xref: /linux/include/rdma/ib_verbs.h (revision fe7fad476ec8153a8b8767a08114e3e4a58a837e)
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/dma-mapping.h>
19 #include <linux/kref.h>
20 #include <linux/list.h>
21 #include <linux/rwsem.h>
22 #include <linux/workqueue.h>
23 #include <linux/irq_poll.h>
24 #include <uapi/linux/if_ether.h>
25 #include <net/ipv6.h>
26 #include <net/ip.h>
27 #include <linux/string.h>
28 #include <linux/slab.h>
29 #include <linux/netdevice.h>
30 #include <linux/refcount.h>
31 #include <linux/if_link.h>
32 #include <linux/atomic.h>
33 #include <linux/mmu_notifier.h>
34 #include <linux/uaccess.h>
35 #include <linux/cgroup_rdma.h>
36 #include <linux/irqflags.h>
37 #include <linux/preempt.h>
38 #include <linux/dim.h>
39 #include <uapi/rdma/ib_user_verbs.h>
40 #include <rdma/rdma_counter.h>
41 #include <rdma/restrack.h>
42 #include <rdma/signature.h>
43 #include <uapi/rdma/rdma_user_ioctl.h>
44 #include <uapi/rdma/ib_user_ioctl_verbs.h>
45 
46 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
47 
48 struct ib_umem_odp;
49 struct ib_uqp_object;
50 struct ib_usrq_object;
51 struct ib_uwq_object;
52 struct rdma_cm_id;
53 struct ib_port;
54 struct hw_stats_device_data;
55 
56 extern struct workqueue_struct *ib_wq;
57 extern struct workqueue_struct *ib_comp_wq;
58 extern struct workqueue_struct *ib_comp_unbound_wq;
59 
60 struct ib_ucq_object;
61 
62 __printf(2, 3) __cold
63 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
64 __printf(2, 3) __cold
65 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
66 __printf(2, 3) __cold
67 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
68 __printf(2, 3) __cold
69 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
70 __printf(2, 3) __cold
71 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
72 __printf(2, 3) __cold
73 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
74 __printf(2, 3) __cold
75 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
76 
77 #if defined(CONFIG_DYNAMIC_DEBUG) || \
78 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
79 #define ibdev_dbg(__dev, format, args...)                       \
80 	dynamic_ibdev_dbg(__dev, format, ##args)
81 #else
82 __printf(2, 3) __cold
83 static inline
84 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
85 #endif
86 
87 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...)           \
88 do {                                                                    \
89 	static DEFINE_RATELIMIT_STATE(_rs,                              \
90 				      DEFAULT_RATELIMIT_INTERVAL,       \
91 				      DEFAULT_RATELIMIT_BURST);         \
92 	if (__ratelimit(&_rs))                                          \
93 		ibdev_level(ibdev, fmt, ##__VA_ARGS__);                 \
94 } while (0)
95 
96 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
97 	ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
98 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
99 	ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
100 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
101 	ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
102 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
103 	ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
104 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
105 	ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
106 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
107 	ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
108 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
109 	ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
110 
111 #if defined(CONFIG_DYNAMIC_DEBUG) || \
112 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
113 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
114 #define ibdev_dbg_ratelimited(ibdev, fmt, ...)                          \
115 do {                                                                    \
116 	static DEFINE_RATELIMIT_STATE(_rs,                              \
117 				      DEFAULT_RATELIMIT_INTERVAL,       \
118 				      DEFAULT_RATELIMIT_BURST);         \
119 	DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt);                 \
120 	if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs))      \
121 		__dynamic_ibdev_dbg(&descriptor, ibdev, fmt,            \
122 				    ##__VA_ARGS__);                     \
123 } while (0)
124 #else
125 __printf(2, 3) __cold
126 static inline
127 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
128 #endif
129 
130 union ib_gid {
131 	u8	raw[16];
132 	struct {
133 		__be64	subnet_prefix;
134 		__be64	interface_id;
135 	} global;
136 };
137 
138 extern union ib_gid zgid;
139 
140 enum ib_gid_type {
141 	IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
142 	IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
143 	IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
144 	IB_GID_TYPE_SIZE
145 };
146 
147 #define ROCE_V2_UDP_DPORT      4791
148 struct ib_gid_attr {
149 	struct net_device __rcu	*ndev;
150 	struct ib_device	*device;
151 	union ib_gid		gid;
152 	enum ib_gid_type	gid_type;
153 	u16			index;
154 	u32			port_num;
155 };
156 
157 enum {
158 	/* set the local administered indication */
159 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
160 };
161 
162 enum rdma_transport_type {
163 	RDMA_TRANSPORT_IB,
164 	RDMA_TRANSPORT_IWARP,
165 	RDMA_TRANSPORT_USNIC,
166 	RDMA_TRANSPORT_USNIC_UDP,
167 	RDMA_TRANSPORT_UNSPECIFIED,
168 };
169 
170 enum rdma_protocol_type {
171 	RDMA_PROTOCOL_IB,
172 	RDMA_PROTOCOL_IBOE,
173 	RDMA_PROTOCOL_IWARP,
174 	RDMA_PROTOCOL_USNIC_UDP
175 };
176 
177 __attribute_const__ enum rdma_transport_type
178 rdma_node_get_transport(unsigned int node_type);
179 
180 enum rdma_network_type {
181 	RDMA_NETWORK_IB,
182 	RDMA_NETWORK_ROCE_V1,
183 	RDMA_NETWORK_IPV4,
184 	RDMA_NETWORK_IPV6
185 };
186 
187 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
188 {
189 	if (network_type == RDMA_NETWORK_IPV4 ||
190 	    network_type == RDMA_NETWORK_IPV6)
191 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
192 	else if (network_type == RDMA_NETWORK_ROCE_V1)
193 		return IB_GID_TYPE_ROCE;
194 	else
195 		return IB_GID_TYPE_IB;
196 }
197 
198 static inline enum rdma_network_type
199 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
200 {
201 	if (attr->gid_type == IB_GID_TYPE_IB)
202 		return RDMA_NETWORK_IB;
203 
204 	if (attr->gid_type == IB_GID_TYPE_ROCE)
205 		return RDMA_NETWORK_ROCE_V1;
206 
207 	if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
208 		return RDMA_NETWORK_IPV4;
209 	else
210 		return RDMA_NETWORK_IPV6;
211 }
212 
213 enum rdma_link_layer {
214 	IB_LINK_LAYER_UNSPECIFIED,
215 	IB_LINK_LAYER_INFINIBAND,
216 	IB_LINK_LAYER_ETHERNET,
217 };
218 
219 enum ib_device_cap_flags {
220 	IB_DEVICE_RESIZE_MAX_WR = IB_UVERBS_DEVICE_RESIZE_MAX_WR,
221 	IB_DEVICE_BAD_PKEY_CNTR = IB_UVERBS_DEVICE_BAD_PKEY_CNTR,
222 	IB_DEVICE_BAD_QKEY_CNTR = IB_UVERBS_DEVICE_BAD_QKEY_CNTR,
223 	IB_DEVICE_RAW_MULTI = IB_UVERBS_DEVICE_RAW_MULTI,
224 	IB_DEVICE_AUTO_PATH_MIG = IB_UVERBS_DEVICE_AUTO_PATH_MIG,
225 	IB_DEVICE_CHANGE_PHY_PORT = IB_UVERBS_DEVICE_CHANGE_PHY_PORT,
226 	IB_DEVICE_UD_AV_PORT_ENFORCE = IB_UVERBS_DEVICE_UD_AV_PORT_ENFORCE,
227 	IB_DEVICE_CURR_QP_STATE_MOD = IB_UVERBS_DEVICE_CURR_QP_STATE_MOD,
228 	IB_DEVICE_SHUTDOWN_PORT = IB_UVERBS_DEVICE_SHUTDOWN_PORT,
229 	/* IB_DEVICE_INIT_TYPE = IB_UVERBS_DEVICE_INIT_TYPE, (not in use) */
230 	IB_DEVICE_PORT_ACTIVE_EVENT = IB_UVERBS_DEVICE_PORT_ACTIVE_EVENT,
231 	IB_DEVICE_SYS_IMAGE_GUID = IB_UVERBS_DEVICE_SYS_IMAGE_GUID,
232 	IB_DEVICE_RC_RNR_NAK_GEN = IB_UVERBS_DEVICE_RC_RNR_NAK_GEN,
233 	IB_DEVICE_SRQ_RESIZE = IB_UVERBS_DEVICE_SRQ_RESIZE,
234 	IB_DEVICE_N_NOTIFY_CQ = IB_UVERBS_DEVICE_N_NOTIFY_CQ,
235 
236 	/* Reserved, old SEND_W_INV = 1 << 16,*/
237 	IB_DEVICE_MEM_WINDOW = IB_UVERBS_DEVICE_MEM_WINDOW,
238 	/*
239 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
240 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
241 	 * messages and can verify the validity of checksum for
242 	 * incoming messages.  Setting this flag implies that the
243 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
244 	 */
245 	IB_DEVICE_UD_IP_CSUM = IB_UVERBS_DEVICE_UD_IP_CSUM,
246 	IB_DEVICE_XRC = IB_UVERBS_DEVICE_XRC,
247 
248 	/*
249 	 * This device supports the IB "base memory management extension",
250 	 * which includes support for fast registrations (IB_WR_REG_MR,
251 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
252 	 * also be set by any iWarp device which must support FRs to comply
253 	 * to the iWarp verbs spec.  iWarp devices also support the
254 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
255 	 * stag.
256 	 */
257 	IB_DEVICE_MEM_MGT_EXTENSIONS = IB_UVERBS_DEVICE_MEM_MGT_EXTENSIONS,
258 	IB_DEVICE_MEM_WINDOW_TYPE_2A = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2A,
259 	IB_DEVICE_MEM_WINDOW_TYPE_2B = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2B,
260 	IB_DEVICE_RC_IP_CSUM = IB_UVERBS_DEVICE_RC_IP_CSUM,
261 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
262 	IB_DEVICE_RAW_IP_CSUM = IB_UVERBS_DEVICE_RAW_IP_CSUM,
263 	IB_DEVICE_MANAGED_FLOW_STEERING =
264 		IB_UVERBS_DEVICE_MANAGED_FLOW_STEERING,
265 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
266 	IB_DEVICE_RAW_SCATTER_FCS = IB_UVERBS_DEVICE_RAW_SCATTER_FCS,
267 	/* The device supports padding incoming writes to cacheline. */
268 	IB_DEVICE_PCI_WRITE_END_PADDING =
269 		IB_UVERBS_DEVICE_PCI_WRITE_END_PADDING,
270 	/* Placement type attributes */
271 	IB_DEVICE_FLUSH_GLOBAL = IB_UVERBS_DEVICE_FLUSH_GLOBAL,
272 	IB_DEVICE_FLUSH_PERSISTENT = IB_UVERBS_DEVICE_FLUSH_PERSISTENT,
273 	IB_DEVICE_ATOMIC_WRITE = IB_UVERBS_DEVICE_ATOMIC_WRITE,
274 };
275 
276 enum ib_kernel_cap_flags {
277 	/*
278 	 * This device supports a per-device lkey or stag that can be
279 	 * used without performing a memory registration for the local
280 	 * memory.  Note that ULPs should never check this flag, but
281 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
282 	 * which will always contain a usable lkey.
283 	 */
284 	IBK_LOCAL_DMA_LKEY = 1 << 0,
285 	/* IB_QP_CREATE_INTEGRITY_EN is supported to implement T10-PI */
286 	IBK_INTEGRITY_HANDOVER = 1 << 1,
287 	/* IB_ACCESS_ON_DEMAND is supported during reg_user_mr() */
288 	IBK_ON_DEMAND_PAGING = 1 << 2,
289 	/* IB_MR_TYPE_SG_GAPS is supported */
290 	IBK_SG_GAPS_REG = 1 << 3,
291 	/* Driver supports RDMA_NLDEV_CMD_DELLINK */
292 	IBK_ALLOW_USER_UNREG = 1 << 4,
293 
294 	/* ipoib will use IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK */
295 	IBK_BLOCK_MULTICAST_LOOPBACK = 1 << 5,
296 	/* iopib will use IB_QP_CREATE_IPOIB_UD_LSO for its QPs */
297 	IBK_UD_TSO = 1 << 6,
298 	/* iopib will use the device ops:
299 	 *   get_vf_config
300 	 *   get_vf_guid
301 	 *   get_vf_stats
302 	 *   set_vf_guid
303 	 *   set_vf_link_state
304 	 */
305 	IBK_VIRTUAL_FUNCTION = 1 << 7,
306 	/* ipoib will use IB_QP_CREATE_NETDEV_USE for its QPs */
307 	IBK_RDMA_NETDEV_OPA = 1 << 8,
308 };
309 
310 enum ib_atomic_cap {
311 	IB_ATOMIC_NONE,
312 	IB_ATOMIC_HCA,
313 	IB_ATOMIC_GLOB
314 };
315 
316 enum ib_odp_general_cap_bits {
317 	IB_ODP_SUPPORT		= 1 << 0,
318 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
319 };
320 
321 enum ib_odp_transport_cap_bits {
322 	IB_ODP_SUPPORT_SEND	= 1 << 0,
323 	IB_ODP_SUPPORT_RECV	= 1 << 1,
324 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
325 	IB_ODP_SUPPORT_READ	= 1 << 3,
326 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
327 	IB_ODP_SUPPORT_SRQ_RECV	= 1 << 5,
328 };
329 
330 struct ib_odp_caps {
331 	uint64_t general_caps;
332 	struct {
333 		uint32_t  rc_odp_caps;
334 		uint32_t  uc_odp_caps;
335 		uint32_t  ud_odp_caps;
336 		uint32_t  xrc_odp_caps;
337 	} per_transport_caps;
338 };
339 
340 struct ib_rss_caps {
341 	/* Corresponding bit will be set if qp type from
342 	 * 'enum ib_qp_type' is supported, e.g.
343 	 * supported_qpts |= 1 << IB_QPT_UD
344 	 */
345 	u32 supported_qpts;
346 	u32 max_rwq_indirection_tables;
347 	u32 max_rwq_indirection_table_size;
348 };
349 
350 enum ib_tm_cap_flags {
351 	/*  Support tag matching with rendezvous offload for RC transport */
352 	IB_TM_CAP_RNDV_RC = 1 << 0,
353 };
354 
355 struct ib_tm_caps {
356 	/* Max size of RNDV header */
357 	u32 max_rndv_hdr_size;
358 	/* Max number of entries in tag matching list */
359 	u32 max_num_tags;
360 	/* From enum ib_tm_cap_flags */
361 	u32 flags;
362 	/* Max number of outstanding list operations */
363 	u32 max_ops;
364 	/* Max number of SGE in tag matching entry */
365 	u32 max_sge;
366 };
367 
368 struct ib_cq_init_attr {
369 	unsigned int	cqe;
370 	u32		comp_vector;
371 	u32		flags;
372 };
373 
374 enum ib_cq_attr_mask {
375 	IB_CQ_MODERATE = 1 << 0,
376 };
377 
378 struct ib_cq_caps {
379 	u16     max_cq_moderation_count;
380 	u16     max_cq_moderation_period;
381 };
382 
383 struct ib_dm_mr_attr {
384 	u64		length;
385 	u64		offset;
386 	u32		access_flags;
387 };
388 
389 struct ib_dm_alloc_attr {
390 	u64	length;
391 	u32	alignment;
392 	u32	flags;
393 };
394 
395 struct ib_device_attr {
396 	u64			fw_ver;
397 	__be64			sys_image_guid;
398 	u64			max_mr_size;
399 	u64			page_size_cap;
400 	u32			vendor_id;
401 	u32			vendor_part_id;
402 	u32			hw_ver;
403 	int			max_qp;
404 	int			max_qp_wr;
405 	u64			device_cap_flags;
406 	u64			kernel_cap_flags;
407 	int			max_send_sge;
408 	int			max_recv_sge;
409 	int			max_sge_rd;
410 	int			max_cq;
411 	int			max_cqe;
412 	int			max_mr;
413 	int			max_pd;
414 	int			max_qp_rd_atom;
415 	int			max_ee_rd_atom;
416 	int			max_res_rd_atom;
417 	int			max_qp_init_rd_atom;
418 	int			max_ee_init_rd_atom;
419 	enum ib_atomic_cap	atomic_cap;
420 	enum ib_atomic_cap	masked_atomic_cap;
421 	int			max_ee;
422 	int			max_rdd;
423 	int			max_mw;
424 	int			max_raw_ipv6_qp;
425 	int			max_raw_ethy_qp;
426 	int			max_mcast_grp;
427 	int			max_mcast_qp_attach;
428 	int			max_total_mcast_qp_attach;
429 	int			max_ah;
430 	int			max_srq;
431 	int			max_srq_wr;
432 	int			max_srq_sge;
433 	unsigned int		max_fast_reg_page_list_len;
434 	unsigned int		max_pi_fast_reg_page_list_len;
435 	u16			max_pkeys;
436 	u8			local_ca_ack_delay;
437 	int			sig_prot_cap;
438 	int			sig_guard_cap;
439 	struct ib_odp_caps	odp_caps;
440 	uint64_t		timestamp_mask;
441 	uint64_t		hca_core_clock; /* in KHZ */
442 	struct ib_rss_caps	rss_caps;
443 	u32			max_wq_type_rq;
444 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
445 	struct ib_tm_caps	tm_caps;
446 	struct ib_cq_caps       cq_caps;
447 	u64			max_dm_size;
448 	/* Max entries for sgl for optimized performance per READ */
449 	u32			max_sgl_rd;
450 };
451 
452 enum ib_mtu {
453 	IB_MTU_256  = 1,
454 	IB_MTU_512  = 2,
455 	IB_MTU_1024 = 3,
456 	IB_MTU_2048 = 4,
457 	IB_MTU_4096 = 5
458 };
459 
460 enum opa_mtu {
461 	OPA_MTU_8192 = 6,
462 	OPA_MTU_10240 = 7
463 };
464 
465 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
466 {
467 	switch (mtu) {
468 	case IB_MTU_256:  return  256;
469 	case IB_MTU_512:  return  512;
470 	case IB_MTU_1024: return 1024;
471 	case IB_MTU_2048: return 2048;
472 	case IB_MTU_4096: return 4096;
473 	default: 	  return -1;
474 	}
475 }
476 
477 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
478 {
479 	if (mtu >= 4096)
480 		return IB_MTU_4096;
481 	else if (mtu >= 2048)
482 		return IB_MTU_2048;
483 	else if (mtu >= 1024)
484 		return IB_MTU_1024;
485 	else if (mtu >= 512)
486 		return IB_MTU_512;
487 	else
488 		return IB_MTU_256;
489 }
490 
491 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
492 {
493 	switch (mtu) {
494 	case OPA_MTU_8192:
495 		return 8192;
496 	case OPA_MTU_10240:
497 		return 10240;
498 	default:
499 		return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
500 	}
501 }
502 
503 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
504 {
505 	if (mtu >= 10240)
506 		return OPA_MTU_10240;
507 	else if (mtu >= 8192)
508 		return OPA_MTU_8192;
509 	else
510 		return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
511 }
512 
513 enum ib_port_state {
514 	IB_PORT_NOP		= 0,
515 	IB_PORT_DOWN		= 1,
516 	IB_PORT_INIT		= 2,
517 	IB_PORT_ARMED		= 3,
518 	IB_PORT_ACTIVE		= 4,
519 	IB_PORT_ACTIVE_DEFER	= 5
520 };
521 
522 enum ib_port_phys_state {
523 	IB_PORT_PHYS_STATE_SLEEP = 1,
524 	IB_PORT_PHYS_STATE_POLLING = 2,
525 	IB_PORT_PHYS_STATE_DISABLED = 3,
526 	IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
527 	IB_PORT_PHYS_STATE_LINK_UP = 5,
528 	IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
529 	IB_PORT_PHYS_STATE_PHY_TEST = 7,
530 };
531 
532 enum ib_port_width {
533 	IB_WIDTH_1X	= 1,
534 	IB_WIDTH_2X	= 16,
535 	IB_WIDTH_4X	= 2,
536 	IB_WIDTH_8X	= 4,
537 	IB_WIDTH_12X	= 8
538 };
539 
540 static inline int ib_width_enum_to_int(enum ib_port_width width)
541 {
542 	switch (width) {
543 	case IB_WIDTH_1X:  return  1;
544 	case IB_WIDTH_2X:  return  2;
545 	case IB_WIDTH_4X:  return  4;
546 	case IB_WIDTH_8X:  return  8;
547 	case IB_WIDTH_12X: return 12;
548 	default: 	  return -1;
549 	}
550 }
551 
552 enum ib_port_speed {
553 	IB_SPEED_SDR	= 1,
554 	IB_SPEED_DDR	= 2,
555 	IB_SPEED_QDR	= 4,
556 	IB_SPEED_FDR10	= 8,
557 	IB_SPEED_FDR	= 16,
558 	IB_SPEED_EDR	= 32,
559 	IB_SPEED_HDR	= 64,
560 	IB_SPEED_NDR	= 128,
561 	IB_SPEED_XDR	= 256,
562 };
563 
564 enum ib_stat_flag {
565 	IB_STAT_FLAG_OPTIONAL = 1 << 0,
566 };
567 
568 /**
569  * struct rdma_stat_desc
570  * @name - The name of the counter
571  * @flags - Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL
572  * @priv - Driver private information; Core code should not use
573  */
574 struct rdma_stat_desc {
575 	const char *name;
576 	unsigned int flags;
577 	const void *priv;
578 };
579 
580 /**
581  * struct rdma_hw_stats
582  * @lock - Mutex to protect parallel write access to lifespan and values
583  *    of counters, which are 64bits and not guaranteed to be written
584  *    atomicaly on 32bits systems.
