified QP.
4025  * @qp: The QP to query.
4026  * @qp_attr: The attributes of the specified QP.
4027  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
4028  * @qp_init_attr: Additional attributes of the selected QP.
4029  *
4030  * The qp_attr_mask may be used to limit the query to gathering only the
4031  * selected attributes.
4032  */
4033 int ib_query_qp(struct ib_qp *qp,
4034 		struct ib_qp_attr *qp_attr,
4035 		int qp_attr_mask,
4036 		struct ib_qp_init_attr *qp_init_attr);
4037 
4038 /**
4039  * ib_destroy_qp - Destroys the specified QP.
4040  * @qp: The QP to destroy.
4041  * @udata: Valid udata or NULL for kernel objects
4042  */
4043 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
4044 
4045 /**
4046  * ib_destroy_qp - Destroys the specified kernel QP.
4047  * @qp: The QP to destroy.
4048  *
4049  * NOTE: for user qp use ib_destroy_qp_user with valid udata!
4050  */
4051 static inline int ib_destroy_qp(struct ib_qp *qp)
4052 {
4053 	return ib_destroy_qp_user(qp, NULL);
4054 }
4055 
4056 /**
4057  * ib_open_qp - Obtain a reference to an existing sharable QP.
4058  * @xrcd: XRC domain
4059  * @qp_open_attr: Attributes identifying the QP to open.
4060  *
4061  * Returns a reference to a sharable QP.
4062  */
4063 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
4064 			 struct ib_qp_open_attr *qp_open_attr);
4065 
4066 /**
4067  * ib_close_qp - Release an external reference to a QP.
4068  * @qp: The QP handle to release
4069  *
4070  * The opened QP handle is released by the caller.  The underlying
4071  * shared QP is not destroyed until all internal references are released.
4072  */
4073 int ib_close_qp(struct ib_qp *qp);
4074 
4075 /**
4076  * ib_post_send - Posts a list of work requests to the send queue of
4077  *   the specified QP.
4078  * @qp: The QP to post the work request on.
4079  * @send_wr: A list of work requests to post on the send queue.
4080  * @bad_send_wr: On an immediate failure, this parameter will reference
4081  *   the work request that failed to be posted on the QP.
4082  *
4083  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
4084  * error is returned, the QP state shall not be affected,
4085  * ib_post_send() will return an immediate error after queueing any
4086  * earlier work requests in the list.
4087  */
4088 static inline int ib_post_send(struct ib_qp *qp,
4089 			       const struct ib_send_wr *send_wr,
4090 			       const struct ib_send_wr **bad_send_wr)
4091 {
4092 	const struct ib_send_wr *dummy;
4093 
4094 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
4095 }
4096 
4097 /**
4098  * ib_post_recv - Posts a list of work requests to the receive queue of
4099  *   the specified QP.
4100  * @qp: The QP to post the work request on.
4101  * @recv_wr: A list of work requests to post on the receive queue.
4102  * @bad_recv_wr: On an immediate failure, this parameter will reference
4103  *   the work request that failed to be posted on the QP.
4104  */
4105 static inline int ib_post_recv(struct ib_qp *qp,
4106 			       const struct ib_recv_wr *recv_wr,
4107 			       const struct ib_recv_wr **bad_recv_wr)
4108 {
4109 	const struct ib_recv_wr *dummy;
4110 
4111 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
4112 }
4113 
4114 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
4115 			    int comp_vector, enum ib_poll_context poll_ctx,
4116 			    const char *caller);
4117 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
4118 					int nr_cqe, int comp_vector,
4119 					enum ib_poll_context poll_ctx)
4120 {
4121 	return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
4122 			     KBUILD_MODNAME);
4123 }
4124 
4125 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
4126 				int nr_cqe, enum ib_poll_context poll_ctx,
4127 				const char *caller);
4128 
4129 /**
4130  * ib_alloc_cq_any: Allocate kernel CQ
4131  * @dev: The IB device
4132  * @private: Private data attached to the CQE
4133  * @nr_cqe: Number of CQEs in the CQ
4134  * @poll_ctx: Context used for polling the CQ
4135  */
4136 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
4137 					    void *private, int nr_cqe,
4138 					    enum ib_poll_context poll_ctx)
4139 {
4140 	return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
4141 				 KBUILD_MODNAME);
4142 }
4143 
4144 void ib_free_cq(struct ib_cq *cq);
4145 int ib_process_cq_direct(struct ib_cq *cq, int budget);
4146 
4147 /**
4148  * ib_create_cq - Creates a CQ on the specified device.
4149  * @device: The device on which to create the CQ.
