xref: /linux/include/rdma/ib_verbs.h (revision c21614d5c0e76cf1d7a864f15d495fd3e5876acc)
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(3, 4) __cold
63 void ibdev_printk(const char *level, const struct ib_device *ibdev,
64 		  const char *format, ...);
65 __printf(2, 3) __cold
66 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
67 __printf(2, 3) __cold
68 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
69 __printf(2, 3) __cold
70 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
71 __printf(2, 3) __cold
72 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
73 __printf(2, 3) __cold
74 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
75 __printf(2, 3) __cold
76 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
77 __printf(2, 3) __cold
78 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
79 
80 #if defined(CONFIG_DYNAMIC_DEBUG) || \
81 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
82 #define ibdev_dbg(__dev, format, args...)                       \
83 	dynamic_ibdev_dbg(__dev, format, ##args)
84 #else
85 __printf(2, 3) __cold
86 static inline
87 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
88 #endif
89 
90 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...)           \
91 do {                                                                    \
92 	static DEFINE_RATELIMIT_STATE(_rs,                              \
93 				      DEFAULT_RATELIMIT_INTERVAL,       \
94 				      DEFAULT_RATELIMIT_BURST);         \
95 	if (__ratelimit(&_rs))                                          \
96 		ibdev_level(ibdev, fmt, ##__VA_ARGS__);                 \
97 } while (0)
98 
99 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
100 	ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
101 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
102 	ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
103 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
104 	ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
105 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
106 	ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
107 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
108 	ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
109 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
110 	ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
111 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
112 	ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
113 
114 #if defined(CONFIG_DYNAMIC_DEBUG) || \
115 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
116 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
117 #define ibdev_dbg_ratelimited(ibdev, fmt, ...)                          \
118 do {                                                                    \
119 	static DEFINE_RATELIMIT_STATE(_rs,                              \
120 				      DEFAULT_RATELIMIT_INTERVAL,       \
121 				      DEFAULT_RATELIMIT_BURST);         \
122 	DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt);                 \
123 	if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs))      \
124 		__dynamic_ibdev_dbg(&descriptor, ibdev, fmt,            \
125 				    ##__VA_ARGS__);                     \
126 } while (0)
127 #else
128 __printf(2, 3) __cold
129 static inline
130 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
131 #endif
132 
133 union ib_gid {
134 	u8	raw[16];
135 	struct {
136 		__be64	subnet_prefix;
137 		__be64	interface_id;
138 	} global;
139 };
140 
141 extern union ib_gid zgid;
142 
143 enum ib_gid_type {
144 	IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
145 	IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
146 	IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
147 	IB_GID_TYPE_SIZE
148 };
149 
150 #define ROCE_V2_UDP_DPORT      4791
151 struct ib_gid_attr {
152 	struct net_device __rcu	*ndev;
153 	struct ib_device	*device;
154 	union ib_gid		gid;
155 	enum ib_gid_type	gid_type;
156 	u16			index;
157 	u32			port_num;
158 };
159 
160 enum {
161 	/* set the local administered indication */
162 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
163 };
164 
165 enum rdma_transport_type {
166 	RDMA_TRANSPORT_IB,
167 	RDMA_TRANSPORT_IWARP,
168 	RDMA_TRANSPORT_USNIC,
169 	RDMA_TRANSPORT_USNIC_UDP,
170 	RDMA_TRANSPORT_UNSPECIFIED,
171 };
172 
173 enum rdma_protocol_type {
174 	RDMA_PROTOCOL_IB,
175 	RDMA_PROTOCOL_IBOE,
176 	RDMA_PROTOCOL_IWARP,
177 	RDMA_PROTOCOL_USNIC_UDP
178 };
179 
180 __attribute_const__ enum rdma_transport_type
181 rdma_node_get_transport(unsigned int node_type);
182 
183 enum rdma_network_type {
184 	RDMA_NETWORK_IB,
185 	RDMA_NETWORK_ROCE_V1,
186 	RDMA_NETWORK_IPV4,
187 	RDMA_NETWORK_IPV6
188 };
189 
190 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
191 {
192 	if (network_type == RDMA_NETWORK_IPV4 ||
193 	    network_type == RDMA_NETWORK_IPV6)
194 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
195 	else if (network_type == RDMA_NETWORK_ROCE_V1)
196 		return IB_GID_TYPE_ROCE;
197 	else
198 		return IB_GID_TYPE_IB;
199 }
200 
201 static inline enum rdma_network_type
202 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
203 {
204 	if (attr->gid_type == IB_GID_TYPE_IB)
205 		return RDMA_NETWORK_IB;
206 
207 	if (attr->gid_type == IB_GID_TYPE_ROCE)
208 		return RDMA_NETWORK_ROCE_V1;
209 
210 	if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
211 		return RDMA_NETWORK_IPV4;
212 	else
213 		return RDMA_NETWORK_IPV6;
214 }
215 
216 enum rdma_link_layer {
217 	IB_LINK_LAYER_UNSPECIFIED,
218 	IB_LINK_LAYER_INFINIBAND,
219 	IB_LINK_LAYER_ETHERNET,
220 };
221 
222 enum ib_device_cap_flags {
223 	IB_DEVICE_RESIZE_MAX_WR = IB_UVERBS_DEVICE_RESIZE_MAX_WR,
224 	IB_DEVICE_BAD_PKEY_CNTR = IB_UVERBS_DEVICE_BAD_PKEY_CNTR,
225 	IB_DEVICE_BAD_QKEY_CNTR = IB_UVERBS_DEVICE_BAD_QKEY_CNTR,
226 	IB_DEVICE_RAW_MULTI = IB_UVERBS_DEVICE_RAW_MULTI,
227 	IB_DEVICE_AUTO_PATH_MIG = IB_UVERBS_DEVICE_AUTO_PATH_MIG,
228 	IB_DEVICE_CHANGE_PHY_PORT = IB_UVERBS_DEVICE_CHANGE_PHY_PORT,
229 	IB_DEVICE_UD_AV_PORT_ENFORCE = IB_UVERBS_DEVICE_UD_AV_PORT_ENFORCE,
230 	IB_DEVICE_CURR_QP_STATE_MOD = IB_UVERBS_DEVICE_CURR_QP_STATE_MOD,
231 	IB_DEVICE_SHUTDOWN_PORT = IB_UVERBS_DEVICE_SHUTDOWN_PORT,
232 	/* IB_DEVICE_INIT_TYPE = IB_UVERBS_DEVICE_INIT_TYPE, (not in use) */
233 	IB_DEVICE_PORT_ACTIVE_EVENT = IB_UVERBS_DEVICE_PORT_ACTIVE_EVENT,
234 	IB_DEVICE_SYS_IMAGE_GUID = IB_UVERBS_DEVICE_SYS_IMAGE_GUID,
235 	IB_DEVICE_RC_RNR_NAK_GEN = IB_UVERBS_DEVICE_RC_RNR_NAK_GEN,
236 	IB_DEVICE_SRQ_RESIZE = IB_UVERBS_DEVICE_SRQ_RESIZE,
237 	IB_DEVICE_N_NOTIFY_CQ = IB_UVERBS_DEVICE_N_NOTIFY_CQ,
238 
239 	/* Reserved, old SEND_W_INV = 1 << 16,*/
240 	IB_DEVICE_MEM_WINDOW = IB_UVERBS_DEVICE_MEM_WINDOW,
241 	/*
242 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
243 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
244 	 * messages and can verify the validity of checksum for
245 	 * incoming messages.  Setting this flag implies that the
246 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
247 	 */
248 	IB_DEVICE_UD_IP_CSUM = IB_UVERBS_DEVICE_UD_IP_CSUM,
249 	IB_DEVICE_XRC = IB_UVERBS_DEVICE_XRC,
250 
251 	/*
252 	 * This device supports the IB "base memory management extension",
253 	 * which includes support for fast registrations (IB_WR_REG_MR,
254 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
255 	 * also be set by any iWarp device which must support FRs to comply
256 	 * to the iWarp verbs spec.  iWarp devices also support the
257 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
258 	 * stag.
259 	 */
260 	IB_DEVICE_MEM_MGT_EXTENSIONS = IB_UVERBS_DEVICE_MEM_MGT_EXTENSIONS,
261 	IB_DEVICE_MEM_WINDOW_TYPE_2A = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2A,
262 	IB_DEVICE_MEM_WINDOW_TYPE_2B = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2B,
263 	IB_DEVICE_RC_IP_CSUM = IB_UVERBS_DEVICE_RC_IP_CSUM,
264 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
265 	IB_DEVICE_RAW_IP_CSUM = IB_UVERBS_DEVICE_RAW_IP_CSUM,
266 	IB_DEVICE_MANAGED_FLOW_STEERING =
267 		IB_UVERBS_DEVICE_MANAGED_FLOW_STEERING,
268 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
269 	IB_DEVICE_RAW_SCATTER_FCS = IB_UVERBS_DEVICE_RAW_SCATTER_FCS,
270 	/* The device supports padding incoming writes to cacheline. */
271 	IB_DEVICE_PCI_WRITE_END_PADDING =
272 		IB_UVERBS_DEVICE_PCI_WRITE_END_PADDING,
273 	/* Placement type attributes */
274 	IB_DEVICE_FLUSH_GLOBAL = IB_UVERBS_DEVICE_FLUSH_GLOBAL,
275 	IB_DEVICE_FLUSH_PERSISTENT = IB_UVERBS_DEVICE_FLUSH_PERSISTENT,
276 	IB_DEVICE_ATOMIC_WRITE = IB_UVERBS_DEVICE_ATOMIC_WRITE,
277 };
278 
279 enum ib_kernel_cap_flags {
280 	/*
281 	 * This device supports a per-device lkey or stag that can be
282 	 * used without performing a memory registration for the local
283 	 * memory.  Note that ULPs should never check this flag, but
284 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
285 	 * which will always contain a usable lkey.
286 	 */
287 	IBK_LOCAL_DMA_LKEY = 1 << 0,
288 	/* IB_QP_CREATE_INTEGRITY_EN is supported to implement T10-PI */
289 	IBK_INTEGRITY_HANDOVER = 1 << 1,
290 	/* IB_ACCESS_ON_DEMAND is supported during reg_user_mr() */
291 	IBK_ON_DEMAND_PAGING = 1 << 2,
292 	/* IB_MR_TYPE_SG_GAPS is supported */
293 	IBK_SG_GAPS_REG = 1 << 3,
294 	/* Driver supports RDMA_NLDEV_CMD_DELLINK */
295 	IBK_ALLOW_USER_UNREG = 1 << 4,
296 
297 	/* ipoib will use IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK */
298 	IBK_BLOCK_MULTICAST_LOOPBACK = 1 << 5,
299 	/* iopib will use IB_QP_CREATE_IPOIB_UD_LSO for its QPs */
300 	IBK_UD_TSO = 1 << 6,
301 	/* iopib will use the device ops:
302 	 *   get_vf_config
303 	 *   get_vf_guid
304 	 *   get_vf_stats
305 	 *   set_vf_guid
306 	 *   set_vf_link_state
307 	 */
308 	IBK_VIRTUAL_FUNCTION = 1 << 7,
309 	/* ipoib will use IB_QP_CREATE_NETDEV_USE for its QPs */
310 	IBK_RDMA_NETDEV_OPA = 1 << 8,
311 };
312 
313 enum ib_atomic_cap {
314 	IB_ATOMIC_NONE,
315 	IB_ATOMIC_HCA,
316 	IB_ATOMIC_GLOB
317 };
318 
319 enum ib_odp_general_cap_bits {
320 	IB_ODP_SUPPORT		= 1 << 0,
321 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
322 };
323 
324 enum ib_odp_transport_cap_bits {
325 	IB_ODP_SUPPORT_SEND	= 1 << 0,
326 	IB_ODP_SUPPORT_RECV	= 1 << 1,
327 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
328 	IB_ODP_SUPPORT_READ	= 1 << 3,
329 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
330 	IB_ODP_SUPPORT_SRQ_RECV	= 1 << 5,
331 };
332 
333 struct ib_odp_caps {
334 	uint64_t general_caps;
335 	struct {
336 		uint32_t  rc_odp_caps;
337 		uint32_t  uc_odp_caps;
338 		uint32_t  ud_odp_caps;
339 		uint32_t  xrc_odp_caps;
340 	} per_transport_caps;
341 };
342 
343 struct ib_rss_caps {
344 	/* Corresponding bit will be set if qp type from
345 	 * 'enum ib_qp_type' is supported, e.g.
346 	 * supported_qpts |= 1 << IB_QPT_UD
347 	 */
348 	u32 supported_qpts;
349 	u32 max_rwq_indirection_tables;
350 	u32 max_rwq_indirection_table_size;
351 };
352 
353 enum ib_tm_cap_flags {
354 	/*  Support tag matching with rendezvous offload for RC transport */
355 	IB_TM_CAP_RNDV_RC = 1 << 0,
356 };
357 
358 struct ib_tm_caps {
359 	/* Max size of RNDV header */
360 	u32 max_rndv_hdr_size;
361 	/* Max number of entries in tag matching list */
362 	u32 max_num_tags;
363 	/* From enum ib_tm_cap_flags */
364 	u32 flags;
365 	/* Max number of outstanding list operations */
366 	u32 max_ops;
367 	/* Max number of SGE in tag matching entry */
368 	u32 max_sge;
369 };
370 
371 struct ib_cq_init_attr {
372 	unsigned int	cqe;
373 	u32		comp_vector;
374 	u32		flags;
375 };
376 
377 enum ib_cq_attr_mask {
378 	IB_CQ_MODERATE = 1 << 0,
379 };
380 
381 struct ib_cq_caps {
382 	u16     max_cq_moderation_count;
383 	u16     max_cq_moderation_period;
384 };
385 
386 struct ib_dm_mr_attr {
387 	u64		length;
388 	u64		offset;
389 	u32		access_flags;
390 };
391 
392 struct ib_dm_alloc_attr {
393 	u64	length;
394 	u32	alignment;
395 	u32	flags;
396 };
397 
398 struct ib_device_attr {
399 	u64			fw_ver;
400 	__be64			sys_image_guid;
401 	u64			max_mr_size;
402 	u64			page_size_cap;
403 	u32			vendor_id;
404 	u32			vendor_part_id;
405 	u32			hw_ver;
406 	int			max_qp;
407 	int			max_qp_wr;
408 	u64			device_cap_flags;
409 	u64			kernel_cap_flags;
410 	int			max_send_sge;
411 	int			max_recv_sge;
412 	int			max_sge_rd;
413 	int			max_cq;
414 	int			max_cqe;
415 	int			max_mr;
416 	int			max_pd;
417 	int			max_qp_rd_atom;
418 	int			max_ee_rd_atom;
419 	int			max_res_rd_atom;
420 	int			max_qp_init_rd_atom;
421 	int			max_ee_init_rd_atom;
422 	enum ib_atomic_cap	atomic_cap;
423 	enum ib_atomic_cap	masked_atomic_cap;
424 	int			max_ee;
425 	int			max_rdd;
426 	int			max_mw;
427 	int			max_raw_ipv6_qp;
428 	int			max_raw_ethy_qp;
429 	int			max_mcast_grp;
430 	int			max_mcast_qp_attach;
431 	int			max_total_mcast_qp_attach;
432 	int			max_ah;
433 	int			max_srq;
434 	int			max_srq_wr;
435 	int			max_srq_sge;
436 	unsigned int		max_fast_reg_page_list_len;
437 	unsigned int		max_pi_fast_reg_page_list_len;
438 	u16			max_pkeys;
439 	u8			local_ca_ack_delay;
440 	int			sig_prot_cap;
441 	int			sig_guard_cap;
442 	struct ib_odp_caps	odp_caps;
443 	uint64_t		timestamp_mask;
444 	uint64_t		hca_core_clock; /* in KHZ */
445 	struct ib_rss_caps	rss_caps;
446 	u32			max_wq_type_rq;
447 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
448 	struct ib_tm_caps	tm_caps;
449 	struct ib_cq_caps       cq_caps;
450 	u64			max_dm_size;
451 	/* Max entries for sgl for optimized performance per READ */
452 	u32			max_sgl_rd;
453 };
454 
455 enum ib_mtu {
456 	IB_MTU_256  = 1,
457 	IB_MTU_512  = 2,
458 	IB_MTU_1024 = 3,
459 	IB_MTU_2048 = 4,
460 	IB_MTU_4096 = 5
461 };
462 
463 enum opa_mtu {
464 	OPA_MTU_8192 = 6,
465 	OPA_MTU_10240 = 7
466 };
467 
468 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
469 {
470 	switch (mtu) {
471 	case IB_MTU_256:  return  256;
472 	case IB_MTU_512:  return  512;
473 	case IB_MTU_1024: return 1024;
474 	case IB_MTU_2048: return 2048;
475 	case IB_MTU_4096: return 4096;
476 	default: 	  return -1;
477 	}
478 }
479 
480 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
481 {
482 	if (mtu >= 4096)
483 		return IB_MTU_4096;
484 	else if (mtu >= 2048)
485 		return IB_MTU_2048;
486 	else if (mtu >= 1024)
487 		return IB_MTU_1024;
488 	else if (mtu >= 512)
489 		return IB_MTU_512;
490 	else
491 		return IB_MTU_256;
492 }
493 
494 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
495 {
496 	switch (mtu) {
497 	case OPA_MTU_8192:
498 		return 8192;
499 	case OPA_MTU_10240:
500 		return 10240;
501 	default:
502 		return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
503 	}
504 }
505 
506 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
507 {
508 	if (mtu >= 10240)
509 		return OPA_MTU_10240;
510 	else if (mtu >= 8192)
511 		return OPA_MTU_8192;
512 	else
513 		return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
514 }
515 
516 enum ib_port_state {
517 	IB_PORT_NOP		= 0,
518 	IB_PORT_DOWN		= 1,
519 	IB_PORT_INIT		= 2,
520 	IB_PORT_ARMED		= 3,
521 	IB_PORT_ACTIVE		= 4,
522 	IB_PORT_ACTIVE_DEFER	= 5
523 };
524 
525 enum ib_port_phys_state {
526 	IB_PORT_PHYS_STATE_SLEEP = 1,
527 	IB_PORT_PHYS_STATE_POLLING = 2,
528 	IB_PORT_PHYS_STATE_DISABLED = 3,
529 	IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
530 	IB_PORT_PHYS_STATE_LINK_UP = 5,
531 	IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
532 	IB_PORT_PHYS_STATE_PHY_TEST = 7,
533 };
534 
535 enum ib_port_width {
536 	IB_WIDTH_1X	= 1,
537 	IB_WIDTH_2X	= 16,
538 	IB_WIDTH_4X	= 2,
539 	IB_WIDTH_8X	= 4,
540 	IB_WIDTH_12X	= 8
541 };
542 
543 static inline int ib_width_enum_to_int(enum ib_port_width width)
544 {
545 	switch (width) {
546 	case IB_WIDTH_1X:  return  1;
547 	case IB_WIDTH_2X:  return  2;
548 	case IB_WIDTH_4X:  return  4;
549 	case IB_WIDTH_8X:  return  8;
550 	case IB_WIDTH_12X: return 12;
551 	default: 	  return -1;
552 	}
553 }
554 
555 enum ib_port_speed {
556 	IB_SPEED_SDR	= 1,
557 	IB_SPEED_DDR	= 2,
558 	IB_SPEED_QDR	= 4,
559 	IB_SPEED_FDR10	= 8,
560 	IB_SPEED_FDR	= 16,
561 	IB_SPEED_EDR	= 32,
562 	IB_SPEED_HDR	= 64,
563 	IB_SPEED_NDR	= 128,
564 	IB_SPEED_XDR	= 256,
565 };
566 
567 enum ib_stat_flag {
568 	IB_STAT_FLAG_OPTIONAL = 1 << 0,
569 };
570 
571 /**
572  * struct rdma_stat_desc
573  * @name - The name of the counter
574  * @flags - Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL
575  * @priv - Driver private information; Core code should not use
576  */
577 struct rdma_stat_desc {
578 	const char *name;
579 	unsigned int flags;
580 	const void *priv;
581 };
582 
583 /**
584  * struct rdma_hw_stats
585  * @lock - Mutex to protect parallel write access to lifespan and values
586  *    of counters, which are 64bits and not guaranteed to be written
587  *    atomicaly on 32bits systems.