585  * @timestamp - Used by the core code to track when the last update was
586  * @lifespan - Used by the core code to determine how old the counters
587  *   should be before being updated again.  Stored in jiffies, defaults
588  *   to 10 milliseconds, drivers can override the default be specifying
589  *   their own value during their allocation routine.
590  * @descs - Array of pointers to static descriptors used for the counters
591  *   in directory.
592  * @is_disabled - A bitmap to indicate each counter is currently disabled
593  *   or not.
594  * @num_counters - How many hardware counters there are.  If name is
595  *   shorter than this number, a kernel oops will result.  Driver authors
596  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
597  *   in their code to prevent this.
598  * @value - Array of u64 counters that are accessed by the sysfs code and
599  *   filled in by the drivers get_stats routine
600  */
601 struct rdma_hw_stats {
602 	struct mutex	lock; /* Protect lifespan and values[] */
603 	unsigned long	timestamp;
604 	unsigned long	lifespan;
605 	const struct rdma_stat_desc *descs;
606 	unsigned long	*is_disabled;
607 	int		num_counters;
608 	u64		value[] __counted_by(num_counters);
609 };
610 
611 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
612 
613 struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
614 	const struct rdma_stat_desc *descs, int num_counters,
615 	unsigned long lifespan);
616 
617 void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
618 
619 /* Define bits for the various functionality this port needs to be supported by
620  * the core.
621  */
622 /* Management                           0x00000FFF */
623 #define RDMA_CORE_CAP_IB_MAD            0x00000001
624 #define RDMA_CORE_CAP_IB_SMI            0x00000002
625 #define RDMA_CORE_CAP_IB_CM             0x00000004
626 #define RDMA_CORE_CAP_IW_CM             0x00000008
627 #define RDMA_CORE_CAP_IB_SA             0x00000010
628 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
629 
630 /* Address format                       0x000FF000 */
631 #define RDMA_CORE_CAP_AF_IB             0x00001000
632 #define RDMA_CORE_CAP_ETH_AH            0x00002000
633 #define RDMA_CORE_CAP_OPA_AH            0x00004000
634 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
635 
636 /* Protocol                             0xFFF00000 */
637 #define RDMA_CORE_CAP_PROT_IB           0x00100000
638 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
639 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
640 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
641 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
642 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
643 
644 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
645 					| RDMA_CORE_CAP_PROT_ROCE     \
646 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
647 
648 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
649 					| RDMA_CORE_CAP_IB_MAD \
650 					| RDMA_CORE_CAP_IB_SMI \
651 					| RDMA_CORE_CAP_IB_CM  \
652 					| RDMA_CORE_CAP_IB_SA  \
653 					| RDMA_CORE_CAP_AF_IB)
654 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
655 					| RDMA_CORE_CAP_IB_MAD  \
656 					| RDMA_CORE_CAP_IB_CM   \
657 					| RDMA_CORE_CAP_AF_IB   \
658 					| RDMA_CORE_CAP_ETH_AH)
659 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
660 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
661 					| RDMA_CORE_CAP_IB_MAD  \
662 					| RDMA_CORE_CAP_IB_CM   \
663 					| RDMA_CORE_CAP_AF_IB   \
664 					| RDMA_CORE_CAP_ETH_AH)
665 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
666 					| RDMA_CORE_CAP_IW_CM)
667 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
668 					| RDMA_CORE_CAP_OPA_MAD)
669 
670 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
671 
672 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
673 
674 struct ib_port_attr {
675 	u64			subnet_prefix;
676 	enum ib_port_state	state;
677 	enum ib_mtu		max_mtu;
678 	enum ib_mtu		active_mtu;
679 	u32                     phys_mtu;
680 	int			gid_tbl_len;
681 	unsigned int		ip_gids:1;
682 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
683 	u32			port_cap_flags;
684 	u32			max_msg_sz;
685 	u32			bad_pkey_cntr;
686 	u32			qkey_viol_cntr;
687 	u16			pkey_tbl_len;
688 	u32			sm_lid;
689 	u32			lid;
690 	u8			lmc;
691 	u8			max_vl_num;
692 	u8			sm_sl;
693 	u8			subnet_timeout;
694 	u8			init_type_reply;
695 	u8			active_width;
696 	u16			active_speed;
697 	u8                      phys_state;
698 	u16			port_cap_flags2;
699 };
700 
701 enum ib_device_modify_flags {
702 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
703 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
704 };
705 
706 #define IB_DEVICE_NODE_DESC_MAX 64
707 
708 struct ib_device_modify {
709 	u64	sys_image_guid;
710 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
711 };
712 
713 enum ib_port_modify_flags {
714 	IB_PORT_SHUTDOWN		= 1,
715 	IB_PORT_INIT_TYPE		= (1<<2),
716 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
717 	IB_PORT_OPA_MASK_CHG		= (1<<4)
718 };
719 
720 struct ib_port_modify {
721 	u32	set_port_cap_mask;
722 	u32	clr_port_cap_mask;
723 	u8	init_type;
724 };
725 
726 enum ib_event_type {
727 	IB_EVENT_CQ_ERR,
728 	IB_EVENT_QP_FATAL,
729 	IB_EVENT_QP_REQ_ERR,
730 	IB_EVENT_QP_ACCESS_ERR,
731 	IB_EVENT_COMM_EST,
732 	IB_EVENT_SQ_DRAINED,
733 	IB_EVENT_PATH_MIG,
734 	IB_EVENT_PATH_MIG_ERR,
735 	IB_EVENT_DEVICE_FATAL,
736 	IB_EVENT_PORT_ACTIVE,
737 	IB_EVENT_PORT_ERR,
738 	IB_EVENT_LID_CHANGE,
739 	IB_EVENT_PKEY_CHANGE,
740 	IB_EVENT_SM_CHANGE,
741 	IB_EVENT_SRQ_ERR,
742 	IB_EVENT_SRQ_LIMIT_REACHED,
743 	IB_EVENT_QP_LAST_WQE_REACHED,
744 	IB_EVENT_CLIENT_REREGISTER,
745 	IB_EVENT_GID_CHANGE,
746 	IB_EVENT_WQ_FATAL,
747 };
748 
749 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
750 
751 struct ib_event {
752 	struct ib_device	*device;
753 	union {
754 		struct ib_cq	*cq;
755 		struct ib_qp	*qp;
756 		struct ib_srq	*srq;
757 		struct ib_wq	*wq;
758 		u32		port_num;
759 	} element;
760 	enum ib_event_type	event;
761 };
762 
763 struct ib_event_handler {
764 	struct ib_device *device;
765 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
766 	struct list_head  list;
767 };
768 
769 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
770 	do {							\
771 		(_ptr)->device  = _device;			\
772 		(_ptr)->handler = _handler;			\
773 		INIT_LIST_HEAD(&(_ptr)->list);			\
774 	} while (0)
775 
776 struct ib_global_route {
777 	const struct ib_gid_attr *sgid_attr;
778 	union ib_gid	dgid;
779 	u32		flow_label;
780 	u8		sgid_index;
781 	u8		hop_limit;
782 	u8		traffic_class;
783 };
784 
785 struct ib_grh {
786 	__be32		version_tclass_flow;
787 	__be16		paylen;
788 	u8		next_hdr;
789 	u8		hop_limit;
790 	union ib_gid	sgid;
791 	union ib_gid	dgid;
792 };
793 
794 union rdma_network_hdr {
795 	struct ib_grh ibgrh;
796 	struct {
797 		/* The IB spec states that if it's IPv4, the header
798 		 * is located in the last 20 bytes of the header.
799 		 */
800 		u8		reserved[20];
801 		struct iphdr	roce4grh;
802 	};
803 };
804 
805 #define IB_QPN_MASK		0xFFFFFF
806 
807 enum {
808 	IB_MULTICAST_QPN = 0xffffff
809 };
810 
811 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
812 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
813 
814 enum ib_ah_flags {
815 	IB_AH_GRH	= 1
816 };
817 
818 enum ib_rate {
819 	IB_RATE_PORT_CURRENT = 0,
820 	IB_RATE_2_5_GBPS = 2,
821 	IB_RATE_5_GBPS   = 5,
822 	IB_RATE_10_GBPS  = 3,
823 	IB_RATE_20_GBPS  = 6,
824 	IB_RATE_30_GBPS  = 4,
825 	IB_RATE_40_GBPS  = 7,
826 	IB_RATE_60_GBPS  = 8,
827 	IB_RATE_80_GBPS  = 9,
828 	IB_RATE_120_GBPS = 10,
829 	IB_RATE_14_GBPS  = 11,
830 	IB_RATE_56_GBPS  = 12,
831 	IB_RATE_112_GBPS = 13,
832 	IB_RATE_168_GBPS = 14,
833 	IB_RATE_25_GBPS  = 15,
834 	IB_RATE_100_GBPS = 16,
835 	IB_RATE_200_GBPS = 17,
836 	IB_RATE_300_GBPS = 18,
837 	IB_RATE_28_GBPS  = 19,
838 	IB_RATE_50_GBPS  = 20,
839 	IB_RATE_400_GBPS = 21,
840 	IB_RATE_600_GBPS = 22,
841 	IB_RATE_800_GBPS = 23,
842 };
843 
844 /**
845  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
846  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
847  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
848  * @rate: rate to convert.
849  */
850 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
851 
852 /**
853  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
854  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
855  * @rate: rate to convert.
856  */
857 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
858 
859 
860 /**
861  * enum ib_mr_type - memory region type
862  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
863  *                            normal registration
864  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
865  *                            register any arbitrary sg lists (without
866  *                            the normal mr constraints - see
867  *                            ib_map_mr_sg)
868  * @IB_MR_TYPE_DM:            memory region that is used for device
869  *                            memory registration
870  * @IB_MR_TYPE_USER:          memory region that is used for the user-space
871  *                            application
872  * @IB_MR_TYPE_DMA:           memory region that is used for DMA operations
873  *                            without address translations (VA=PA)
874  * @IB_MR_TYPE_INTEGRITY:     memory region that is used for
875  *                            data integrity operations
876  */
877 enum ib_mr_type {
878 	IB_MR_TYPE_MEM_REG,
879 	IB_MR_TYPE_SG_GAPS,
880 	IB_MR_TYPE_DM,
881 	IB_MR_TYPE_USER,
882 	IB_MR_TYPE_DMA,
883 	IB_MR_TYPE_INTEGRITY,
884 };
885 
886 enum ib_mr_status_check {
887 	IB_MR_CHECK_SIG_STATUS = 1,
888 };
889 
890 /**
891  * struct ib_mr_status - Memory region status container
892  *
893  * @fail_status: Bitmask of MR checks status. For each
894  *     failed check a corresponding status bit is set.
895  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
896  *     failure.
897  */
898 struct ib_mr_status {
899 	u32		    fail_status;
900 	struct ib_sig_err   sig_err;
901 };
902 
903 /**
904  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
905  * enum.
906  * @mult: multiple to convert.
907  */
908 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
909 
910 struct rdma_ah_init_attr {
911 	struct rdma_ah_attr *ah_attr;
912 	u32 flags;
913 	struct net_device *xmit_slave;
914 };
915 
916 enum rdma_ah_attr_type {
917 	RDMA_AH_ATTR_TYPE_UNDEFINED,
918 	RDMA_AH_ATTR_TYPE_IB,
919 	RDMA_AH_ATTR_TYPE_ROCE,
920 	RDMA_AH_ATTR_TYPE_OPA,
921 };
922 
923 struct ib_ah_attr {
924 	u16			dlid;
925 	u8			src_path_bits;
926 };
927 
928 struct roce_ah_attr {
929 	u8			dmac[ETH_ALEN];
930 };
931 
932 struct opa_ah_attr {
933 	u32			dlid;
934 	u8			src_path_bits;
935 	bool			make_grd;
936 };
937 
938 struct rdma_ah_attr {
939 	struct ib_global_route	grh;
940 	u8			sl;
941 	u8			static_rate;
942 	u32			port_num;
943 	u8			ah_flags;
944 	enum rdma_ah_attr_type type;
945 	union {
946 		struct ib_ah_attr ib;
947 		struct roce_ah_attr roce;
948 		struct opa_ah_attr opa;
949 	};
950 };
951 
952 enum ib_wc_status {
953 	IB_WC_SUCCESS,
954 	IB_WC_LOC_LEN_ERR,
955 	IB_WC_LOC_QP_OP_ERR,
956 	IB_WC_LOC_EEC_OP_ERR,
957 	IB_WC_LOC_PROT_ERR,
958 	IB_WC_WR_FLUSH_ERR,
959 	IB_WC_MW_BIND_ERR,
960 	IB_WC_BAD_RESP_ERR,
961 	IB_WC_LOC_ACCESS_ERR,
962 	IB_WC_REM_INV_REQ_ERR,
963 	IB_WC_REM_ACCESS_ERR,
964 	IB_WC_REM_OP_ERR,
965 	IB_WC_RETRY_EXC_ERR,
966 	IB_WC_RNR_RETRY_EXC_ERR,
967 	IB_WC_LOC_RDD_VIOL_ERR,
968 	IB_WC_REM_INV_RD_REQ_ERR,
969 	IB_WC_REM_ABORT_ERR,
970 	IB_WC_INV_EECN_ERR,
971 	IB_WC_INV_EEC_STATE_ERR,
972 	IB_WC_FATAL_ERR,
973 	IB_WC_RESP_TIMEOUT_ERR,
974 	IB_WC_GENERAL_ERR
975 };
976 
977 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
978 
979 enum ib_wc_opcode {
980 	IB_WC_SEND = IB_UVERBS_WC_SEND,
981 	IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
982 	IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
983 	IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
984 	IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
985 	IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
986 	IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
987 	IB_WC_LSO = IB_UVERBS_WC_TSO,
988 	IB_WC_ATOMIC_WRITE = IB_UVERBS_WC_ATOMIC_WRITE,
989 	IB_WC_REG_MR,
990 	IB_WC_MASKED_COMP_SWAP,
991 	IB_WC_MASKED_FETCH_ADD,
992 	IB_WC_FLUSH = IB_UVERBS_WC_FLUSH,
993 /*
994  * Set value of IB_WC_RECV so consumers can test if a completion is a
995  * receive by testing (opcode & IB_WC_RECV).
996  */
997 	IB_WC_RECV			= 1 << 7,
998 	IB_WC_RECV_RDMA_WITH_IMM
999 };
1000 
1001 enum ib_wc_flags {
1002 	IB_WC_GRH		= 1,
1003 	IB_WC_WITH_IMM		= (1<<1),
1004 	IB_WC_WITH_INVALIDATE	= (1<<2),
1005 	IB_WC_IP_CSUM_OK	= (1<<3),
1006 	IB_WC_WITH_SMAC		= (1<<4),
1007 	IB_WC_WITH_VLAN		= (1<<5),
1008 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
1009 };
1010 
1011 struct ib_wc {
1012 	union {
1013 		u64		wr_id;
1014 		struct ib_cqe	*wr_cqe;
1015 	};
1016 	enum ib_wc_status	status;
1017 	enum ib_wc_opcode	opcode;
1018 	u32			vendor_err;
1019 	u32			byte_len;
1020 	struct ib_qp	       *qp;
1021 	union {
1022 		__be32		imm_data;
1023 		u32		invalidate_rkey;
1024 	} ex;
1025 	u32			src_qp;
1026 	u32			slid;
1027 	int			wc_flags;
1028 	u16			pkey_index;
1029 	u8			sl;
1030 	u8			dlid_path_bits;
1031 	u32 port_num; /* valid only for DR SMPs on switches */
1032 	u8			smac[ETH_ALEN];
1033 	u16			vlan_id;
1034 	u8			network_hdr_type;
1035 };
1036 
1037 enum ib_cq_notify_flags {
1038 	IB_CQ_SOLICITED			= 1 << 0,
1039 	IB_CQ_NEXT_COMP			= 1 << 1,
1040 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1041 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
1042 };
1043 
1044 enum ib_srq_type {
1045 	IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1046 	IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1047 	IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1048 };
1049 
1050 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1051 {
1052 	return srq_type == IB_SRQT_XRC ||
1053 	       srq_type == IB_SRQT_TM;
1054 }
1055 
1056 enum ib_srq_attr_mask {
1057 	IB_SRQ_MAX_WR	= 1 << 0,
1058 	IB_SRQ_LIMIT	= 1 << 1,
1059 };
1060 
1061 struct ib_srq_attr {
1062 	u32	max_wr;
1063 	u32	max_sge;
1064 	u32	srq_limit;
1065 };
1066 
1067 struct ib_srq_init_attr {
1068 	void		      (*event_handler)(struct ib_event *, void *);
1069 	void		       *srq_context;
1070 	struct ib_srq_attr	attr;
1071 	enum ib_srq_type	srq_type;
1072 
1073 	struct {
1074 		struct ib_cq   *cq;
1075 		union {
1076 			struct {
1077 				struct ib_xrcd *xrcd;
1078 			} xrc;
1079 
1080 			struct {
1081 				u32		max_num_tags;
1082 			} tag_matching;
1083 		};
1084 	} ext;
1085 };
1086 
1087 struct ib_qp_cap {
1088 	u32	max_send_wr;
1089 	u32	max_recv_wr;
1090 	u32	max_send_sge;
1091 	u32	max_recv_sge;
1092 	u32	max_inline_data;
1093 
1094 	/*
1095 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1096 	 * ib_create_qp() will calculate the right amount of needed WRs
1097 	 * and MRs based on this.
1098 	 */
1099 	u32	max_rdma_ctxs;
1100 };
1101 
1102 enum ib_sig_type {
1103 	IB_SIGNAL_ALL_WR,
1104 	IB_SIGNAL_REQ_WR
1105 };
1106 
1107 enum ib_qp_type {
1108 	/*
1109 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1110 	 * here (and in that order) since the MAD layer uses them as
1111 	 * indices into a 2-entry table.
1112 	 */
1113 	IB_QPT_SMI,
1114 	IB_QPT_GSI,
1115 
1116 	IB_QPT_RC = IB_UVERBS_QPT_RC,
1117 	IB_QPT_UC = IB_UVERBS_QPT_UC,
1118 	IB_QPT_UD = IB_UVERBS_QPT_UD,
1119 	IB_QPT_RAW_IPV6,
1120 	IB_QPT_RAW_ETHERTYPE,
1121 	IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1122 	IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1123 	IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1124 	IB_QPT_MAX,
1125 	IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1126 	/* Reserve a range for qp types internal to the low level driver.
1127 	 * These qp types will not be visible at the IB core layer, so the
1128 	 * IB_QPT_MAX usages should not be affected in the core layer
1129 	 */
1130 	IB_QPT_RESERVED1 = 0x1000,
1131 	IB_QPT_RESERVED2,
1132 	IB_QPT_RESERVED3,
1133 	IB_QPT_RESERVED4,
1134 	IB_QPT_RESERVED5,
1135 	IB_QPT_RESERVED6,
1136 	IB_QPT_RESERVED7,
1137 	IB_QPT_RESERVED8,
1138 	IB_QPT_RESERVED9,
1139 	IB_QPT_RESERVED10,
1140 };
1141 
1142 enum ib_qp_create_flags {
1143 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1144 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	=
1145 		IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1146 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1147 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1148 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1149 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1150 	IB_QP_CREATE_INTEGRITY_EN		= 1 << 6,
1151 	IB_QP_CREATE_NETDEV_USE			= 1 << 7,
1152 	IB_QP_CREATE_SCATTER_FCS		=
1153 		IB_UVERBS_QP_CREATE_SCATTER_FCS,
1154 	IB_QP_CREATE_CVLAN_STRIPPING		=
1155 		IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1156 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1157 	IB_QP_CREATE_PCI_WRITE_END_PADDING	=
1158 		IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1159 	/* reserve bits 26-31 for low level drivers' internal use */
1160 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1161 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1162 };
1163 
1164 /*
1165  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1166  * callback to destroy the passed in QP.