4150  * @comp_handler: A user-specified callback that is invoked when a
4151  *   completion event occurs on the CQ.
4152  * @event_handler: A user-specified callback that is invoked when an
4153  *   asynchronous event not associated with a completion occurs on the CQ.
4154  * @cq_context: Context associated with the CQ returned to the user via
4155  *   the associated completion and event handlers.
4156  * @cq_attr: The attributes the CQ should be created upon.
4157  *
4158  * Users can examine the cq structure to determine the actual CQ size.
4159  */
4160 struct ib_cq *__ib_create_cq(struct ib_device *device,
4161 			     ib_comp_handler comp_handler,
4162 			     void (*event_handler)(struct ib_event *, void *),
4163 			     void *cq_context,
4164 			     const struct ib_cq_init_attr *cq_attr,
4165 			     const char *caller);
4166 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
4167 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
4168 
4169 /**
4170  * rdma_set_cq_moderation - Modifies moderation params of the CQ
4171  * @cq: The CQ to modify.
4172  * @cq_count: number of CQEs that will trigger an event
4173  * @cq_period: max period of time in usec before triggering an event
4174  *
4175  */
4176 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
4177 
4178 /**
4179  * ib_destroy_cq_user - Destroys the specified CQ.
4180  * @cq: The CQ to destroy.
4181  * @udata: Valid user data or NULL for kernel objects
4182  */
4183 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
4184 
4185 /**
4186  * ib_destroy_cq - Destroys the specified kernel CQ.
4187  * @cq: The CQ to destroy.
4188  *
4189  * NOTE: for user cq use ib_destroy_cq_user with valid udata!
4190  */
4191 static inline void ib_destroy_cq(struct ib_cq *cq)
4192 {
4193 	int ret = ib_destroy_cq_user(cq, NULL);
4194 
4195 	WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
4196 }
4197 
4198 /**
4199  * ib_poll_cq - poll a CQ for completion(s)
4200  * @cq:the CQ being polled
4201  * @num_entries:maximum number of completions to return
4202  * @wc:array of at least @num_entries &struct ib_wc where completions
4203  *   will be returned
4204  *
4205  * Poll a CQ for (possibly multiple) completions.  If the return value
4206  * is < 0, an error occurred.  If the return value is >= 0, it is the
4207  * number of completions returned.  If the return value is
4208  * non-negative and < num_entries, then the CQ was emptied.
4209  */
4210 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
4211 			     struct ib_wc *wc)
4212 {
4213 	return cq->device->ops.poll_cq(cq, num_entries, wc);
4214 }
4215 
4216 /**
4217  * ib_req_notify_cq - Request completion notification on a CQ.
4218  * @cq: The CQ to generate an event for.
4219  * @flags:
4220  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
4221  *   to request an event on the next solicited event or next work
4222  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
4223  *   may also be |ed in to request a hint about missed events, as
4224  *   described below.
4225  *
4226  * Return Value:
4227  *    < 0 means an error occurred while requesting notification
4228  *   == 0 means notification was requested successfully, and if
4229  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
4230  *        were missed and it is safe to wait for another event.  In
4231  *        this case is it guaranteed that any work completions added
4232  *        to the CQ since the last CQ poll will trigger a completion
4233  *        notification event.
4234  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
4235  *        in.  It means that the consumer must poll the CQ again to
4236  *        make sure it is empty to avoid missing an event because of a
4237  *        race between requesting notification and an entry being
4238  *        added to the CQ.  This return value means it is possible
4239  *        (but not guaranteed) that a work completion has been added
4240  *        to the CQ since the last poll without triggering a
4241  *        completion notification event.
4242  */
4243 static inline int ib_req_notify_cq(struct ib_cq *cq,
4244 				   enum ib_cq_notify_flags flags)
4245 {
4246 	return cq->device->ops.req_notify_cq(cq, flags);
4247 }
4248 
4249 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4250 			     int comp_vector_hint,
4251 			     enum ib_poll_context poll_ctx);
4252 
4253 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4254 
4255 /*
4256  * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
4257  * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
4258  * address into the dma address.
4259  */
4260 static inline bool ib_uses_virt_dma(struct ib_device *dev)
4261 {
4262 	return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
4263 }
4264 
4265 /*
4266  * Check if a IB device's underlying DMA mapping supports P2PDMA transfers.
4267  */
4268 static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev)
4269 {
4270 	if (ib_uses_virt_dma(dev))
4271 		return false;
4272 
4273 	return dma_pci_p2pdma_supported(dev->dma_device);
4274 }
4275 
4276 /**
4277  * ib_virt_dma_to_ptr - Convert a dma_addr to a kernel pointer
4278  * @dma_addr: The DMA address
4279  *
4280  * Used by ib_uses_virt_dma() devices to get back to the kernel pointer after
4281  * going through the dma_addr marshalling.