588  * @timestamp - Used by the core code to track when the last update was
589  * @lifespan - Used by the core code to determine how old the counters
590  *   should be before being updated again.  Stored in jiffies, defaults
591  *   to 10 milliseconds, drivers can override the default be specifying
592  *   their own value during their allocation routine.
593  * @descs - Array of pointers to static descriptors used for the counters
594  *   in directory.
595  * @is_disabled - A bitmap to indicate each counter is currently disabled
596  *   or not.
597  * @num_counters - How many hardware counters there are.  If name is
598  *   shorter than this number, a kernel oops will result.  Driver authors
599  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
600  *   in their code to prevent this.
601  * @value - Array of u64 counters that are accessed by the sysfs code and
602  *   filled in by the drivers get_stats routine
603  */
604 struct rdma_hw_stats {
605 	struct mutex	lock; /* Protect lifespan and values[] */
606 	unsigned long	timestamp;
607 	unsigned long	lifespan;
608 	const struct rdma_stat_desc *descs;
609 	unsigned long	*is_disabled;
610 	int		num_counters;
611 	u64		value[] __counted_by(num_counters);
612 };
613 
614 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
615 
616 struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
617 	const struct rdma_stat_desc *descs, int num_counters,
618 	unsigned long lifespan);
619 
620 void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
621 
622 /* Define bits for the various functionality this port needs to be supported by
623  * the core.
624  */
625 /* Management                           0x00000FFF */
626 #define RDMA_CORE_CAP_IB_MAD            0x00000001
627 #define RDMA_CORE_CAP_IB_SMI            0x00000002
628 #define RDMA_CORE_CAP_IB_CM             0x00000004
629 #define RDMA_CORE_CAP_IW_CM             0x00000008
630 #define RDMA_CORE_CAP_IB_SA             0x00000010
631 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
632 
633 /* Address format                       0x000FF000 */
634 #define RDMA_CORE_CAP_AF_IB             0x00001000
635 #define RDMA_CORE_CAP_ETH_AH            0x00002000
636 #define RDMA_CORE_CAP_OPA_AH            0x00004000
637 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
638 
639 /* Protocol                             0xFFF00000 */
640 #define RDMA_CORE_CAP_PROT_IB           0x00100000
641 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
642 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
643 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
644 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
645 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
646 
647 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
648 					| RDMA_CORE_CAP_PROT_ROCE     \
649 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
650 
651 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
652 					| RDMA_CORE_CAP_IB_MAD \
653 					| RDMA_CORE_CAP_IB_SMI \
654 					| RDMA_CORE_CAP_IB_CM  \
655 					| RDMA_CORE_CAP_IB_SA  \
656 					| RDMA_CORE_CAP_AF_IB)
657 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
658 					| RDMA_CORE_CAP_IB_MAD  \
659 					| RDMA_CORE_CAP_IB_CM   \
660 					| RDMA_CORE_CAP_AF_IB   \
661 					| RDMA_CORE_CAP_ETH_AH)
662 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
663 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
664 					| RDMA_CORE_CAP_IB_MAD  \
665 					| RDMA_CORE_CAP_IB_CM   \
666 					| RDMA_CORE_CAP_AF_IB   \
667 					| RDMA_CORE_CAP_ETH_AH)
668 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
669 					| RDMA_CORE_CAP_IW_CM)
670 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
671 					| RDMA_CORE_CAP_OPA_MAD)
672 
673 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
674 
675 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
676 
677 struct ib_port_attr {
678 	u64			subnet_prefix;
679 	enum ib_port_state	state;
680 	enum ib_mtu		max_mtu;
681 	enum ib_mtu		active_mtu;
682 	u32                     phys_mtu;
683 	int			gid_tbl_len;
684 	unsigned int		ip_gids:1;
685 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
686 	u32			port_cap_flags;
687 	u32			max_msg_sz;
688 	u32			bad_pkey_cntr;
689 	u32			qkey_viol_cntr;
690 	u16			pkey_tbl_len;
691 	u32			sm_lid;
692 	u32			lid;
693 	u8			lmc;
694 	u8			max_vl_num;
695 	u8			sm_sl;
696 	u8			subnet_timeout;
697 	u8			init_type_reply;
698 	u8			active_width;
699 	u16			active_speed;
700 	u8                      phys_state;
701 	u16			port_cap_flags2;
702 };
703 
704 enum ib_device_modify_flags {
705 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
706 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
707 };
708 
709 #define IB_DEVICE_NODE_DESC_MAX 64
710 
711 struct ib_device_modify {
712 	u64	sys_image_guid;
713 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
714 };
715 
716 enum ib_port_modify_flags {
717 	IB_PORT_SHUTDOWN		= 1,
718 	IB_PORT_INIT_TYPE		= (1<<2),
719 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
720 	IB_PORT_OPA_MASK_CHG		= (1<<4)
721 };
722 
723 struct ib_port_modify {
724 	u32	set_port_cap_mask;
725 	u32	clr_port_cap_mask;
726 	u8	init_type;
727 };
728 
729 enum ib_event_type {
730 	IB_EVENT_CQ_ERR,
731 	IB_EVENT_QP_FATAL,
732 	IB_EVENT_QP_REQ_ERR,
733 	IB_EVENT_QP_ACCESS_ERR,
734 	IB_EVENT_COMM_EST,
735 	IB_EVENT_SQ_DRAINED,
736 	IB_EVENT_PATH_MIG,
737 	IB_EVENT_PATH_MIG_ERR,
738 	IB_EVENT_DEVICE_FATAL,
739 	IB_EVENT_PORT_ACTIVE,
740 	IB_EVENT_PORT_ERR,
741 	IB_EVENT_LID_CHANGE,
742 	IB_EVENT_PKEY_CHANGE,
743 	IB_EVENT_SM_CHANGE,
744 	IB_EVENT_SRQ_ERR,
745 	IB_EVENT_SRQ_LIMIT_REACHED,
746 	IB_EVENT_QP_LAST_WQE_REACHED,
747 	IB_EVENT_CLIENT_REREGISTER,
748 	IB_EVENT_GID_CHANGE,
749 	IB_EVENT_WQ_FATAL,
750 };
751 
752 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
753 
754 struct ib_event {
755 	struct ib_device	*device;
756 	union {
757 		struct ib_cq	*cq;
758 		struct ib_qp	*qp;
759 		struct ib_srq	*srq;
760 		struct ib_wq	*wq;
761 		u32		port_num;
762 	} element;
763 	enum ib_event_type	event;
764 };
765 
766 struct ib_event_handler {
767 	struct ib_device *device;
768 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
769 	struct list_head  list;
770 };
771 
772 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
773 	do {							\
774 		(_ptr)->device  = _device;			\
775 		(_ptr)->handler = _handler;			\
776 		INIT_LIST_HEAD(&(_ptr)->list);			\
777 	} while (0)
778 
779 struct ib_global_route {
780 	const struct ib_gid_attr *sgid_attr;
781 	union ib_gid	dgid;
782 	u32		flow_label;
783 	u8		sgid_index;
784 	u8		hop_limit;
785 	u8		traffic_class;
786 };
787 
788 struct ib_grh {
789 	__be32		version_tclass_flow;
790 	__be16		paylen;
791 	u8		next_hdr;
792 	u8		hop_limit;
793 	union ib_gid	sgid;
794 	union ib_gid	dgid;
795 };
796 
797 union rdma_network_hdr {
798 	struct ib_grh ibgrh;
799 	struct {
800 		/* The IB spec states that if it's IPv4, the header
801 		 * is located in the last 20 bytes of the header.
802 		 */
803 		u8		reserved[20];
804 		struct iphdr	roce4grh;
805 	};
806 };
807 
808 #define IB_QPN_MASK		0xFFFFFF
809 
810 enum {
811 	IB_MULTICAST_QPN = 0xffffff
812 };
813 
814 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
815 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
816 
817 enum ib_ah_flags {
818 	IB_AH_GRH	= 1
819 };
820 
821 enum ib_rate {
822 	IB_RATE_PORT_CURRENT = 0,
823 	IB_RATE_2_5_GBPS = 2,
824 	IB_RATE_5_GBPS   = 5,
825 	IB_RATE_10_GBPS  = 3,
826 	IB_RATE_20_GBPS  = 6,
827 	IB_RATE_30_GBPS  = 4,
828 	IB_RATE_40_GBPS  = 7,
829 	IB_RATE_60_GBPS  = 8,
830 	IB_RATE_80_GBPS  = 9,
831 	IB_RATE_120_GBPS = 10,
832 	IB_RATE_14_GBPS  = 11,
833 	IB_RATE_56_GBPS  = 12,
834 	IB_RATE_112_GBPS = 13,
835 	IB_RATE_168_GBPS = 14,
836 	IB_RATE_25_GBPS  = 15,
837 	IB_RATE_100_GBPS = 16,
838 	IB_RATE_200_GBPS = 17,
839 	IB_RATE_300_GBPS = 18,
840 	IB_RATE_28_GBPS  = 19,
841 	IB_RATE_50_GBPS  = 20,
842 	IB_RATE_400_GBPS = 21,
843 	IB_RATE_600_GBPS = 22,
844 	IB_RATE_800_GBPS = 23,
845 };
846 
847 /**
848  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
849  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
850  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
851  * @rate: rate to convert.
852  */
853 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
854 
855 /**
856  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
857  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
858  * @rate: rate to convert.
859  */
860 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
861 
862 
863 /**
864  * enum ib_mr_type - memory region type
865  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
866  *                            normal registration
867  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
868  *                            register any arbitrary sg lists (without
869  *                            the normal mr constraints - see
870  *                            ib_map_mr_sg)
871  * @IB_MR_TYPE_DM:            memory region that is used for device
872  *                            memory registration
873  * @IB_MR_TYPE_USER:          memory region that is used for the user-space
874  *                            application
875  * @IB_MR_TYPE_DMA:           memory region that is used for DMA operations
876  *                            without address translations (VA=PA)
877  * @IB_MR_TYPE_INTEGRITY:     memory region that is used for
878  *                            data integrity operations
879  */
880 enum ib_mr_type {
881 	IB_MR_TYPE_MEM_REG,
882 	IB_MR_TYPE_SG_GAPS,
883 	IB_MR_TYPE_DM,
884 	IB_MR_TYPE_USER,
885 	IB_MR_TYPE_DMA,
886 	IB_MR_TYPE_INTEGRITY,
887 };
888 
889 enum ib_mr_status_check {
890 	IB_MR_CHECK_SIG_STATUS = 1,
891 };
892 
893 /**
894  * struct ib_mr_status - Memory region status container
895  *
896  * @fail_status: Bitmask of MR checks status. For each
897  *     failed check a corresponding status bit is set.
898  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
899  *     failure.
900  */
901 struct ib_mr_status {
902 	u32		    fail_status;
903 	struct ib_sig_err   sig_err;
904 };
905 
906 /**
907  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
908  * enum.
909  * @mult: multiple to convert.
910  */
911 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
912 
913 struct rdma_ah_init_attr {
914 	struct rdma_ah_attr *ah_attr;
915 	u32 flags;
916 	struct net_device *xmit_slave;
917 };
918 
919 enum rdma_ah_attr_type {
920 	RDMA_AH_ATTR_TYPE_UNDEFINED,
921 	RDMA_AH_ATTR_TYPE_IB,
922 	RDMA_AH_ATTR_TYPE_ROCE,
923 	RDMA_AH_ATTR_TYPE_OPA,
924 };
925 
926 struct ib_ah_attr {
927 	u16			dlid;
928 	u8			src_path_bits;
929 };
930 
931 struct roce_ah_attr {
932 	u8			dmac[ETH_ALEN];
933 };
934 
935 struct opa_ah_attr {
936 	u32			dlid;
937 	u8			src_path_bits;
938 	bool			make_grd;
939 };
940 
941 struct rdma_ah_attr {
942 	struct ib_global_route	grh;
943 	u8			sl;
944 	u8			static_rate;
945 	u32			port_num;
946 	u8			ah_flags;
947 	enum rdma_ah_attr_type type;
948 	union {
949 		struct ib_ah_attr ib;
950 		struct roce_ah_attr roce;
951 		struct opa_ah_attr opa;
952 	};
953 };
954 
955 enum ib_wc_status {
956 	IB_WC_SUCCESS,
957 	IB_WC_LOC_LEN_ERR,
958 	IB_WC_LOC_QP_OP_ERR,
959 	IB_WC_LOC_EEC_OP_ERR,
960 	IB_WC_LOC_PROT_ERR,
961 	IB_WC_WR_FLUSH_ERR,
962 	IB_WC_MW_BIND_ERR,
963 	IB_WC_BAD_RESP_ERR,
964 	IB_WC_LOC_ACCESS_ERR,
965 	IB_WC_REM_INV_REQ_ERR,
966 	IB_WC_REM_ACCESS_ERR,
967 	IB_WC_REM_OP_ERR,
968 	IB_WC_RETRY_EXC_ERR,
969 	IB_WC_RNR_RETRY_EXC_ERR,
970 	IB_WC_LOC_RDD_VIOL_ERR,
971 	IB_WC_REM_INV_RD_REQ_ERR,
972 	IB_WC_REM_ABORT_ERR,
973 	IB_WC_INV_EECN_ERR,
974 	IB_WC_INV_EEC_STATE_ERR,
975 	IB_WC_FATAL_ERR,
976 	IB_WC_RESP_TIMEOUT_ERR,
977 	IB_WC_GENERAL_ERR
978 };
979 
980 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
981 
982 enum ib_wc_opcode {
983 	IB_WC_SEND = IB_UVERBS_WC_SEND,
984 	IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
985 	IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
986 	IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
987 	IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
988 	IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
989 	IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
990 	IB_WC_LSO = IB_UVERBS_WC_TSO,
991 	IB_WC_ATOMIC_WRITE = IB_UVERBS_WC_ATOMIC_WRITE,
992 	IB_WC_REG_MR,
993 	IB_WC_MASKED_COMP_SWAP,
994 	IB_WC_MASKED_FETCH_ADD,
995 	IB_WC_FLUSH = IB_UVERBS_WC_FLUSH,
996 /*
997  * Set value of IB_WC_RECV so consumers can test if a completion is a
998  * receive by testing (opcode & IB_WC_RECV).
999  */
1000 	IB_WC_RECV			= 1 << 7,
1001 	IB_WC_RECV_RDMA_WITH_IMM
1002 };
1003 
1004 enum ib_wc_flags {
1005 	IB_WC_GRH		= 1,
1006 	IB_WC_WITH_IMM		= (1<<1),
1007 	IB_WC_WITH_INVALIDATE	= (1<<2),
1008 	IB_WC_IP_CSUM_OK	= (1<<3),
1009 	IB_WC_WITH_SMAC		= (1<<4),
1010 	IB_WC_WITH_VLAN		= (1<<5),
1011 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
1012 };
1013 
1014 struct ib_wc {
1015 	union {
1016 		u64		wr_id;
1017 		struct ib_cqe	*wr_cqe;
1018 	};
1019 	enum ib_wc_status	status;
1020 	enum ib_wc_opcode	opcode;
1021 	u32			vendor_err;
1022 	u32			byte_len;
1023 	struct ib_qp	       *qp;
1024 	union {
1025 		__be32		imm_data;
1026 		u32		invalidate_rkey;
1027 	} ex;
1028 	u32			src_qp;
1029 	u32			slid;
1030 	int			wc_flags;
1031 	u16			pkey_index;
1032 	u8			sl;
1033 	u8			dlid_path_bits;
1034 	u32 port_num; /* valid only for DR SMPs on switches */
1035 	u8			smac[ETH_ALEN];
1036 	u16			vlan_id;
1037 	u8			network_hdr_type;
1038 };
1039 
1040 enum ib_cq_notify_flags {
1041 	IB_CQ_SOLICITED			= 1 << 0,
1042 	IB_CQ_NEXT_COMP			= 1 << 1,
1043 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1044 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
1045 };
1046 
1047 enum ib_srq_type {
1048 	IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1049 	IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1050 	IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1051 };
1052 
1053 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1054 {
1055 	return srq_type == IB_SRQT_XRC ||
1056 	       srq_type == IB_SRQT_TM;
1057 }
1058 
1059 enum ib_srq_attr_mask {
1060 	IB_SRQ_MAX_WR	= 1 << 0,
1061 	IB_SRQ_LIMIT	= 1 << 1,
1062 };
1063 
1064 struct ib_srq_attr {
1065 	u32	max_wr;
1066 	u32	max_sge;
1067 	u32	srq_limit;
1068 };
1069 
1070 struct ib_srq_init_attr {
1071 	void		      (*event_handler)(struct ib_event *, void *);
1072 	void		       *srq_context;
1073 	struct ib_srq_attr	attr;
1074 	enum ib_srq_type	srq_type;
1075 
1076 	struct {
1077 		struct ib_cq   *cq;
1078 		union {
1079 			struct {
1080 				struct ib_xrcd *xrcd;
1081 			} xrc;
1082 
1083 			struct {
1084 				u32		max_num_tags;
1085 			} tag_matching;
1086 		};
1087 	} ext;
1088 };
1089 
1090 struct ib_qp_cap {
1091 	u32	max_send_wr;
1092 	u32	max_recv_wr;
1093 	u32	max_send_sge;
1094 	u32	max_recv_sge;
1095 	u32	max_inline_data;
1096 
1097 	/*
1098 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1099 	 * ib_create_qp() will calculate the right amount of needed WRs
1100 	 * and MRs based on this.
1101 	 */
1102 	u32	max_rdma_ctxs;
1103 };
1104 
1105 enum ib_sig_type {
1106 	IB_SIGNAL_ALL_WR,
1107 	IB_SIGNAL_REQ_WR
1108 };
1109 
1110 enum ib_qp_type {
1111 	/*
1112 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1113 	 * here (and in that order) since the MAD layer uses them as
1114 	 * indices into a 2-entry table.
1115 	 */
1116 	IB_QPT_SMI,
1117 	IB_QPT_GSI,
1118 
1119 	IB_QPT_RC = IB_UVERBS_QPT_RC,
1120 	IB_QPT_UC = IB_UVERBS_QPT_UC,
1121 	IB_QPT_UD = IB_UVERBS_QPT_UD,
1122 	IB_QPT_RAW_IPV6,
1123 	IB_QPT_RAW_ETHERTYPE,
1124 	IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1125 	IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1126 	IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1127 	IB_QPT_MAX,
1128 	IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1129 	/* Reserve a range for qp types internal to the low level driver.