1167  */
1168 
1169 struct ib_qp_init_attr {
1170 	/* This callback occurs in workqueue context */
1171 	void                  (*event_handler)(struct ib_event *, void *);
1172 
1173 	void		       *qp_context;
1174 	struct ib_cq	       *send_cq;
1175 	struct ib_cq	       *recv_cq;
1176 	struct ib_srq	       *srq;
1177 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1178 	struct ib_qp_cap	cap;
1179 	enum ib_sig_type	sq_sig_type;
1180 	enum ib_qp_type		qp_type;
1181 	u32			create_flags;
1182 
1183 	/*
1184 	 * Only needed for special QP types, or when using the RW API.
1185 	 */
1186 	u32			port_num;
1187 	struct ib_rwq_ind_table *rwq_ind_tbl;
1188 	u32			source_qpn;
1189 };
1190 
1191 struct ib_qp_open_attr {
1192 	void                  (*event_handler)(struct ib_event *, void *);
1193 	void		       *qp_context;
1194 	u32			qp_num;
1195 	enum ib_qp_type		qp_type;
1196 };
1197 
1198 enum ib_rnr_timeout {
1199 	IB_RNR_TIMER_655_36 =  0,
1200 	IB_RNR_TIMER_000_01 =  1,
1201 	IB_RNR_TIMER_000_02 =  2,
1202 	IB_RNR_TIMER_000_03 =  3,
1203 	IB_RNR_TIMER_000_04 =  4,
1204 	IB_RNR_TIMER_000_06 =  5,
1205 	IB_RNR_TIMER_000_08 =  6,
1206 	IB_RNR_TIMER_000_12 =  7,
1207 	IB_RNR_TIMER_000_16 =  8,
1208 	IB_RNR_TIMER_000_24 =  9,
1209 	IB_RNR_TIMER_000_32 = 10,
1210 	IB_RNR_TIMER_000_48 = 11,
1211 	IB_RNR_TIMER_000_64 = 12,
1212 	IB_RNR_TIMER_000_96 = 13,
1213 	IB_RNR_TIMER_001_28 = 14,
1214 	IB_RNR_TIMER_001_92 = 15,
1215 	IB_RNR_TIMER_002_56 = 16,
1216 	IB_RNR_TIMER_003_84 = 17,
1217 	IB_RNR_TIMER_005_12 = 18,
1218 	IB_RNR_TIMER_007_68 = 19,
1219 	IB_RNR_TIMER_010_24 = 20,
1220 	IB_RNR_TIMER_015_36 = 21,
1221 	IB_RNR_TIMER_020_48 = 22,
1222 	IB_RNR_TIMER_030_72 = 23,
1223 	IB_RNR_TIMER_040_96 = 24,
1224 	IB_RNR_TIMER_061_44 = 25,
1225 	IB_RNR_TIMER_081_92 = 26,
1226 	IB_RNR_TIMER_122_88 = 27,
1227 	IB_RNR_TIMER_163_84 = 28,
1228 	IB_RNR_TIMER_245_76 = 29,
1229 	IB_RNR_TIMER_327_68 = 30,
1230 	IB_RNR_TIMER_491_52 = 31
1231 };
1232 
1233 enum ib_qp_attr_mask {
1234 	IB_QP_STATE			= 1,
1235 	IB_QP_CUR_STATE			= (1<<1),
1236 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1237 	IB_QP_ACCESS_FLAGS		= (1<<3),
1238 	IB_QP_PKEY_INDEX		= (1<<4),
1239 	IB_QP_PORT			= (1<<5),
1240 	IB_QP_QKEY			= (1<<6),
1241 	IB_QP_AV			= (1<<7),
1242 	IB_QP_PATH_MTU			= (1<<8),
1243 	IB_QP_TIMEOUT			= (1<<9),
1244 	IB_QP_RETRY_CNT			= (1<<10),
1245 	IB_QP_RNR_RETRY			= (1<<11),
1246 	IB_QP_RQ_PSN			= (1<<12),
1247 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1248 	IB_QP_ALT_PATH			= (1<<14),
1249 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1250 	IB_QP_SQ_PSN			= (1<<16),
1251 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1252 	IB_QP_PATH_MIG_STATE		= (1<<18),
1253 	IB_QP_CAP			= (1<<19),
1254 	IB_QP_DEST_QPN			= (1<<20),
1255 	IB_QP_RESERVED1			= (1<<21),
1256 	IB_QP_RESERVED2			= (1<<22),
1257 	IB_QP_RESERVED3			= (1<<23),
1258 	IB_QP_RESERVED4			= (1<<24),
1259 	IB_QP_RATE_LIMIT		= (1<<25),
1260 
1261 	IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1262 };
1263 
1264 enum ib_qp_state {
1265 	IB_QPS_RESET,
1266 	IB_QPS_INIT,
1267 	IB_QPS_RTR,
1268 	IB_QPS_RTS,
1269 	IB_QPS_SQD,
1270 	IB_QPS_SQE,
1271 	IB_QPS_ERR
1272 };
1273 
1274 enum ib_mig_state {
1275 	IB_MIG_MIGRATED,
1276 	IB_MIG_REARM,
1277 	IB_MIG_ARMED
1278 };
1279 
1280 enum ib_mw_type {
1281 	IB_MW_TYPE_1 = 1,
1282 	IB_MW_TYPE_2 = 2
1283 };
1284 
1285 struct ib_qp_attr {
1286 	enum ib_qp_state	qp_state;
1287 	enum ib_qp_state	cur_qp_state;
1288 	enum ib_mtu		path_mtu;
1289 	enum ib_mig_state	path_mig_state;
1290 	u32			qkey;
1291 	u32			rq_psn;
1292 	u32			sq_psn;
1293 	u32			dest_qp_num;
1294 	int			qp_access_flags;
1295 	struct ib_qp_cap	cap;
1296 	struct rdma_ah_attr	ah_attr;
1297 	struct rdma_ah_attr	alt_ah_attr;
1298 	u16			pkey_index;
1299 	u16			alt_pkey_index;
1300 	u8			en_sqd_async_notify;
1301 	u8			sq_draining;
1302 	u8			max_rd_atomic;
1303 	u8			max_dest_rd_atomic;
1304 	u8			min_rnr_timer;
1305 	u32			port_num;
1306 	u8			timeout;
1307 	u8			retry_cnt;
1308 	u8			rnr_retry;
1309 	u32			alt_port_num;
1310 	u8			alt_timeout;
1311 	u32			rate_limit;
1312 	struct net_device	*xmit_slave;
1313 };
1314 
1315 enum ib_wr_opcode {
1316 	/* These are shared with userspace */
1317 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1318 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1319 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1320 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1321 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1322 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1323 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1324 	IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1325 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1326 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1327 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1328 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1329 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1330 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1331 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1332 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1333 	IB_WR_FLUSH = IB_UVERBS_WR_FLUSH,
1334 	IB_WR_ATOMIC_WRITE = IB_UVERBS_WR_ATOMIC_WRITE,
1335 
1336 	/* These are kernel only and can not be issued by userspace */
1337 	IB_WR_REG_MR = 0x20,
1338 	IB_WR_REG_MR_INTEGRITY,
1339 
1340 	/* reserve values for low level drivers' internal use.
1341 	 * These values will not be used at all in the ib core layer.
1342 	 */
1343 	IB_WR_RESERVED1 = 0xf0,
1344 	IB_WR_RESERVED2,
1345 	IB_WR_RESERVED3,
1346 	IB_WR_RESERVED4,
1347 	IB_WR_RESERVED5,
1348 	IB_WR_RESERVED6,
1349 	IB_WR_RESERVED7,
1350 	IB_WR_RESERVED8,
1351 	IB_WR_RESERVED9,
1352 	IB_WR_RESERVED10,
1353 };
1354 
1355 enum ib_send_flags {
1356 	IB_SEND_FENCE		= 1,
1357 	IB_SEND_SIGNALED	= (1<<1),
1358 	IB_SEND_SOLICITED	= (1<<2),
1359 	IB_SEND_INLINE		= (1<<3),
1360 	IB_SEND_IP_CSUM		= (1<<4),
1361 
1362 	/* reserve bits 26-31 for low level drivers' internal use */
1363 	IB_SEND_RESERVED_START	= (1 << 26),
1364 	IB_SEND_RESERVED_END	= (1 << 31),
1365 };
1366 
1367 struct ib_sge {
1368 	u64	addr;
1369 	u32	length;
1370 	u32	lkey;
1371 };
1372 
1373 struct ib_cqe {
1374 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1375 };
1376 
1377 struct ib_send_wr {
1378 	struct ib_send_wr      *next;
1379 	union {
1380 		u64		wr_id;
1381 		struct ib_cqe	*wr_cqe;
1382 	};
1383 	struct ib_sge	       *sg_list;
1384 	int			num_sge;
1385 	enum ib_wr_opcode	opcode;
1386 	int			send_flags;
1387 	union {
1388 		__be32		imm_data;
1389 		u32		invalidate_rkey;
1390 	} ex;
1391 };
1392 
1393 struct ib_rdma_wr {
1394 	struct ib_send_wr	wr;
1395 	u64			remote_addr;
1396 	u32			rkey;
1397 };
1398 
1399 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1400 {
1401 	return container_of(wr, struct ib_rdma_wr, wr);
1402 }
1403 
1404 struct ib_atomic_wr {
1405 	struct ib_send_wr	wr;
1406 	u64			remote_addr;
1407 	u64			compare_add;
1408 	u64			swap;
1409 	u64			compare_add_mask;
1410 	u64			swap_mask;
1411 	u32			rkey;
1412 };
1413 
1414 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1415 {
1416 	return container_of(wr, struct ib_atomic_wr, wr);
1417 }
1418 
1419 struct ib_ud_wr {
1420 	struct ib_send_wr	wr;
1421 	struct ib_ah		*ah;
1422 	void			*header;
1423 	int			hlen;
1424 	int			mss;
1425 	u32			remote_qpn;
1426 	u32			remote_qkey;
1427 	u16			pkey_index; /* valid for GSI only */
1428 	u32			port_num; /* valid for DR SMPs on switch only */
1429 };
1430 
1431 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1432 {
1433 	return container_of(wr, struct ib_ud_wr, wr);
1434 }
1435 
1436 struct ib_reg_wr {
1437 	struct ib_send_wr	wr;
1438 	struct ib_mr		*mr;
1439 	u32			key;
1440 	int			access;
1441 };
1442 
1443 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1444 {
1445 	return container_of(wr, struct ib_reg_wr, wr);
1446 }
1447 
1448 struct ib_recv_wr {
1449 	struct ib_recv_wr      *next;
1450 	union {
1451 		u64		wr_id;
1452 		struct ib_cqe	*wr_cqe;
1453 	};
1454 	struct ib_sge	       *sg_list;
1455 	int			num_sge;
1456 };
1457 
1458 enum ib_access_flags {
1459 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1460 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1461 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1462 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1463 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1464 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1465 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1466 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1467 	IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1468 	IB_ACCESS_FLUSH_GLOBAL = IB_UVERBS_ACCESS_FLUSH_GLOBAL,
1469 	IB_ACCESS_FLUSH_PERSISTENT = IB_UVERBS_ACCESS_FLUSH_PERSISTENT,
1470 
1471 	IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1472 	IB_ACCESS_SUPPORTED =
1473 		((IB_ACCESS_FLUSH_PERSISTENT << 1) - 1) | IB_ACCESS_OPTIONAL,
1474 };
1475 
1476 /*
1477  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1478  * are hidden here instead of a uapi header!
1479  */
1480 enum ib_mr_rereg_flags {
1481 	IB_MR_REREG_TRANS	= 1,
1482 	IB_MR_REREG_PD		= (1<<1),
1483 	IB_MR_REREG_ACCESS	= (1<<2),
1484 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1485 };
1486 
1487 struct ib_umem;
1488 
1489 enum rdma_remove_reason {
1490 	/*
1491 	 * Userspace requested uobject deletion or initial try
1492 	 * to remove uobject via cleanup. Call could fail
1493 	 */
1494 	RDMA_REMOVE_DESTROY,
1495 	/* Context deletion. This call should delete the actual object itself */
1496 	RDMA_REMOVE_CLOSE,
1497 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1498 	RDMA_REMOVE_DRIVER_REMOVE,
1499 	/* uobj is being cleaned-up before being committed */
1500 	RDMA_REMOVE_ABORT,
1501 	/* The driver failed to destroy the uobject and is being disconnected */
1502 	RDMA_REMOVE_DRIVER_FAILURE,
1503 };
1504 
1505 struct ib_rdmacg_object {
1506 #ifdef CONFIG_CGROUP_RDMA
1507 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1508 #endif
1509 };
1510 
1511 struct ib_ucontext {
1512 	struct ib_device       *device;
1513 	struct ib_uverbs_file  *ufile;
1514 
1515 	struct ib_rdmacg_object	cg_obj;
1516 	/*
1517 	 * Implementation details of the RDMA core, don't use in drivers:
1518 	 */
1519 	struct rdma_restrack_entry res;
1520 	struct xarray mmap_xa;
1521 };
1522 
1523 struct ib_uobject {
1524 	u64			user_handle;	/* handle given to us by userspace */
1525 	/* ufile & ucontext owning this object */
1526 	struct ib_uverbs_file  *ufile;
1527 	/* FIXME, save memory: ufile->context == context */
1528 	struct ib_ucontext     *context;	/* associated user context */
1529 	void		       *object;		/* containing object */
1530 	struct list_head	list;		/* link to context's list */
1531 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1532 	int			id;		/* index into kernel idr */
1533 	struct kref		ref;
1534 	atomic_t		usecnt;		/* protects exclusive access */
1535 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1536 
1537 	const struct uverbs_api_object *uapi_object;
1538 };
1539 
1540 struct ib_udata {
1541 	const void __user *inbuf;
1542 	void __user *outbuf;
1543 	size_t       inlen;
1544 	size_t       outlen;
1545 };
1546 
1547 struct ib_pd {
1548 	u32			local_dma_lkey;
1549 	u32			flags;
1550 	struct ib_device       *device;
1551 	struct ib_uobject      *uobject;
1552 	atomic_t          	usecnt; /* count all resources */
1553 
1554 	u32			unsafe_global_rkey;
1555 
1556 	/*
1557 	 * Implementation details of the RDMA core, don't use in drivers:
1558 	 */
1559 	struct ib_mr	       *__internal_mr;
1560 	struct rdma_restrack_entry res;
1561 };
1562 
1563 struct ib_xrcd {
1564 	struct ib_device       *device;
1565 	atomic_t		usecnt; /* count all exposed resources */
1566 	struct inode	       *inode;
1567 	struct rw_semaphore	tgt_qps_rwsem;
1568 	struct xarray		tgt_qps;
1569 };
1570 
1571 struct ib_ah {
1572 	struct ib_device	*device;
1573 	struct ib_pd		*pd;
1574 	struct ib_uobject	*uobject;
1575 	const struct ib_gid_attr *sgid_attr;
1576 	enum rdma_ah_attr_type	type;
1577 };
1578 
1579 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1580 
1581 enum ib_poll_context {
1582 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1583 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1584 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1585 	IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1586 
1587 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1588 };
1589 
1590 struct ib_cq {
1591 	struct ib_device       *device;
1592 	struct ib_ucq_object   *uobject;
1593 	ib_comp_handler   	comp_handler;
1594 	void                  (*event_handler)(struct ib_event *, void *);
1595 	void                   *cq_context;
1596 	int               	cqe;
1597 	unsigned int		cqe_used;
1598 	atomic_t          	usecnt; /* count number of work queues */
1599 	enum ib_poll_context	poll_ctx;
1600 	struct ib_wc		*wc;
1601 	struct list_head        pool_entry;
1602 	union {
1603 		struct irq_poll		iop;
1604 		struct work_struct	work;
1605 	};
1606 	struct workqueue_struct *comp_wq;
1607 	struct dim *dim;
1608 
1609 	/* updated only by trace points */
1610 	ktime_t timestamp;
1611 	u8 interrupt:1;
1612 	u8 shared:1;
1613 	unsigned int comp_vector;
1614 
1615 	/*
1616 	 * Implementation details of the RDMA core, don't use in drivers:
1617 	 */
1618 	struct rdma_restrack_entry res;
1619 };
1620 
1621 struct ib_srq {
1622 	struct ib_device       *device;
1623 	struct ib_pd	       *pd;
1624 	struct ib_usrq_object  *uobject;
1625 	void		      (*event_handler)(struct ib_event *, void *);
1626 	void		       *srq_context;
1627 	enum ib_srq_type	srq_type;
1628 	atomic_t		usecnt;
1629 
1630 	struct {
1631 		struct ib_cq   *cq;
1632 		union {
1633 			struct {
1634 				struct ib_xrcd *xrcd;
1635 				u32		srq_num;
1636 			} xrc;
1637 		};
1638 	} ext;
1639 
1640 	/*
1641 	 * Implementation details of the RDMA core, don't use in drivers:
1642 	 */
1643 	struct rdma_restrack_entry res;
1644 };
1645 
1646 enum ib_raw_packet_caps {
1647 	/*
1648 	 * Strip cvlan from incoming packet and report it in the matching work
1649 	 * completion is supported.
1650 	 */
1651 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING =
1652 		IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING,
1653 	/*
1654 	 * Scatter FCS field of an incoming packet to host memory is supported.
1655 	 */
1656 	IB_RAW_PACKET_CAP_SCATTER_FCS = IB_UVERBS_RAW_PACKET_CAP_SCATTER_FCS,
1657 	/* Checksum offloads are supported (for both send and receive). */
1658 	IB_RAW_PACKET_CAP_IP_CSUM = IB_UVERBS_RAW_PACKET_CAP_IP_CSUM,
1659 	/*
1660 	 * When a packet is received for an RQ with no receive WQEs, the
1661 	 * packet processing is delayed.
1662 	 */
1663 	IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP,
1664 };
1665 
1666 enum ib_wq_type {
1667 	IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1668 };
1669 
1670 enum ib_wq_state {
1671 	IB_WQS_RESET,
1672 	IB_WQS_RDY,
1673 	IB_WQS_ERR
1674 };
1675 
1676 struct ib_wq {
1677 	struct ib_device       *device;
1678 	struct ib_uwq_object   *uobject;
1679 	void		    *wq_context;
1680 	void		    (*event_handler)(struct ib_event *, void *);
1681 	struct ib_pd	       *pd;
1682 	struct ib_cq	       *cq;
1683 	u32		wq_num;
1684 	enum ib_wq_state       state;
1685 	enum ib_wq_type	wq_type;
1686 	atomic_t		usecnt;
1687 };
1688 
1689 enum ib_wq_flags {
1690 	IB_WQ_FLAGS_CVLAN_STRIPPING	= IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1691 	IB_WQ_FLAGS_SCATTER_FCS		= IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1692 	IB_WQ_FLAGS_DELAY_DROP		= IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1693 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1694 				IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1695 };
1696 
1697 struct ib_wq_init_attr {
1698 	void		       *wq_context;
1699 	enum ib_wq_type	wq_type;
1700 	u32		max_wr;
1701 	u32		max_sge;
1702 	struct	ib_cq	       *cq;
1703 	void		    (*event_handler)(struct ib_event *, void *);
1704 	u32		create_flags; /* Use enum ib_wq_flags */
1705 };
1706 
1707 enum ib_wq_attr_mask {
1708 	IB_WQ_STATE		= 1 << 0,
1709 	IB_WQ_CUR_STATE		= 1 << 1,
1710 	IB_WQ_FLAGS		= 1 << 2,
1711 };
1712 
1713 struct ib_wq_attr {
1714 	enum	ib_wq_state	wq_state;
1715 	enum	ib_wq_state	curr_wq_state;
1716 	u32			flags; /* Use enum ib_wq_flags */
1717 	u32			flags_mask; /* Use enum ib_wq_flags */
1718 };
1719 
1720 struct ib_rwq_ind_table {
1721 	struct ib_device	*device;
1722 	struct ib_uobject      *uobject;
1723 	atomic_t		usecnt;
1724 	u32		ind_tbl_num;
1725 	u32		log_ind_tbl_size;
1726 	struct ib_wq	**ind_tbl;
1727 };
1728 
1729 struct ib_rwq_ind_table_init_attr {
1730 	u32		log_ind_tbl_size;
1731 	/* Each entry is a pointer to Receive Work Queue */
1732 	struct ib_wq	**ind_tbl;
1733 };
1734 
1735 enum port_pkey_state {
1736 	IB_PORT_PKEY_NOT_VALID = 0,
1737 	IB_PORT_PKEY_VALID = 1,
1738 	IB_PORT_PKEY_LISTED = 2,
1739 };
1740 
1741 struct ib_qp_security;
1742 
1743 struct ib_port_pkey {
1744 	enum port_pkey_state	state;
1745 	u16			pkey_index;
1746 	u32			port_num;
1747 	struct list_head	qp_list;
1748 	struct list_head	to_error_list;
1749 	struct ib_qp_security  *sec;
1750 };
1751 
1752 struct ib_ports_pkeys {
1753 	struct ib_port_pkey	main;
1754 	struct ib_port_pkey	alt;
1755 };
1756 
1757 struct ib_qp_security {
1758 	struct ib_qp	       *qp;
1759 	struct ib_device       *dev;
1760 	/* Hold this mutex when changing port and pkey settings. */
1761 	struct mutex		mutex;
1762 	struct ib_ports_pkeys  *ports_pkeys;
1763 	/* A list of all open shared QP handles.  Required to enforce security
1764 	 * properly for all users of a shared QP.