4282  */
4283 static inline void *ib_virt_dma_to_ptr(u64 dma_addr)
4284 {
4285 	/* virt_dma mode maps the kvs's directly into the dma addr */
4286 	return (void *)(uintptr_t)dma_addr;
4287 }
4288 
4289 /**
4290  * ib_virt_dma_to_page - Convert a dma_addr to a struct page
4291  * @dma_addr: The DMA address
4292  *
4293  * Used by ib_uses_virt_dma() device to get back to the struct page after going
4294  * through the dma_addr marshalling.
4295  */
4296 static inline struct page *ib_virt_dma_to_page(u64 dma_addr)
4297 {
4298 	return virt_to_page(ib_virt_dma_to_ptr(dma_addr));
4299 }
4300 
4301 /**
4302  * ib_dma_mapping_error - check a DMA addr for error
4303  * @dev: The device for which the dma_addr was created
4304  * @dma_addr: The DMA address to check
4305  */
4306 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4307 {
4308 	if (ib_uses_virt_dma(dev))
4309 		return 0;
4310 	return dma_mapping_error(dev->dma_device, dma_addr);
4311 }
4312 
4313 /**
4314  * ib_dma_map_single - Map a kernel virtual address to DMA address
4315  * @dev: The device for which the dma_addr is to be created
4316  * @cpu_addr: The kernel virtual address
4317  * @size: The size of the region in bytes
4318  * @direction: The direction of the DMA
4319  */
4320 static inline u64 ib_dma_map_single(struct ib_device *dev,
4321 				    void *cpu_addr, size_t size,
4322 				    enum dma_data_direction direction)
4323 {
4324 	if (ib_uses_virt_dma(dev))
4325 		return (uintptr_t)cpu_addr;
4326 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4327 }
4328 
4329 /**
4330  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4331  * @dev: The device for which the DMA address was created
4332  * @addr: The DMA address
4333  * @size: The size of the region in bytes
4334  * @direction: The direction of the DMA
4335  */
4336 static inline void ib_dma_unmap_single(struct ib_device *dev,
4337 				       u64 addr, size_t size,
4338 				       enum dma_data_direction direction)
4339 {
4340 	if (!ib_uses_virt_dma(dev))
4341 		dma_unmap_single(dev->dma_device, addr, size, direction);
4342 }
4343 
4344 /**
4345  * ib_dma_map_page - Map a physical page to DMA address
4346  * @dev: The device for which the dma_addr is to be created
4347  * @page: The page to be mapped
4348  * @offset: The offset within the page
4349  * @size: The size of the region in bytes
4350  * @direction: The direction of the DMA
4351  */
4352 static inline u64 ib_dma_map_page(struct ib_device *dev,
4353 				  struct page *page,
4354 				  unsigned long offset,
4355 				  size_t size,
4356 					 enum dma_data_direction direction)
4357 {
4358 	if (ib_uses_virt_dma(dev))
4359 		return (uintptr_t)(page_address(page) + offset);
4360 	return dma_map_page(dev->dma_device, page, offset, size, direction);
4361 }
4362 
4363 /**
4364  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4365  * @dev: The device for which the DMA address was created
4366  * @addr: The DMA address
4367  * @size: The size of the region in bytes
4368  * @direction: The direction of the DMA
4369  */
4370 static inline void ib_dma_unmap_page(struct ib_device *dev,
4371 				     u64 addr, size_t size,
4372 				     enum dma_data_direction direction)
4373 {
4374 	if (!ib_uses_virt_dma(dev))
4375 		dma_unmap_page(dev->dma_device, addr, size, direction);
4376 }
4377 
4378 /**
4379  * ib_dma_map_bvec - Map a bio_vec to DMA address
4380  * @dev: The device for which the dma_addr is to be created
4381  * @bvec: The bio_vec to map
4382  * @direction: The direction of the DMA
4383  *
4384  * Returns a DMA address for the bio_vec. The caller must check the
4385  * result with ib_dma_mapping_error() before use; a failed mapping
4386  * must not be passed to ib_dma_unmap_bvec().
4387  *
4388  * For software RDMA devices (rxe, siw), returns a virtual address
4389  * and no actual DMA mapping occurs.