1130 	 * These qp types will not be visible at the IB core layer, so the
1131 	 * IB_QPT_MAX usages should not be affected in the core layer
1132 	 */
1133 	IB_QPT_RESERVED1 = 0x1000,
1134 	IB_QPT_RESERVED2,
1135 	IB_QPT_RESERVED3,
1136 	IB_QPT_RESERVED4,
1137 	IB_QPT_RESERVED5,
1138 	IB_QPT_RESERVED6,
1139 	IB_QPT_RESERVED7,
1140 	IB_QPT_RESERVED8,
1141 	IB_QPT_RESERVED9,
1142 	IB_QPT_RESERVED10,
1143 };
1144 
1145 enum ib_qp_create_flags {
1146 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1147 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	=
1148 		IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1149 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1150 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1151 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1152 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1153 	IB_QP_CREATE_INTEGRITY_EN		= 1 << 6,
1154 	IB_QP_CREATE_NETDEV_USE			= 1 << 7,
1155 	IB_QP_CREATE_SCATTER_FCS		=
1156 		IB_UVERBS_QP_CREATE_SCATTER_FCS,
1157 	IB_QP_CREATE_CVLAN_STRIPPING		=
1158 		IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1159 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1160 	IB_QP_CREATE_PCI_WRITE_END_PADDING	=
1161 		IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1162 	/* reserve bits 26-31 for low level drivers' internal use */
1163 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1164 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1165 };
1166 
1167 /*
1168  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1169  * callback to destroy the passed in QP.
1170  */
1171 
1172 struct ib_qp_init_attr {
1173 	/* This callback occurs in workqueue context */
1174 	void                  (*event_handler)(struct ib_event *, void *);
1175 
1176 	void		       *qp_context;
1177 	struct ib_cq	       *send_cq;
1178 	struct ib_cq	       *recv_cq;
1179 	struct ib_srq	       *srq;
1180 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1181 	struct ib_qp_cap	cap;
1182 	enum ib_sig_type	sq_sig_type;
1183 	enum ib_qp_type		qp_type;
1184 	u32			create_flags;
1185 
1186 	/*
1187 	 * Only needed for special QP types, or when using the RW API.
1188 	 */
1189 	u32			port_num;
1190 	struct ib_rwq_ind_table *rwq_ind_tbl;
1191 	u32			source_qpn;
1192 };
1193 
1194 struct ib_qp_open_attr {
1195 	void                  (*event_handler)(struct ib_event *, void *);
1196 	void		       *qp_context;
1197 	u32			qp_num;
1198 	enum ib_qp_type		qp_type;
1199 };
1200 
1201 enum ib_rnr_timeout {
1202 	IB_RNR_TIMER_655_36 =  0,
1203 	IB_RNR_TIMER_000_01 =  1,
1204 	IB_RNR_TIMER_000_02 =  2,
1205 	IB_RNR_TIMER_000_03 =  3,
1206 	IB_RNR_TIMER_000_04 =  4,
1207 	IB_RNR_TIMER_000_06 =  5,
1208 	IB_RNR_TIMER_000_08 =  6,
1209 	IB_RNR_TIMER_000_12 =  7,
1210 	IB_RNR_TIMER_000_16 =  8,
1211 	IB_RNR_TIMER_000_24 =  9,
1212 	IB_RNR_TIMER_000_32 = 10,
1213 	IB_RNR_TIMER_000_48 = 11,
1214 	IB_RNR_TIMER_000_64 = 12,
1215 	IB_RNR_TIMER_000_96 = 13,
1216 	IB_RNR_TIMER_001_28 = 14,
1217 	IB_RNR_TIMER_001_92 = 15,
1218 	IB_RNR_TIMER_002_56 = 16,
1219 	IB_RNR_TIMER_003_84 = 17,
1220 	IB_RNR_TIMER_005_12 = 18,
1221 	IB_RNR_TIMER_007_68 = 19,
1222 	IB_RNR_TIMER_010_24 = 20,
1223 	IB_RNR_TIMER_015_36 = 21,
1224 	IB_RNR_TIMER_020_48 = 22,
1225 	IB_RNR_TIMER_030_72 = 23,
1226 	IB_RNR_TIMER_040_96 = 24,
1227 	IB_RNR_TIMER_061_44 = 25,
1228 	IB_RNR_TIMER_081_92 = 26,
1229 	IB_RNR_TIMER_122_88 = 27,
1230 	IB_RNR_TIMER_163_84 = 28,
1231 	IB_RNR_TIMER_245_76 = 29,
1232 	IB_RNR_TIMER_327_68 = 30,
1233 	IB_RNR_TIMER_491_52 = 31
1234 };
1235 
1236 enum ib_qp_attr_mask {
1237 	IB_QP_STATE			= 1,
1238 	IB_QP_CUR_STATE			= (1<<1),
1239 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1240 	IB_QP_ACCESS_FLAGS		= (1<<3),
1241 	IB_QP_PKEY_INDEX		= (1<<4),
1242 	IB_QP_PORT			= (1<<5),
1243 	IB_QP_QKEY			= (1<<6),
1244 	IB_QP_AV			= (1<<7),
1245 	IB_QP_PATH_MTU			= (1<<8),
1246 	IB_QP_TIMEOUT			= (1<<9),
1247 	IB_QP_RETRY_CNT			= (1<<10),
1248 	IB_QP_RNR_RETRY			= (1<<11),
1249 	IB_QP_RQ_PSN			= (1<<12),
1250 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1251 	IB_QP_ALT_PATH			= (1<<14),
1252 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1253 	IB_QP_SQ_PSN			= (1<<16),
1254 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1255 	IB_QP_PATH_MIG_STATE		= (1<<18),
1256 	IB_QP_CAP			= (1<<19),
1257 	IB_QP_DEST_QPN			= (1<<20),
1258 	IB_QP_RESERVED1			= (1<<21),
1259 	IB_QP_RESERVED2			= (1<<22),
1260 	IB_QP_RESERVED3			= (1<<23),
1261 	IB_QP_RESERVED4			= (1<<24),
1262 	IB_QP_RATE_LIMIT		= (1<<25),
1263 
1264 	IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1265 };
1266 
1267 enum ib_qp_state {
1268 	IB_QPS_RESET,
1269 	IB_QPS_INIT,
1270 	IB_QPS_RTR,
1271 	IB_QPS_RTS,
1272 	IB_QPS_SQD,
1273 	IB_QPS_SQE,
1274 	IB_QPS_ERR
1275 };
1276 
1277 enum ib_mig_state {
1278 	IB_MIG_MIGRATED,
1279 	IB_MIG_REARM,
1280 	IB_MIG_ARMED
1281 };
1282 
1283 enum ib_mw_type {
1284 	IB_MW_TYPE_1 = 1,
1285 	IB_MW_TYPE_2 = 2
1286 };
1287 
1288 struct ib_qp_attr {
1289 	enum ib_qp_state	qp_state;
1290 	enum ib_qp_state	cur_qp_state;
1291 	enum ib_mtu		path_mtu;
1292 	enum ib_mig_state	path_mig_state;
1293 	u32			qkey;
1294 	u32			rq_psn;
1295 	u32			sq_psn;
1296 	u32			dest_qp_num;
1297 	int			qp_access_flags;
1298 	struct ib_qp_cap	cap;
1299 	struct rdma_ah_attr	ah_attr;
1300 	struct rdma_ah_attr	alt_ah_attr;
1301 	u16			pkey_index;
1302 	u16			alt_pkey_index;
1303 	u8			en_sqd_async_notify;
1304 	u8			sq_draining;
1305 	u8			max_rd_atomic;
1306 	u8			max_dest_rd_atomic;
1307 	u8			min_rnr_timer;
1308 	u32			port_num;
1309 	u8			timeout;
1310 	u8			retry_cnt;
1311 	u8			rnr_retry;
1312 	u32			alt_port_num;
1313 	u8			alt_timeout;
1314 	u32			rate_limit;
1315 	struct net_device	*xmit_slave;
1316 };
1317 
1318 enum ib_wr_opcode {
1319 	/* These are shared with userspace */
1320 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1321 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1322 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1323 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1324 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1325 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1326 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1327 	IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1328 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1329 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1330 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1331 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1332 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1333 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1334 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1335 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1336 	IB_WR_FLUSH = IB_UVERBS_WR_FLUSH,
1337 	IB_WR_ATOMIC_WRITE = IB_UVERBS_WR_ATOMIC_WRITE,
1338 
1339 	/* These are kernel only and can not be issued by userspace */
1340 	IB_WR_REG_MR = 0x20,
1341 	IB_WR_REG_MR_INTEGRITY,
1342 
1343 	/* reserve values for low level drivers' internal use.
1344 	 * These values will not be used at all in the ib core layer.
1345 	 */
1346 	IB_WR_RESERVED1 = 0xf0,
1347 	IB_WR_RESERVED2,
1348 	IB_WR_RESERVED3,
1349 	IB_WR_RESERVED4,
1350 	IB_WR_RESERVED5,
1351 	IB_WR_RESERVED6,
1352 	IB_WR_RESERVED7,
1353 	IB_WR_RESERVED8,
1354 	IB_WR_RESERVED9,
1355 	IB_WR_RESERVED10,
1356 };
1357 
1358 enum ib_send_flags {
1359 	IB_SEND_FENCE		= 1,
1360 	IB_SEND_SIGNALED	= (1<<1),
1361 	IB_SEND_SOLICITED	= (1<<2),
1362 	IB_SEND_INLINE		= (1<<3),
1363 	IB_SEND_IP_CSUM		= (1<<4),
1364 
1365 	/* reserve bits 26-31 for low level drivers' internal use */
1366 	IB_SEND_RESERVED_START	= (1 << 26),
1367 	IB_SEND_RESERVED_END	= (1 << 31),
1368 };
1369 
1370 struct ib_sge {
1371 	u64	addr;
1372 	u32	length;
1373 	u32	lkey;
1374 };
1375 
1376 struct ib_cqe {
1377 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1378 };
1379 
1380 struct ib_send_wr {
1381 	struct ib_send_wr      *next;
1382 	union {
1383 		u64		wr_id;
1384 		struct ib_cqe	*wr_cqe;
1385 	};
1386 	struct ib_sge	       *sg_list;
1387 	int			num_sge;
1388 	enum ib_wr_opcode	opcode;
1389 	int			send_flags;
1390 	union {
1391 		__be32		imm_data;
1392 		u32		invalidate_rkey;
1393 	} ex;
1394 };
1395 
1396 struct ib_rdma_wr {
1397 	struct ib_send_wr	wr;
1398 	u64			remote_addr;
1399 	u32			rkey;
1400 };
1401 
1402 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1403 {
1404 	return container_of(wr, struct ib_rdma_wr, wr);
1405 }
1406 
1407 struct ib_atomic_wr {
1408 	struct ib_send_wr	wr;
1409 	u64			remote_addr;
1410 	u64			compare_add;
1411 	u64			swap;
1412 	u64			compare_add_mask;
1413 	u64			swap_mask;
1414 	u32			rkey;
1415 };
1416 
1417 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1418 {
1419 	return container_of(wr, struct ib_atomic_wr, wr);
1420 }
1421 
1422 struct ib_ud_wr {
1423 	struct ib_send_wr	wr;
1424 	struct ib_ah		*ah;
1425 	void			*header;
1426 	int			hlen;
1427 	int			mss;
1428 	u32			remote_qpn;
1429 	u32			remote_qkey;
1430 	u16			pkey_index; /* valid for GSI only */
1431 	u32			port_num; /* valid for DR SMPs on switch only */
1432 };
1433 
1434 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1435 {
1436 	return container_of(wr, struct ib_ud_wr, wr);
1437 }
1438 
1439 struct ib_reg_wr {
1440 	struct ib_send_wr	wr;
1441 	struct ib_mr		*mr;
1442 	u32			key;
1443 	int			access;
1444 };
1445 
1446 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1447 {
1448 	return container_of(wr, struct ib_reg_wr, wr);
1449 }
1450 
1451 struct ib_recv_wr {
1452 	struct ib_recv_wr      *next;
1453 	union {
1454 		u64		wr_id;
1455 		struct ib_cqe	*wr_cqe;
1456 	};
1457 	struct ib_sge	       *sg_list;
1458 	int			num_sge;
1459 };
1460 
1461 enum ib_access_flags {
1462 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1463 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1464 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1465 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1466 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1467 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1468 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1469 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1470 	IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1471 	IB_ACCESS_FLUSH_GLOBAL = IB_UVERBS_ACCESS_FLUSH_GLOBAL,
1472 	IB_ACCESS_FLUSH_PERSISTENT = IB_UVERBS_ACCESS_FLUSH_PERSISTENT,
1473 
1474 	IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1475 	IB_ACCESS_SUPPORTED =
1476 		((IB_ACCESS_FLUSH_PERSISTENT << 1) - 1) | IB_ACCESS_OPTIONAL,
1477 };
1478 
1479 /*
1480  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1481  * are hidden here instead of a uapi header!
1482  */
1483 enum ib_mr_rereg_flags {
1484 	IB_MR_REREG_TRANS	= 1,
1485 	IB_MR_REREG_PD		= (1<<1),
1486 	IB_MR_REREG_ACCESS	= (1<<2),
1487 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1488 };
1489 
1490 struct ib_umem;
1491 
1492 enum rdma_remove_reason {
1493 	/*
1494 	 * Userspace requested uobject deletion or initial try
1495 	 * to remove uobject via cleanup. Call could fail
1496 	 */
1497 	RDMA_REMOVE_DESTROY,
1498 	/* Context deletion. This call should delete the actual object itself */
1499 	RDMA_REMOVE_CLOSE,
1500 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1501 	RDMA_REMOVE_DRIVER_REMOVE,
1502 	/* uobj is being cleaned-up before being committed */
1503 	RDMA_REMOVE_ABORT,
1504 	/* The driver failed to destroy the uobject and is being disconnected */
1505 	RDMA_REMOVE_DRIVER_FAILURE,
1506 };
1507 
1508 struct ib_rdmacg_object {
1509 #ifdef CONFIG_CGROUP_RDMA
1510 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1511 #endif
1512 };
1513 
1514 struct ib_ucontext {
1515 	struct ib_device       *device;
1516 	struct ib_uverbs_file  *ufile;
1517 
1518 	struct ib_rdmacg_object	cg_obj;
1519 	/*
1520 	 * Implementation details of the RDMA core, don't use in drivers:
1521 	 */
1522 	struct rdma_restrack_entry res;
1523 	struct xarray mmap_xa;
1524 };
1525 
1526 struct ib_uobject {
1527 	u64			user_handle;	/* handle given to us by userspace */
1528 	/* ufile & ucontext owning this object */
1529 	struct ib_uverbs_file  *ufile;
1530 	/* FIXME, save memory: ufile->context == context */
1531 	struct ib_ucontext     *context;	/* associated user context */
1532 	void		       *object;		/* containing object */
1533 	struct list_head	list;		/* link to context's list */
1534 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1535 	int			id;		/* index into kernel idr */
1536 	struct kref		ref;
1537 	atomic_t		usecnt;		/* protects exclusive access */
1538 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1539 
1540 	const struct uverbs_api_object *uapi_object;
1541 };
1542 
1543 struct ib_udata {
1544 	const void __user *inbuf;
1545 	void __user *outbuf;
1546 	size_t       inlen;
1547 	size_t       outlen;
1548 };
1549 
1550 struct ib_pd {
1551 	u32			local_dma_lkey;
1552 	u32			flags;
1553 	struct ib_device       *device;
1554 	struct ib_uobject      *uobject;
1555 	atomic_t          	usecnt; /* count all resources */
1556 
1557 	u32			unsafe_global_rkey;
1558 
1559 	/*
1560 	 * Implementation details of the RDMA core, don't use in drivers:
1561 	 */
1562 	struct ib_mr	       *__internal_mr;
1563 	struct rdma_restrack_entry res;
1564 };
1565 
1566 struct ib_xrcd {
1567 	struct ib_device       *device;
1568 	atomic_t		usecnt; /* count all exposed resources */
1569 	struct inode	       *inode;
1570 	struct rw_semaphore	tgt_qps_rwsem;
1571 	struct xarray		tgt_qps;
1572 };
1573 
1574 struct ib_ah {
1575 	struct ib_device	*device;
1576 	struct ib_pd		*pd;
1577 	struct ib_uobject	*uobject;
1578 	const struct ib_gid_attr *sgid_attr;
1579 	enum rdma_ah_attr_type	type;
1580 };
1581 
1582 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1583 
1584 enum ib_poll_context {
1585 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1586 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1587 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1588 	IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1589 
1590 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1591 };
1592 
1593 struct ib_cq {
1594 	struct ib_device       *device;
1595 	struct ib_ucq_object   *uobject;
1596 	ib_comp_handler   	comp_handler;
1597 	void                  (*event_handler)(struct ib_event *, void *);
1598 	void                   *cq_context;
1599 	int               	cqe;
1600 	unsigned int		cqe_used;
1601 	atomic_t          	usecnt; /* count number of work queues */
1602 	enum ib_poll_context	poll_ctx;
1603 	struct ib_wc		*wc;
1604 	struct list_head        pool_entry;
1605 	union {
1606 		struct irq_poll		iop;
1607 		struct work_struct	work;
1608 	};
1609 	struct workqueue_struct *comp_wq;
1610 	struct dim *dim;
1611 
1612 	/* updated only by trace points */
1613 	ktime_t timestamp;
1614 	u8 interrupt:1;
1615 	u8 shared:1;
1616 	unsigned int comp_vector;
1617 
1618 	/*
1619 	 * Implementation details of the RDMA core, don't use in drivers:
1620 	 */
1621 	struct rdma_restrack_entry res;
1622 };
1623 
1624 struct ib_srq {
1625 	struct ib_device       *device;
1626 	struct ib_pd	       *pd;
1627 	struct ib_usrq_object  *uobject;
1628 	void		      (*event_handler)(struct ib_event *, void *);
1629 	void		       *srq_context;
1630 	enum ib_srq_type	srq_type;
1631 	atomic_t		usecnt;
1632 
1633 	struct {
1634 		struct ib_cq   *cq;
1635 		union {
1636 			struct {
1637 				struct ib_xrcd *xrcd;
1638 				u32		srq_num;
1639 			} xrc;
1640 		};
1641 	} ext;
1642 
1643 	/*
1644 	 * Implementation details of the RDMA core, don't use in drivers:
1645 	 */
1646 	struct rdma_restrack_entry res;
1647 };
1648 
1649 enum ib_raw_packet_caps {
1650 	/*
1651 	 * Strip cvlan from incoming packet and report it in the matching work
1652 	 * completion is supported.
1653 	 */
1654 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING =
1655 		IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING,
1656 	/*
1657 	 * Scatter FCS field of an incoming packet to host memory is supported.
1658 	 */
1659 	IB_RAW_PACKET_CAP_SCATTER_FCS = IB_UVERBS_RAW_PACKET_CAP_SCATTER_FCS,
1660 	/* Checksum offloads are supported (for both send and receive). */
1661 	IB_RAW_PACKET_CAP_IP_CSUM = IB_UVERBS_RAW_PACKET_CAP_IP_CSUM,
1662 	/*
1663 	 * When a packet is received for an RQ with no receive WQEs, the
1664 	 * packet processing is delayed.