1765 	 */
1766 	struct list_head        shared_qp_list;
1767 	void                   *security;
1768 	bool			destroying;
1769 	atomic_t		error_list_count;
1770 	struct completion	error_complete;
1771 	int			error_comps_pending;
1772 };
1773 
1774 /*
1775  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1776  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1777  */
1778 struct ib_qp {
1779 	struct ib_device       *device;
1780 	struct ib_pd	       *pd;
1781 	struct ib_cq	       *send_cq;
1782 	struct ib_cq	       *recv_cq;
1783 	spinlock_t		mr_lock;
1784 	int			mrs_used;
1785 	struct list_head	rdma_mrs;
1786 	struct list_head	sig_mrs;
1787 	struct ib_srq	       *srq;
1788 	struct completion	srq_completion;
1789 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1790 	struct list_head	xrcd_list;
1791 
1792 	/* count times opened, mcast attaches, flow attaches */
1793 	atomic_t		usecnt;
1794 	struct list_head	open_list;
1795 	struct ib_qp           *real_qp;
1796 	struct ib_uqp_object   *uobject;
1797 	void                  (*event_handler)(struct ib_event *, void *);
1798 	void                  (*registered_event_handler)(struct ib_event *, void *);
1799 	void		       *qp_context;
1800 	/* sgid_attrs associated with the AV's */
1801 	const struct ib_gid_attr *av_sgid_attr;
1802 	const struct ib_gid_attr *alt_path_sgid_attr;
1803 	u32			qp_num;
1804 	u32			max_write_sge;
1805 	u32			max_read_sge;
1806 	enum ib_qp_type		qp_type;
1807 	struct ib_rwq_ind_table *rwq_ind_tbl;
1808 	struct ib_qp_security  *qp_sec;
1809 	u32			port;
1810 
1811 	bool			integrity_en;
1812 	/*
1813 	 * Implementation details of the RDMA core, don't use in drivers:
1814 	 */
1815 	struct rdma_restrack_entry     res;
1816 
1817 	/* The counter the qp is bind to */
1818 	struct rdma_counter    *counter;
1819 };
1820 
1821 struct ib_dm {
1822 	struct ib_device  *device;
1823 	u32		   length;
1824 	u32		   flags;
1825 	struct ib_uobject *uobject;
1826 	atomic_t	   usecnt;
1827 };
1828 
1829 struct ib_mr {
1830 	struct ib_device  *device;
1831 	struct ib_pd	  *pd;
1832 	u32		   lkey;
1833 	u32		   rkey;
1834 	u64		   iova;
1835 	u64		   length;
1836 	unsigned int	   page_size;
1837 	enum ib_mr_type	   type;
1838 	bool		   need_inval;
1839 	union {
1840 		struct ib_uobject	*uobject;	/* user */
1841 		struct list_head	qp_entry;	/* FR */
1842 	};
1843 
1844 	struct ib_dm      *dm;
1845 	struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1846 	/*
1847 	 * Implementation details of the RDMA core, don't use in drivers:
1848 	 */
1849 	struct rdma_restrack_entry res;
1850 };
1851 
1852 struct ib_mw {
1853 	struct ib_device	*device;
1854 	struct ib_pd		*pd;
1855 	struct ib_uobject	*uobject;
1856 	u32			rkey;
1857 	enum ib_mw_type         type;
1858 };
1859 
1860 /* Supported steering options */
1861 enum ib_flow_attr_type {
1862 	/* steering according to rule specifications */
1863 	IB_FLOW_ATTR_NORMAL		= 0x0,
1864 	/* default unicast and multicast rule -
1865 	 * receive all Eth traffic which isn't steered to any QP
1866 	 */
1867 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1868 	/* default multicast rule -
1869 	 * receive all Eth multicast traffic which isn't steered to any QP
1870 	 */
1871 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1872 	/* sniffer rule - receive all port traffic */
1873 	IB_FLOW_ATTR_SNIFFER		= 0x3
1874 };
1875 
1876 /* Supported steering header types */
1877 enum ib_flow_spec_type {
1878 	/* L2 headers*/
1879 	IB_FLOW_SPEC_ETH		= 0x20,
1880 	IB_FLOW_SPEC_IB			= 0x22,
1881 	/* L3 header*/
1882 	IB_FLOW_SPEC_IPV4		= 0x30,
1883 	IB_FLOW_SPEC_IPV6		= 0x31,
1884 	IB_FLOW_SPEC_ESP                = 0x34,
1885 	/* L4 headers*/
1886 	IB_FLOW_SPEC_TCP		= 0x40,
1887 	IB_FLOW_SPEC_UDP		= 0x41,
1888 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1889 	IB_FLOW_SPEC_GRE		= 0x51,
1890 	IB_FLOW_SPEC_MPLS		= 0x60,
1891 	IB_FLOW_SPEC_INNER		= 0x100,
1892 	/* Actions */
1893 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1894 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1895 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1896 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1897 };
1898 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1899 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1900 
1901 enum ib_flow_flags {
1902 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1903 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1904 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1905 };
1906 
1907 struct ib_flow_eth_filter {
1908 	u8	dst_mac[6];
1909 	u8	src_mac[6];
1910 	__be16	ether_type;
1911 	__be16	vlan_tag;
1912 };
1913 
1914 struct ib_flow_spec_eth {
1915 	u32			  type;
1916 	u16			  size;
1917 	struct ib_flow_eth_filter val;
1918 	struct ib_flow_eth_filter mask;
1919 };
1920 
1921 struct ib_flow_ib_filter {
1922 	__be16 dlid;
1923 	__u8   sl;
1924 };
1925 
1926 struct ib_flow_spec_ib {
1927 	u32			 type;
1928 	u16			 size;
1929 	struct ib_flow_ib_filter val;
1930 	struct ib_flow_ib_filter mask;
1931 };
1932 
1933 /* IPv4 header flags */
1934 enum ib_ipv4_flags {
1935 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1936 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1937 				    last have this flag set */
1938 };
1939 
1940 struct ib_flow_ipv4_filter {
1941 	__be32	src_ip;
1942 	__be32	dst_ip;
1943 	u8	proto;
1944 	u8	tos;
1945 	u8	ttl;
1946 	u8	flags;
1947 };
1948 
1949 struct ib_flow_spec_ipv4 {
1950 	u32			   type;
1951 	u16			   size;
1952 	struct ib_flow_ipv4_filter val;
1953 	struct ib_flow_ipv4_filter mask;
1954 };
1955 
1956 struct ib_flow_ipv6_filter {
1957 	u8	src_ip[16];
1958 	u8	dst_ip[16];
1959 	__be32	flow_label;
1960 	u8	next_hdr;
1961 	u8	traffic_class;
1962 	u8	hop_limit;
1963 } __packed;
1964 
1965 struct ib_flow_spec_ipv6 {
1966 	u32			   type;
1967 	u16			   size;
1968 	struct ib_flow_ipv6_filter val;
1969 	struct ib_flow_ipv6_filter mask;
1970 };
1971 
1972 struct ib_flow_tcp_udp_filter {
1973 	__be16	dst_port;
1974 	__be16	src_port;
1975 };
1976 
1977 struct ib_flow_spec_tcp_udp {
1978 	u32			      type;
1979 	u16			      size;
1980 	struct ib_flow_tcp_udp_filter val;
1981 	struct ib_flow_tcp_udp_filter mask;
1982 };
1983 
1984 struct ib_flow_tunnel_filter {
1985 	__be32	tunnel_id;
1986 };
1987 
1988 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1989  * the tunnel_id from val has the vni value
1990  */
1991 struct ib_flow_spec_tunnel {
1992 	u32			      type;
1993 	u16			      size;
1994 	struct ib_flow_tunnel_filter  val;
1995 	struct ib_flow_tunnel_filter  mask;
1996 };
1997 
1998 struct ib_flow_esp_filter {
1999 	__be32	spi;
2000 	__be32  seq;
2001 };
2002 
2003 struct ib_flow_spec_esp {
2004 	u32                           type;
2005 	u16			      size;
2006 	struct ib_flow_esp_filter     val;
2007 	struct ib_flow_esp_filter     mask;
2008 };
2009 
2010 struct ib_flow_gre_filter {
2011 	__be16 c_ks_res0_ver;
2012 	__be16 protocol;
2013 	__be32 key;
2014 };
2015 
2016 struct ib_flow_spec_gre {
2017 	u32                           type;
2018 	u16			      size;
2019 	struct ib_flow_gre_filter     val;
2020 	struct ib_flow_gre_filter     mask;
2021 };
2022 
2023 struct ib_flow_mpls_filter {
2024 	__be32 tag;
2025 };
2026 
2027 struct ib_flow_spec_mpls {
2028 	u32                           type;
2029 	u16			      size;
2030 	struct ib_flow_mpls_filter     val;
2031 	struct ib_flow_mpls_filter     mask;
2032 };
2033 
2034 struct ib_flow_spec_action_tag {
2035 	enum ib_flow_spec_type	      type;
2036 	u16			      size;
2037 	u32                           tag_id;
2038 };
2039 
2040 struct ib_flow_spec_action_drop {
2041 	enum ib_flow_spec_type	      type;
2042 	u16			      size;
2043 };
2044 
2045 struct ib_flow_spec_action_handle {
2046 	enum ib_flow_spec_type	      type;
2047 	u16			      size;
2048 	struct ib_flow_action	     *act;
2049 };
2050 
2051 enum ib_counters_description {
2052 	IB_COUNTER_PACKETS,
2053 	IB_COUNTER_BYTES,
2054 };
2055 
2056 struct ib_flow_spec_action_count {
2057 	enum ib_flow_spec_type type;
2058 	u16 size;
2059 	struct ib_counters *counters;
2060 };
2061 
2062 union ib_flow_spec {
2063 	struct {
2064 		u32			type;
2065 		u16			size;
2066 	};
2067 	struct ib_flow_spec_eth		eth;
2068 	struct ib_flow_spec_ib		ib;
2069 	struct ib_flow_spec_ipv4        ipv4;
2070 	struct ib_flow_spec_tcp_udp	tcp_udp;
2071 	struct ib_flow_spec_ipv6        ipv6;
2072 	struct ib_flow_spec_tunnel      tunnel;
2073 	struct ib_flow_spec_esp		esp;
2074 	struct ib_flow_spec_gre		gre;
2075 	struct ib_flow_spec_mpls	mpls;
2076 	struct ib_flow_spec_action_tag  flow_tag;
2077 	struct ib_flow_spec_action_drop drop;
2078 	struct ib_flow_spec_action_handle action;
2079 	struct ib_flow_spec_action_count flow_count;
2080 };
2081 
2082 struct ib_flow_attr {
2083 	enum ib_flow_attr_type type;
2084 	u16	     size;
2085 	u16	     priority;
2086 	u32	     flags;
2087 	u8	     num_of_specs;
2088 	u32	     port;
2089 	union ib_flow_spec flows[];
2090 };
2091 
2092 struct ib_flow {
2093 	struct ib_qp		*qp;
2094 	struct ib_device	*device;
2095 	struct ib_uobject	*uobject;
2096 };
2097 
2098 enum ib_flow_action_type {
2099 	IB_FLOW_ACTION_UNSPECIFIED,
2100 	IB_FLOW_ACTION_ESP = 1,
2101 };
2102 
2103 struct ib_flow_action_attrs_esp_keymats {
2104 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2105 	union {
2106 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2107 	} keymat;
2108 };
2109 
2110 struct ib_flow_action_attrs_esp_replays {
2111 	enum ib_uverbs_flow_action_esp_replay			protocol;
2112 	union {
2113 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2114 	} replay;
2115 };
2116 
2117 enum ib_flow_action_attrs_esp_flags {
2118 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2119 	 * This is done in order to share the same flags between user-space and
2120 	 * kernel and spare an unnecessary translation.
2121 	 */
2122 
2123 	/* Kernel flags */
2124 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2125 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2126 };
2127 
2128 struct ib_flow_spec_list {
2129 	struct ib_flow_spec_list	*next;
2130 	union ib_flow_spec		spec;
2131 };
2132 
2133 struct ib_flow_action_attrs_esp {
2134 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2135 	struct ib_flow_action_attrs_esp_replays		*replay;
2136 	struct ib_flow_spec_list			*encap;
2137 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2138 	 * Value of 0 is a valid value.
2139 	 */
2140 	u32						esn;
2141 	u32						spi;
2142 	u32						seq;
2143 	u32						tfc_pad;
2144 	/* Use enum ib_flow_action_attrs_esp_flags */
2145 	u64						flags;
2146 	u64						hard_limit_pkts;
2147 };
2148 
2149 struct ib_flow_action {
2150 	struct ib_device		*device;
2151 	struct ib_uobject		*uobject;
2152 	enum ib_flow_action_type	type;
2153 	atomic_t			usecnt;
2154 };
2155 
2156 struct ib_mad;
2157 
2158 enum ib_process_mad_flags {
2159 	IB_MAD_IGNORE_MKEY	= 1,
2160 	IB_MAD_IGNORE_BKEY	= 2,
2161 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2162 };
2163 
2164 enum ib_mad_result {
2165 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2166 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2167 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2168 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2169 };
2170 
2171 struct ib_port_cache {
2172 	u64		      subnet_prefix;
2173 	struct ib_pkey_cache  *pkey;
2174 	struct ib_gid_table   *gid;
2175 	u8                     lmc;
2176 	enum ib_port_state     port_state;
2177 	enum ib_port_state     last_port_state;
2178 };
2179 
2180 struct ib_port_immutable {
2181 	int                           pkey_tbl_len;
2182 	int                           gid_tbl_len;
2183 	u32                           core_cap_flags;
2184 	u32                           max_mad_size;
2185 };
2186 
2187 struct ib_port_data {
2188 	struct ib_device *ib_dev;
2189 
2190 	struct ib_port_immutable immutable;
2191 
2192 	spinlock_t pkey_list_lock;
2193 
2194 	spinlock_t netdev_lock;
2195 
2196 	struct list_head pkey_list;
2197 
2198 	struct ib_port_cache cache;
2199 
2200 	struct net_device __rcu *netdev;
2201 	netdevice_tracker netdev_tracker;
2202 	struct hlist_node ndev_hash_link;
2203 	struct rdma_port_counter port_counter;
2204 	struct ib_port *sysfs;
2205 };
2206 
2207 /* rdma netdev type - specifies protocol type */
2208 enum rdma_netdev_t {
2209 	RDMA_NETDEV_OPA_VNIC,
2210 	RDMA_NETDEV_IPOIB,
2211 };
2212 
2213 /**
2214  * struct rdma_netdev - rdma netdev
2215  * For cases where netstack interfacing is required.
2216  */
2217 struct rdma_netdev {
2218 	void              *clnt_priv;
2219 	struct ib_device  *hca;
2220 	u32		   port_num;
2221 	int                mtu;
2222 
2223 	/*
2224 	 * cleanup function must be specified.
2225 	 * FIXME: This is only used for OPA_VNIC and that usage should be
2226 	 * removed too.
2227 	 */
2228 	void (*free_rdma_netdev)(struct net_device *netdev);
2229 
2230 	/* control functions */
2231 	void (*set_id)(struct net_device *netdev, int id);
2232 	/* send packet */
2233 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2234 		    struct ib_ah *address, u32 dqpn);
2235 	/* multicast */
2236 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2237 			    union ib_gid *gid, u16 mlid,
2238 			    int set_qkey, u32 qkey);
2239 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2240 			    union ib_gid *gid, u16 mlid);
2241 	/* timeout */
2242 	void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2243 };
2244 
2245 struct rdma_netdev_alloc_params {
2246 	size_t sizeof_priv;
2247 	unsigned int txqs;
2248 	unsigned int rxqs;
2249 	void *param;
2250 
2251 	int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2252 				      struct net_device *netdev, void *param);
2253 };
2254 
2255 struct ib_odp_counters {
2256 	atomic64_t faults;
2257 	atomic64_t faults_handled;
2258 	atomic64_t invalidations;
2259 	atomic64_t invalidations_handled;
2260 	atomic64_t prefetch;
2261 };
2262 
2263 struct ib_counters {
2264 	struct ib_device	*device;
2265 	struct ib_uobject	*uobject;
2266 	/* num of objects attached */
2267 	atomic_t	usecnt;
2268 };
2269 
2270 struct ib_counters_read_attr {
2271 	u64	*counters_buff;
2272 	u32	ncounters;
2273 	u32	flags; /* use enum ib_read_counters_flags */
2274 };
2275 
2276 struct uverbs_attr_bundle;
2277 struct iw_cm_id;
2278 struct iw_cm_conn_param;
2279 
2280 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member)                      \
2281 	.size_##ib_struct =                                                    \
2282 		(sizeof(struct drv_struct) +                                   \
2283 		 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) +      \
2284 		 BUILD_BUG_ON_ZERO(                                            \
2285 			 !__same_type(((struct drv_struct *)NULL)->member,     \
2286 				      struct ib_struct)))
2287 
2288 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp)                          \
2289 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2290 					   gfp, false))
2291 
2292 #define rdma_zalloc_drv_obj_numa(ib_dev, ib_type)                              \
2293 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2294 					   GFP_KERNEL, true))
2295 
2296 #define rdma_zalloc_drv_obj(ib_dev, ib_type)                                   \
2297 	rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2298 
2299 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2300 
2301 struct rdma_user_mmap_entry {
2302 	struct kref ref;
2303 	struct ib_ucontext *ucontext;
2304 	unsigned long start_pgoff;
2305 	size_t npages;
2306 	bool driver_removed;
2307 };
2308 
2309 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2310 static inline u64
2311 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2312 {
2313 	return (u64)entry->start_pgoff << PAGE_SHIFT;
2314 }
2315 
2316 /**
2317  * struct ib_device_ops - InfiniBand device operations
2318  * This structure defines all the InfiniBand device operations, providers will
2319  * need to define the supported operations, otherwise they will be set to null.
2320  */
2321 struct ib_device_ops {
2322 	struct module *owner;
2323 	enum rdma_driver_id driver_id;
2324 	u32 uverbs_abi_ver;
2325 	unsigned int uverbs_no_driver_id_binding:1;
2326 
2327 	/*
2328 	 * NOTE: New drivers should not make use of device_group; instead new
2329 	 * device parameter should be exposed via netlink command. This
2330 	 * mechanism exists only for existing drivers.
2331 	 */
2332 	const struct attribute_group *device_group;
2333 	const struct attribute_group **port_groups;
2334 
2335 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2336 			 const struct ib_send_wr **bad_send_wr);
2337 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2338 			 const struct ib_recv_wr **bad_recv_wr);
2339 	void (*drain_rq)(struct ib_qp *qp);
2340 	void (*drain_sq)(struct ib_qp *qp);
2341 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2342 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2343 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2344 	int (*post_srq_recv)(struct ib_srq *srq,
2345 			     const struct ib_recv_wr *recv_wr,
2346 			     const struct ib_recv_wr **bad_recv_wr);
2347 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2348 			   u32 port_num, const struct ib_wc *in_wc,
2349 			   const struct ib_grh *in_grh,
2350 			   const struct ib_mad *in_mad, struct ib_mad *out_mad,
2351 			   size_t *out_mad_size, u16 *out_mad_pkey_index);
2352 	int (*query_device)(struct ib_device *device,
2353 			    struct ib_device_attr *device_attr,
2354 			    struct ib_udata *udata);
2355 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2356 			     struct ib_device_modify *device_modify);
2357 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2358 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2359 						     int comp_vector);
2360 	int (*query_port)(struct ib_device *device, u32 port_num,
2361 			  struct ib_port_attr *port_attr);
2362 	int (*modify_port)(struct ib_device *device, u32 port_num,
2363 			   int port_modify_mask,
2364 			   struct ib_port_modify *port_modify);
2365 	/**
2366 	 * The following mandatory functions are used only at device
2367 	 * registration.  Keep functions such as these at the end of this
2368 	 * structure to avoid cache line misses when accessing struct ib_device
2369 	 * in fast paths.