4390  */
4391 static inline u64 ib_dma_map_bvec(struct ib_device *dev,
4392 				  struct bio_vec *bvec,
4393 				  enum dma_data_direction direction)
4394 {
4395 	if (ib_uses_virt_dma(dev))
4396 		return (uintptr_t)bvec_virt(bvec);
4397 	return dma_map_phys(dev->dma_device, bvec_phys(bvec),
4398 			    bvec->bv_len, direction, 0);
4399 }
4400 
4401 /**
4402  * ib_dma_unmap_bvec - Unmap a bio_vec DMA mapping
4403  * @dev: The device for which the DMA address was created
4404  * @addr: The DMA address returned by ib_dma_map_bvec()
4405  * @size: The size of the region in bytes
4406  * @direction: The direction of the DMA
4407  *
4408  * Releases a DMA mapping created by ib_dma_map_bvec(). For software
4409  * RDMA devices this is a no-op since no actual mapping occurred.
4410  */
4411 static inline void ib_dma_unmap_bvec(struct ib_device *dev,
4412 				     u64 addr, size_t size,
4413 				     enum dma_data_direction direction)
4414 {
4415 	if (!ib_uses_virt_dma(dev))
4416 		dma_unmap_phys(dev->dma_device, addr, size, direction, 0);
4417 }
4418 
4419 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4420 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4421 				      struct scatterlist *sg, int nents,
4422 				      enum dma_data_direction direction,
4423 				      unsigned long dma_attrs)
4424 {
4425 	if (ib_uses_virt_dma(dev))
4426 		return ib_dma_virt_map_sg(dev, sg, nents);
4427 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4428 				dma_attrs);
4429 }
4430 
4431 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4432 					 struct scatterlist *sg, int nents,
4433 					 enum dma_data_direction direction,
4434 					 unsigned long dma_attrs)
4435 {
4436 	if (!ib_uses_virt_dma(dev))
4437 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4438 				   dma_attrs);
4439 }
4440 
4441 /**
4442  * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4443  * @dev: The device for which the DMA addresses are to be created
4444  * @sgt: The sg_table object describing the buffer
4445  * @direction: The direction of the DMA
4446  * @dma_attrs: Optional DMA attributes for the map operation
4447  */
4448 static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4449 					   struct sg_table *sgt,
4450 					   enum dma_data_direction direction,
4451 					   unsigned long dma_attrs)
4452 {
4453 	int nents;
4454 
4455 	if (ib_uses_virt_dma(dev)) {
4456 		nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4457 		if (!nents)
4458 			return -EIO;
4459 		sgt->nents = nents;
4460 		return 0;
4461 	}
4462 	return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4463 }
4464 
4465 static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4466 					      struct sg_table *sgt,
4467 					      enum dma_data_direction direction,
4468 					      unsigned long dma_attrs)
4469 {
4470 	if (!ib_uses_virt_dma(dev))
4471 		dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4472 }
4473 
4474 /**
4475  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4476  * @dev: The device for which the DMA addresses are to be created
4477  * @sg: The array of scatter/gather entries
4478  * @nents: The number of scatter/gather entries
4479  * @direction: The direction of the DMA
4480  */
4481 static inline int ib_dma_map_sg(struct ib_device *dev,
4482 				struct scatterlist *sg, int nents,
4483 				enum dma_data_direction direction)
4484 {
4485 	return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4486 }
4487 
4488 /**
4489  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4490  * @dev: The device for which the DMA addresses were created
4491  * @sg: The array of scatter/gather entries
4492  * @nents: The number of scatter/gather entries
4493  * @direction: The direction of the DMA
4494  */
4495 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4496 				   struct scatterlist *sg, int nents,
4497 				   enum dma_data_direction direction)
4498 {
4499 	ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4500 }
4501 
4502 /**
4503  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4504  * @dev: The device to query
4505  *
4506  * The returned value represents a size in bytes.
4507  */
4508 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4509 {
4510 	if (ib_uses_virt_dma(dev))
4511 		return UINT_MAX;
4512 	return dma_get_max_seg_size(dev->dma_device);
4513 }
4514 
4515 /**
4516  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4517  * @dev: The device for which the DMA address was created
4518  * @addr: The DMA address
4519  * @size: The size of the region in bytes
4520  * @dir: The direction of the DMA
4521  */
4522 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4523 					      u64 addr,
4524 					      size_t size,
4525 					      enum dma_data_direction dir)
4526 {
4527 	if (!ib_uses_virt_dma(dev))
4528 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4529 }
4530 
4531 /**
4532  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4533  * @dev: The device for which the DMA address was created
4534  * @addr: The DMA address
4535  * @size: The size of the region in bytes
4536  * @dir: The direction of the DMA
4537  */
4538 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4539 						 u64 addr,
4540 						 size_t size,
4541 						 enum dma_data_direction dir)
4542 {
4543 	if (!ib_uses_virt_dma(dev))
4544 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4545 }
4546 
4547 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4548  * space. This function should be called when 'current' is the owning MM.