1665 	 */
1666 	IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP,
1667 };
1668 
1669 enum ib_wq_type {
1670 	IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1671 };
1672 
1673 enum ib_wq_state {
1674 	IB_WQS_RESET,
1675 	IB_WQS_RDY,
1676 	IB_WQS_ERR
1677 };
1678 
1679 struct ib_wq {
1680 	struct ib_device       *device;
1681 	struct ib_uwq_object   *uobject;
1682 	void		    *wq_context;
1683 	void		    (*event_handler)(struct ib_event *, void *);
1684 	struct ib_pd	       *pd;
1685 	struct ib_cq	       *cq;
1686 	u32		wq_num;
1687 	enum ib_wq_state       state;
1688 	enum ib_wq_type	wq_type;
1689 	atomic_t		usecnt;
1690 };
1691 
1692 enum ib_wq_flags {
1693 	IB_WQ_FLAGS_CVLAN_STRIPPING	= IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1694 	IB_WQ_FLAGS_SCATTER_FCS		= IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1695 	IB_WQ_FLAGS_DELAY_DROP		= IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1696 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1697 				IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1698 };
1699 
1700 struct ib_wq_init_attr {
1701 	void		       *wq_context;
1702 	enum ib_wq_type	wq_type;
1703 	u32		max_wr;
1704 	u32		max_sge;
1705 	struct	ib_cq	       *cq;
1706 	void		    (*event_handler)(struct ib_event *, void *);
1707 	u32		create_flags; /* Use enum ib_wq_flags */
1708 };
1709 
1710 enum ib_wq_attr_mask {
1711 	IB_WQ_STATE		= 1 << 0,
1712 	IB_WQ_CUR_STATE		= 1 << 1,
1713 	IB_WQ_FLAGS		= 1 << 2,
1714 };
1715 
1716 struct ib_wq_attr {
1717 	enum	ib_wq_state	wq_state;
1718 	enum	ib_wq_state	curr_wq_state;
1719 	u32			flags; /* Use enum ib_wq_flags */
1720 	u32			flags_mask; /* Use enum ib_wq_flags */
1721 };
1722 
1723 struct ib_rwq_ind_table {
1724 	struct ib_device	*device;
1725 	struct ib_uobject      *uobject;
1726 	atomic_t		usecnt;
1727 	u32		ind_tbl_num;
1728 	u32		log_ind_tbl_size;
1729 	struct ib_wq	**ind_tbl;
1730 };
1731 
1732 struct ib_rwq_ind_table_init_attr {
1733 	u32		log_ind_tbl_size;
1734 	/* Each entry is a pointer to Receive Work Queue */
1735 	struct ib_wq	**ind_tbl;
1736 };
1737 
1738 enum port_pkey_state {
1739 	IB_PORT_PKEY_NOT_VALID = 0,
1740 	IB_PORT_PKEY_VALID = 1,
1741 	IB_PORT_PKEY_LISTED = 2,
1742 };
1743 
1744 struct ib_qp_security;
1745 
1746 struct ib_port_pkey {
1747 	enum port_pkey_state	state;
1748 	u16			pkey_index;
1749 	u32			port_num;
1750 	struct list_head	qp_list;
1751 	struct list_head	to_error_list;
1752 	struct ib_qp_security  *sec;
1753 };
1754 
1755 struct ib_ports_pkeys {
1756 	struct ib_port_pkey	main;
1757 	struct ib_port_pkey	alt;
1758 };
1759 
1760 struct ib_qp_security {
1761 	struct ib_qp	       *qp;
1762 	struct ib_device       *dev;
1763 	/* Hold this mutex when changing port and pkey settings. */
1764 	struct mutex		mutex;
1765 	struct ib_ports_pkeys  *ports_pkeys;
1766 	/* A list of all open shared QP handles.  Required to enforce security
1767 	 * properly for all users of a shared QP.
1768 	 */
1769 	struct list_head        shared_qp_list;
1770 	void                   *security;
1771 	bool			destroying;
1772 	atomic_t		error_list_count;
1773 	struct completion	error_complete;
1774 	int			error_comps_pending;
1775 };
1776 
1777 /*
1778  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1779  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1780  */
1781 struct ib_qp {
1782 	struct ib_device       *device;
1783 	struct ib_pd	       *pd;
1784 	struct ib_cq	       *send_cq;
1785 	struct ib_cq	       *recv_cq;
1786 	spinlock_t		mr_lock;
1787 	int			mrs_used;
1788 	struct list_head	rdma_mrs;
1789 	struct list_head	sig_mrs;
1790 	struct ib_srq	       *srq;
1791 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1792 	struct list_head	xrcd_list;
1793 
1794 	/* count times opened, mcast attaches, flow attaches */
1795 	atomic_t		usecnt;
1796 	struct list_head	open_list;
1797 	struct ib_qp           *real_qp;
1798 	struct ib_uqp_object   *uobject;
1799 	void                  (*event_handler)(struct ib_event *, void *);
1800 	void		       *qp_context;
1801 	/* sgid_attrs associated with the AV's */
1802 	const struct ib_gid_attr *av_sgid_attr;
1803 	const struct ib_gid_attr *alt_path_sgid_attr;
1804 	u32			qp_num;
1805 	u32			max_write_sge;
1806 	u32			max_read_sge;
1807 	enum ib_qp_type		qp_type;
1808 	struct ib_rwq_ind_table *rwq_ind_tbl;
1809 	struct ib_qp_security  *qp_sec;
1810 	u32			port;
1811 
1812 	bool			integrity_en;
1813 	/*
1814 	 * Implementation details of the RDMA core, don't use in drivers:
1815 	 */
1816 	struct rdma_restrack_entry     res;
1817 
1818 	/* The counter the qp is bind to */
1819 	struct rdma_counter    *counter;
1820 };
1821 
1822 struct ib_dm {
1823 	struct ib_device  *device;
1824 	u32		   length;
1825 	u32		   flags;
1826 	struct ib_uobject *uobject;
1827 	atomic_t	   usecnt;
1828 };
1829 
1830 struct ib_mr {
1831 	struct ib_device  *device;
1832 	struct ib_pd	  *pd;
1833 	u32		   lkey;
1834 	u32		   rkey;
1835 	u64		   iova;
1836 	u64		   length;
1837 	unsigned int	   page_size;
1838 	enum ib_mr_type	   type;
1839 	bool		   need_inval;
1840 	union {
1841 		struct ib_uobject	*uobject;	/* user */
1842 		struct list_head	qp_entry;	/* FR */
1843 	};
1844 
1845 	struct ib_dm      *dm;
1846 	struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1847 	/*
1848 	 * Implementation details of the RDMA core, don't use in drivers:
1849 	 */
1850 	struct rdma_restrack_entry res;
1851 };
1852 
1853 struct ib_mw {
1854 	struct ib_device	*device;
1855 	struct ib_pd		*pd;
1856 	struct ib_uobject	*uobject;
1857 	u32			rkey;
1858 	enum ib_mw_type         type;
1859 };
1860 
1861 /* Supported steering options */
1862 enum ib_flow_attr_type {
1863 	/* steering according to rule specifications */
1864 	IB_FLOW_ATTR_NORMAL		= 0x0,
1865 	/* default unicast and multicast rule -
1866 	 * receive all Eth traffic which isn't steered to any QP
1867 	 */
1868 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1869 	/* default multicast rule -
1870 	 * receive all Eth multicast traffic which isn't steered to any QP
1871 	 */
1872 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1873 	/* sniffer rule - receive all port traffic */
1874 	IB_FLOW_ATTR_SNIFFER		= 0x3
1875 };
1876 
1877 /* Supported steering header types */
1878 enum ib_flow_spec_type {
1879 	/* L2 headers*/
1880 	IB_FLOW_SPEC_ETH		= 0x20,
1881 	IB_FLOW_SPEC_IB			= 0x22,
1882 	/* L3 header*/
1883 	IB_FLOW_SPEC_IPV4		= 0x30,
1884 	IB_FLOW_SPEC_IPV6		= 0x31,
1885 	IB_FLOW_SPEC_ESP                = 0x34,
1886 	/* L4 headers*/
1887 	IB_FLOW_SPEC_TCP		= 0x40,
1888 	IB_FLOW_SPEC_UDP		= 0x41,
1889 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1890 	IB_FLOW_SPEC_GRE		= 0x51,
1891 	IB_FLOW_SPEC_MPLS		= 0x60,
1892 	IB_FLOW_SPEC_INNER		= 0x100,
1893 	/* Actions */
1894 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1895 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1896 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1897 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1898 };
1899 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1900 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1901 
1902 enum ib_flow_flags {
1903 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1904 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1905 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1906 };
1907 
1908 struct ib_flow_eth_filter {
1909 	u8	dst_mac[6];
1910 	u8	src_mac[6];
1911 	__be16	ether_type;
1912 	__be16	vlan_tag;
1913 };
1914 
1915 struct ib_flow_spec_eth {
1916 	u32			  type;
1917 	u16			  size;
1918 	struct ib_flow_eth_filter val;
1919 	struct ib_flow_eth_filter mask;
1920 };
1921 
1922 struct ib_flow_ib_filter {
1923 	__be16 dlid;
1924 	__u8   sl;
1925 };
1926 
1927 struct ib_flow_spec_ib {
1928 	u32			 type;
1929 	u16			 size;
1930 	struct ib_flow_ib_filter val;
1931 	struct ib_flow_ib_filter mask;
1932 };
1933 
1934 /* IPv4 header flags */
1935 enum ib_ipv4_flags {
1936 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1937 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1938 				    last have this flag set */
1939 };
1940 
1941 struct ib_flow_ipv4_filter {
1942 	__be32	src_ip;
1943 	__be32	dst_ip;
1944 	u8	proto;
1945 	u8	tos;
1946 	u8	ttl;
1947 	u8	flags;
1948 };
1949 
1950 struct ib_flow_spec_ipv4 {
1951 	u32			   type;
1952 	u16			   size;
1953 	struct ib_flow_ipv4_filter val;
1954 	struct ib_flow_ipv4_filter mask;
1955 };
1956 
1957 struct ib_flow_ipv6_filter {
1958 	u8	src_ip[16];
1959 	u8	dst_ip[16];
1960 	__be32	flow_label;
1961 	u8	next_hdr;
1962 	u8	traffic_class;
1963 	u8	hop_limit;
1964 } __packed;
1965 
1966 struct ib_flow_spec_ipv6 {
1967 	u32			   type;
1968 	u16			   size;
1969 	struct ib_flow_ipv6_filter val;
1970 	struct ib_flow_ipv6_filter mask;
1971 };
1972 
1973 struct ib_flow_tcp_udp_filter {
1974 	__be16	dst_port;
1975 	__be16	src_port;
1976 };
1977 
1978 struct ib_flow_spec_tcp_udp {
1979 	u32			      type;
1980 	u16			      size;
1981 	struct ib_flow_tcp_udp_filter val;
1982 	struct ib_flow_tcp_udp_filter mask;
1983 };
1984 
1985 struct ib_flow_tunnel_filter {
1986 	__be32	tunnel_id;
1987 };
1988 
1989 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1990  * the tunnel_id from val has the vni value
1991  */
1992 struct ib_flow_spec_tunnel {
1993 	u32			      type;
1994 	u16			      size;
1995 	struct ib_flow_tunnel_filter  val;
1996 	struct ib_flow_tunnel_filter  mask;
1997 };
1998 
1999 struct ib_flow_esp_filter {
2000 	__be32	spi;
2001 	__be32  seq;
2002 };
2003 
2004 struct ib_flow_spec_esp {
2005 	u32                           type;
2006 	u16			      size;
2007 	struct ib_flow_esp_filter     val;
2008 	struct ib_flow_esp_filter     mask;
2009 };
2010 
2011 struct ib_flow_gre_filter {
2012 	__be16 c_ks_res0_ver;
2013 	__be16 protocol;
2014 	__be32 key;
2015 };
2016 
2017 struct ib_flow_spec_gre {
2018 	u32                           type;
2019 	u16			      size;
2020 	struct ib_flow_gre_filter     val;
2021 	struct ib_flow_gre_filter     mask;
2022 };
2023 
2024 struct ib_flow_mpls_filter {
2025 	__be32 tag;
2026 };
2027 
2028 struct ib_flow_spec_mpls {
2029 	u32                           type;
2030 	u16			      size;
2031 	struct ib_flow_mpls_filter     val;
2032 	struct ib_flow_mpls_filter     mask;
2033 };
2034 
2035 struct ib_flow_spec_action_tag {
2036 	enum ib_flow_spec_type	      type;
2037 	u16			      size;
2038 	u32                           tag_id;
2039 };
2040 
2041 struct ib_flow_spec_action_drop {
2042 	enum ib_flow_spec_type	      type;
2043 	u16			      size;
2044 };
2045 
2046 struct ib_flow_spec_action_handle {
2047 	enum ib_flow_spec_type	      type;
2048 	u16			      size;
2049 	struct ib_flow_action	     *act;
2050 };
2051 
2052 enum ib_counters_description {
2053 	IB_COUNTER_PACKETS,
2054 	IB_COUNTER_BYTES,
2055 };
2056 
2057 struct ib_flow_spec_action_count {
2058 	enum ib_flow_spec_type type;
2059 	u16 size;
2060 	struct ib_counters *counters;
2061 };
2062 
2063 union ib_flow_spec {
2064 	struct {
2065 		u32			type;
2066 		u16			size;
2067 	};
2068 	struct ib_flow_spec_eth		eth;
2069 	struct ib_flow_spec_ib		ib;
2070 	struct ib_flow_spec_ipv4        ipv4;
2071 	struct ib_flow_spec_tcp_udp	tcp_udp;
2072 	struct ib_flow_spec_ipv6        ipv6;
2073 	struct ib_flow_spec_tunnel      tunnel;
2074 	struct ib_flow_spec_esp		esp;
2075 	struct ib_flow_spec_gre		gre;
2076 	struct ib_flow_spec_mpls	mpls;
2077 	struct ib_flow_spec_action_tag  flow_tag;
2078 	struct ib_flow_spec_action_drop drop;
2079 	struct ib_flow_spec_action_handle action;
2080 	struct ib_flow_spec_action_count flow_count;
2081 };
2082 
2083 struct ib_flow_attr {
2084 	enum ib_flow_attr_type type;
2085 	u16	     size;
2086 	u16	     priority;
2087 	u32	     flags;
2088 	u8	     num_of_specs;
2089 	u32	     port;
2090 	union ib_flow_spec flows[];
2091 };
2092 
2093 struct ib_flow {
2094 	struct ib_qp		*qp;
2095 	struct ib_device	*device;
2096 	struct ib_uobject	*uobject;
2097 };
2098 
2099 enum ib_flow_action_type {
2100 	IB_FLOW_ACTION_UNSPECIFIED,
2101 	IB_FLOW_ACTION_ESP = 1,
2102 };
2103 
2104 struct ib_flow_action_attrs_esp_keymats {
2105 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2106 	union {
2107 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2108 	} keymat;
2109 };
2110 
2111 struct ib_flow_action_attrs_esp_replays {
2112 	enum ib_uverbs_flow_action_esp_replay			protocol;
2113 	union {
2114 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2115 	} replay;
2116 };
2117 
2118 enum ib_flow_action_attrs_esp_flags {
2119 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2120 	 * This is done in order to share the same flags between user-space and
2121 	 * kernel and spare an unnecessary translation.
2122 	 */
2123 
2124 	/* Kernel flags */
2125 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2126 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2127 };
2128 
2129 struct ib_flow_spec_list {
2130 	struct ib_flow_spec_list	*next;
2131 	union ib_flow_spec		spec;
2132 };
2133 
2134 struct ib_flow_action_attrs_esp {
2135 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2136 	struct ib_flow_action_attrs_esp_replays		*replay;
2137 	struct ib_flow_spec_list			*encap;
2138 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2139 	 * Value of 0 is a valid value.
2140 	 */
2141 	u32						esn;
2142 	u32						spi;
2143 	u32						seq;
2144 	u32						tfc_pad;
2145 	/* Use enum ib_flow_action_attrs_esp_flags */
2146 	u64						flags;
2147 	u64						hard_limit_pkts;
2148 };
2149 
2150 struct ib_flow_action {
2151 	struct ib_device		*device;
2152 	struct ib_uobject		*uobject;
2153 	enum ib_flow_action_type	type;
2154 	atomic_t			usecnt;
2155 };
2156 
2157 struct ib_mad;
2158 
2159 enum ib_process_mad_flags {
2160 	IB_MAD_IGNORE_MKEY	= 1,
2161 	IB_MAD_IGNORE_BKEY	= 2,
2162 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2163 };
2164 
2165 enum ib_mad_result {
2166 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2167 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2168 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2169 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2170 };
2171 
2172 struct ib_port_cache {
2173 	u64		      subnet_prefix;
2174 	struct ib_pkey_cache  *pkey;
2175 	struct ib_gid_table   *gid;
2176 	u8                     lmc;
2177 	enum ib_port_state     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 invalidations;
2258 	atomic64_t prefetch;
2259 };
2260 
2261 struct ib_counters {
2262 	struct ib_device	*device;
2263 	struct ib_uobject	*uobject;
2264 	/* num of objects attached */
2265 	atomic_t	usecnt;
2266 };
2267 
2268 struct ib_counters_read_attr {
2269 	u64	*counters_buff;
2270 	u32	ncounters;
2271 	u32	flags; /* use enum ib_read_counters_flags */
2272 };
2273 
2274 struct uverbs_attr_bundle;
2275 struct iw_cm_id;
2276 struct iw_cm_conn_param;
2277 
2278 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member)                      \
2279 	.size_##ib_struct =                                                    \
2280 		(sizeof(struct drv_struct) +                                   \
2281 		 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) +      \
2282 		 BUILD_BUG_ON_ZERO(                                            \
2283 			 !__same_type(((struct drv_struct *)NULL)->member,     \
2284 				      struct ib_struct)))
2285 
2286 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp)                          \
2287 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2288 					   gfp, false))
2289 
2290 #define rdma_zalloc_drv_obj_numa(ib_dev, ib_type)                              \
2291 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2292 					   GFP_KERNEL, true))
2293 
2294 #define rdma_zalloc_drv_obj(ib_dev, ib_type)                                   \
2295 	rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2296 
2297 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2298 
2299 struct rdma_user_mmap_entry {
2300 	struct kref ref;
2301 	struct ib_ucontext *ucontext;
2302 	unsigned long start_pgoff;
2303 	size_t npages;
2304 	bool driver_removed;
2305 };
2306 
2307 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2308 static inline u64
2309 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2310 {
2311 	return (u64)entry->start_pgoff << PAGE_SHIFT;
2312 }
2313 
2314 /**
2315  * struct ib_device_ops - InfiniBand device operations
2316  * This structure defines all the InfiniBand device operations, providers will
2317  * need to define the supported operations, otherwise they will be set to null.
2318  */
2319 struct ib_device_ops {
2320 	struct module *owner;
2321 	enum rdma_driver_id driver_id;
2322 	u32 uverbs_abi_ver;
2323 	unsigned int uverbs_no_driver_id_binding:1;
2324 
2325 	/*
2326 	 * NOTE: New drivers should not make use of device_group; instead new
2327 	 * device parameter should be exposed via netlink command. This
2328 	 * mechanism exists only for existing drivers.