2370 	 */
2371 	int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2372 				  struct ib_port_immutable *immutable);
2373 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2374 					       u32 port_num);
2375 	/**
2376 	 * When calling get_netdev, the HW vendor's driver should return the
2377 	 * net device of device @device at port @port_num or NULL if such
2378 	 * a net device doesn't exist. The vendor driver should call dev_hold
2379 	 * on this net device. The HW vendor's device driver must guarantee
2380 	 * that this function returns NULL before the net device has finished
2381 	 * NETDEV_UNREGISTER state.
2382 	 */
2383 	struct net_device *(*get_netdev)(struct ib_device *device,
2384 					 u32 port_num);
2385 	/**
2386 	 * rdma netdev operation
2387 	 *
2388 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2389 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2390 	 */
2391 	struct net_device *(*alloc_rdma_netdev)(
2392 		struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2393 		const char *name, unsigned char name_assign_type,
2394 		void (*setup)(struct net_device *));
2395 
2396 	int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2397 				      enum rdma_netdev_t type,
2398 				      struct rdma_netdev_alloc_params *params);
2399 	/**
2400 	 * query_gid should be return GID value for @device, when @port_num
2401 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2402 	 * is RoCE link layer.
2403 	 */
2404 	int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2405 			 union ib_gid *gid);
2406 	/**
2407 	 * When calling add_gid, the HW vendor's driver should add the gid
2408 	 * of device of port at gid index available at @attr. Meta-info of
2409 	 * that gid (for example, the network device related to this gid) is
2410 	 * available at @attr. @context allows the HW vendor driver to store
2411 	 * extra information together with a GID entry. The HW vendor driver may
2412 	 * allocate memory to contain this information and store it in @context
2413 	 * when a new GID entry is written to. Params are consistent until the
2414 	 * next call of add_gid or delete_gid. The function should return 0 on
2415 	 * success or error otherwise. The function could be called
2416 	 * concurrently for different ports. This function is only called when
2417 	 * roce_gid_table is used.
2418 	 */
2419 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2420 	/**
2421 	 * When calling del_gid, the HW vendor's driver should delete the
2422 	 * gid of device @device at gid index gid_index of port port_num
2423 	 * available in @attr.
2424 	 * Upon the deletion of a GID entry, the HW vendor must free any
2425 	 * allocated memory. The caller will clear @context afterwards.
2426 	 * This function is only called when roce_gid_table is used.
2427 	 */
2428 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2429 	int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2430 			  u16 *pkey);
2431 	int (*alloc_ucontext)(struct ib_ucontext *context,
2432 			      struct ib_udata *udata);
2433 	void (*dealloc_ucontext)(struct ib_ucontext *context);
2434 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2435 	/**
2436 	 * This will be called once refcount of an entry in mmap_xa reaches
2437 	 * zero. The type of the memory that was mapped may differ between
2438 	 * entries and is opaque to the rdma_user_mmap interface.
2439 	 * Therefore needs to be implemented by the driver in mmap_free.
2440 	 */
2441 	void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2442 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2443 	int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2444 	int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2445 	int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2446 			 struct ib_udata *udata);
2447 	int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2448 			      struct ib_udata *udata);
2449 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2450 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2451 	int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2452 	int (*create_srq)(struct ib_srq *srq,
2453 			  struct ib_srq_init_attr *srq_init_attr,
2454 			  struct ib_udata *udata);
2455 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2456 			  enum ib_srq_attr_mask srq_attr_mask,
2457 			  struct ib_udata *udata);
2458 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2459 	int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2460 	int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2461 			 struct ib_udata *udata);
2462 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2463 			 int qp_attr_mask, struct ib_udata *udata);
2464 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2465 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2466 	int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2467 	int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2468 			 struct uverbs_attr_bundle *attrs);
2469 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2470 	int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2471 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2472 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2473 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2474 				     u64 virt_addr, int mr_access_flags,
2475 				     struct ib_udata *udata);
2476 	struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2477 					    u64 length, u64 virt_addr, int fd,
2478 					    int mr_access_flags,
2479 					    struct uverbs_attr_bundle *attrs);
2480 	struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2481 				       u64 length, u64 virt_addr,
2482 				       int mr_access_flags, struct ib_pd *pd,
2483 				       struct ib_udata *udata);
2484 	int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2485 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2486 				  u32 max_num_sg);
2487 	struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2488 					    u32 max_num_data_sg,
2489 					    u32 max_num_meta_sg);
2490 	int (*advise_mr)(struct ib_pd *pd,
2491 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2492 			 struct ib_sge *sg_list, u32 num_sge,
2493 			 struct uverbs_attr_bundle *attrs);
2494 
2495 	/*
2496 	 * Kernel users should universally support relaxed ordering (RO), as
2497 	 * they are designed to read data only after observing the CQE and use
2498 	 * the DMA API correctly.
2499 	 *
2500 	 * Some drivers implicitly enable RO if platform supports it.
2501 	 */
2502 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2503 			 unsigned int *sg_offset);
2504 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2505 			       struct ib_mr_status *mr_status);
2506 	int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2507 	int (*dealloc_mw)(struct ib_mw *mw);
2508 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2509 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2510 	int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2511 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2512 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2513 				       struct ib_flow_attr *flow_attr,
2514 				       struct ib_udata *udata);
2515 	int (*destroy_flow)(struct ib_flow *flow_id);
2516 	int (*destroy_flow_action)(struct ib_flow_action *action);
2517 	int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2518 				 int state);
2519 	int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2520 			     struct ifla_vf_info *ivf);
2521 	int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2522 			    struct ifla_vf_stats *stats);
2523 	int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2524 			    struct ifla_vf_guid *node_guid,
2525 			    struct ifla_vf_guid *port_guid);
2526 	int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2527 			   int type);
2528 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2529 				   struct ib_wq_init_attr *init_attr,
2530 				   struct ib_udata *udata);
2531 	int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2532 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2533 			 u32 wq_attr_mask, struct ib_udata *udata);
2534 	int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2535 				    struct ib_rwq_ind_table_init_attr *init_attr,
2536 				    struct ib_udata *udata);
2537 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2538 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2539 				  struct ib_ucontext *context,
2540 				  struct ib_dm_alloc_attr *attr,
2541 				  struct uverbs_attr_bundle *attrs);
2542 	int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2543 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2544 				   struct ib_dm_mr_attr *attr,
2545 				   struct uverbs_attr_bundle *attrs);
2546 	int (*create_counters)(struct ib_counters *counters,
2547 			       struct uverbs_attr_bundle *attrs);
2548 	int (*destroy_counters)(struct ib_counters *counters);
2549 	int (*read_counters)(struct ib_counters *counters,
2550 			     struct ib_counters_read_attr *counters_read_attr,
2551 			     struct uverbs_attr_bundle *attrs);
2552 	int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2553 			    int data_sg_nents, unsigned int *data_sg_offset,
2554 			    struct scatterlist *meta_sg, int meta_sg_nents,
2555 			    unsigned int *meta_sg_offset);
2556 
2557 	/**
2558 	 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2559 	 *   fill in the driver initialized data.  The struct is kfree()'ed by
2560 	 *   the sysfs core when the device is removed.  A lifespan of -1 in the
2561 	 *   return struct tells the core to set a default lifespan.
2562 	 */
2563 	struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2564 	struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2565 						     u32 port_num);
2566 	/**
2567 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2568 	 * @index - The index in the value array we wish to have updated, or
2569 	 *   num_counters if we want all stats updated
2570 	 * Return codes -
2571 	 *   < 0 - Error, no counters updated
2572 	 *   index - Updated the single counter pointed to by index
2573 	 *   num_counters - Updated all counters (will reset the timestamp
2574 	 *     and prevent further calls for lifespan milliseconds)
2575 	 * Drivers are allowed to update all counters in leiu of just the
2576 	 *   one given in index at their option
2577 	 */
2578 	int (*get_hw_stats)(struct ib_device *device,
2579 			    struct rdma_hw_stats *stats, u32 port, int index);
2580 
2581 	/**
2582 	 * modify_hw_stat - Modify the counter configuration
2583 	 * @enable: true/false when enable/disable a counter
2584 	 * Return codes - 0 on success or error code otherwise.
2585 	 */
2586 	int (*modify_hw_stat)(struct ib_device *device, u32 port,
2587 			      unsigned int counter_index, bool enable);
2588 	/**
2589 	 * Allows rdma drivers to add their own restrack attributes.
2590 	 */
2591 	int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2592 	int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2593 	int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2594 	int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2595 	int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2596 	int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2597 	int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2598 	int (*fill_res_srq_entry)(struct sk_buff *msg, struct ib_srq *ib_srq);
2599 	int (*fill_res_srq_entry_raw)(struct sk_buff *msg, struct ib_srq *ib_srq);
2600 
2601 	/* Device lifecycle callbacks */
2602 	/*
2603 	 * Called after the device becomes registered, before clients are
2604 	 * attached
2605 	 */
2606 	int (*enable_driver)(struct ib_device *dev);
2607 	/*
2608 	 * This is called as part of ib_dealloc_device().
2609 	 */
2610 	void (*dealloc_driver)(struct ib_device *dev);
2611 
2612 	/* iWarp CM callbacks */
2613 	void (*iw_add_ref)(struct ib_qp *qp);
2614 	void (*iw_rem_ref)(struct ib_qp *qp);
2615 	struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2616 	int (*iw_connect)(struct iw_cm_id *cm_id,
2617 			  struct iw_cm_conn_param *conn_param);
2618 	int (*iw_accept)(struct iw_cm_id *cm_id,
2619 			 struct iw_cm_conn_param *conn_param);
2620 	int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2621 			 u8 pdata_len);
2622 	int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2623 	int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2624 	/**
2625 	 * counter_bind_qp - Bind a QP to a counter.
2626 	 * @counter - The counter to be bound. If counter->id is zero then
2627 	 *   the driver needs to allocate a new counter and set counter->id
2628 	 */
2629 	int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2630 	/**
2631 	 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2632 	 *   counter and bind it onto the default one
2633 	 */
2634 	int (*counter_unbind_qp)(struct ib_qp *qp);
2635 	/**
2636 	 * counter_dealloc -De-allocate the hw counter
2637 	 */
2638 	int (*counter_dealloc)(struct rdma_counter *counter);
2639 	/**
2640 	 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2641 	 * the driver initialized data.
2642 	 */
2643 	struct rdma_hw_stats *(*counter_alloc_stats)(
2644 		struct rdma_counter *counter);
2645 	/**
2646 	 * counter_update_stats - Query the stats value of this counter
2647 	 */
2648 	int (*counter_update_stats)(struct rdma_counter *counter);
2649 
2650 	/**
2651 	 * Allows rdma drivers to add their own restrack attributes
2652 	 * dumped via 'rdma stat' iproute2 command.
2653 	 */
2654 	int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2655 
2656 	/* query driver for its ucontext properties */
2657 	int (*query_ucontext)(struct ib_ucontext *context,
2658 			      struct uverbs_attr_bundle *attrs);
2659 
2660 	/*
2661 	 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2662 	 * Everyone else relies on Linux memory management model.
2663 	 */
2664 	int (*get_numa_node)(struct ib_device *dev);
2665 
2666 	/**
2667 	 * add_sub_dev - Add a sub IB device
2668 	 */
2669 	struct ib_device *(*add_sub_dev)(struct ib_device *parent,
2670 					 enum rdma_nl_dev_type type,
2671 					 const char *name);
2672 
2673 	/**
2674 	 * del_sub_dev - Delete a sub IB device
2675 	 */
2676 	void (*del_sub_dev)(struct ib_device *sub_dev);
2677 
2678 	/**
2679 	 * ufile_cleanup - Attempt to cleanup ubojects HW resources inside
2680 	 * the ufile.
2681 	 */
2682 	void (*ufile_hw_cleanup)(struct ib_uverbs_file *ufile);
2683 
2684 	/**
2685 	 * report_port_event - Drivers need to implement this if they have
2686 	 * some private stuff to handle when link status changes.
2687 	 */
2688 	void (*report_port_event)(struct ib_device *ibdev,
2689 				  struct net_device *ndev, unsigned long event);
2690 
2691 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
2692 	DECLARE_RDMA_OBJ_SIZE(ib_counters);
2693 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
2694 	DECLARE_RDMA_OBJ_SIZE(ib_mw);
2695 	DECLARE_RDMA_OBJ_SIZE(ib_pd);
2696 	DECLARE_RDMA_OBJ_SIZE(ib_qp);
2697 	DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2698 	DECLARE_RDMA_OBJ_SIZE(ib_srq);
2699 	DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2700 	DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2701 };
2702 
2703 struct ib_core_device {
2704 	/* device must be the first element in structure until,
2705 	 * union of ib_core_device and device exists in ib_device.
2706 	 */
2707 	struct device dev;
2708 	possible_net_t rdma_net;
2709 	struct kobject *ports_kobj;
2710 	struct list_head port_list;
2711 	struct ib_device *owner; /* reach back to owner ib_device */
2712 };
2713 
2714 struct rdma_restrack_root;
2715 struct ib_device {
2716 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2717 	struct device                *dma_device;
2718 	struct ib_device_ops	     ops;
2719 	char                          name[IB_DEVICE_NAME_MAX];
2720 	struct rcu_head rcu_head;
2721 
2722 	struct list_head              event_handler_list;
2723 	/* Protects event_handler_list */
2724 	struct rw_semaphore event_handler_rwsem;
2725 
2726 	/* Protects QP's event_handler calls and open_qp list */
2727 	spinlock_t qp_open_list_lock;
2728 
2729 	struct rw_semaphore	      client_data_rwsem;
2730 	struct xarray                 client_data;
2731 	struct mutex                  unregistration_lock;
2732 
2733 	/* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2734 	rwlock_t cache_lock;
2735 	/**
2736 	 * port_data is indexed by port number
2737 	 */
2738 	struct ib_port_data *port_data;
2739 
2740 	int			      num_comp_vectors;
2741 
2742 	union {
2743 		struct device		dev;
2744 		struct ib_core_device	coredev;
2745 	};
2746 
2747 	/* First group is for device attributes,
2748 	 * Second group is for driver provided attributes (optional).
2749 	 * Third group is for the hw_stats
2750 	 * It is a NULL terminated array.
2751 	 */
2752 	const struct attribute_group	*groups[4];
2753 
2754 	u64			     uverbs_cmd_mask;
2755 
2756 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2757 	__be64			     node_guid;
2758 	u32			     local_dma_lkey;
2759 	u16                          is_switch:1;
2760 	/* Indicates kernel verbs support, should not be used in drivers */
2761 	u16                          kverbs_provider:1;
2762 	/* CQ adaptive moderation (RDMA DIM) */
2763 	u16                          use_cq_dim:1;
2764 	u8                           node_type;
2765 	u32			     phys_port_cnt;
2766 	struct ib_device_attr        attrs;
2767 	struct hw_stats_device_data *hw_stats_data;
2768 
2769 #ifdef CONFIG_CGROUP_RDMA
2770 	struct rdmacg_device         cg_device;
2771 #endif
2772 
2773 	u32                          index;
2774 
2775 	spinlock_t                   cq_pools_lock;
2776 	struct list_head             cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2777 
2778 	struct rdma_restrack_root *res;
2779 
2780 	const struct uapi_definition   *driver_def;
2781 
2782 	/*
2783 	 * Positive refcount indicates that the device is currently
2784 	 * registered and cannot be unregistered.
2785 	 */
2786 	refcount_t refcount;
2787 	struct completion unreg_completion;
2788 	struct work_struct unregistration_work;
2789 
2790 	const struct rdma_link_ops *link_ops;
2791 
2792 	/* Protects compat_devs xarray modifications */
2793 	struct mutex compat_devs_mutex;
2794 	/* Maintains compat devices for each net namespace */
2795 	struct xarray compat_devs;
2796 
2797 	/* Used by iWarp CM */
2798 	char iw_ifname[IFNAMSIZ];
2799 	u32 iw_driver_flags;
2800 	u32 lag_flags;
2801 
2802 	/* A parent device has a list of sub-devices */
2803 	struct mutex subdev_lock;
2804 	struct list_head subdev_list_head;
2805 
2806 	/* A sub device has a type and a parent */
2807 	enum rdma_nl_dev_type type;
2808 	struct ib_device *parent;
2809 	struct list_head subdev_list;
2810 
2811 	enum rdma_nl_name_assign_type name_assign_type;
2812 };
2813 
2814 static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
2815 				    gfp_t gfp, bool is_numa_aware)
2816 {
2817 	if (is_numa_aware && dev->ops.get_numa_node)
2818 		return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
2819 
2820 	return kzalloc(size, gfp);
2821 }
2822 
2823 struct ib_client_nl_info;
2824 struct ib_client {
2825 	const char *name;
2826 	int (*add)(struct ib_device *ibdev);
2827 	void (*remove)(struct ib_device *, void *client_data);
2828 	void (*rename)(struct ib_device *dev, void *client_data);
2829 	int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2830 			   struct ib_client_nl_info *res);
2831 	int (*get_global_nl_info)(struct ib_client_nl_info *res);
2832 
2833 	/* Returns the net_dev belonging to this ib_client and matching the
2834 	 * given parameters.
2835 	 * @dev:	 An RDMA device that the net_dev use for communication.
2836 	 * @port:	 A physical port number on the RDMA device.
2837 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2838 	 * @gid:	 A GID that the net_dev uses to communicate.
2839 	 * @addr:	 An IP address the net_dev is configured with.
2840 	 * @client_data: The device's client data set by ib_set_client_data().
2841 	 *
2842 	 * An ib_client that implements a net_dev on top of RDMA devices
2843 	 * (such as IP over IB) should implement this callback, allowing the
2844 	 * rdma_cm module to find the right net_dev for a given request.
2845 	 *
2846 	 * The caller is responsible for calling dev_put on the returned
2847 	 * netdev. */
2848 	struct net_device *(*get_net_dev_by_params)(
2849 			struct ib_device *dev,
2850 			u32 port,
2851 			u16 pkey,
2852 			const union ib_gid *gid,
2853 			const struct sockaddr *addr,
2854 			void *client_data);
2855 
2856 	refcount_t uses;
2857 	struct completion uses_zero;
2858 	u32 client_id;
2859 
2860 	/* kverbs are not required by the client */
2861 	u8 no_kverbs_req:1;
2862 };
2863 
2864 /*
2865  * IB block DMA iterator
2866  *
2867  * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2868  * to a HW supported page size.
2869  */
2870 struct ib_block_iter {
2871 	/* internal states */
2872 	struct scatterlist *__sg;	/* sg holding the current aligned block */
2873 	dma_addr_t __dma_addr;		/* unaligned DMA address of this block */
2874 	size_t __sg_numblocks;		/* ib_umem_num_dma_blocks() */
2875 	unsigned int __sg_nents;	/* number of SG entries */
2876 	unsigned int __sg_advance;	/* number of bytes to advance in sg in next step */
2877 	unsigned int __pg_bit;		/* alignment of current block */
2878 };
2879 
2880 struct ib_device *_ib_alloc_device(size_t size);
2881 #define ib_alloc_device(drv_struct, member)                                    \
2882 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2883 				      BUILD_BUG_ON_ZERO(offsetof(              \
2884 					      struct drv_struct, member))),    \
2885 		     struct drv_struct, member)
2886 
2887 void ib_dealloc_device(struct ib_device *device);
2888 
2889 void ib_get_device_fw_str(struct ib_device *device, char *str);
2890 
2891 int ib_register_device(struct ib_device *device, const char *name,
2892 		       struct device *dma_device);
2893 void ib_unregister_device(struct ib_device *device);
2894 void ib_unregister_driver(enum rdma_driver_id driver_id);
2895 void ib_unregister_device_and_put(struct ib_device *device);
2896 void ib_unregister_device_queued(struct ib_device *ib_dev);
2897 
2898 int ib_register_client   (struct ib_client *client);
2899 void ib_unregister_client(struct ib_client *client);
2900 
2901 void __rdma_block_iter_start(struct ib_block_iter *biter,
2902 			     struct scatterlist *sglist,
2903 			     unsigned int nents,
2904 			     unsigned long pgsz);
2905 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2906 
2907 /**
2908  * rdma_block_iter_dma_address - get the aligned dma address of the current
2909  * block held by the block iterator.
2910  * @biter: block iterator holding the memory block
2911  */
2912 static inline dma_addr_t
2913 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2914 {
2915 	return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2916 }
2917 
2918 /**
2919  * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2920  * @sglist: sglist to iterate over
2921  * @biter: block iterator holding the memory block
2922  * @nents: maximum number of sg entries to iterate over
2923  * @pgsz: best HW supported page size to use
2924  *
2925  * Callers may use rdma_block_iter_dma_address() to get each
2926  * blocks aligned DMA address.