4549  */
4550 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4551 			     u64 virt_addr, int mr_access_flags);
4552 
4553 /* ib_advise_mr -  give an advice about an address range in a memory region */
4554 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4555 		 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4556 /**
4557  * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4558  *   HCA translation table.
4559  * @mr: The memory region to deregister.
4560  * @udata: Valid user data or NULL for kernel object
4561  *
4562  * This function can fail, if the memory region has memory windows bound to it.
4563  */
4564 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4565 
4566 /**
4567  * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4568  *   HCA translation table.
4569  * @mr: The memory region to deregister.
4570  *
4571  * This function can fail, if the memory region has memory windows bound to it.
4572  *
4573  * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4574  */
4575 static inline int ib_dereg_mr(struct ib_mr *mr)
4576 {
4577 	return ib_dereg_mr_user(mr, NULL);
4578 }
4579 
4580 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4581 			  u32 max_num_sg);
4582 
4583 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4584 				    u32 max_num_data_sg,
4585 				    u32 max_num_meta_sg);
4586 
4587 /**
4588  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4589  *   R_Key and L_Key.
4590  * @mr: struct ib_mr pointer to be updated.
4591  * @newkey: new key to be used.
4592  */
4593 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4594 {
4595 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4596 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4597 }
4598 
4599 /**
4600  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4601  * for calculating a new rkey for type 2 memory windows.
4602  * @rkey: the rkey to increment.
4603  */
4604 static inline u32 ib_inc_rkey(u32 rkey)
4605 {
4606 	const u32 mask = 0x000000ff;
4607 	return ((rkey + 1) & mask) | (rkey & ~mask);
4608 }
4609 
4610 /**
4611  * ib_attach_mcast - Attaches the specified QP to a multicast group.
4612  * @qp: QP to attach to the multicast group.  The QP must be type
4613  *   IB_QPT_UD.
4614  * @gid: Multicast group GID.
4615  * @lid: Multicast group LID in host byte order.
4616  *
4617  * In order to send and receive multicast packets, subnet
4618  * administration must have created the multicast group and configured
4619  * the fabric appropriately.  The port associated with the specified
4620  * QP must also be a member of the multicast group.
4621  */
4622 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4623 
4624 /**
4625  * ib_detach_mcast - Detaches the specified QP from a multicast group.
4626  * @qp: QP to detach from the multicast group.
4627  * @gid: Multicast group GID.
4628  * @lid: Multicast group LID in host byte order.
4629  */
4630 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4631 
4632 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4633 				   struct inode *inode, struct ib_udata *udata);
4634 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4635 
4636 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4637 				     unsigned int flags)
4638 {
4639 	u64 device_cap = ib_dev->attrs.device_cap_flags;
4640 
4641 	/*
4642 	 * Local write permission is required if remote write or
4643 	 * remote atomic permission is also requested.
4644 	 */
4645 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4646 	    !(flags & IB_ACCESS_LOCAL_WRITE))
4647 		return -EINVAL;
4648 
4649 	if (flags & ~IB_ACCESS_SUPPORTED)
4650 		return -EINVAL;
4651 
4652 	if (flags & IB_ACCESS_ON_DEMAND &&
4653 	    !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING))
4654 		return -EOPNOTSUPP;
4655 
4656 	if ((flags & IB_ACCESS_FLUSH_GLOBAL &&
4657 	    !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) ||
4658 	    (flags & IB_ACCESS_FLUSH_PERSISTENT &&
4659 	    !(device_cap & IB_DEVICE_FLUSH_PERSISTENT)))
4660 		return -EOPNOTSUPP;
4661 
4662 	return 0;
4663 }
4664 
4665 static inline bool ib_access_writable(int access_flags)
4666 {
4667 	/*
4668 	 * We have writable memory backing the MR if any of the following
4669 	 * access flags are set.  "Local write" and "remote write" obviously
4670 	 * require write access.  "Remote atomic" can do things like fetch and
4671 	 * add, which will modify memory, and "MW bind" can change permissions
4672 	 * by binding a window.
4673 	 */
4674 	return access_flags &
4675 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
4676 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4677 }
4678 
4679 /**
4680  * ib_check_mr_status: lightweight check of MR status.
4681  *     This routine may provide status checks on a selected
4682  *     ib_mr. first use is for signature status check.
4683  *
4684  * @mr: A memory region.
4685  * @check_mask: Bitmask of which checks to perform from
4686  *     ib_mr_status_check enumeration.
4687  * @mr_status: The container of relevant status checks.