2329 	 */
2330 	const struct attribute_group *device_group;
2331 	const struct attribute_group **port_groups;
2332 
2333 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2334 			 const struct ib_send_wr **bad_send_wr);
2335 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2336 			 const struct ib_recv_wr **bad_recv_wr);
2337 	void (*drain_rq)(struct ib_qp *qp);
2338 	void (*drain_sq)(struct ib_qp *qp);
2339 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2340 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2341 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2342 	int (*post_srq_recv)(struct ib_srq *srq,
2343 			     const struct ib_recv_wr *recv_wr,
2344 			     const struct ib_recv_wr **bad_recv_wr);
2345 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2346 			   u32 port_num, const struct ib_wc *in_wc,
2347 			   const struct ib_grh *in_grh,
2348 			   const struct ib_mad *in_mad, struct ib_mad *out_mad,
2349 			   size_t *out_mad_size, u16 *out_mad_pkey_index);
2350 	int (*query_device)(struct ib_device *device,
2351 			    struct ib_device_attr *device_attr,
2352 			    struct ib_udata *udata);
2353 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2354 			     struct ib_device_modify *device_modify);
2355 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2356 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2357 						     int comp_vector);
2358 	int (*query_port)(struct ib_device *device, u32 port_num,
2359 			  struct ib_port_attr *port_attr);
2360 	int (*modify_port)(struct ib_device *device, u32 port_num,
2361 			   int port_modify_mask,
2362 			   struct ib_port_modify *port_modify);
2363 	/**
2364 	 * The following mandatory functions are used only at device
2365 	 * registration.  Keep functions such as these at the end of this
2366 	 * structure to avoid cache line misses when accessing struct ib_device
2367 	 * in fast paths.
2368 	 */
2369 	int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2370 				  struct ib_port_immutable *immutable);
2371 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2372 					       u32 port_num);
2373 	/**
2374 	 * When calling get_netdev, the HW vendor's driver should return the
2375 	 * net device of device @device at port @port_num or NULL if such
2376 	 * a net device doesn't exist. The vendor driver should call dev_hold
2377 	 * on this net device. The HW vendor's device driver must guarantee
2378 	 * that this function returns NULL before the net device has finished
2379 	 * NETDEV_UNREGISTER state.
2380 	 */
2381 	struct net_device *(*get_netdev)(struct ib_device *device,
2382 					 u32 port_num);
2383 	/**
2384 	 * rdma netdev operation
2385 	 *
2386 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2387 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2388 	 */
2389 	struct net_device *(*alloc_rdma_netdev)(
2390 		struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2391 		const char *name, unsigned char name_assign_type,
2392 		void (*setup)(struct net_device *));
2393 
2394 	int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2395 				      enum rdma_netdev_t type,
2396 				      struct rdma_netdev_alloc_params *params);
2397 	/**
2398 	 * query_gid should be return GID value for @device, when @port_num
2399 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2400 	 * is RoCE link layer.
2401 	 */
2402 	int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2403 			 union ib_gid *gid);
2404 	/**
2405 	 * When calling add_gid, the HW vendor's driver should add the gid
2406 	 * of device of port at gid index available at @attr. Meta-info of
2407 	 * that gid (for example, the network device related to this gid) is
2408 	 * available at @attr. @context allows the HW vendor driver to store
2409 	 * extra information together with a GID entry. The HW vendor driver may
2410 	 * allocate memory to contain this information and store it in @context
2411 	 * when a new GID entry is written to. Params are consistent until the
2412 	 * next call of add_gid or delete_gid. The function should return 0 on
2413 	 * success or error otherwise. The function could be called
2414 	 * concurrently for different ports. This function is only called when
2415 	 * roce_gid_table is used.
2416 	 */
2417 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2418 	/**
2419 	 * When calling del_gid, the HW vendor's driver should delete the
2420 	 * gid of device @device at gid index gid_index of port port_num
2421 	 * available in @attr.
2422 	 * Upon the deletion of a GID entry, the HW vendor must free any
2423 	 * allocated memory. The caller will clear @context afterwards.
2424 	 * This function is only called when roce_gid_table is used.
2425 	 */
2426 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2427 	int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2428 			  u16 *pkey);
2429 	int (*alloc_ucontext)(struct ib_ucontext *context,
2430 			      struct ib_udata *udata);
2431 	void (*dealloc_ucontext)(struct ib_ucontext *context);
2432 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2433 	/**
2434 	 * This will be called once refcount of an entry in mmap_xa reaches
2435 	 * zero. The type of the memory that was mapped may differ between
2436 	 * entries and is opaque to the rdma_user_mmap interface.
2437 	 * Therefore needs to be implemented by the driver in mmap_free.
2438 	 */
2439 	void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2440 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2441 	int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2442 	int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2443 	int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2444 			 struct ib_udata *udata);
2445 	int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2446 			      struct ib_udata *udata);
2447 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2448 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2449 	int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2450 	int (*create_srq)(struct ib_srq *srq,
2451 			  struct ib_srq_init_attr *srq_init_attr,
2452 			  struct ib_udata *udata);
2453 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2454 			  enum ib_srq_attr_mask srq_attr_mask,
2455 			  struct ib_udata *udata);
2456 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2457 	int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2458 	int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2459 			 struct ib_udata *udata);
2460 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2461 			 int qp_attr_mask, struct ib_udata *udata);
2462 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2463 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2464 	int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2465 	int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2466 			 struct ib_udata *udata);
2467 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2468 	int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2469 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2470 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2471 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2472 				     u64 virt_addr, int mr_access_flags,
2473 				     struct ib_udata *udata);
2474 	struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2475 					    u64 length, u64 virt_addr, int fd,
2476 					    int mr_access_flags,
2477 					    struct ib_udata *udata);
2478 	struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2479 				       u64 length, u64 virt_addr,
2480 				       int mr_access_flags, struct ib_pd *pd,
2481 				       struct ib_udata *udata);
2482 	int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2483 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2484 				  u32 max_num_sg);
2485 	struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2486 					    u32 max_num_data_sg,
2487 					    u32 max_num_meta_sg);
2488 	int (*advise_mr)(struct ib_pd *pd,
2489 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2490 			 struct ib_sge *sg_list, u32 num_sge,
2491 			 struct uverbs_attr_bundle *attrs);
2492 
2493 	/*
2494 	 * Kernel users should universally support relaxed ordering (RO), as
2495 	 * they are designed to read data only after observing the CQE and use
2496 	 * the DMA API correctly.
2497 	 *
2498 	 * Some drivers implicitly enable RO if platform supports it.
2499 	 */
2500 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2501 			 unsigned int *sg_offset);
2502 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2503 			       struct ib_mr_status *mr_status);
2504 	int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2505 	int (*dealloc_mw)(struct ib_mw *mw);
2506 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2507 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2508 	int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2509 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2510 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2511 				       struct ib_flow_attr *flow_attr,
2512 				       struct ib_udata *udata);
2513 	int (*destroy_flow)(struct ib_flow *flow_id);
2514 	int (*destroy_flow_action)(struct ib_flow_action *action);
2515 	int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2516 				 int state);
2517 	int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2518 			     struct ifla_vf_info *ivf);
2519 	int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2520 			    struct ifla_vf_stats *stats);
2521 	int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2522 			    struct ifla_vf_guid *node_guid,
2523 			    struct ifla_vf_guid *port_guid);
2524 	int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2525 			   int type);
2526 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2527 				   struct ib_wq_init_attr *init_attr,
2528 				   struct ib_udata *udata);
2529 	int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2530 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2531 			 u32 wq_attr_mask, struct ib_udata *udata);
2532 	int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2533 				    struct ib_rwq_ind_table_init_attr *init_attr,
2534 				    struct ib_udata *udata);
2535 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2536 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2537 				  struct ib_ucontext *context,
2538 				  struct ib_dm_alloc_attr *attr,
2539 				  struct uverbs_attr_bundle *attrs);
2540 	int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2541 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2542 				   struct ib_dm_mr_attr *attr,
2543 				   struct uverbs_attr_bundle *attrs);
2544 	int (*create_counters)(struct ib_counters *counters,
2545 			       struct uverbs_attr_bundle *attrs);
2546 	int (*destroy_counters)(struct ib_counters *counters);
2547 	int (*read_counters)(struct ib_counters *counters,
2548 			     struct ib_counters_read_attr *counters_read_attr,
2549 			     struct uverbs_attr_bundle *attrs);
2550 	int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2551 			    int data_sg_nents, unsigned int *data_sg_offset,
2552 			    struct scatterlist *meta_sg, int meta_sg_nents,
2553 			    unsigned int *meta_sg_offset);
2554 
2555 	/**
2556 	 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2557 	 *   fill in the driver initialized data.  The struct is kfree()'ed by
2558 	 *   the sysfs core when the device is removed.  A lifespan of -1 in the
2559 	 *   return struct tells the core to set a default lifespan.
2560 	 */
2561 	struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2562 	struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2563 						     u32 port_num);
2564 	/**
2565 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2566 	 * @index - The index in the value array we wish to have updated, or
2567 	 *   num_counters if we want all stats updated
2568 	 * Return codes -
2569 	 *   < 0 - Error, no counters updated
2570 	 *   index - Updated the single counter pointed to by index
2571 	 *   num_counters - Updated all counters (will reset the timestamp
2572 	 *     and prevent further calls for lifespan milliseconds)
2573 	 * Drivers are allowed to update all counters in leiu of just the
2574 	 *   one given in index at their option
2575 	 */
2576 	int (*get_hw_stats)(struct ib_device *device,
2577 			    struct rdma_hw_stats *stats, u32 port, int index);
2578 
2579 	/**
2580 	 * modify_hw_stat - Modify the counter configuration
2581 	 * @enable: true/false when enable/disable a counter
2582 	 * Return codes - 0 on success or error code otherwise.
2583 	 */
2584 	int (*modify_hw_stat)(struct ib_device *device, u32 port,
2585 			      unsigned int counter_index, bool enable);
2586 	/**
2587 	 * Allows rdma drivers to add their own restrack attributes.
2588 	 */
2589 	int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2590 	int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2591 	int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2592 	int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2593 	int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2594 	int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2595 	int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2596 	int (*fill_res_srq_entry)(struct sk_buff *msg, struct ib_srq *ib_srq);
2597 	int (*fill_res_srq_entry_raw)(struct sk_buff *msg, struct ib_srq *ib_srq);
2598 
2599 	/* Device lifecycle callbacks */
2600 	/*
2601 	 * Called after the device becomes registered, before clients are
2602 	 * attached
2603 	 */
2604 	int (*enable_driver)(struct ib_device *dev);
2605 	/*
2606 	 * This is called as part of ib_dealloc_device().
2607 	 */
2608 	void (*dealloc_driver)(struct ib_device *dev);
2609 
2610 	/* iWarp CM callbacks */
2611 	void (*iw_add_ref)(struct ib_qp *qp);
2612 	void (*iw_rem_ref)(struct ib_qp *qp);
2613 	struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2614 	int (*iw_connect)(struct iw_cm_id *cm_id,
2615 			  struct iw_cm_conn_param *conn_param);
2616 	int (*iw_accept)(struct iw_cm_id *cm_id,
2617 			 struct iw_cm_conn_param *conn_param);
2618 	int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2619 			 u8 pdata_len);
2620 	int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2621 	int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2622 	/**
2623 	 * counter_bind_qp - Bind a QP to a counter.
2624 	 * @counter - The counter to be bound. If counter->id is zero then
2625 	 *   the driver needs to allocate a new counter and set counter->id
2626 	 */
2627 	int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2628 	/**
2629 	 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2630 	 *   counter and bind it onto the default one
2631 	 */
2632 	int (*counter_unbind_qp)(struct ib_qp *qp);
2633 	/**
2634 	 * counter_dealloc -De-allocate the hw counter
2635 	 */
2636 	int (*counter_dealloc)(struct rdma_counter *counter);
2637 	/**
2638 	 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2639 	 * the driver initialized data.
2640 	 */
2641 	struct rdma_hw_stats *(*counter_alloc_stats)(
2642 		struct rdma_counter *counter);
2643 	/**
2644 	 * counter_update_stats - Query the stats value of this counter
2645 	 */
2646 	int (*counter_update_stats)(struct rdma_counter *counter);
2647 
2648 	/**
2649 	 * Allows rdma drivers to add their own restrack attributes
2650 	 * dumped via 'rdma stat' iproute2 command.
2651 	 */
2652 	int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2653 
2654 	/* query driver for its ucontext properties */
2655 	int (*query_ucontext)(struct ib_ucontext *context,
2656 			      struct uverbs_attr_bundle *attrs);
2657 
2658 	/*
2659 	 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2660 	 * Everyone else relies on Linux memory management model.
2661 	 */
2662 	int (*get_numa_node)(struct ib_device *dev);
2663 
2664 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
2665 	DECLARE_RDMA_OBJ_SIZE(ib_counters);
2666 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
2667 	DECLARE_RDMA_OBJ_SIZE(ib_mw);
2668 	DECLARE_RDMA_OBJ_SIZE(ib_pd);
2669 	DECLARE_RDMA_OBJ_SIZE(ib_qp);
2670 	DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2671 	DECLARE_RDMA_OBJ_SIZE(ib_srq);
2672 	DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2673 	DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2674 };
2675 
2676 struct ib_core_device {
2677 	/* device must be the first element in structure until,
2678 	 * union of ib_core_device and device exists in ib_device.
2679 	 */
2680 	struct device dev;
2681 	possible_net_t rdma_net;
2682 	struct kobject *ports_kobj;
2683 	struct list_head port_list;
2684 	struct ib_device *owner; /* reach back to owner ib_device */
2685 };
2686 
2687 struct rdma_restrack_root;
2688 struct ib_device {
2689 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2690 	struct device                *dma_device;
2691 	struct ib_device_ops	     ops;
2692 	char                          name[IB_DEVICE_NAME_MAX];
2693 	struct rcu_head rcu_head;
2694 
2695 	struct list_head              event_handler_list;
2696 	/* Protects event_handler_list */
2697 	struct rw_semaphore event_handler_rwsem;
2698 
2699 	/* Protects QP's event_handler calls and open_qp list */
2700 	spinlock_t qp_open_list_lock;
2701 
2702 	struct rw_semaphore	      client_data_rwsem;
2703 	struct xarray                 client_data;
2704 	struct mutex                  unregistration_lock;
2705 
2706 	/* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2707 	rwlock_t cache_lock;
2708 	/**
2709 	 * port_data is indexed by port number
2710 	 */
2711 	struct ib_port_data *port_data;
2712 
2713 	int			      num_comp_vectors;
2714 
2715 	union {
2716 		struct device		dev;
2717 		struct ib_core_device	coredev;
2718 	};
2719 
2720 	/* First group is for device attributes,
2721 	 * Second group is for driver provided attributes (optional).
2722 	 * Third group is for the hw_stats
2723 	 * It is a NULL terminated array.
2724 	 */
2725 	const struct attribute_group	*groups[4];
2726 
2727 	u64			     uverbs_cmd_mask;
2728 
2729 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2730 	__be64			     node_guid;
2731 	u32			     local_dma_lkey;
2732 	u16                          is_switch:1;
2733 	/* Indicates kernel verbs support, should not be used in drivers */
2734 	u16                          kverbs_provider:1;
2735 	/* CQ adaptive moderation (RDMA DIM) */
2736 	u16                          use_cq_dim:1;
2737 	u8                           node_type;
2738 	u32			     phys_port_cnt;
2739 	struct ib_device_attr        attrs;
2740 	struct hw_stats_device_data *hw_stats_data;
2741 
2742 #ifdef CONFIG_CGROUP_RDMA
2743 	struct rdmacg_device         cg_device;
2744 #endif
2745 
2746 	u32                          index;
2747 
2748 	spinlock_t                   cq_pools_lock;
2749 	struct list_head             cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2750 
2751 	struct rdma_restrack_root *res;
2752 
2753 	const struct uapi_definition   *driver_def;
2754 
2755 	/*
2756 	 * Positive refcount indicates that the device is currently
2757 	 * registered and cannot be unregistered.
2758 	 */
2759 	refcount_t refcount;
2760 	struct completion unreg_completion;
2761 	struct work_struct unregistration_work;
2762 
2763 	const struct rdma_link_ops *link_ops;
2764 
2765 	/* Protects compat_devs xarray modifications */
2766 	struct mutex compat_devs_mutex;
2767 	/* Maintains compat devices for each net namespace */
2768 	struct xarray compat_devs;
2769 
2770 	/* Used by iWarp CM */
2771 	char iw_ifname[IFNAMSIZ];
2772 	u32 iw_driver_flags;
2773 	u32 lag_flags;
2774 };
2775 
2776 static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
2777 				    gfp_t gfp, bool is_numa_aware)
2778 {
2779 	if (is_numa_aware && dev->ops.get_numa_node)
2780 		return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
2781 
2782 	return kzalloc(size, gfp);
2783 }
2784 
2785 struct ib_client_nl_info;
2786 struct ib_client {
2787 	const char *name;
2788 	int (*add)(struct ib_device *ibdev);
2789 	void (*remove)(struct ib_device *, void *client_data);
2790 	void (*rename)(struct ib_device *dev, void *client_data);
2791 	int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2792 			   struct ib_client_nl_info *res);
2793 	int (*get_global_nl_info)(struct ib_client_nl_info *res);
2794 
2795 	/* Returns the net_dev belonging to this ib_client and matching the
2796 	 * given parameters.
2797 	 * @dev:	 An RDMA device that the net_dev use for communication.
2798 	 * @port:	 A physical port number on the RDMA device.
2799 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2800 	 * @gid:	 A GID that the net_dev uses to communicate.
2801 	 * @addr:	 An IP address the net_dev is configured with.
2802 	 * @client_data: The device's client data set by ib_set_client_data().
2803 	 *
2804 	 * An ib_client that implements a net_dev on top of RDMA devices
2805 	 * (such as IP over IB) should implement this callback, allowing the
2806 	 * rdma_cm module to find the right net_dev for a given request.
2807 	 *
2808 	 * The caller is responsible for calling dev_put on the returned
2809 	 * netdev. */
2810 	struct net_device *(*get_net_dev_by_params)(
2811 			struct ib_device *dev,
2812 			u32 port,
2813 			u16 pkey,
2814 			const union ib_gid *gid,
2815 			const struct sockaddr *addr,
2816 			void *client_data);
2817 
2818 	refcount_t uses;
2819 	struct completion uses_zero;
2820 	u32 client_id;
2821 
2822 	/* kverbs are not required by the client */
2823 	u8 no_kverbs_req:1;
2824 };
2825 
2826 /*
2827  * IB block DMA iterator
2828  *
2829  * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2830  * to a HW supported page size.
2831  */
2832 struct ib_block_iter {
2833 	/* internal states */
2834 	struct scatterlist *__sg;	/* sg holding the current aligned block */
2835 	dma_addr_t __dma_addr;		/* unaligned DMA address of this block */
2836 	size_t __sg_numblocks;		/* ib_umem_num_dma_blocks() */
2837 	unsigned int __sg_nents;	/* number of SG entries */
2838 	unsigned int __sg_advance;	/* number of bytes to advance in sg in next step */
2839 	unsigned int __pg_bit;		/* alignment of current block */
2840 };
2841 
2842 struct ib_device *_ib_alloc_device(size_t size);
2843 #define ib_alloc_device(drv_struct, member)                                    \
2844 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2845 				      BUILD_BUG_ON_ZERO(offsetof(              \
2846 					      struct drv_struct, member))),    \
2847 		     struct drv_struct, member)
2848 
2849 void ib_dealloc_device(struct ib_device *device);
2850 
2851 void ib_get_device_fw_str(struct ib_device *device, char *str);
2852 
2853 int ib_register_device(struct ib_device *device, const char *name,
2854 		       struct device *dma_device);
2855 void ib_unregister_device(struct ib_device *device);
2856 void ib_unregister_driver(enum rdma_driver_id driver_id);
2857 void ib_unregister_device_and_put(struct ib_device *device);
2858 void ib_unregister_device_queued(struct ib_device *ib_dev);
2859 
2860 int ib_register_client   (struct ib_client *client);
2861 void ib_unregister_client(struct ib_client *client);
2862 
2863 void __rdma_block_iter_start(struct ib_block_iter *biter,
2864 			     struct scatterlist *sglist,
2865 			     unsigned int nents,
2866 			     unsigned long pgsz);
2867 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2868 
2869 /**
2870  * rdma_block_iter_dma_address - get the aligned dma address of the current
2871  * block held by the block iterator.