2927  */
2928 #define rdma_for_each_block(sglist, biter, nents, pgsz)		\
2929 	for (__rdma_block_iter_start(biter, sglist, nents,	\
2930 				     pgsz);			\
2931 	     __rdma_block_iter_next(biter);)
2932 
2933 /**
2934  * ib_get_client_data - Get IB client context
2935  * @device:Device to get context for
2936  * @client:Client to get context for
2937  *
2938  * ib_get_client_data() returns the client context data set with
2939  * ib_set_client_data(). This can only be called while the client is
2940  * registered to the device, once the ib_client remove() callback returns this
2941  * cannot be called.
2942  */
2943 static inline void *ib_get_client_data(struct ib_device *device,
2944 				       struct ib_client *client)
2945 {
2946 	return xa_load(&device->client_data, client->client_id);
2947 }
2948 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2949 			 void *data);
2950 void ib_set_device_ops(struct ib_device *device,
2951 		       const struct ib_device_ops *ops);
2952 
2953 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2954 		      unsigned long pfn, unsigned long size, pgprot_t prot,
2955 		      struct rdma_user_mmap_entry *entry);
2956 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2957 				struct rdma_user_mmap_entry *entry,
2958 				size_t length);
2959 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2960 				      struct rdma_user_mmap_entry *entry,
2961 				      size_t length, u32 min_pgoff,
2962 				      u32 max_pgoff);
2963 
2964 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2965 void rdma_user_mmap_disassociate(struct ib_device *device);
2966 #else
2967 static inline void rdma_user_mmap_disassociate(struct ib_device *device)
2968 {
2969 }
2970 #endif
2971 
2972 static inline int
2973 rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
2974 				  struct rdma_user_mmap_entry *entry,
2975 				  size_t length, u32 pgoff)
2976 {
2977 	return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff,
2978 						 pgoff);
2979 }
2980 
2981 struct rdma_user_mmap_entry *
2982 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2983 			       unsigned long pgoff);
2984 struct rdma_user_mmap_entry *
2985 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2986 			 struct vm_area_struct *vma);
2987 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2988 
2989 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2990 
2991 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2992 {
2993 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2994 }
2995 
2996 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2997 {
2998 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2999 }
3000 
3001 static inline bool ib_is_buffer_cleared(const void __user *p,
3002 					size_t len)
3003 {
3004 	bool ret;
3005 	u8 *buf;
3006 
3007 	if (len > USHRT_MAX)
3008 		return false;
3009 
3010 	buf = memdup_user(p, len);
3011 	if (IS_ERR(buf))
3012 		return false;
3013 
3014 	ret = !memchr_inv(buf, 0, len);
3015 	kfree(buf);
3016 	return ret;
3017 }
3018 
3019 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
3020 				       size_t offset,
3021 				       size_t len)
3022 {
3023 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
3024 }
3025 
3026 /**
3027  * ib_modify_qp_is_ok - Check that the supplied attribute mask
3028  * contains all required attributes and no attributes not allowed for
3029  * the given QP state transition.
3030  * @cur_state: Current QP state
3031  * @next_state: Next QP state
3032  * @type: QP type
3033  * @mask: Mask of supplied QP attributes
3034  *
3035  * This function is a helper function that a low-level driver's
3036  * modify_qp method can use to validate the consumer's input.  It
3037  * checks that cur_state and next_state are valid QP states, that a
3038  * transition from cur_state to next_state is allowed by the IB spec,
3039  * and that the attribute mask supplied is allowed for the transition.
3040  */
3041 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
3042 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
3043 
3044 void ib_register_event_handler(struct ib_event_handler *event_handler);
3045 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
3046 void ib_dispatch_event(const struct ib_event *event);
3047 
3048 int ib_query_port(struct ib_device *device,
3049 		  u32 port_num, struct ib_port_attr *port_attr);
3050 
3051 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3052 					       u32 port_num);
3053 
3054 /**
3055  * rdma_cap_ib_switch - Check if the device is IB switch
3056  * @device: Device to check
3057  *
3058  * Device driver is responsible for setting is_switch bit on
3059  * in ib_device structure at init time.
3060  *
3061  * Return: true if the device is IB switch.
3062  */
3063 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3064 {
3065 	return device->is_switch;
3066 }
3067 
3068 /**
3069  * rdma_start_port - Return the first valid port number for the device
3070  * specified
3071  *
3072  * @device: Device to be checked
3073  *
3074  * Return start port number
3075  */
3076 static inline u32 rdma_start_port(const struct ib_device *device)
3077 {
3078 	return rdma_cap_ib_switch(device) ? 0 : 1;
3079 }
3080 
3081 /**
3082  * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3083  * @device - The struct ib_device * to iterate over
3084  * @iter - The unsigned int to store the port number
3085  */
3086 #define rdma_for_each_port(device, iter)                                       \
3087 	for (iter = rdma_start_port(device +				       \
3088 				    BUILD_BUG_ON_ZERO(!__same_type(u32,	       \
3089 								   iter)));    \
3090 	     iter <= rdma_end_port(device); iter++)
3091 
3092 /**
3093  * rdma_end_port - Return the last valid port number for the device
3094  * specified
3095  *
3096  * @device: Device to be checked
3097  *
3098  * Return last port number
3099  */
3100 static inline u32 rdma_end_port(const struct ib_device *device)
3101 {
3102 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3103 }
3104 
3105 static inline int rdma_is_port_valid(const struct ib_device *device,
3106 				     unsigned int port)
3107 {
3108 	return (port >= rdma_start_port(device) &&
3109 		port <= rdma_end_port(device));
3110 }
3111 
3112 static inline bool rdma_is_grh_required(const struct ib_device *device,
3113 					u32 port_num)
3114 {
3115 	return device->port_data[port_num].immutable.core_cap_flags &
3116 	       RDMA_CORE_PORT_IB_GRH_REQUIRED;
3117 }
3118 
3119 static inline bool rdma_protocol_ib(const struct ib_device *device,
3120 				    u32 port_num)
3121 {
3122 	return device->port_data[port_num].immutable.core_cap_flags &
3123 	       RDMA_CORE_CAP_PROT_IB;
3124 }
3125 
3126 static inline bool rdma_protocol_roce(const struct ib_device *device,
3127 				      u32 port_num)
3128 {
3129 	return device->port_data[port_num].immutable.core_cap_flags &
3130 	       (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3131 }
3132 
3133 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3134 						u32 port_num)
3135 {
3136 	return device->port_data[port_num].immutable.core_cap_flags &
3137 	       RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3138 }
3139 
3140 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3141 						u32 port_num)
3142 {
3143 	return device->port_data[port_num].immutable.core_cap_flags &
3144 	       RDMA_CORE_CAP_PROT_ROCE;
3145 }
3146 
3147 static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3148 				       u32 port_num)
3149 {
3150 	return device->port_data[port_num].immutable.core_cap_flags &
3151 	       RDMA_CORE_CAP_PROT_IWARP;
3152 }
3153 
3154 static inline bool rdma_ib_or_roce(const struct ib_device *device,
3155 				   u32 port_num)
3156 {
3157 	return rdma_protocol_ib(device, port_num) ||
3158 		rdma_protocol_roce(device, port_num);
3159 }
3160 
3161 static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3162 					    u32 port_num)
3163 {
3164 	return device->port_data[port_num].immutable.core_cap_flags &
3165 	       RDMA_CORE_CAP_PROT_RAW_PACKET;
3166 }
3167 
3168 static inline bool rdma_protocol_usnic(const struct ib_device *device,
3169 				       u32 port_num)
3170 {
3171 	return device->port_data[port_num].immutable.core_cap_flags &
3172 	       RDMA_CORE_CAP_PROT_USNIC;
3173 }
3174 
3175 /**
3176  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3177  * Management Datagrams.
3178  * @device: Device to check
3179  * @port_num: Port number to check
3180  *
3181  * Management Datagrams (MAD) are a required part of the InfiniBand
3182  * specification and are supported on all InfiniBand devices.  A slightly
3183  * extended version are also supported on OPA interfaces.
3184  *
3185  * Return: true if the port supports sending/receiving of MAD packets.
3186  */
3187 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3188 {
3189 	return device->port_data[port_num].immutable.core_cap_flags &
3190 	       RDMA_CORE_CAP_IB_MAD;
3191 }
3192 
3193 /**
3194  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3195  * Management Datagrams.
3196  * @device: Device to check
3197  * @port_num: Port number to check
3198  *
3199  * Intel OmniPath devices extend and/or replace the InfiniBand Management
3200  * datagrams with their own versions.  These OPA MADs share many but not all of
3201  * the characteristics of InfiniBand MADs.
3202  *
3203  * OPA MADs differ in the following ways:
3204  *
3205  *    1) MADs are variable size up to 2K
3206  *       IBTA defined MADs remain fixed at 256 bytes
3207  *    2) OPA SMPs must carry valid PKeys
3208  *    3) OPA SMP packets are a different format
3209  *
3210  * Return: true if the port supports OPA MAD packet formats.
3211  */
3212 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3213 {
3214 	return device->port_data[port_num].immutable.core_cap_flags &
3215 		RDMA_CORE_CAP_OPA_MAD;
3216 }
3217 
3218 /**
3219  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3220  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3221  * @device: Device to check
3222  * @port_num: Port number to check
3223  *
3224  * Each InfiniBand node is required to provide a Subnet Management Agent
3225  * that the subnet manager can access.  Prior to the fabric being fully
3226  * configured by the subnet manager, the SMA is accessed via a well known
3227  * interface called the Subnet Management Interface (SMI).  This interface
3228  * uses directed route packets to communicate with the SM to get around the
3229  * chicken and egg problem of the SM needing to know what's on the fabric
3230  * in order to configure the fabric, and needing to configure the fabric in
3231  * order to send packets to the devices on the fabric.  These directed
3232  * route packets do not need the fabric fully configured in order to reach
3233  * their destination.  The SMI is the only method allowed to send
3234  * directed route packets on an InfiniBand fabric.
3235  *
3236  * Return: true if the port provides an SMI.
3237  */
3238 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3239 {
3240 	return device->port_data[port_num].immutable.core_cap_flags &
3241 	       RDMA_CORE_CAP_IB_SMI;
3242 }
3243 
3244 /**
3245  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3246  * Communication Manager.
3247  * @device: Device to check
3248  * @port_num: Port number to check
3249  *
3250  * The InfiniBand Communication Manager is one of many pre-defined General
3251  * Service Agents (GSA) that are accessed via the General Service
3252  * Interface (GSI).  It's role is to facilitate establishment of connections
3253  * between nodes as well as other management related tasks for established
3254  * connections.
3255  *
3256  * Return: true if the port supports an IB CM (this does not guarantee that
3257  * a CM is actually running however).
3258  */
3259 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3260 {
3261 	return device->port_data[port_num].immutable.core_cap_flags &
3262 	       RDMA_CORE_CAP_IB_CM;
3263 }
3264 
3265 /**
3266  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3267  * Communication Manager.
3268  * @device: Device to check
3269  * @port_num: Port number to check
3270  *
3271  * Similar to above, but specific to iWARP connections which have a different
3272  * managment protocol than InfiniBand.
3273  *
3274  * Return: true if the port supports an iWARP CM (this does not guarantee that
3275  * a CM is actually running however).
3276  */
3277 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3278 {
3279 	return device->port_data[port_num].immutable.core_cap_flags &
3280 	       RDMA_CORE_CAP_IW_CM;
3281 }
3282 
3283 /**
3284  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3285  * Subnet Administration.
3286  * @device: Device to check
3287  * @port_num: Port number to check
3288  *
3289  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3290  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
3291  * fabrics, devices should resolve routes to other hosts by contacting the
3292  * SA to query the proper route.
3293  *
3294  * Return: true if the port should act as a client to the fabric Subnet
3295  * Administration interface.  This does not imply that the SA service is
3296  * running locally.
3297  */
3298 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3299 {
3300 	return device->port_data[port_num].immutable.core_cap_flags &
3301 	       RDMA_CORE_CAP_IB_SA;
3302 }
3303 
3304 /**
3305  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3306  * Multicast.
3307  * @device: Device to check
3308  * @port_num: Port number to check
3309  *
3310  * InfiniBand multicast registration is more complex than normal IPv4 or
3311  * IPv6 multicast registration.  Each Host Channel Adapter must register
3312  * with the Subnet Manager when it wishes to join a multicast group.  It
3313  * should do so only once regardless of how many queue pairs it subscribes
3314  * to this group.  And it should leave the group only after all queue pairs
3315  * attached to the group have been detached.
3316  *
3317  * Return: true if the port must undertake the additional adminstrative
3318  * overhead of registering/unregistering with the SM and tracking of the
3319  * total number of queue pairs attached to the multicast group.
3320  */
3321 static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3322 				     u32 port_num)
3323 {
3324 	return rdma_cap_ib_sa(device, port_num);
3325 }
3326 
3327 /**
3328  * rdma_cap_af_ib - Check if the port of device has the capability
3329  * Native Infiniband Address.
3330  * @device: Device to check
3331  * @port_num: Port number to check
3332  *
3333  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3334  * GID.  RoCE uses a different mechanism, but still generates a GID via
3335  * a prescribed mechanism and port specific data.
3336  *
3337  * Return: true if the port uses a GID address to identify devices on the
3338  * network.
3339  */
3340 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3341 {
3342 	return device->port_data[port_num].immutable.core_cap_flags &
3343 	       RDMA_CORE_CAP_AF_IB;
3344 }
3345 
3346 /**
3347  * rdma_cap_eth_ah - Check if the port of device has the capability
3348  * Ethernet Address Handle.
3349  * @device: Device to check
3350  * @port_num: Port number to check
3351  *
3352  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3353  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3354  * port.  Normally, packet headers are generated by the sending host
3355  * adapter, but when sending connectionless datagrams, we must manually
3356  * inject the proper headers for the fabric we are communicating over.
3357  *
3358  * Return: true if we are running as a RoCE port and must force the
3359  * addition of a Global Route Header built from our Ethernet Address
3360  * Handle into our header list for connectionless packets.
3361  */
3362 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3363 {
3364 	return device->port_data[port_num].immutable.core_cap_flags &
3365 	       RDMA_CORE_CAP_ETH_AH;
3366 }
3367 
3368 /**
3369  * rdma_cap_opa_ah - Check if the port of device supports
3370  * OPA Address handles
3371  * @device: Device to check
3372  * @port_num: Port number to check
3373  *
3374  * Return: true if we are running on an OPA device which supports
3375  * the extended OPA addressing.
3376  */
3377 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3378 {
3379 	return (device->port_data[port_num].immutable.core_cap_flags &
3380 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3381 }
3382 
3383 /**
3384  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3385  *
3386  * @device: Device
3387  * @port_num: Port number
3388  *
3389  * This MAD size includes the MAD headers and MAD payload.  No other headers
3390  * are included.
3391  *
3392  * Return the max MAD size required by the Port.  Will return 0 if the port
3393  * does not support MADs
3394  */
3395 static inline size_t rdma_max_mad_size(const struct ib_device *device,
3396 				       u32 port_num)
3397 {
3398 	return device->port_data[port_num].immutable.max_mad_size;
3399 }
3400 
3401 /**
3402  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3403  * @device: Device to check
3404  * @port_num: Port number to check
3405  *
3406  * RoCE GID table mechanism manages the various GIDs for a device.
3407  *
3408  * NOTE: if allocating the port's GID table has failed, this call will still
3409  * return true, but any RoCE GID table API will fail.
3410  *
3411  * Return: true if the port uses RoCE GID table mechanism in order to manage
3412  * its GIDs.
3413  */
3414 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3415 					   u32 port_num)
3416 {
3417 	return rdma_protocol_roce(device, port_num) &&
3418 		device->ops.add_gid && device->ops.del_gid;
3419 }
3420 
3421 /*
3422  * Check if the device supports READ W/ INVALIDATE.
3423  */
3424 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3425 {
3426 	/*
3427 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3428 	 * has support for it yet.
3429 	 */
3430 	return rdma_protocol_iwarp(dev, port_num);
3431 }
3432 
3433 /**
3434  * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3435  * @device: Device
3436  * @port_num: 1 based Port number
3437  *
3438  * Return true if port is an Intel OPA port , false if not
3439  */
3440 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3441 					  u32 port_num)
3442 {
3443 	return (device->port_data[port_num].immutable.core_cap_flags &
3444 		RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3445 }
3446 
3447 /**
3448  * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3449  * @device: Device
3450  * @port_num: Port number
3451  * @mtu: enum value of MTU
3452  *
3453  * Return the MTU size supported by the port as an integer value. Will return
3454  * -1 if enum value of mtu is not supported.
3455  */
3456 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3457 				       int mtu)
3458 {
3459 	if (rdma_core_cap_opa_port(device, port))
3460 		return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3461 	else
3462 		return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3463 }
3464 
3465 /**
3466  * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3467  * @device: Device
3468  * @port_num: Port number
3469  * @attr: port attribute
3470  *
3471  * Return the MTU size supported by the port as an integer value.
3472  */
3473 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3474 				     struct ib_port_attr *attr)
3475 {
3476 	if (rdma_core_cap_opa_port(device, port))
3477 		return attr->phys_mtu;
3478 	else
3479 		return ib_mtu_enum_to_int(attr->max_mtu);
3480 }
3481 
3482 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3483 			 int state);
3484 int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3485 		     struct ifla_vf_info *info);
3486 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3487 		    struct ifla_vf_stats *stats);
3488 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3489 		    struct ifla_vf_guid *node_guid,
3490 		    struct ifla_vf_guid *port_guid);
3491 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3492 		   int type);
3493 
3494 int ib_query_pkey(struct ib_device *device,
3495 		  u32 port_num, u16 index, u16 *pkey);
3496 
3497 int ib_modify_device(struct ib_device *device,
3498 		     int device_modify_mask,
3499 		     struct ib_device_modify *device_modify);
3500 
3501 int ib_modify_port(struct ib_device *device,
3502 		   u32 port_num, int port_modify_mask,
3503 		   struct ib_port_modify *port_modify);
3504 
3505 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3506 		u32 *port_num, u16 *index);
3507 
3508 int ib_find_pkey(struct ib_device *device,
3509 		 u32 port_num, u16 pkey, u16 *index);
3510 
3511 enum ib_pd_flags {
3512 	/*
3513 	 * Create a memory registration for all memory in the system and place
3514 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3515 	 * ULPs to avoid the overhead of dynamic MRs.
3516 	 *
3517 	 * This flag is generally considered unsafe and must only be used in
3518 	 * extremly trusted environments.  Every use of it will log a warning
3519 	 * in the kernel log.
3520 	 */
3521 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3522 };
3523 
3524 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3525 		const char *caller);
3526 
3527 /**
3528  * ib_alloc_pd - Allocates an unused protection domain.
3529  * @device: The device on which to allocate the protection domain.
3530  * @flags: protection domain flags
3531  *
3532  * A protection domain object provides an association between QPs, shared
3533  * receive queues, address handles, memory regions, and memory windows.
3534  *
3535  * Every PD has a local_dma_lkey which can be used as the lkey value for local
3536  * memory operations.
3537  */
3538 #define ib_alloc_pd(device, flags) \
3539 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3540 
3541 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3542 
3543 /**
3544  * ib_dealloc_pd - Deallocate kernel PD
3545  * @pd: The protection domain
3546  *
3547  * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3548  */
3549 static inline void ib_dealloc_pd(struct ib_pd *pd)
3550 {
3551 	int ret = ib_dealloc_pd_user(pd, NULL);
3552 
3553 	WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3554 }
3555 
3556 enum rdma_create_ah_flags {
3557 	/* In a sleepable context */
3558 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3559 };
3560 
3561 /**
3562  * rdma_create_ah - Creates an address handle for the given address vector.
3563  * @pd: The protection domain associated with the address handle.
3564  * @ah_attr: The attributes of the address vector.
3565  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3566  *
3567  * The address handle is used to reference a local or global destination
3568  * in all UD QP post sends.
3569  */
3570 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3571 			     u32 flags);
3572 
3573 /**
3574  * rdma_create_user_ah - Creates an address handle for the given address vector.
3575  * It resolves destination mac address for ah attribute of RoCE type.
3576  * @pd: The protection domain associated with the address handle.
3577  * @ah_attr: The attributes of the address vector.
3578  * @udata: pointer to user's input output buffer information need by
3579  *         provider driver.
3580  *
3581  * It returns 0 on success and returns appropriate error code on error.
3582  * The address handle is used to reference a local or global destination
3583  * in all UD QP post sends.