4688  *     failed checks will be indicated in the status bitmask
4689  *     and the relevant info shall be in the error item.
4690  */
4691 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4692 		       struct ib_mr_status *mr_status);
4693 
4694 /**
4695  * ib_device_try_get: Hold a registration lock
4696  * @dev: The device to lock
4697  *
4698  * A device under an active registration lock cannot become unregistered. It
4699  * is only possible to obtain a registration lock on a device that is fully
4700  * registered, otherwise this function returns false.
4701  *
4702  * The registration lock is only necessary for actions which require the
4703  * device to still be registered. Uses that only require the device pointer to
4704  * be valid should use get_device(&ibdev->dev) to hold the memory.
4705  *
4706  */
4707 static inline bool ib_device_try_get(struct ib_device *dev)
4708 {
4709 	return refcount_inc_not_zero(&dev->refcount);
4710 }
4711 
4712 void ib_device_put(struct ib_device *device);
4713 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4714 					  enum rdma_driver_id driver_id);
4715 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4716 					    u16 pkey, const union ib_gid *gid,
4717 					    const struct sockaddr *addr);
4718 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4719 			 unsigned int port);
4720 struct net_device *ib_device_get_netdev(struct ib_device *ib_dev,
4721 					u32 port);
4722 int ib_query_netdev_port(struct ib_device *ibdev, struct net_device *ndev,
4723 			 u32 *port);
4724 
4725 static inline enum ib_port_state ib_get_curr_port_state(struct net_device *net_dev)
4726 {
4727 	return (netif_running(net_dev) && netif_carrier_ok(net_dev)) ?
4728 		IB_PORT_ACTIVE : IB_PORT_DOWN;
4729 }
4730 
4731 void ib_dispatch_port_state_event(struct ib_device *ibdev,
4732 				  struct net_device *ndev);
4733 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4734 			   struct ib_wq_init_attr *init_attr);
4735 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4736 
4737 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4738 		 unsigned int *sg_offset, unsigned int page_size);
4739 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4740 		    int data_sg_nents, unsigned int *data_sg_offset,
4741 		    struct scatterlist *meta_sg, int meta_sg_nents,
4742 		    unsigned int *meta_sg_offset, unsigned int page_size);
4743 
4744 static inline int
4745 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4746 		  unsigned int *sg_offset, unsigned int page_size)
4747 {
4748 	int n;
4749 
4750 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4751 	mr->iova = 0;
4752 
4753 	return n;
4754 }
4755 
4756 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4757 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4758 
4759 void ib_drain_rq(struct ib_qp *qp);
4760 void ib_drain_sq(struct ib_qp *qp);
4761 void ib_drain_qp(struct ib_qp *qp);
4762 
4763 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4764 		     u8 *width);
4765 
4766 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4767 {
4768 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4769 		return attr->roce.dmac;
4770 	return NULL;
4771 }
4772 
4773 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4774 {
4775 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4776 		attr->ib.dlid = (u16)dlid;
4777 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4778 		attr->opa.dlid = dlid;
4779 }
4780 
4781 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4782 {
4783 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4784 		return attr->ib.dlid;
4785 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4786 		return attr->opa.dlid;
4787 	return 0;
4788 }
4789 
4790 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4791 {
4792 	attr->sl = sl;
4793 }
4794 
4795 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4796 {
4797 	return attr->sl;
4798 }
4799 
4800 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4801 					 u8 src_path_bits)
4802 {
4803 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4804 		attr->ib.src_path_bits = src_path_bits;
4805 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4806 		attr->opa.src_path_bits = src_path_bits;
4807 }
4808 
4809 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4810 {
4811 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4812 		return attr->ib.src_path_bits;
4813 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4814 		return attr->opa.src_path_bits;
4815 	return 0;
4816 }
4817 
4818 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4819 					bool make_grd)
4820 {
4821 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4822 		attr->opa.make_grd = make_grd;
4823 }
4824 
4825 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4826 {
4827 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4828 		return attr->opa.make_grd;
4829 	return false;
4830 }
4831 
4832 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4833 {
4834 	attr->port_num = port_num;
4835 }
4836 
4837 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4838 {
4839 	return attr->port_num;
4840 }
4841 
4842 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4843 					   u8 static_rate)
4844 {
4845 	attr->static_rate = static_rate;
4846 }
4847 
4848 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4849 {
4850 	return attr->static_rate;
4851 }
4852 
4853 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4854 					enum ib_ah_flags flag)
4855 {