2872  * @biter: block iterator holding the memory block
2873  */
2874 static inline dma_addr_t
2875 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2876 {
2877 	return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2878 }
2879 
2880 /**
2881  * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2882  * @sglist: sglist to iterate over
2883  * @biter: block iterator holding the memory block
2884  * @nents: maximum number of sg entries to iterate over
2885  * @pgsz: best HW supported page size to use
2886  *
2887  * Callers may use rdma_block_iter_dma_address() to get each
2888  * blocks aligned DMA address.
2889  */
2890 #define rdma_for_each_block(sglist, biter, nents, pgsz)		\
2891 	for (__rdma_block_iter_start(biter, sglist, nents,	\
2892 				     pgsz);			\
2893 	     __rdma_block_iter_next(biter);)
2894 
2895 /**
2896  * ib_get_client_data - Get IB client context
2897  * @device:Device to get context for
2898  * @client:Client to get context for
2899  *
2900  * ib_get_client_data() returns the client context data set with
2901  * ib_set_client_data(). This can only be called while the client is
2902  * registered to the device, once the ib_client remove() callback returns this
2903  * cannot be called.
2904  */
2905 static inline void *ib_get_client_data(struct ib_device *device,
2906 				       struct ib_client *client)
2907 {
2908 	return xa_load(&device->client_data, client->client_id);
2909 }
2910 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2911 			 void *data);
2912 void ib_set_device_ops(struct ib_device *device,
2913 		       const struct ib_device_ops *ops);
2914 
2915 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2916 		      unsigned long pfn, unsigned long size, pgprot_t prot,
2917 		      struct rdma_user_mmap_entry *entry);
2918 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2919 				struct rdma_user_mmap_entry *entry,
2920 				size_t length);
2921 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2922 				      struct rdma_user_mmap_entry *entry,
2923 				      size_t length, u32 min_pgoff,
2924 				      u32 max_pgoff);
2925 
2926 static inline int
2927 rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
2928 				  struct rdma_user_mmap_entry *entry,
2929 				  size_t length, u32 pgoff)
2930 {
2931 	return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff,
2932 						 pgoff);
2933 }
2934 
2935 struct rdma_user_mmap_entry *
2936 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2937 			       unsigned long pgoff);
2938 struct rdma_user_mmap_entry *
2939 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2940 			 struct vm_area_struct *vma);
2941 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2942 
2943 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2944 
2945 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2946 {
2947 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2948 }
2949 
2950 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2951 {
2952 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2953 }
2954 
2955 static inline bool ib_is_buffer_cleared(const void __user *p,
2956 					size_t len)
2957 {
2958 	bool ret;
2959 	u8 *buf;
2960 
2961 	if (len > USHRT_MAX)
2962 		return false;
2963 
2964 	buf = memdup_user(p, len);
2965 	if (IS_ERR(buf))
2966 		return false;
2967 
2968 	ret = !memchr_inv(buf, 0, len);
2969 	kfree(buf);
2970 	return ret;
2971 }
2972 
2973 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2974 				       size_t offset,
2975 				       size_t len)
2976 {
2977 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
2978 }
2979 
2980 /**
2981  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2982  * contains all required attributes and no attributes not allowed for
2983  * the given QP state transition.
2984  * @cur_state: Current QP state
2985  * @next_state: Next QP state
2986  * @type: QP type
2987  * @mask: Mask of supplied QP attributes
2988  *
2989  * This function is a helper function that a low-level driver's
2990  * modify_qp method can use to validate the consumer's input.  It
2991  * checks that cur_state and next_state are valid QP states, that a
2992  * transition from cur_state to next_state is allowed by the IB spec,
2993  * and that the attribute mask supplied is allowed for the transition.
2994  */
2995 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2996 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
2997 
2998 void ib_register_event_handler(struct ib_event_handler *event_handler);
2999 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
3000 void ib_dispatch_event(const struct ib_event *event);
3001 
3002 int ib_query_port(struct ib_device *device,
3003 		  u32 port_num, struct ib_port_attr *port_attr);
3004 
3005 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3006 					       u32 port_num);
3007 
3008 /**
3009  * rdma_cap_ib_switch - Check if the device is IB switch
3010  * @device: Device to check
3011  *
3012  * Device driver is responsible for setting is_switch bit on
3013  * in ib_device structure at init time.
3014  *
3015  * Return: true if the device is IB switch.
3016  */
3017 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3018 {
3019 	return device->is_switch;
3020 }
3021 
3022 /**
3023  * rdma_start_port - Return the first valid port number for the device
3024  * specified
3025  *
3026  * @device: Device to be checked
3027  *
3028  * Return start port number
3029  */
3030 static inline u32 rdma_start_port(const struct ib_device *device)
3031 {
3032 	return rdma_cap_ib_switch(device) ? 0 : 1;
3033 }
3034 
3035 /**
3036  * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3037  * @device - The struct ib_device * to iterate over
3038  * @iter - The unsigned int to store the port number
3039  */
3040 #define rdma_for_each_port(device, iter)                                       \
3041 	for (iter = rdma_start_port(device +				       \
3042 				    BUILD_BUG_ON_ZERO(!__same_type(u32,	       \
3043 								   iter)));    \
3044 	     iter <= rdma_end_port(device); iter++)
3045 
3046 /**
3047  * rdma_end_port - Return the last valid port number for the device
3048  * specified
3049  *
3050  * @device: Device to be checked
3051  *
3052  * Return last port number
3053  */
3054 static inline u32 rdma_end_port(const struct ib_device *device)
3055 {
3056 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3057 }
3058 
3059 static inline int rdma_is_port_valid(const struct ib_device *device,
3060 				     unsigned int port)
3061 {
3062 	return (port >= rdma_start_port(device) &&
3063 		port <= rdma_end_port(device));
3064 }
3065 
3066 static inline bool rdma_is_grh_required(const struct ib_device *device,
3067 					u32 port_num)
3068 {
3069 	return device->port_data[port_num].immutable.core_cap_flags &
3070 	       RDMA_CORE_PORT_IB_GRH_REQUIRED;
3071 }
3072 
3073 static inline bool rdma_protocol_ib(const struct ib_device *device,
3074 				    u32 port_num)
3075 {
3076 	return device->port_data[port_num].immutable.core_cap_flags &
3077 	       RDMA_CORE_CAP_PROT_IB;
3078 }
3079 
3080 static inline bool rdma_protocol_roce(const struct ib_device *device,
3081 				      u32 port_num)
3082 {
3083 	return device->port_data[port_num].immutable.core_cap_flags &
3084 	       (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3085 }
3086 
3087 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3088 						u32 port_num)
3089 {
3090 	return device->port_data[port_num].immutable.core_cap_flags &
3091 	       RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3092 }
3093 
3094 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3095 						u32 port_num)
3096 {
3097 	return device->port_data[port_num].immutable.core_cap_flags &
3098 	       RDMA_CORE_CAP_PROT_ROCE;
3099 }
3100 
3101 static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3102 				       u32 port_num)
3103 {
3104 	return device->port_data[port_num].immutable.core_cap_flags &
3105 	       RDMA_CORE_CAP_PROT_IWARP;
3106 }
3107 
3108 static inline bool rdma_ib_or_roce(const struct ib_device *device,
3109 				   u32 port_num)
3110 {
3111 	return rdma_protocol_ib(device, port_num) ||
3112 		rdma_protocol_roce(device, port_num);
3113 }
3114 
3115 static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3116 					    u32 port_num)
3117 {
3118 	return device->port_data[port_num].immutable.core_cap_flags &
3119 	       RDMA_CORE_CAP_PROT_RAW_PACKET;
3120 }
3121 
3122 static inline bool rdma_protocol_usnic(const struct ib_device *device,
3123 				       u32 port_num)
3124 {
3125 	return device->port_data[port_num].immutable.core_cap_flags &
3126 	       RDMA_CORE_CAP_PROT_USNIC;
3127 }
3128 
3129 /**
3130  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3131  * Management Datagrams.
3132  * @device: Device to check
3133  * @port_num: Port number to check
3134  *
3135  * Management Datagrams (MAD) are a required part of the InfiniBand
3136  * specification and are supported on all InfiniBand devices.  A slightly
3137  * extended version are also supported on OPA interfaces.
3138  *
3139  * Return: true if the port supports sending/receiving of MAD packets.
3140  */
3141 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3142 {
3143 	return device->port_data[port_num].immutable.core_cap_flags &
3144 	       RDMA_CORE_CAP_IB_MAD;
3145 }
3146 
3147 /**
3148  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3149  * Management Datagrams.
3150  * @device: Device to check
3151  * @port_num: Port number to check
3152  *
3153  * Intel OmniPath devices extend and/or replace the InfiniBand Management
3154  * datagrams with their own versions.  These OPA MADs share many but not all of
3155  * the characteristics of InfiniBand MADs.
3156  *
3157  * OPA MADs differ in the following ways:
3158  *
3159  *    1) MADs are variable size up to 2K
3160  *       IBTA defined MADs remain fixed at 256 bytes
3161  *    2) OPA SMPs must carry valid PKeys
3162  *    3) OPA SMP packets are a different format
3163  *
3164  * Return: true if the port supports OPA MAD packet formats.
3165  */
3166 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3167 {
3168 	return device->port_data[port_num].immutable.core_cap_flags &
3169 		RDMA_CORE_CAP_OPA_MAD;
3170 }
3171 
3172 /**
3173  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3174  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3175  * @device: Device to check
3176  * @port_num: Port number to check
3177  *
3178  * Each InfiniBand node is required to provide a Subnet Management Agent
3179  * that the subnet manager can access.  Prior to the fabric being fully
3180  * configured by the subnet manager, the SMA is accessed via a well known
3181  * interface called the Subnet Management Interface (SMI).  This interface
3182  * uses directed route packets to communicate with the SM to get around the
3183  * chicken and egg problem of the SM needing to know what's on the fabric
3184  * in order to configure the fabric, and needing to configure the fabric in
3185  * order to send packets to the devices on the fabric.  These directed
3186  * route packets do not need the fabric fully configured in order to reach
3187  * their destination.  The SMI is the only method allowed to send
3188  * directed route packets on an InfiniBand fabric.
3189  *
3190  * Return: true if the port provides an SMI.
3191  */
3192 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3193 {
3194 	return device->port_data[port_num].immutable.core_cap_flags &
3195 	       RDMA_CORE_CAP_IB_SMI;
3196 }
3197 
3198 /**
3199  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3200  * Communication Manager.
3201  * @device: Device to check
3202  * @port_num: Port number to check
3203  *
3204  * The InfiniBand Communication Manager is one of many pre-defined General
3205  * Service Agents (GSA) that are accessed via the General Service
3206  * Interface (GSI).  It's role is to facilitate establishment of connections
3207  * between nodes as well as other management related tasks for established
3208  * connections.
3209  *
3210  * Return: true if the port supports an IB CM (this does not guarantee that
3211  * a CM is actually running however).
3212  */
3213 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3214 {
3215 	return device->port_data[port_num].immutable.core_cap_flags &
3216 	       RDMA_CORE_CAP_IB_CM;
3217 }
3218 
3219 /**
3220  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3221  * Communication Manager.
3222  * @device: Device to check
3223  * @port_num: Port number to check
3224  *
3225  * Similar to above, but specific to iWARP connections which have a different
3226  * managment protocol than InfiniBand.
3227  *
3228  * Return: true if the port supports an iWARP CM (this does not guarantee that
3229  * a CM is actually running however).
3230  */
3231 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3232 {
3233 	return device->port_data[port_num].immutable.core_cap_flags &
3234 	       RDMA_CORE_CAP_IW_CM;
3235 }
3236 
3237 /**
3238  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3239  * Subnet Administration.
3240  * @device: Device to check
3241  * @port_num: Port number to check
3242  *
3243  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3244  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
3245  * fabrics, devices should resolve routes to other hosts by contacting the
3246  * SA to query the proper route.
3247  *
3248  * Return: true if the port should act as a client to the fabric Subnet
3249  * Administration interface.  This does not imply that the SA service is
3250  * running locally.
3251  */
3252 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3253 {
3254 	return device->port_data[port_num].immutable.core_cap_flags &
3255 	       RDMA_CORE_CAP_IB_SA;
3256 }
3257 
3258 /**
3259  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3260  * Multicast.
3261  * @device: Device to check
3262  * @port_num: Port number to check
3263  *
3264  * InfiniBand multicast registration is more complex than normal IPv4 or
3265  * IPv6 multicast registration.  Each Host Channel Adapter must register
3266  * with the Subnet Manager when it wishes to join a multicast group.  It
3267  * should do so only once regardless of how many queue pairs it subscribes
3268  * to this group.  And it should leave the group only after all queue pairs
3269  * attached to the group have been detached.
3270  *
3271  * Return: true if the port must undertake the additional adminstrative
3272  * overhead of registering/unregistering with the SM and tracking of the
3273  * total number of queue pairs attached to the multicast group.
3274  */
3275 static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3276 				     u32 port_num)
3277 {
3278 	return rdma_cap_ib_sa(device, port_num);
3279 }
3280 
3281 /**
3282  * rdma_cap_af_ib - Check if the port of device has the capability
3283  * Native Infiniband Address.
3284  * @device: Device to check
3285  * @port_num: Port number to check
3286  *
3287  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3288  * GID.  RoCE uses a different mechanism, but still generates a GID via
3289  * a prescribed mechanism and port specific data.
3290  *
3291  * Return: true if the port uses a GID address to identify devices on the
3292  * network.
3293  */
3294 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3295 {
3296 	return device->port_data[port_num].immutable.core_cap_flags &
3297 	       RDMA_CORE_CAP_AF_IB;
3298 }
3299 
3300 /**
3301  * rdma_cap_eth_ah - Check if the port of device has the capability
3302  * Ethernet Address Handle.
3303  * @device: Device to check
3304  * @port_num: Port number to check
3305  *
3306  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3307  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3308  * port.  Normally, packet headers are generated by the sending host
3309  * adapter, but when sending connectionless datagrams, we must manually
3310  * inject the proper headers for the fabric we are communicating over.
3311  *
3312  * Return: true if we are running as a RoCE port and must force the
3313  * addition of a Global Route Header built from our Ethernet Address
3314  * Handle into our header list for connectionless packets.
3315  */
3316 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3317 {
3318 	return device->port_data[port_num].immutable.core_cap_flags &
3319 	       RDMA_CORE_CAP_ETH_AH;
3320 }
3321 
3322 /**
3323  * rdma_cap_opa_ah - Check if the port of device supports
3324  * OPA Address handles
3325  * @device: Device to check
3326  * @port_num: Port number to check
3327  *
3328  * Return: true if we are running on an OPA device which supports
3329  * the extended OPA addressing.
3330  */
3331 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3332 {
3333 	return (device->port_data[port_num].immutable.core_cap_flags &
3334 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3335 }
3336 
3337 /**
3338  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3339  *
3340  * @device: Device
3341  * @port_num: Port number
3342  *
3343  * This MAD size includes the MAD headers and MAD payload.  No other headers
3344  * are included.
3345  *
3346  * Return the max MAD size required by the Port.  Will return 0 if the port
3347  * does not support MADs
3348  */
3349 static inline size_t rdma_max_mad_size(const struct ib_device *device,
3350 				       u32 port_num)
3351 {
3352 	return device->port_data[port_num].immutable.max_mad_size;
3353 }
3354 
3355 /**
3356  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3357  * @device: Device to check
3358  * @port_num: Port number to check
3359  *
3360  * RoCE GID table mechanism manages the various GIDs for a device.
3361  *
3362  * NOTE: if allocating the port's GID table has failed, this call will still
3363  * return true, but any RoCE GID table API will fail.
3364  *
3365  * Return: true if the port uses RoCE GID table mechanism in order to manage
3366  * its GIDs.
3367  */
3368 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3369 					   u32 port_num)
3370 {
3371 	return rdma_protocol_roce(device, port_num) &&
3372 		device->ops.add_gid && device->ops.del_gid;
3373 }
3374 
3375 /*
3376  * Check if the device supports READ W/ INVALIDATE.
3377  */
3378 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3379 {
3380 	/*
3381 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3382 	 * has support for it yet.
3383 	 */
3384 	return rdma_protocol_iwarp(dev, port_num);
3385 }
3386 
3387 /**
3388  * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3389  * @device: Device
3390  * @port_num: 1 based Port number
3391  *
3392  * Return true if port is an Intel OPA port , false if not
3393  */
3394 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3395 					  u32 port_num)
3396 {
3397 	return (device->port_data[port_num].immutable.core_cap_flags &
3398 		RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3399 }
3400 
3401 /**
3402  * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3403  * @device: Device
3404  * @port_num: Port number
3405  * @mtu: enum value of MTU
3406  *
3407  * Return the MTU size supported by the port as an integer value. Will return
3408  * -1 if enum value of mtu is not supported.
3409  */
3410 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3411 				       int mtu)
3412 {
3413 	if (rdma_core_cap_opa_port(device, port))
3414 		return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3415 	else
3416 		return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3417 }
3418 
3419 /**
3420  * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3421  * @device: Device
3422  * @port_num: Port number
3423  * @attr: port attribute
3424  *
3425  * Return the MTU size supported by the port as an integer value.
3426  */
3427 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3428 				     struct ib_port_attr *attr)
3429 {
3430 	if (rdma_core_cap_opa_port(device, port))
3431 		return attr->phys_mtu;
3432 	else
3433 		return ib_mtu_enum_to_int(attr->max_mtu);
3434 }
3435 
3436 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3437 			 int state);
3438 int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3439 		     struct ifla_vf_info *info);
3440 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3441 		    struct ifla_vf_stats *stats);
3442 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3443 		    struct ifla_vf_guid *node_guid,
3444 		    struct ifla_vf_guid *port_guid);
3445 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3446 		   int type);
3447 
3448 int ib_query_pkey(struct ib_device *device,
3449 		  u32 port_num, u16 index, u16 *pkey);
3450 
3451 int ib_modify_device(struct ib_device *device,
3452 		     int device_modify_mask,
3453 		     struct ib_device_modify *device_modify);
3454 
3455 int ib_modify_port(struct ib_device *device,
3456 		   u32 port_num, int port_modify_mask,
3457 		   struct ib_port_modify *port_modify);
3458 
3459 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3460 		u32 *port_num, u16 *index);
3461 
3462 int ib_find_pkey(struct ib_device *device,
3463 		 u32 port_num, u16 pkey, u16 *index);
3464 
3465 enum ib_pd_flags {
3466 	/*
3467 	 * Create a memory registration for all memory in the system and place
3468 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3469 	 * ULPs to avoid the overhead of dynamic MRs.
3470 	 *
3471 	 * This flag is generally considered unsafe and must only be used in
3472 	 * extremly trusted environments.  Every use of it will log a warning
3473 	 * in the kernel log.