3584  */
3585 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3586 				  struct rdma_ah_attr *ah_attr,
3587 				  struct ib_udata *udata);
3588 /**
3589  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3590  *   work completion.
3591  * @hdr: the L3 header to parse
3592  * @net_type: type of header to parse
3593  * @sgid: place to store source gid
3594  * @dgid: place to store destination gid
3595  */
3596 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3597 			      enum rdma_network_type net_type,
3598 			      union ib_gid *sgid, union ib_gid *dgid);
3599 
3600 /**
3601  * ib_get_rdma_header_version - Get the header version
3602  * @hdr: the L3 header to parse
3603  */
3604 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3605 
3606 /**
3607  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3608  *   work completion.
3609  * @device: Device on which the received message arrived.
3610  * @port_num: Port on which the received message arrived.
3611  * @wc: Work completion associated with the received message.
3612  * @grh: References the received global route header.  This parameter is
3613  *   ignored unless the work completion indicates that the GRH is valid.
3614  * @ah_attr: Returned attributes that can be used when creating an address
3615  *   handle for replying to the message.
3616  * When ib_init_ah_attr_from_wc() returns success,
3617  * (a) for IB link layer it optionally contains a reference to SGID attribute
3618  * when GRH is present for IB link layer.
3619  * (b) for RoCE link layer it contains a reference to SGID attribute.
3620  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3621  * attributes which are initialized using ib_init_ah_attr_from_wc().
3622  *
3623  */
3624 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3625 			    const struct ib_wc *wc, const struct ib_grh *grh,
3626 			    struct rdma_ah_attr *ah_attr);
3627 
3628 /**
3629  * ib_create_ah_from_wc - Creates an address handle associated with the
3630  *   sender of the specified work completion.
3631  * @pd: The protection domain associated with the address handle.
3632  * @wc: Work completion information associated with a received message.
3633  * @grh: References the received global route header.  This parameter is
3634  *   ignored unless the work completion indicates that the GRH is valid.
3635  * @port_num: The outbound port number to associate with the address.
3636  *
3637  * The address handle is used to reference a local or global destination
3638  * in all UD QP post sends.
3639  */
3640 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3641 				   const struct ib_grh *grh, u32 port_num);
3642 
3643 /**
3644  * rdma_modify_ah - Modifies the address vector associated with an address
3645  *   handle.
3646  * @ah: The address handle to modify.
3647  * @ah_attr: The new address vector attributes to associate with the
3648  *   address handle.
3649  */
3650 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3651 
3652 /**
3653  * rdma_query_ah - Queries the address vector associated with an address
3654  *   handle.
3655  * @ah: The address handle to query.
3656  * @ah_attr: The address vector attributes associated with the address
3657  *   handle.
3658  */
3659 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3660 
3661 enum rdma_destroy_ah_flags {
3662 	/* In a sleepable context */
3663 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3664 };
3665 
3666 /**
3667  * rdma_destroy_ah_user - Destroys an address handle.
3668  * @ah: The address handle to destroy.
3669  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3670  * @udata: Valid user data or NULL for kernel objects
3671  */
3672 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3673 
3674 /**
3675  * rdma_destroy_ah - Destroys an kernel address handle.
3676  * @ah: The address handle to destroy.
3677  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3678  *
3679  * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3680  */
3681 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3682 {
3683 	int ret = rdma_destroy_ah_user(ah, flags, NULL);
3684 
3685 	WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3686 }
3687 
3688 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3689 				  struct ib_srq_init_attr *srq_init_attr,
3690 				  struct ib_usrq_object *uobject,
3691 				  struct ib_udata *udata);
3692 static inline struct ib_srq *
3693 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3694 {
3695 	if (!pd->device->ops.create_srq)
3696 		return ERR_PTR(-EOPNOTSUPP);
3697 
3698 	return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3699 }
3700 
3701 /**
3702  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3703  * @srq: The SRQ to modify.
3704  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3705  *   the current values of selected SRQ attributes are returned.
3706  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3707  *   are being modified.
3708  *
3709  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3710  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3711  * the number of receives queued drops below the limit.
3712  */
3713 int ib_modify_srq(struct ib_srq *srq,
3714 		  struct ib_srq_attr *srq_attr,
3715 		  enum ib_srq_attr_mask srq_attr_mask);
3716 
3717 /**
3718  * ib_query_srq - Returns the attribute list and current values for the
3719  *   specified SRQ.
3720  * @srq: The SRQ to query.
3721  * @srq_attr: The attributes of the specified SRQ.
3722  */
3723 int ib_query_srq(struct ib_srq *srq,
3724 		 struct ib_srq_attr *srq_attr);
3725 
3726 /**
3727  * ib_destroy_srq_user - Destroys the specified SRQ.
3728  * @srq: The SRQ to destroy.
3729  * @udata: Valid user data or NULL for kernel objects
3730  */
3731 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3732 
3733 /**
3734  * ib_destroy_srq - Destroys the specified kernel SRQ.
3735  * @srq: The SRQ to destroy.
3736  *
3737  * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3738  */
3739 static inline void ib_destroy_srq(struct ib_srq *srq)
3740 {
3741 	int ret = ib_destroy_srq_user(srq, NULL);
3742 
3743 	WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3744 }
3745 
3746 /**
3747  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3748  * @srq: The SRQ to post the work request on.
3749  * @recv_wr: A list of work requests to post on the receive queue.
3750  * @bad_recv_wr: On an immediate failure, this parameter will reference
3751  *   the work request that failed to be posted on the QP.
3752  */
3753 static inline int ib_post_srq_recv(struct ib_srq *srq,
3754 				   const struct ib_recv_wr *recv_wr,
3755 				   const struct ib_recv_wr **bad_recv_wr)
3756 {
3757 	const struct ib_recv_wr *dummy;
3758 
3759 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3760 					      bad_recv_wr ? : &dummy);
3761 }
3762 
3763 struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3764 				  struct ib_qp_init_attr *qp_init_attr,
3765 				  const char *caller);
3766 /**
3767  * ib_create_qp - Creates a kernel QP associated with the specific protection
3768  * domain.
3769  * @pd: The protection domain associated with the QP.
3770  * @init_attr: A list of initial attributes required to create the
3771  *   QP.  If QP creation succeeds, then the attributes are updated to
3772  *   the actual capabilities of the created QP.
3773  */
3774 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3775 					 struct ib_qp_init_attr *init_attr)
3776 {
3777 	return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
3778 }
3779 
3780 /**
3781  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3782  * @qp: The QP to modify.
3783  * @attr: On input, specifies the QP attributes to modify.  On output,
3784  *   the current values of selected QP attributes are returned.
3785  * @attr_mask: A bit-mask used to specify which attributes of the QP
3786  *   are being modified.
3787  * @udata: pointer to user's input output buffer information
3788  *   are being modified.
3789  * It returns 0 on success and returns appropriate error code on error.
3790  */
3791 int ib_modify_qp_with_udata(struct ib_qp *qp,
3792 			    struct ib_qp_attr *attr,
3793 			    int attr_mask,
3794 			    struct ib_udata *udata);
3795 
3796 /**
3797  * ib_modify_qp - Modifies the attributes for the specified QP and then
3798  *   transitions the QP to the given state.
3799  * @qp: The QP to modify.
3800  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3801  *   the current values of selected QP attributes are returned.
3802  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3803  *   are being modified.
3804  */
3805 int ib_modify_qp(struct ib_qp *qp,
3806 		 struct ib_qp_attr *qp_attr,
3807 		 int qp_attr_mask);
3808 
3809 /**
3810  * ib_query_qp - Returns the attribute list and current values for the
3811  *   specified QP.
3812  * @qp: The QP to query.
3813  * @qp_attr: The attributes of the specified QP.
3814  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3815  * @qp_init_attr: Additional attributes of the selected QP.
3816  *
3817  * The qp_attr_mask may be used to limit the query to gathering only the
3818  * selected attributes.
3819  */
3820 int ib_query_qp(struct ib_qp *qp,
3821 		struct ib_qp_attr *qp_attr,
3822 		int qp_attr_mask,
3823 		struct ib_qp_init_attr *qp_init_attr);
3824 
3825 /**
3826  * ib_destroy_qp - Destroys the specified QP.
3827  * @qp: The QP to destroy.
3828  * @udata: Valid udata or NULL for kernel objects
3829  */
3830 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3831 
3832 /**
3833  * ib_destroy_qp - Destroys the specified kernel QP.
3834  * @qp: The QP to destroy.
3835  *
3836  * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3837  */
3838 static inline int ib_destroy_qp(struct ib_qp *qp)
3839 {
3840 	return ib_destroy_qp_user(qp, NULL);
3841 }
3842 
3843 /**
3844  * ib_open_qp - Obtain a reference to an existing sharable QP.
3845  * @xrcd - XRC domain
3846  * @qp_open_attr: Attributes identifying the QP to open.
3847  *
3848  * Returns a reference to a sharable QP.
3849  */
3850 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3851 			 struct ib_qp_open_attr *qp_open_attr);
3852 
3853 /**
3854  * ib_close_qp - Release an external reference to a QP.
3855  * @qp: The QP handle to release
3856  *
3857  * The opened QP handle is released by the caller.  The underlying
3858  * shared QP is not destroyed until all internal references are released.
3859  */
3860 int ib_close_qp(struct ib_qp *qp);
3861 
3862 /**
3863  * ib_post_send - Posts a list of work requests to the send queue of
3864  *   the specified QP.
3865  * @qp: The QP to post the work request on.
3866  * @send_wr: A list of work requests to post on the send queue.
3867  * @bad_send_wr: On an immediate failure, this parameter will reference
3868  *   the work request that failed to be posted on the QP.
3869  *
3870  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3871  * error is returned, the QP state shall not be affected,
3872  * ib_post_send() will return an immediate error after queueing any
3873  * earlier work requests in the list.
3874  */
3875 static inline int ib_post_send(struct ib_qp *qp,
3876 			       const struct ib_send_wr *send_wr,
3877 			       const struct ib_send_wr **bad_send_wr)
3878 {
3879 	const struct ib_send_wr *dummy;
3880 
3881 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3882 }
3883 
3884 /**
3885  * ib_post_recv - Posts a list of work requests to the receive queue of
3886  *   the specified QP.
3887  * @qp: The QP to post the work request on.
3888  * @recv_wr: A list of work requests to post on the receive queue.
3889  * @bad_recv_wr: On an immediate failure, this parameter will reference
3890  *   the work request that failed to be posted on the QP.
3891  */
3892 static inline int ib_post_recv(struct ib_qp *qp,
3893 			       const struct ib_recv_wr *recv_wr,
3894 			       const struct ib_recv_wr **bad_recv_wr)
3895 {
3896 	const struct ib_recv_wr *dummy;
3897 
3898 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3899 }
3900 
3901 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3902 			    int comp_vector, enum ib_poll_context poll_ctx,
3903 			    const char *caller);
3904 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3905 					int nr_cqe, int comp_vector,
3906 					enum ib_poll_context poll_ctx)
3907 {
3908 	return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3909 			     KBUILD_MODNAME);
3910 }
3911 
3912 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3913 				int nr_cqe, enum ib_poll_context poll_ctx,
3914 				const char *caller);
3915 
3916 /**
3917  * ib_alloc_cq_any: Allocate kernel CQ
3918  * @dev: The IB device
3919  * @private: Private data attached to the CQE
3920  * @nr_cqe: Number of CQEs in the CQ
3921  * @poll_ctx: Context used for polling the CQ
3922  */
3923 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3924 					    void *private, int nr_cqe,
3925 					    enum ib_poll_context poll_ctx)
3926 {
3927 	return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3928 				 KBUILD_MODNAME);
3929 }
3930 
3931 void ib_free_cq(struct ib_cq *cq);
3932 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3933 
3934 /**
3935  * ib_create_cq - Creates a CQ on the specified device.
3936  * @device: The device on which to create the CQ.
3937  * @comp_handler: A user-specified callback that is invoked when a
3938  *   completion event occurs on the CQ.
3939  * @event_handler: A user-specified callback that is invoked when an
3940  *   asynchronous event not associated with a completion occurs on the CQ.
3941  * @cq_context: Context associated with the CQ returned to the user via
3942  *   the associated completion and event handlers.
3943  * @cq_attr: The attributes the CQ should be created upon.
3944  *
3945  * Users can examine the cq structure to determine the actual CQ size.
3946  */
3947 struct ib_cq *__ib_create_cq(struct ib_device *device,
3948 			     ib_comp_handler comp_handler,
3949 			     void (*event_handler)(struct ib_event *, void *),
3950 			     void *cq_context,
3951 			     const struct ib_cq_init_attr *cq_attr,
3952 			     const char *caller);
3953 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3954 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3955 
3956 /**
3957  * ib_resize_cq - Modifies the capacity of the CQ.
3958  * @cq: The CQ to resize.
3959  * @cqe: The minimum size of the CQ.
3960  *
3961  * Users can examine the cq structure to determine the actual CQ size.
3962  */
3963 int ib_resize_cq(struct ib_cq *cq, int cqe);
3964 
3965 /**
3966  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3967  * @cq: The CQ to modify.
3968  * @cq_count: number of CQEs that will trigger an event
3969  * @cq_period: max period of time in usec before triggering an event
3970  *
3971  */
3972 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3973 
3974 /**
3975  * ib_destroy_cq_user - Destroys the specified CQ.
3976  * @cq: The CQ to destroy.
3977  * @udata: Valid user data or NULL for kernel objects
3978  */
3979 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3980 
3981 /**
3982  * ib_destroy_cq - Destroys the specified kernel CQ.
3983  * @cq: The CQ to destroy.
3984  *
3985  * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3986  */
3987 static inline void ib_destroy_cq(struct ib_cq *cq)
3988 {
3989 	int ret = ib_destroy_cq_user(cq, NULL);
3990 
3991 	WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3992 }
3993 
3994 /**
3995  * ib_poll_cq - poll a CQ for completion(s)
3996  * @cq:the CQ being polled
3997  * @num_entries:maximum number of completions to return
3998  * @wc:array of at least @num_entries &struct ib_wc where completions
3999  *   will be returned
4000  *
4001  * Poll a CQ for (possibly multiple) completions.  If the return value
4002  * is < 0, an error occurred.  If the return value is >= 0, it is the
4003  * number of completions returned.  If the return value is
4004  * non-negative and < num_entries, then the CQ was emptied.
4005  */
4006 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
4007 			     struct ib_wc *wc)
4008 {
4009 	return cq->device->ops.poll_cq(cq, num_entries, wc);
4010 }
4011 
4012 /**
4013  * ib_req_notify_cq - Request completion notification on a CQ.
4014  * @cq: The CQ to generate an event for.
4015  * @flags:
4016  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
4017  *   to request an event on the next solicited event or next work
4018  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
4019  *   may also be |ed in to request a hint about missed events, as
4020  *   described below.
4021  *
4022  * Return Value:
4023  *    < 0 means an error occurred while requesting notification
4024  *   == 0 means notification was requested successfully, and if
4025  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
4026  *        were missed and it is safe to wait for another event.  In
4027  *        this case is it guaranteed that any work completions added
4028  *        to the CQ since the last CQ poll will trigger a completion
4029  *        notification event.
4030  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
4031  *        in.  It means that the consumer must poll the CQ again to
4032  *        make sure it is empty to avoid missing an event because of a
4033  *        race between requesting notification and an entry being
4034  *        added to the CQ.  This return value means it is possible
4035  *        (but not guaranteed) that a work completion has been added
4036  *        to the CQ since the last poll without triggering a
4037  *        completion notification event.
4038  */
4039 static inline int ib_req_notify_cq(struct ib_cq *cq,
4040 				   enum ib_cq_notify_flags flags)
4041 {
4042 	return cq->device->ops.req_notify_cq(cq, flags);
4043 }
4044 
4045 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4046 			     int comp_vector_hint,
4047 			     enum ib_poll_context poll_ctx);
4048 
4049 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4050 
4051 /*
4052  * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
4053  * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
4054  * address into the dma address.
4055  */
4056 static inline bool ib_uses_virt_dma(struct ib_device *dev)
4057 {
4058 	return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
4059 }
4060 
4061 /*
4062  * Check if a IB device's underlying DMA mapping supports P2PDMA transfers.
4063  */
4064 static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev)
4065 {
4066 	if (ib_uses_virt_dma(dev))
4067 		return false;
4068 
4069 	return dma_pci_p2pdma_supported(dev->dma_device);
4070 }
4071 
4072 /**
4073  * ib_virt_dma_to_ptr - Convert a dma_addr to a kernel pointer
4074  * @dma_addr: The DMA address
4075  *
4076  * Used by ib_uses_virt_dma() devices to get back to the kernel pointer after
4077  * going through the dma_addr marshalling.
4078  */
4079 static inline void *ib_virt_dma_to_ptr(u64 dma_addr)
4080 {
4081 	/* virt_dma mode maps the kvs's directly into the dma addr */
4082 	return (void *)(uintptr_t)dma_addr;
4083 }
4084 
4085 /**
4086  * ib_virt_dma_to_page - Convert a dma_addr to a struct page
4087  * @dma_addr: The DMA address
4088  *
4089  * Used by ib_uses_virt_dma() device to get back to the struct page after going
4090  * through the dma_addr marshalling.
4091  */
4092 static inline struct page *ib_virt_dma_to_page(u64 dma_addr)
4093 {
4094 	return virt_to_page(ib_virt_dma_to_ptr(dma_addr));
4095 }
4096 
4097 /**
4098  * ib_dma_mapping_error - check a DMA addr for error
4099  * @dev: The device for which the dma_addr was created
4100  * @dma_addr: The DMA address to check
4101  */
4102 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4103 {
4104 	if (ib_uses_virt_dma(dev))
4105 		return 0;
4106 	return dma_mapping_error(dev->dma_device, dma_addr);
4107 }
4108 
4109 /**
4110  * ib_dma_map_single - Map a kernel virtual address to DMA address
4111  * @dev: The device for which the dma_addr is to be created
4112  * @cpu_addr: The kernel virtual address
4113  * @size: The size of the region in bytes
4114  * @direction: The direction of the DMA
4115  */
4116 static inline u64 ib_dma_map_single(struct ib_device *dev,
4117 				    void *cpu_addr, size_t size,
4118 				    enum dma_data_direction direction)
4119 {
4120 	if (ib_uses_virt_dma(dev))
4121 		return (uintptr_t)cpu_addr;
4122 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4123 }
4124 
4125 /**
4126  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4127  * @dev: The device for which the DMA address was created
4128  * @addr: The DMA address
4129  * @size: The size of the region in bytes
4130  * @direction: The direction of the DMA
4131  */
4132 static inline void ib_dma_unmap_single(struct ib_device *dev,
4133 				       u64 addr, size_t size,
4134 				       enum dma_data_direction direction)
4135 {
4136 	if (!ib_uses_virt_dma(dev))
4137 		dma_unmap_single(dev->dma_device, addr, size, direction);
4138 }
4139 
4140 /**
4141  * ib_dma_map_page - Map a physical page to DMA address
4142  * @dev: The device for which the dma_addr is to be created
4143  * @page: The page to be mapped
4144  * @offset: The offset within the page
4145  * @size: The size of the region in bytes
4146  * @direction: The direction of the DMA
4147  */
4148 static inline u64 ib_dma_map_page(struct ib_device *dev,
4149 				  struct page *page,
4150 				  unsigned long offset,
4151 				  size_t size,
4152 					 enum dma_data_direction direction)
4153 {
4154 	if (ib_uses_virt_dma(dev))
4155 		return (uintptr_t)(page_address(page) + offset);
4156 	return dma_map_page(dev->dma_device, page, offset, size, direction);
4157 }
4158 
4159 /**
4160  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4161  * @dev: The device for which the DMA address was created
4162  * @addr: The DMA address
4163  * @size: The size of the region in bytes
4164  * @direction: The direction of the DMA
4165  */
4166 static inline void ib_dma_unmap_page(struct ib_device *dev,
4167 				     u64 addr, size_t size,
4168 				     enum dma_data_direction direction)
4169 {
4170 	if (!ib_uses_virt_dma(dev))
4171 		dma_unmap_page(dev->dma_device, addr, size, direction);
4172 }
4173 
4174 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4175 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4176 				      struct scatterlist *sg, int nents,
4177 				      enum dma_data_direction direction,
4178 				      unsigned long dma_attrs)
4179 {
4180 	if (ib_uses_virt_dma(dev))
4181 		return ib_dma_virt_map_sg(dev, sg, nents);
4182 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4183 				dma_attrs);
4184 }
4185 
4186 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4187 					 struct scatterlist *sg, int nents,
4188 					 enum dma_data_direction direction,
4189 					 unsigned long dma_attrs)
4190 {
4191 	if (!ib_uses_virt_dma(dev))
4192 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4193 				   dma_attrs);
4194 }
4195 
4196 /**
4197  * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4198  * @dev: The device for which the DMA addresses are to be created
4199  * @sg: The sg_table object describing the buffer
4200  * @direction: The direction of the DMA
4201  * @attrs: Optional DMA attributes for the map operation
4202  */
4203 static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4204 					   struct sg_table *sgt,
4205 					   enum dma_data_direction direction,
4206 					   unsigned long dma_attrs)
4207 {
4208 	int nents;
4209 
4210 	if (ib_uses_virt_dma(dev)) {
4211 		nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4212 		if (!nents)
4213 			return -EIO;
4214 		sgt->nents = nents;
4215 		return 0;
4216 	}
4217 	return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4218 }
4219 
4220 static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4221 					      struct sg_table *sgt,
4222 					      enum dma_data_direction direction,
4223 					      unsigned long dma_attrs)
4224 {
4225 	if (!ib_uses_virt_dma(dev))
4226 		dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4227 }
4228 
4229 /**
4230  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4231  * @dev: The device for which the DMA addresses are to be created
4232  * @sg: The array of scatter/gather entries
4233  * @nents: The number of scatter/gather entries
4234  * @direction: The direction of the DMA
4235  */
4236 static inline int ib_dma_map_sg(struct ib_device *dev,
4237 				struct scatterlist *sg, int nents,
4238 				enum dma_data_direction direction)
4239 {
4240 	return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4241 }
4242 
4243 /**
4244  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4245  * @dev: The device for which the DMA addresses were created
4246  * @sg: The array of scatter/gather entries
4247  * @nents: The number of scatter/gather entries
4248  * @direction: The direction of the DMA
4249  */
4250 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4251 				   struct scatterlist *sg, int nents,
4252 				   enum dma_data_direction direction)
4253 {
4254 	ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4255 }
4256 
4257 /**
4258  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4259  * @dev: The device to query
4260  *
4261  * The returned value represents a size in bytes.