4856 	attr->ah_flags = flag;
4857 }
4858 
4859 static inline enum ib_ah_flags
4860 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4861 {
4862 	return attr->ah_flags;
4863 }
4864 
4865 static inline const struct ib_global_route
4866 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4867 {
4868 	return &attr->grh;
4869 }
4870 
4871 /*To retrieve and modify the grh */
4872 static inline struct ib_global_route
4873 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4874 {
4875 	return &attr->grh;
4876 }
4877 
4878 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4879 {
4880 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4881 
4882 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4883 }
4884 
4885 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4886 					     __be64 prefix)
4887 {
4888 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4889 
4890 	grh->dgid.global.subnet_prefix = prefix;
4891 }
4892 
4893 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4894 					    __be64 if_id)
4895 {
4896 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4897 
4898 	grh->dgid.global.interface_id = if_id;
4899 }
4900 
4901 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4902 				   union ib_gid *dgid, u32 flow_label,
4903 				   u8 sgid_index, u8 hop_limit,
4904 				   u8 traffic_class)
4905 {
4906 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4907 
4908 	attr->ah_flags = IB_AH_GRH;
4909 	if (dgid)
4910 		grh->dgid = *dgid;
4911 	grh->flow_label = flow_label;
4912 	grh->sgid_index = sgid_index;
4913 	grh->hop_limit = hop_limit;
4914 	grh->traffic_class = traffic_class;
4915 	grh->sgid_attr = NULL;
4916 }
4917 
4918 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4919 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4920 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4921 			     const struct ib_gid_attr *sgid_attr);
4922 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4923 		       const struct rdma_ah_attr *src);
4924 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4925 			  const struct rdma_ah_attr *new);
4926 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4927 
4928 /**
4929  * rdma_ah_find_type - Return address handle type.
4930  *
4931  * @dev: Device to be checked
4932  * @port_num: Port number
4933  */
4934 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4935 						       u32 port_num)
4936 {
4937 	if (rdma_protocol_roce(dev, port_num))
4938 		return RDMA_AH_ATTR_TYPE_ROCE;
4939 	if (rdma_protocol_ib(dev, port_num)) {
4940 		if (rdma_cap_opa_ah(dev, port_num))
4941 			return RDMA_AH_ATTR_TYPE_OPA;
4942 		return RDMA_AH_ATTR_TYPE_IB;
4943 	}
4944 	if (dev->type == RDMA_DEVICE_TYPE_SMI)
4945 		return RDMA_AH_ATTR_TYPE_IB;
4946 
4947 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4948 }
4949 
4950 /**
4951  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4952  *     In the current implementation the only way to
4953  *     get the 32bit lid is from other sources for OPA.
4954  *     For IB, lids will always be 16bits so cast the
4955  *     value accordingly.
4956  *
4957  * @lid: A 32bit LID
4958  */
4959 static inline u16 ib_lid_cpu16(u32 lid)
4960 {
4961 	WARN_ON_ONCE(lid & 0xFFFF0000);
4962 	return (u16)lid;
4963 }
4964 
4965 /**
4966  * ib_lid_be16 - Return lid in 16bit BE encoding.
4967  *
4968  * @lid: A 32bit LID
4969  */
4970 static inline __be16 ib_lid_be16(u32 lid)
4971 {
4972 	WARN_ON_ONCE(lid & 0xFFFF0000);
4973 	return cpu_to_be16((u16)lid);
4974 }
4975 
4976 /**
4977  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4978  * and add their gids, as needed, to the relevant RoCE devices.
4979  *
4980  * @ibdev:         the rdma device
4981  */
4982 void rdma_roce_rescan_device(struct ib_device *ibdev);
4983 void rdma_roce_rescan_port(struct ib_device *ib_dev, u32 port);
4984 void roce_del_all_netdev_gids(struct ib_device *ib_dev,
4985 			      u32 port, struct net_device *ndev);
4986 
4987 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4988 
4989 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
4990 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4991 bool rdma_uattrs_has_raw_cap(const struct uverbs_attr_bundle *attrs);
4992 #else
4993 static inline int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs)
4994 {
4995 	return 0;
4996 }
4997 static inline bool
4998 rdma_uattrs_has_raw_cap(const struct uverbs_attr_bundle *attrs)
4999 {
5000 	return false;
5001 }
5002 #endif
5003 
5004 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
5005 				     enum rdma_netdev_t type, const char *name,
5006 				     unsigned char name_assign_type,
5007 				     void (*setup)(struct net_device *));
5008 
5009 int rdma_init_netdev(struct ib_device *device, u32 port_num,
5010 		     enum rdma_netdev_t type, const char *name,
5011 		     unsigned char name_assign_type,
5012 		     void (*setup)(struct net_device *),
5013 		     struct net_device *netdev);
5014 
5015 /**
5016  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
5017  *
5018  * @device:	device pointer for which ib_device pointer to retrieve
5019  *
5020  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
5021  *
5022  */
5023 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
5024 {
5025 	struct ib_core_device *coredev =
5026 		container_of(device, struct ib_core_device, dev);
5027 
5028 	return coredev->owner;
5029 }
5030 
5031 /**
5032  * ibdev_to_node - return the NUMA node for a given ib_device
5033  * @ibdev:	device to get the NUMA node for.