3474 	 */
3475 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3476 };
3477 
3478 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3479 		const char *caller);
3480 
3481 /**
3482  * ib_alloc_pd - Allocates an unused protection domain.
3483  * @device: The device on which to allocate the protection domain.
3484  * @flags: protection domain flags
3485  *
3486  * A protection domain object provides an association between QPs, shared
3487  * receive queues, address handles, memory regions, and memory windows.
3488  *
3489  * Every PD has a local_dma_lkey which can be used as the lkey value for local
3490  * memory operations.
3491  */
3492 #define ib_alloc_pd(device, flags) \
3493 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3494 
3495 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3496 
3497 /**
3498  * ib_dealloc_pd - Deallocate kernel PD
3499  * @pd: The protection domain
3500  *
3501  * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3502  */
3503 static inline void ib_dealloc_pd(struct ib_pd *pd)
3504 {
3505 	int ret = ib_dealloc_pd_user(pd, NULL);
3506 
3507 	WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3508 }
3509 
3510 enum rdma_create_ah_flags {
3511 	/* In a sleepable context */
3512 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3513 };
3514 
3515 /**
3516  * rdma_create_ah - Creates an address handle for the given address vector.
3517  * @pd: The protection domain associated with the address handle.
3518  * @ah_attr: The attributes of the address vector.
3519  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3520  *
3521  * The address handle is used to reference a local or global destination
3522  * in all UD QP post sends.
3523  */
3524 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3525 			     u32 flags);
3526 
3527 /**
3528  * rdma_create_user_ah - Creates an address handle for the given address vector.
3529  * It resolves destination mac address for ah attribute of RoCE type.
3530  * @pd: The protection domain associated with the address handle.
3531  * @ah_attr: The attributes of the address vector.
3532  * @udata: pointer to user's input output buffer information need by
3533  *         provider driver.
3534  *
3535  * It returns 0 on success and returns appropriate error code on error.
3536  * The address handle is used to reference a local or global destination
3537  * in all UD QP post sends.
3538  */
3539 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3540 				  struct rdma_ah_attr *ah_attr,
3541 				  struct ib_udata *udata);
3542 /**
3543  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3544  *   work completion.
3545  * @hdr: the L3 header to parse
3546  * @net_type: type of header to parse
3547  * @sgid: place to store source gid
3548  * @dgid: place to store destination gid
3549  */
3550 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3551 			      enum rdma_network_type net_type,
3552 			      union ib_gid *sgid, union ib_gid *dgid);
3553 
3554 /**
3555  * ib_get_rdma_header_version - Get the header version
3556  * @hdr: the L3 header to parse
3557  */
3558 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3559 
3560 /**
3561  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3562  *   work completion.
3563  * @device: Device on which the received message arrived.
3564  * @port_num: Port on which the received message arrived.
3565  * @wc: Work completion associated with the received message.
3566  * @grh: References the received global route header.  This parameter is
3567  *   ignored unless the work completion indicates that the GRH is valid.
3568  * @ah_attr: Returned attributes that can be used when creating an address
3569  *   handle for replying to the message.
3570  * When ib_init_ah_attr_from_wc() returns success,
3571  * (a) for IB link layer it optionally contains a reference to SGID attribute
3572  * when GRH is present for IB link layer.
3573  * (b) for RoCE link layer it contains a reference to SGID attribute.
3574  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3575  * attributes which are initialized using ib_init_ah_attr_from_wc().
3576  *
3577  */
3578 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3579 			    const struct ib_wc *wc, const struct ib_grh *grh,
3580 			    struct rdma_ah_attr *ah_attr);
3581 
3582 /**
3583  * ib_create_ah_from_wc - Creates an address handle associated with the
3584  *   sender of the specified work completion.
3585  * @pd: The protection domain associated with the address handle.
3586  * @wc: Work completion information associated with a received message.
3587  * @grh: References the received global route header.  This parameter is
3588  *   ignored unless the work completion indicates that the GRH is valid.
3589  * @port_num: The outbound port number to associate with the address.
3590  *
3591  * The address handle is used to reference a local or global destination
3592  * in all UD QP post sends.
3593  */
3594 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3595 				   const struct ib_grh *grh, u32 port_num);
3596 
3597 /**
3598  * rdma_modify_ah - Modifies the address vector associated with an address
3599  *   handle.
3600  * @ah: The address handle to modify.
3601  * @ah_attr: The new address vector attributes to associate with the
3602  *   address handle.
3603  */
3604 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3605 
3606 /**
3607  * rdma_query_ah - Queries the address vector associated with an address
3608  *   handle.
3609  * @ah: The address handle to query.
3610  * @ah_attr: The address vector attributes associated with the address
3611  *   handle.
3612  */
3613 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3614 
3615 enum rdma_destroy_ah_flags {
3616 	/* In a sleepable context */
3617 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3618 };
3619 
3620 /**
3621  * rdma_destroy_ah_user - Destroys an address handle.
3622  * @ah: The address handle to destroy.
3623  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3624  * @udata: Valid user data or NULL for kernel objects
3625  */
3626 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3627 
3628 /**
3629  * rdma_destroy_ah - Destroys an kernel address handle.
3630  * @ah: The address handle to destroy.
3631  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3632  *
3633  * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3634  */
3635 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3636 {
3637 	int ret = rdma_destroy_ah_user(ah, flags, NULL);
3638 
3639 	WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3640 }
3641 
3642 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3643 				  struct ib_srq_init_attr *srq_init_attr,
3644 				  struct ib_usrq_object *uobject,
3645 				  struct ib_udata *udata);
3646 static inline struct ib_srq *
3647 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3648 {
3649 	if (!pd->device->ops.create_srq)
3650 		return ERR_PTR(-EOPNOTSUPP);
3651 
3652 	return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3653 }
3654 
3655 /**
3656  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3657  * @srq: The SRQ to modify.
3658  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3659  *   the current values of selected SRQ attributes are returned.
3660  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3661  *   are being modified.
3662  *
3663  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3664  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3665  * the number of receives queued drops below the limit.
3666  */
3667 int ib_modify_srq(struct ib_srq *srq,
3668 		  struct ib_srq_attr *srq_attr,
3669 		  enum ib_srq_attr_mask srq_attr_mask);
3670 
3671 /**
3672  * ib_query_srq - Returns the attribute list and current values for the
3673  *   specified SRQ.
3674  * @srq: The SRQ to query.
3675  * @srq_attr: The attributes of the specified SRQ.
3676  */
3677 int ib_query_srq(struct ib_srq *srq,
3678 		 struct ib_srq_attr *srq_attr);
3679 
3680 /**
3681  * ib_destroy_srq_user - Destroys the specified SRQ.
3682  * @srq: The SRQ to destroy.
3683  * @udata: Valid user data or NULL for kernel objects
3684  */
3685 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3686 
3687 /**
3688  * ib_destroy_srq - Destroys the specified kernel SRQ.
3689  * @srq: The SRQ to destroy.
3690  *
3691  * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3692  */
3693 static inline void ib_destroy_srq(struct ib_srq *srq)
3694 {
3695 	int ret = ib_destroy_srq_user(srq, NULL);
3696 
3697 	WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3698 }
3699 
3700 /**
3701  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3702  * @srq: The SRQ to post the work request on.
3703  * @recv_wr: A list of work requests to post on the receive queue.
3704  * @bad_recv_wr: On an immediate failure, this parameter will reference
3705  *   the work request that failed to be posted on the QP.
3706  */
3707 static inline int ib_post_srq_recv(struct ib_srq *srq,
3708 				   const struct ib_recv_wr *recv_wr,
3709 				   const struct ib_recv_wr **bad_recv_wr)
3710 {
3711 	const struct ib_recv_wr *dummy;
3712 
3713 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3714 					      bad_recv_wr ? : &dummy);
3715 }
3716 
3717 struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3718 				  struct ib_qp_init_attr *qp_init_attr,
3719 				  const char *caller);
3720 /**
3721  * ib_create_qp - Creates a kernel QP associated with the specific protection
3722  * domain.
3723  * @pd: The protection domain associated with the QP.
3724  * @init_attr: A list of initial attributes required to create the
3725  *   QP.  If QP creation succeeds, then the attributes are updated to
3726  *   the actual capabilities of the created QP.
3727  */
3728 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3729 					 struct ib_qp_init_attr *init_attr)
3730 {
3731 	return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
3732 }
3733 
3734 /**
3735  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3736  * @qp: The QP to modify.
3737  * @attr: On input, specifies the QP attributes to modify.  On output,
3738  *   the current values of selected QP attributes are returned.
3739  * @attr_mask: A bit-mask used to specify which attributes of the QP
3740  *   are being modified.
3741  * @udata: pointer to user's input output buffer information
3742  *   are being modified.
3743  * It returns 0 on success and returns appropriate error code on error.
3744  */
3745 int ib_modify_qp_with_udata(struct ib_qp *qp,
3746 			    struct ib_qp_attr *attr,
3747 			    int attr_mask,
3748 			    struct ib_udata *udata);
3749 
3750 /**
3751  * ib_modify_qp - Modifies the attributes for the specified QP and then
3752  *   transitions the QP to the given state.
3753  * @qp: The QP to modify.
3754  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3755  *   the current values of selected QP attributes are returned.
3756  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3757  *   are being modified.
3758  */
3759 int ib_modify_qp(struct ib_qp *qp,
3760 		 struct ib_qp_attr *qp_attr,
3761 		 int qp_attr_mask);
3762 
3763 /**
3764  * ib_query_qp - Returns the attribute list and current values for the
3765  *   specified QP.
3766  * @qp: The QP to query.
3767  * @qp_attr: The attributes of the specified QP.
3768  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3769  * @qp_init_attr: Additional attributes of the selected QP.
3770  *
3771  * The qp_attr_mask may be used to limit the query to gathering only the
3772  * selected attributes.
3773  */
3774 int ib_query_qp(struct ib_qp *qp,
3775 		struct ib_qp_attr *qp_attr,
3776 		int qp_attr_mask,
3777 		struct ib_qp_init_attr *qp_init_attr);
3778 
3779 /**
3780  * ib_destroy_qp - Destroys the specified QP.
3781  * @qp: The QP to destroy.
3782  * @udata: Valid udata or NULL for kernel objects
3783  */
3784 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3785 
3786 /**
3787  * ib_destroy_qp - Destroys the specified kernel QP.
3788  * @qp: The QP to destroy.
3789  *
3790  * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3791  */
3792 static inline int ib_destroy_qp(struct ib_qp *qp)
3793 {
3794 	return ib_destroy_qp_user(qp, NULL);
3795 }
3796 
3797 /**
3798  * ib_open_qp - Obtain a reference to an existing sharable QP.
3799  * @xrcd - XRC domain
3800  * @qp_open_attr: Attributes identifying the QP to open.
3801  *
3802  * Returns a reference to a sharable QP.
3803  */
3804 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3805 			 struct ib_qp_open_attr *qp_open_attr);
3806 
3807 /**
3808  * ib_close_qp - Release an external reference to a QP.
3809  * @qp: The QP handle to release
3810  *
3811  * The opened QP handle is released by the caller.  The underlying
3812  * shared QP is not destroyed until all internal references are released.
3813  */
3814 int ib_close_qp(struct ib_qp *qp);
3815 
3816 /**
3817  * ib_post_send - Posts a list of work requests to the send queue of
3818  *   the specified QP.
3819  * @qp: The QP to post the work request on.
3820  * @send_wr: A list of work requests to post on the send queue.
3821  * @bad_send_wr: On an immediate failure, this parameter will reference
3822  *   the work request that failed to be posted on the QP.
3823  *
3824  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3825  * error is returned, the QP state shall not be affected,
3826  * ib_post_send() will return an immediate error after queueing any
3827  * earlier work requests in the list.
3828  */
3829 static inline int ib_post_send(struct ib_qp *qp,
3830 			       const struct ib_send_wr *send_wr,
3831 			       const struct ib_send_wr **bad_send_wr)
3832 {
3833 	const struct ib_send_wr *dummy;
3834 
3835 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3836 }
3837 
3838 /**
3839  * ib_post_recv - Posts a list of work requests to the receive queue of
3840  *   the specified QP.
3841  * @qp: The QP to post the work request on.
3842  * @recv_wr: A list of work requests to post on the receive queue.
3843  * @bad_recv_wr: On an immediate failure, this parameter will reference
3844  *   the work request that failed to be posted on the QP.
3845  */
3846 static inline int ib_post_recv(struct ib_qp *qp,
3847 			       const struct ib_recv_wr *recv_wr,
3848 			       const struct ib_recv_wr **bad_recv_wr)
3849 {
3850 	const struct ib_recv_wr *dummy;
3851 
3852 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3853 }
3854 
3855 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3856 			    int comp_vector, enum ib_poll_context poll_ctx,
3857 			    const char *caller);
3858 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3859 					int nr_cqe, int comp_vector,
3860 					enum ib_poll_context poll_ctx)
3861 {
3862 	return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3863 			     KBUILD_MODNAME);
3864 }
3865 
3866 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3867 				int nr_cqe, enum ib_poll_context poll_ctx,
3868 				const char *caller);
3869 
3870 /**
3871  * ib_alloc_cq_any: Allocate kernel CQ
3872  * @dev: The IB device
3873  * @private: Private data attached to the CQE
3874  * @nr_cqe: Number of CQEs in the CQ
3875  * @poll_ctx: Context used for polling the CQ
3876  */
3877 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3878 					    void *private, int nr_cqe,
3879 					    enum ib_poll_context poll_ctx)
3880 {
3881 	return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3882 				 KBUILD_MODNAME);
3883 }
3884 
3885 void ib_free_cq(struct ib_cq *cq);
3886 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3887 
3888 /**
3889  * ib_create_cq - Creates a CQ on the specified device.
3890  * @device: The device on which to create the CQ.
3891  * @comp_handler: A user-specified callback that is invoked when a
3892  *   completion event occurs on the CQ.
3893  * @event_handler: A user-specified callback that is invoked when an
3894  *   asynchronous event not associated with a completion occurs on the CQ.
3895  * @cq_context: Context associated with the CQ returned to the user via
3896  *   the associated completion and event handlers.
3897  * @cq_attr: The attributes the CQ should be created upon.
3898  *
3899  * Users can examine the cq structure to determine the actual CQ size.
3900  */
3901 struct ib_cq *__ib_create_cq(struct ib_device *device,
3902 			     ib_comp_handler comp_handler,
3903 			     void (*event_handler)(struct ib_event *, void *),
3904 			     void *cq_context,
3905 			     const struct ib_cq_init_attr *cq_attr,
3906 			     const char *caller);
3907 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3908 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3909 
3910 /**
3911  * ib_resize_cq - Modifies the capacity of the CQ.
3912  * @cq: The CQ to resize.
3913  * @cqe: The minimum size of the CQ.
3914  *
3915  * Users can examine the cq structure to determine the actual CQ size.
3916  */
3917 int ib_resize_cq(struct ib_cq *cq, int cqe);
3918 
3919 /**
3920  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3921  * @cq: The CQ to modify.
3922  * @cq_count: number of CQEs that will trigger an event
3923  * @cq_period: max period of time in usec before triggering an event
3924  *
3925  */
3926 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3927 
3928 /**
3929  * ib_destroy_cq_user - Destroys the specified CQ.
3930  * @cq: The CQ to destroy.
3931  * @udata: Valid user data or NULL for kernel objects
3932  */
3933 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3934 
3935 /**
3936  * ib_destroy_cq - Destroys the specified kernel CQ.
3937  * @cq: The CQ to destroy.
3938  *
3939  * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3940  */
3941 static inline void ib_destroy_cq(struct ib_cq *cq)
3942 {
3943 	int ret = ib_destroy_cq_user(cq, NULL);
3944 
3945 	WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3946 }
3947 
3948 /**
3949  * ib_poll_cq - poll a CQ for completion(s)
3950  * @cq:the CQ being polled
3951  * @num_entries:maximum number of completions to return
3952  * @wc:array of at least @num_entries &struct ib_wc where completions
3953  *   will be returned
3954  *
3955  * Poll a CQ for (possibly multiple) completions.  If the return value
3956  * is < 0, an error occurred.  If the return value is >= 0, it is the
3957  * number of completions returned.  If the return value is
3958  * non-negative and < num_entries, then the CQ was emptied.
3959  */
3960 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3961 			     struct ib_wc *wc)
3962 {
3963 	return cq->device->ops.poll_cq(cq, num_entries, wc);
3964 }
3965 
3966 /**
3967  * ib_req_notify_cq - Request completion notification on a CQ.
3968  * @cq: The CQ to generate an event for.
3969  * @flags:
3970  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3971  *   to request an event on the next solicited event or next work
3972  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3973  *   may also be |ed in to request a hint about missed events, as
3974  *   described below.
3975  *
3976  * Return Value:
3977  *    < 0 means an error occurred while requesting notification
3978  *   == 0 means notification was requested successfully, and if
3979  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3980  *        were missed and it is safe to wait for another event.  In
3981  *        this case is it guaranteed that any work completions added
3982  *        to the CQ since the last CQ poll will trigger a completion
3983  *        notification event.
3984  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3985  *        in.  It means that the consumer must poll the CQ again to
3986  *        make sure it is empty to avoid missing an event because of a
3987  *        race between requesting notification and an entry being
3988  *        added to the CQ.  This return value means it is possible
3989  *        (but not guaranteed) that a work completion has been added
3990  *        to the CQ since the last poll without triggering a
3991  *        completion notification event.
3992  */
3993 static inline int ib_req_notify_cq(struct ib_cq *cq,
3994 				   enum ib_cq_notify_flags flags)
3995 {
3996 	return cq->device->ops.req_notify_cq(cq, flags);
3997 }
3998 
3999 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4000 			     int comp_vector_hint,
4001 			     enum ib_poll_context poll_ctx);
4002 
4003 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4004 
4005 /*
4006  * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
4007  * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
4008  * address into the dma address.
4009  */
4010 static inline bool ib_uses_virt_dma(struct ib_device *dev)
4011 {
4012 	return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
4013 }
4014 
4015 /*
4016  * Check if a IB device's underlying DMA mapping supports P2PDMA transfers.
4017  */
4018 static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev)
4019 {
4020 	if (ib_uses_virt_dma(dev))
4021 		return false;
4022 
4023 	return dma_pci_p2pdma_supported(dev->dma_device);
4024 }
4025 
4026 /**
4027  * ib_virt_dma_to_ptr - Convert a dma_addr to a kernel pointer
4028  * @dma_addr: The DMA address
4029  *
4030  * Used by ib_uses_virt_dma() devices to get back to the kernel pointer after
4031  * going through the dma_addr marshalling.
4032  */
4033 static inline void *ib_virt_dma_to_ptr(u64 dma_addr)
4034 {
4035 	/* virt_dma mode maps the kvs's directly into the dma addr */
4036 	return (void *)(uintptr_t)dma_addr;
4037 }
4038 
4039 /**
4040  * ib_virt_dma_to_page - Convert a dma_addr to a struct page
4041  * @dma_addr: The DMA address
4042  *
4043  * Used by ib_uses_virt_dma() device to get back to the struct page after going
4044  * through the dma_addr marshalling.