4262  */
4263 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4264 {
4265 	if (ib_uses_virt_dma(dev))
4266 		return UINT_MAX;
4267 	return dma_get_max_seg_size(dev->dma_device);
4268 }
4269 
4270 /**
4271  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4272  * @dev: The device for which the DMA address was created
4273  * @addr: The DMA address
4274  * @size: The size of the region in bytes
4275  * @dir: The direction of the DMA
4276  */
4277 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4278 					      u64 addr,
4279 					      size_t size,
4280 					      enum dma_data_direction dir)
4281 {
4282 	if (!ib_uses_virt_dma(dev))
4283 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4284 }
4285 
4286 /**
4287  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4288  * @dev: The device for which the DMA address was created
4289  * @addr: The DMA address
4290  * @size: The size of the region in bytes
4291  * @dir: The direction of the DMA
4292  */
4293 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4294 						 u64 addr,
4295 						 size_t size,
4296 						 enum dma_data_direction dir)
4297 {
4298 	if (!ib_uses_virt_dma(dev))
4299 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4300 }
4301 
4302 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4303  * space. This function should be called when 'current' is the owning MM.
4304  */
4305 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4306 			     u64 virt_addr, int mr_access_flags);
4307 
4308 /* ib_advise_mr -  give an advice about an address range in a memory region */
4309 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4310 		 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4311 /**
4312  * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4313  *   HCA translation table.
4314  * @mr: The memory region to deregister.
4315  * @udata: Valid user data or NULL for kernel object
4316  *
4317  * This function can fail, if the memory region has memory windows bound to it.
4318  */
4319 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4320 
4321 /**
4322  * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4323  *   HCA translation table.
4324  * @mr: The memory region to deregister.
4325  *
4326  * This function can fail, if the memory region has memory windows bound to it.
4327  *
4328  * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4329  */
4330 static inline int ib_dereg_mr(struct ib_mr *mr)
4331 {
4332 	return ib_dereg_mr_user(mr, NULL);
4333 }
4334 
4335 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4336 			  u32 max_num_sg);
4337 
4338 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4339 				    u32 max_num_data_sg,
4340 				    u32 max_num_meta_sg);
4341 
4342 /**
4343  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4344  *   R_Key and L_Key.
4345  * @mr - struct ib_mr pointer to be updated.
4346  * @newkey - new key to be used.
4347  */
4348 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4349 {
4350 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4351 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4352 }
4353 
4354 /**
4355  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4356  * for calculating a new rkey for type 2 memory windows.
4357  * @rkey - the rkey to increment.
4358  */
4359 static inline u32 ib_inc_rkey(u32 rkey)
4360 {
4361 	const u32 mask = 0x000000ff;
4362 	return ((rkey + 1) & mask) | (rkey & ~mask);
4363 }
4364 
4365 /**
4366  * ib_attach_mcast - Attaches the specified QP to a multicast group.
4367  * @qp: QP to attach to the multicast group.  The QP must be type
4368  *   IB_QPT_UD.
4369  * @gid: Multicast group GID.
4370  * @lid: Multicast group LID in host byte order.
4371  *
4372  * In order to send and receive multicast packets, subnet
4373  * administration must have created the multicast group and configured
4374  * the fabric appropriately.  The port associated with the specified
4375  * QP must also be a member of the multicast group.
4376  */
4377 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4378 
4379 /**
4380  * ib_detach_mcast - Detaches the specified QP from a multicast group.
4381  * @qp: QP to detach from the multicast group.
4382  * @gid: Multicast group GID.
4383  * @lid: Multicast group LID in host byte order.
4384  */
4385 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4386 
4387 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4388 				   struct inode *inode, struct ib_udata *udata);
4389 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4390 
4391 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4392 				     unsigned int flags)
4393 {
4394 	u64 device_cap = ib_dev->attrs.device_cap_flags;
4395 
4396 	/*
4397 	 * Local write permission is required if remote write or
4398 	 * remote atomic permission is also requested.
4399 	 */
4400 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4401 	    !(flags & IB_ACCESS_LOCAL_WRITE))
4402 		return -EINVAL;
4403 
4404 	if (flags & ~IB_ACCESS_SUPPORTED)
4405 		return -EINVAL;
4406 
4407 	if (flags & IB_ACCESS_ON_DEMAND &&
4408 	    !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING))
4409 		return -EOPNOTSUPP;
4410 
4411 	if ((flags & IB_ACCESS_FLUSH_GLOBAL &&
4412 	    !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) ||
4413 	    (flags & IB_ACCESS_FLUSH_PERSISTENT &&
4414 	    !(device_cap & IB_DEVICE_FLUSH_PERSISTENT)))
4415 		return -EOPNOTSUPP;
4416 
4417 	return 0;
4418 }
4419 
4420 static inline bool ib_access_writable(int access_flags)
4421 {
4422 	/*
4423 	 * We have writable memory backing the MR if any of the following
4424 	 * access flags are set.  "Local write" and "remote write" obviously
4425 	 * require write access.  "Remote atomic" can do things like fetch and
4426 	 * add, which will modify memory, and "MW bind" can change permissions
4427 	 * by binding a window.
4428 	 */
4429 	return access_flags &
4430 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
4431 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4432 }
4433 
4434 /**
4435  * ib_check_mr_status: lightweight check of MR status.
4436  *     This routine may provide status checks on a selected
4437  *     ib_mr. first use is for signature status check.
4438  *
4439  * @mr: A memory region.
4440  * @check_mask: Bitmask of which checks to perform from
4441  *     ib_mr_status_check enumeration.
4442  * @mr_status: The container of relevant status checks.
4443  *     failed checks will be indicated in the status bitmask
4444  *     and the relevant info shall be in the error item.
4445  */
4446 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4447 		       struct ib_mr_status *mr_status);
4448 
4449 /**
4450  * ib_device_try_get: Hold a registration lock
4451  * device: The device to lock
4452  *
4453  * A device under an active registration lock cannot become unregistered. It
4454  * is only possible to obtain a registration lock on a device that is fully
4455  * registered, otherwise this function returns false.
4456  *
4457  * The registration lock is only necessary for actions which require the
4458  * device to still be registered. Uses that only require the device pointer to
4459  * be valid should use get_device(&ibdev->dev) to hold the memory.
4460  *
4461  */
4462 static inline bool ib_device_try_get(struct ib_device *dev)
4463 {
4464 	return refcount_inc_not_zero(&dev->refcount);
4465 }
4466 
4467 void ib_device_put(struct ib_device *device);
4468 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4469 					  enum rdma_driver_id driver_id);
4470 struct ib_device *ib_device_get_by_name(const char *name,
4471 					enum rdma_driver_id driver_id);
4472 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4473 					    u16 pkey, const union ib_gid *gid,
4474 					    const struct sockaddr *addr);
4475 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4476 			 unsigned int port);
4477 struct net_device *ib_device_get_netdev(struct ib_device *ib_dev,
4478 					u32 port);
4479 int ib_query_netdev_port(struct ib_device *ibdev, struct net_device *ndev,
4480 			 u32 *port);
4481 
4482 static inline enum ib_port_state ib_get_curr_port_state(struct net_device *net_dev)
4483 {
4484 	return (netif_running(net_dev) && netif_carrier_ok(net_dev)) ?
4485 		IB_PORT_ACTIVE : IB_PORT_DOWN;
4486 }
4487 
4488 void ib_dispatch_port_state_event(struct ib_device *ibdev,
4489 				  struct net_device *ndev);
4490 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4491 			   struct ib_wq_init_attr *init_attr);
4492 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4493 
4494 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4495 		 unsigned int *sg_offset, unsigned int page_size);
4496 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4497 		    int data_sg_nents, unsigned int *data_sg_offset,
4498 		    struct scatterlist *meta_sg, int meta_sg_nents,
4499 		    unsigned int *meta_sg_offset, unsigned int page_size);
4500 
4501 static inline int
4502 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4503 		  unsigned int *sg_offset, unsigned int page_size)
4504 {
4505 	int n;
4506 
4507 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4508 	mr->iova = 0;
4509 
4510 	return n;
4511 }
4512 
4513 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4514 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4515 
4516 void ib_drain_rq(struct ib_qp *qp);
4517 void ib_drain_sq(struct ib_qp *qp);
4518 void ib_drain_qp(struct ib_qp *qp);
4519 
4520 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4521 		     u8 *width);
4522 
4523 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4524 {
4525 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4526 		return attr->roce.dmac;
4527 	return NULL;
4528 }
4529 
4530 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4531 {
4532 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4533 		attr->ib.dlid = (u16)dlid;
4534 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4535 		attr->opa.dlid = dlid;
4536 }
4537 
4538 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4539 {
4540 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4541 		return attr->ib.dlid;
4542 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4543 		return attr->opa.dlid;
4544 	return 0;
4545 }
4546 
4547 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4548 {
4549 	attr->sl = sl;
4550 }
4551 
4552 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4553 {
4554 	return attr->sl;
4555 }
4556 
4557 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4558 					 u8 src_path_bits)
4559 {
4560 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4561 		attr->ib.src_path_bits = src_path_bits;
4562 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4563 		attr->opa.src_path_bits = src_path_bits;
4564 }
4565 
4566 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4567 {
4568 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4569 		return attr->ib.src_path_bits;
4570 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4571 		return attr->opa.src_path_bits;
4572 	return 0;
4573 }
4574 
4575 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4576 					bool make_grd)
4577 {
4578 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4579 		attr->opa.make_grd = make_grd;
4580 }
4581 
4582 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4583 {
4584 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4585 		return attr->opa.make_grd;
4586 	return false;
4587 }
4588 
4589 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4590 {
4591 	attr->port_num = port_num;
4592 }
4593 
4594 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4595 {
4596 	return attr->port_num;
4597 }
4598 
4599 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4600 					   u8 static_rate)
4601 {
4602 	attr->static_rate = static_rate;
4603 }
4604 
4605 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4606 {
4607 	return attr->static_rate;
4608 }
4609 
4610 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4611 					enum ib_ah_flags flag)
4612 {
4613 	attr->ah_flags = flag;
4614 }
4615 
4616 static inline enum ib_ah_flags
4617 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4618 {
4619 	return attr->ah_flags;
4620 }
4621 
4622 static inline const struct ib_global_route
4623 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4624 {
4625 	return &attr->grh;
4626 }
4627 
4628 /*To retrieve and modify the grh */
4629 static inline struct ib_global_route
4630 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4631 {
4632 	return &attr->grh;
4633 }
4634 
4635 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4636 {
4637 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4638 
4639 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4640 }
4641 
4642 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4643 					     __be64 prefix)
4644 {
4645 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4646 
4647 	grh->dgid.global.subnet_prefix = prefix;
4648 }
4649 
4650 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4651 					    __be64 if_id)
4652 {
4653 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4654 
4655 	grh->dgid.global.interface_id = if_id;
4656 }
4657 
4658 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4659 				   union ib_gid *dgid, u32 flow_label,
4660 				   u8 sgid_index, u8 hop_limit,
4661 				   u8 traffic_class)
4662 {
4663 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4664 
4665 	attr->ah_flags = IB_AH_GRH;
4666 	if (dgid)
4667 		grh->dgid = *dgid;
4668 	grh->flow_label = flow_label;
4669 	grh->sgid_index = sgid_index;
4670 	grh->hop_limit = hop_limit;
4671 	grh->traffic_class = traffic_class;
4672 	grh->sgid_attr = NULL;
4673 }
4674 
4675 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4676 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4677 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4678 			     const struct ib_gid_attr *sgid_attr);
4679 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4680 		       const struct rdma_ah_attr *src);
4681 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4682 			  const struct rdma_ah_attr *new);
4683 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4684 
4685 /**
4686  * rdma_ah_find_type - Return address handle type.
4687  *
4688  * @dev: Device to be checked
4689  * @port_num: Port number
4690  */
4691 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4692 						       u32 port_num)
4693 {
4694 	if (rdma_protocol_roce(dev, port_num))
4695 		return RDMA_AH_ATTR_TYPE_ROCE;
4696 	if (rdma_protocol_ib(dev, port_num)) {
4697 		if (rdma_cap_opa_ah(dev, port_num))
4698 			return RDMA_AH_ATTR_TYPE_OPA;
4699 		return RDMA_AH_ATTR_TYPE_IB;
4700 	}
4701 	if (dev->type == RDMA_DEVICE_TYPE_SMI)
4702 		return RDMA_AH_ATTR_TYPE_IB;
4703 
4704 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4705 }
4706 
4707 /**
4708  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4709  *     In the current implementation the only way to
4710  *     get the 32bit lid is from other sources for OPA.
4711  *     For IB, lids will always be 16bits so cast the
4712  *     value accordingly.
4713  *
4714  * @lid: A 32bit LID
4715  */
4716 static inline u16 ib_lid_cpu16(u32 lid)
4717 {
4718 	WARN_ON_ONCE(lid & 0xFFFF0000);
4719 	return (u16)lid;
4720 }
4721 
4722 /**
4723  * ib_lid_be16 - Return lid in 16bit BE encoding.
4724  *
4725  * @lid: A 32bit LID
4726  */
4727 static inline __be16 ib_lid_be16(u32 lid)
4728 {
4729 	WARN_ON_ONCE(lid & 0xFFFF0000);
4730 	return cpu_to_be16((u16)lid);
4731 }
4732 
4733 /**
4734  * ib_get_vector_affinity - Get the affinity mappings of a given completion
4735  *   vector
4736  * @device:         the rdma device
4737  * @comp_vector:    index of completion vector
4738  *
4739  * Returns NULL on failure, otherwise a corresponding cpu map of the
4740  * completion vector (returns all-cpus map if the device driver doesn't
4741  * implement get_vector_affinity).
4742  */
4743 static inline const struct cpumask *
4744 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4745 {
4746 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4747 	    !device->ops.get_vector_affinity)
4748 		return NULL;
4749 
4750 	return device->ops.get_vector_affinity(device, comp_vector);
4751 
4752 }
4753 
4754 /**
4755  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4756  * and add their gids, as needed, to the relevant RoCE devices.
4757  *
4758  * @device:         the rdma device
4759  */
4760 void rdma_roce_rescan_device(struct ib_device *ibdev);
4761 void rdma_roce_rescan_port(struct ib_device *ib_dev, u32 port);
4762 void roce_del_all_netdev_gids(struct ib_device *ib_dev,
4763 			      u32 port, struct net_device *ndev);
4764 
4765 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4766 
4767 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4768 
4769 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4770 				     enum rdma_netdev_t type, const char *name,
4771 				     unsigned char name_assign_type,
4772 				     void (*setup)(struct net_device *));
4773 
4774 int rdma_init_netdev(struct ib_device *device, u32 port_num,
4775 		     enum rdma_netdev_t type, const char *name,
4776 		     unsigned char name_assign_type,
4777 		     void (*setup)(struct net_device *),
4778 		     struct net_device *netdev);
4779 
4780 /**
4781  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4782  *
4783  * @device:	device pointer for which ib_device pointer to retrieve
4784  *
4785  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4786  *
4787  */
4788 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4789 {
4790 	struct ib_core_device *coredev =
4791 		container_of(device, struct ib_core_device, dev);
4792 
4793 	return coredev->owner;
4794 }
4795 
4796 /**
4797  * ibdev_to_node - return the NUMA node for a given ib_device
4798  * @dev:	device to get the NUMA node for.
4799  */
4800 static inline int ibdev_to_node(struct ib_device *ibdev)
4801 {
4802 	struct device *parent = ibdev->dev.parent;
4803 
4804 	if (!parent)
4805 		return NUMA_NO_NODE;
4806 	return dev_to_node(parent);
4807 }
4808 
4809 /**
4810  * rdma_device_to_drv_device - Helper macro to reach back to driver's
4811  *			       ib_device holder structure from device pointer.
4812  *
4813  * NOTE: New drivers should not make use of this API; This API is only for
4814  * existing drivers who have exposed sysfs entries using
4815  * ops->device_group.
4816  */
4817 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
4818 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4819 
4820 bool rdma_dev_access_netns(const struct ib_device *device,
4821 			   const struct net *net);
4822 
4823 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4824 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4825 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4826 
4827 /**
4828  * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4829  *                               on the flow_label
4830  *
4831  * This function will convert the 20 bit flow_label input to a valid RoCE v2
4832  * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4833  * convention.
4834  */
4835 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4836 {
4837 	u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4838 
4839 	fl_low ^= fl_high >> 14;
4840 	return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4841 }
4842 
4843 /**
4844  * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4845  *                        local and remote qpn values
4846  *
4847  * This function folded the multiplication results of two qpns, 24 bit each,
4848  * fields, and converts it to a 20 bit results.
4849  *
4850  * This function will create symmetric flow_label value based on the local
4851  * and remote qpn values. this will allow both the requester and responder
4852  * to calculate the same flow_label for a given connection.
4853  *
4854  * This helper function should be used by driver in case the upper layer
4855  * provide a zero flow_label value. This is to improve entropy of RDMA
4856  * traffic in the network.
4857  */
4858 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4859 {
4860 	u64 v = (u64)lqpn * rqpn;
4861 
4862 	v ^= v >> 20;
4863 	v ^= v >> 40;
4864 
4865 	return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4866 }
4867 
4868 /**
4869  * rdma_get_udp_sport - Calculate and set UDP source port based on the flow
4870  *                      label. If flow label is not defined in GRH then
4871  *                      calculate it based on lqpn/rqpn.
4872  *
4873  * @fl:                 flow label from GRH
4874  * @lqpn:               local qp number
4875  * @rqpn:               remote qp number
4876  */
4877 static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
4878 {
4879 	if (!fl)
4880 		fl = rdma_calc_flow_label(lqpn, rqpn);
4881 
4882 	return rdma_flow_label_to_udp_sport(fl);
4883 }
4884 
4885 const struct ib_port_immutable*
4886 ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4887 
4888 /** ib_add_sub_device - Add a sub IB device on an existing one
4889  *
4890  * @parent: The IB device that needs to add a sub device
4891  * @type: The type of the new sub device
4892  * @name: The name of the new sub device
4893  *
4894  *
4895  * Return 0 on success, an error code otherwise
4896  */
4897 int ib_add_sub_device(struct ib_device *parent,
4898 		      enum rdma_nl_dev_type type,
4899 		      const char *name);
4900 
4901 
4902 /** ib_del_sub_device_and_put - Delect an IB sub device while holding a 'get'
4903  *
4904  * @sub: The sub device that is going to be deleted
4905  *
4906  * Return 0 on success, an error code otherwise
4907  */
4908 int ib_del_sub_device_and_put(struct ib_device *sub);
4909 
4910 static inline void ib_mark_name_assigned_by_user(struct ib_device *ibdev)
4911 {
4912 	ibdev->name_assign_type = RDMA_NAME_ASSIGN_TYPE_USER;
4913 }
4914 
4915 #endif /* IB_VERBS_H */
4916