5034  */
5035 static inline int ibdev_to_node(struct ib_device *ibdev)
5036 {
5037 	struct device *parent = ibdev->dev.parent;
5038 
5039 	if (!parent)
5040 		return NUMA_NO_NODE;
5041 	return dev_to_node(parent);
5042 }
5043 
5044 /**
5045  * rdma_device_to_drv_device - Helper macro to reach back to driver's
5046  *			       ib_device holder structure from device pointer.
5047  *
5048  * NOTE: New drivers should not make use of this API; This API is only for
5049  * existing drivers who have exposed sysfs entries using
5050  * ops->device_group.
5051  */
5052 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
5053 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
5054 
5055 bool rdma_dev_access_netns(const struct ib_device *device,
5056 			   const struct net *net);
5057 
5058 bool rdma_dev_has_raw_cap(const struct ib_device *dev);
5059 static inline struct net *rdma_dev_net(struct ib_device *device)
5060 {
5061 	return read_pnet(&device->coredev.rdma_net);
5062 }
5063 
5064 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
5065 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
5066 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
5067 
5068 /**
5069  * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
5070  *                               on the flow_label
5071  * @fl: flow_label value
5072  *
5073  * This function will convert the 20 bit flow_label input to a valid RoCE v2
5074  * UDP src port 14 bit value. All RoCE V2 drivers should use this same
5075  * convention.
5076  */
5077 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
5078 {
5079 	u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
5080 
5081 	fl_low ^= fl_high >> 14;
5082 	return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
5083 }
5084 
5085 /**
5086  * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
5087  *                        local and remote qpn values
5088  *
5089  * This function folded the multiplication results of two qpns, 24 bit each,
5090  * fields, and converts it to a 20 bit results.
5091  *
5092  * This function will create symmetric flow_label value based on the local
5093  * and remote qpn values. this will allow both the requester and responder
5094  * to calculate the same flow_label for a given connection.
5095  *
5096  * This helper function should be used by driver in case the upper layer
5097  * provide a zero flow_label value. This is to improve entropy of RDMA
5098  * traffic in the network.
5099  */
5100 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
5101 {
5102 	u64 v = (u64)lqpn * rqpn;
5103 
5104 	v ^= v >> 20;
5105 	v ^= v >> 40;
5106 
5107 	return (u32)(v & IB_GRH_FLOWLABEL_MASK);
5108 }
5109 
5110 /**
5111  * rdma_get_udp_sport - Calculate and set UDP source port based on the flow
5112  *                      label. If flow label is not defined in GRH then
5113  *                      calculate it based on lqpn/rqpn.
5114  *
5115  * @fl:                 flow label from GRH
5116  * @lqpn:               local qp number
5117  * @rqpn:               remote qp number
5118  */
5119 static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
5120 {
5121 	if (!fl)
5122 		fl = rdma_calc_flow_label(lqpn, rqpn);
5123 
5124 	return rdma_flow_label_to_udp_sport(fl);
5125 }
5126 
5127 const struct ib_port_immutable*
5128 ib_port_immutable_read(struct ib_device *dev, unsigned int port);
5129 
5130 /** ib_add_sub_device - Add a sub IB device on an existing one
5131  *
5132  * @parent: The IB device that needs to add a sub device
5133  * @type: The type of the new sub device
5134  * @name: The name of the new sub device
5135  *
5136  *
5137  * Return 0 on success, an error code otherwise
5138  */
5139 int ib_add_sub_device(struct ib_device *parent,
5140 		      enum rdma_nl_dev_type type,
5141 		      const char *name);
5142 
5143 
5144 /** ib_del_sub_device_and_put - Delect an IB sub device while holding a 'get'
5145  *
5146  * @sub: The sub device that is going to be deleted
5147  *
5148  * Return 0 on success, an error code otherwise
5149  */
5150 int ib_del_sub_device_and_put(struct ib_device *sub);
5151 
5152 static inline void ib_mark_name_assigned_by_user(struct ib_device *ibdev)
5153 {
5154 	ibdev->name_assign_type = RDMA_NAME_ASSIGN_TYPE_USER;
5155 }
5156 
5157 #endif /* IB_VERBS_H */
5158