4045  */
4046 static inline struct page *ib_virt_dma_to_page(u64 dma_addr)
4047 {
4048 	return virt_to_page(ib_virt_dma_to_ptr(dma_addr));
4049 }
4050 
4051 /**
4052  * ib_dma_mapping_error - check a DMA addr for error
4053  * @dev: The device for which the dma_addr was created
4054  * @dma_addr: The DMA address to check
4055  */
4056 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4057 {
4058 	if (ib_uses_virt_dma(dev))
4059 		return 0;
4060 	return dma_mapping_error(dev->dma_device, dma_addr);
4061 }
4062 
4063 /**
4064  * ib_dma_map_single - Map a kernel virtual address to DMA address
4065  * @dev: The device for which the dma_addr is to be created
4066  * @cpu_addr: The kernel virtual address
4067  * @size: The size of the region in bytes
4068  * @direction: The direction of the DMA
4069  */
4070 static inline u64 ib_dma_map_single(struct ib_device *dev,
4071 				    void *cpu_addr, size_t size,
4072 				    enum dma_data_direction direction)
4073 {
4074 	if (ib_uses_virt_dma(dev))
4075 		return (uintptr_t)cpu_addr;
4076 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4077 }
4078 
4079 /**
4080  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4081  * @dev: The device for which the DMA address was created
4082  * @addr: The DMA address
4083  * @size: The size of the region in bytes
4084  * @direction: The direction of the DMA
4085  */
4086 static inline void ib_dma_unmap_single(struct ib_device *dev,
4087 				       u64 addr, size_t size,
4088 				       enum dma_data_direction direction)
4089 {
4090 	if (!ib_uses_virt_dma(dev))
4091 		dma_unmap_single(dev->dma_device, addr, size, direction);
4092 }
4093 
4094 /**
4095  * ib_dma_map_page - Map a physical page to DMA address
4096  * @dev: The device for which the dma_addr is to be created
4097  * @page: The page to be mapped
4098  * @offset: The offset within the page
4099  * @size: The size of the region in bytes
4100  * @direction: The direction of the DMA
4101  */
4102 static inline u64 ib_dma_map_page(struct ib_device *dev,
4103 				  struct page *page,
4104 				  unsigned long offset,
4105 				  size_t size,
4106 					 enum dma_data_direction direction)
4107 {
4108 	if (ib_uses_virt_dma(dev))
4109 		return (uintptr_t)(page_address(page) + offset);
4110 	return dma_map_page(dev->dma_device, page, offset, size, direction);
4111 }
4112 
4113 /**
4114  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4115  * @dev: The device for which the DMA address was created
4116  * @addr: The DMA address
4117  * @size: The size of the region in bytes
4118  * @direction: The direction of the DMA
4119  */
4120 static inline void ib_dma_unmap_page(struct ib_device *dev,
4121 				     u64 addr, size_t size,
4122 				     enum dma_data_direction direction)
4123 {
4124 	if (!ib_uses_virt_dma(dev))
4125 		dma_unmap_page(dev->dma_device, addr, size, direction);
4126 }
4127 
4128 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4129 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4130 				      struct scatterlist *sg, int nents,
4131 				      enum dma_data_direction direction,
4132 				      unsigned long dma_attrs)
4133 {
4134 	if (ib_uses_virt_dma(dev))
4135 		return ib_dma_virt_map_sg(dev, sg, nents);
4136 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4137 				dma_attrs);
4138 }
4139 
4140 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4141 					 struct scatterlist *sg, int nents,
4142 					 enum dma_data_direction direction,
4143 					 unsigned long dma_attrs)
4144 {
4145 	if (!ib_uses_virt_dma(dev))
4146 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4147 				   dma_attrs);
4148 }
4149 
4150 /**
4151  * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4152  * @dev: The device for which the DMA addresses are to be created
4153  * @sg: The sg_table object describing the buffer
4154  * @direction: The direction of the DMA
4155  * @attrs: Optional DMA attributes for the map operation
4156  */
4157 static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4158 					   struct sg_table *sgt,
4159 					   enum dma_data_direction direction,
4160 					   unsigned long dma_attrs)
4161 {
4162 	int nents;
4163 
4164 	if (ib_uses_virt_dma(dev)) {
4165 		nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4166 		if (!nents)
4167 			return -EIO;
4168 		sgt->nents = nents;
4169 		return 0;
4170 	}
4171 	return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4172 }
4173 
4174 static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4175 					      struct sg_table *sgt,
4176 					      enum dma_data_direction direction,
4177 					      unsigned long dma_attrs)
4178 {
4179 	if (!ib_uses_virt_dma(dev))
4180 		dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4181 }
4182 
4183 /**
4184  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4185  * @dev: The device for which the DMA addresses are to be created
4186  * @sg: The array of scatter/gather entries
4187  * @nents: The number of scatter/gather entries
4188  * @direction: The direction of the DMA
4189  */
4190 static inline int ib_dma_map_sg(struct ib_device *dev,
4191 				struct scatterlist *sg, int nents,
4192 				enum dma_data_direction direction)
4193 {
4194 	return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4195 }
4196 
4197 /**
4198  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4199  * @dev: The device for which the DMA addresses were created
4200  * @sg: The array of scatter/gather entries
4201  * @nents: The number of scatter/gather entries
4202  * @direction: The direction of the DMA
4203  */
4204 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4205 				   struct scatterlist *sg, int nents,
4206 				   enum dma_data_direction direction)
4207 {
4208 	ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4209 }
4210 
4211 /**
4212  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4213  * @dev: The device to query
4214  *
4215  * The returned value represents a size in bytes.
4216  */
4217 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4218 {
4219 	if (ib_uses_virt_dma(dev))
4220 		return UINT_MAX;
4221 	return dma_get_max_seg_size(dev->dma_device);
4222 }
4223 
4224 /**
4225  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4226  * @dev: The device for which the DMA address was created
4227  * @addr: The DMA address
4228  * @size: The size of the region in bytes
4229  * @dir: The direction of the DMA
4230  */
4231 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4232 					      u64 addr,
4233 					      size_t size,
4234 					      enum dma_data_direction dir)
4235 {
4236 	if (!ib_uses_virt_dma(dev))
4237 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4238 }
4239 
4240 /**
4241  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4242  * @dev: The device for which the DMA address was created
4243  * @addr: The DMA address
4244  * @size: The size of the region in bytes
4245  * @dir: The direction of the DMA
4246  */
4247 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4248 						 u64 addr,
4249 						 size_t size,
4250 						 enum dma_data_direction dir)
4251 {
4252 	if (!ib_uses_virt_dma(dev))
4253 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4254 }
4255 
4256 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4257  * space. This function should be called when 'current' is the owning MM.
4258  */
4259 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4260 			     u64 virt_addr, int mr_access_flags);
4261 
4262 /* ib_advise_mr -  give an advice about an address range in a memory region */
4263 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4264 		 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4265 /**
4266  * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4267  *   HCA translation table.
4268  * @mr: The memory region to deregister.
4269  * @udata: Valid user data or NULL for kernel object
4270  *
4271  * This function can fail, if the memory region has memory windows bound to it.
4272  */
4273 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4274 
4275 /**
4276  * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4277  *   HCA translation table.
4278  * @mr: The memory region to deregister.
4279  *
4280  * This function can fail, if the memory region has memory windows bound to it.
4281  *
4282  * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4283  */
4284 static inline int ib_dereg_mr(struct ib_mr *mr)
4285 {
4286 	return ib_dereg_mr_user(mr, NULL);
4287 }
4288 
4289 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4290 			  u32 max_num_sg);
4291 
4292 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4293 				    u32 max_num_data_sg,
4294 				    u32 max_num_meta_sg);
4295 
4296 /**
4297  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4298  *   R_Key and L_Key.
4299  * @mr - struct ib_mr pointer to be updated.
4300  * @newkey - new key to be used.
4301  */
4302 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4303 {
4304 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4305 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4306 }
4307 
4308 /**
4309  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4310  * for calculating a new rkey for type 2 memory windows.
4311  * @rkey - the rkey to increment.
4312  */
4313 static inline u32 ib_inc_rkey(u32 rkey)
4314 {
4315 	const u32 mask = 0x000000ff;
4316 	return ((rkey + 1) & mask) | (rkey & ~mask);
4317 }
4318 
4319 /**
4320  * ib_attach_mcast - Attaches the specified QP to a multicast group.
4321  * @qp: QP to attach to the multicast group.  The QP must be type
4322  *   IB_QPT_UD.
4323  * @gid: Multicast group GID.
4324  * @lid: Multicast group LID in host byte order.
4325  *
4326  * In order to send and receive multicast packets, subnet
4327  * administration must have created the multicast group and configured
4328  * the fabric appropriately.  The port associated with the specified
4329  * QP must also be a member of the multicast group.
4330  */
4331 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4332 
4333 /**
4334  * ib_detach_mcast - Detaches the specified QP from a multicast group.
4335  * @qp: QP to detach from the multicast group.
4336  * @gid: Multicast group GID.
4337  * @lid: Multicast group LID in host byte order.
4338  */
4339 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4340 
4341 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4342 				   struct inode *inode, struct ib_udata *udata);
4343 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4344 
4345 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4346 				     unsigned int flags)
4347 {
4348 	u64 device_cap = ib_dev->attrs.device_cap_flags;
4349 
4350 	/*
4351 	 * Local write permission is required if remote write or
4352 	 * remote atomic permission is also requested.
4353 	 */
4354 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4355 	    !(flags & IB_ACCESS_LOCAL_WRITE))
4356 		return -EINVAL;
4357 
4358 	if (flags & ~IB_ACCESS_SUPPORTED)
4359 		return -EINVAL;
4360 
4361 	if (flags & IB_ACCESS_ON_DEMAND &&
4362 	    !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING))
4363 		return -EOPNOTSUPP;
4364 
4365 	if ((flags & IB_ACCESS_FLUSH_GLOBAL &&
4366 	    !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) ||
4367 	    (flags & IB_ACCESS_FLUSH_PERSISTENT &&
4368 	    !(device_cap & IB_DEVICE_FLUSH_PERSISTENT)))
4369 		return -EOPNOTSUPP;
4370 
4371 	return 0;
4372 }
4373 
4374 static inline bool ib_access_writable(int access_flags)
4375 {
4376 	/*
4377 	 * We have writable memory backing the MR if any of the following
4378 	 * access flags are set.  "Local write" and "remote write" obviously
4379 	 * require write access.  "Remote atomic" can do things like fetch and
4380 	 * add, which will modify memory, and "MW bind" can change permissions
4381 	 * by binding a window.
4382 	 */
4383 	return access_flags &
4384 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
4385 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4386 }
4387 
4388 /**
4389  * ib_check_mr_status: lightweight check of MR status.
4390  *     This routine may provide status checks on a selected
4391  *     ib_mr. first use is for signature status check.
4392  *
4393  * @mr: A memory region.
4394  * @check_mask: Bitmask of which checks to perform from
4395  *     ib_mr_status_check enumeration.
4396  * @mr_status: The container of relevant status checks.
4397  *     failed checks will be indicated in the status bitmask
4398  *     and the relevant info shall be in the error item.
4399  */
4400 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4401 		       struct ib_mr_status *mr_status);
4402 
4403 /**
4404  * ib_device_try_get: Hold a registration lock
4405  * device: The device to lock
4406  *
4407  * A device under an active registration lock cannot become unregistered. It
4408  * is only possible to obtain a registration lock on a device that is fully
4409  * registered, otherwise this function returns false.
4410  *
4411  * The registration lock is only necessary for actions which require the
4412  * device to still be registered. Uses that only require the device pointer to
4413  * be valid should use get_device(&ibdev->dev) to hold the memory.
4414  *
4415  */
4416 static inline bool ib_device_try_get(struct ib_device *dev)
4417 {
4418 	return refcount_inc_not_zero(&dev->refcount);
4419 }
4420 
4421 void ib_device_put(struct ib_device *device);
4422 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4423 					  enum rdma_driver_id driver_id);
4424 struct ib_device *ib_device_get_by_name(const char *name,
4425 					enum rdma_driver_id driver_id);
4426 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4427 					    u16 pkey, const union ib_gid *gid,
4428 					    const struct sockaddr *addr);
4429 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4430 			 unsigned int port);
4431 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4432 			   struct ib_wq_init_attr *init_attr);
4433 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4434 
4435 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4436 		 unsigned int *sg_offset, unsigned int page_size);
4437 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4438 		    int data_sg_nents, unsigned int *data_sg_offset,
4439 		    struct scatterlist *meta_sg, int meta_sg_nents,
4440 		    unsigned int *meta_sg_offset, unsigned int page_size);
4441 
4442 static inline int
4443 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4444 		  unsigned int *sg_offset, unsigned int page_size)
4445 {
4446 	int n;
4447 
4448 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4449 	mr->iova = 0;
4450 
4451 	return n;
4452 }
4453 
4454 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4455 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4456 
4457 void ib_drain_rq(struct ib_qp *qp);
4458 void ib_drain_sq(struct ib_qp *qp);
4459 void ib_drain_qp(struct ib_qp *qp);
4460 
4461 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4462 		     u8 *width);
4463 
4464 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4465 {
4466 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4467 		return attr->roce.dmac;
4468 	return NULL;
4469 }
4470 
4471 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4472 {
4473 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4474 		attr->ib.dlid = (u16)dlid;
4475 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4476 		attr->opa.dlid = dlid;
4477 }
4478 
4479 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4480 {
4481 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4482 		return attr->ib.dlid;
4483 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4484 		return attr->opa.dlid;
4485 	return 0;
4486 }
4487 
4488 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4489 {
4490 	attr->sl = sl;
4491 }
4492 
4493 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4494 {
4495 	return attr->sl;
4496 }
4497 
4498 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4499 					 u8 src_path_bits)
4500 {
4501 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4502 		attr->ib.src_path_bits = src_path_bits;
4503 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4504 		attr->opa.src_path_bits = src_path_bits;
4505 }
4506 
4507 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4508 {
4509 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4510 		return attr->ib.src_path_bits;
4511 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4512 		return attr->opa.src_path_bits;
4513 	return 0;
4514 }
4515 
4516 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4517 					bool make_grd)
4518 {
4519 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4520 		attr->opa.make_grd = make_grd;
4521 }
4522 
4523 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4524 {
4525 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4526 		return attr->opa.make_grd;
4527 	return false;
4528 }
4529 
4530 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4531 {
4532 	attr->port_num = port_num;
4533 }
4534 
4535 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4536 {
4537 	return attr->port_num;
4538 }
4539 
4540 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4541 					   u8 static_rate)
4542 {
4543 	attr->static_rate = static_rate;
4544 }
4545 
4546 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4547 {
4548 	return attr->static_rate;
4549 }
4550 
4551 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4552 					enum ib_ah_flags flag)
4553 {
4554 	attr->ah_flags = flag;
4555 }
4556 
4557 static inline enum ib_ah_flags
4558 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4559 {
4560 	return attr->ah_flags;
4561 }
4562 
4563 static inline const struct ib_global_route
4564 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4565 {
4566 	return &attr->grh;
4567 }
4568 
4569 /*To retrieve and modify the grh */
4570 static inline struct ib_global_route
4571 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4572 {
4573 	return &attr->grh;
4574 }
4575 
4576 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4577 {
4578 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4579 
4580 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4581 }
4582 
4583 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4584 					     __be64 prefix)
4585 {
4586 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4587 
4588 	grh->dgid.global.subnet_prefix = prefix;
4589 }
4590 
4591 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4592 					    __be64 if_id)
4593 {
4594 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4595 
4596 	grh->dgid.global.interface_id = if_id;
4597 }
4598 
4599 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4600 				   union ib_gid *dgid, u32 flow_label,
4601 				   u8 sgid_index, u8 hop_limit,
4602 				   u8 traffic_class)
4603 {
4604 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4605 
4606 	attr->ah_flags = IB_AH_GRH;
4607 	if (dgid)
4608 		grh->dgid = *dgid;
4609 	grh->flow_label = flow_label;
4610 	grh->sgid_index = sgid_index;
4611 	grh->hop_limit = hop_limit;
4612 	grh->traffic_class = traffic_class;
4613 	grh->sgid_attr = NULL;
4614 }
4615 
4616 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4617 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4618 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4619 			     const struct ib_gid_attr *sgid_attr);
4620 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4621 		       const struct rdma_ah_attr *src);
4622 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4623 			  const struct rdma_ah_attr *new);
4624 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4625 
4626 /**
4627  * rdma_ah_find_type - Return address handle type.
4628  *
4629  * @dev: Device to be checked
4630  * @port_num: Port number
4631  */
4632 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4633 						       u32 port_num)
4634 {
4635 	if (rdma_protocol_roce(dev, port_num))
4636 		return RDMA_AH_ATTR_TYPE_ROCE;
4637 	if (rdma_protocol_ib(dev, port_num)) {
4638 		if (rdma_cap_opa_ah(dev, port_num))
4639 			return RDMA_AH_ATTR_TYPE_OPA;
4640 		return RDMA_AH_ATTR_TYPE_IB;
4641 	}
4642 
4643 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4644 }
4645 
4646 /**
4647  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4648  *     In the current implementation the only way to
4649  *     get the 32bit lid is from other sources for OPA.
4650  *     For IB, lids will always be 16bits so cast the
4651  *     value accordingly.
4652  *
4653  * @lid: A 32bit LID
4654  */
4655 static inline u16 ib_lid_cpu16(u32 lid)
4656 {
4657 	WARN_ON_ONCE(lid & 0xFFFF0000);
4658 	return (u16)lid;
4659 }
4660 
4661 /**
4662  * ib_lid_be16 - Return lid in 16bit BE encoding.
4663  *
4664  * @lid: A 32bit LID
4665  */
4666 static inline __be16 ib_lid_be16(u32 lid)
4667 {
4668 	WARN_ON_ONCE(lid & 0xFFFF0000);
4669 	return cpu_to_be16((u16)lid);
4670 }
4671 
4672 /**
4673  * ib_get_vector_affinity - Get the affinity mappings of a given completion
4674  *   vector
4675  * @device:         the rdma device
4676  * @comp_vector:    index of completion vector
4677  *
4678  * Returns NULL on failure, otherwise a corresponding cpu map of the
4679  * completion vector (returns all-cpus map if the device driver doesn't
4680  * implement get_vector_affinity).
4681  */
4682 static inline const struct cpumask *
4683 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4684 {
4685 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4686 	    !device->ops.get_vector_affinity)
4687 		return NULL;
4688 
4689 	return device->ops.get_vector_affinity(device, comp_vector);
4690 
4691 }
4692 
4693 /**
4694  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4695  * and add their gids, as needed, to the relevant RoCE devices.
4696  *
4697  * @device:         the rdma device
4698  */
4699 void rdma_roce_rescan_device(struct ib_device *ibdev);
4700 
4701 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4702 
4703 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4704 
4705 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4706 				     enum rdma_netdev_t type, const char *name,
4707 				     unsigned char name_assign_type,
4708 				     void (*setup)(struct net_device *));
4709 
4710 int rdma_init_netdev(struct ib_device *device, u32 port_num,
4711 		     enum rdma_netdev_t type, const char *name,
4712 		     unsigned char name_assign_type,
4713 		     void (*setup)(struct net_device *),
4714 		     struct net_device *netdev);
4715 
4716 /**
4717  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4718  *
4719  * @device:	device pointer for which ib_device pointer to retrieve
4720  *
4721  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4722  *
4723  */
4724 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4725 {
4726 	struct ib_core_device *coredev =
4727 		container_of(device, struct ib_core_device, dev);
4728 
4729 	return coredev->owner;
4730 }
4731 
4732 /**
4733  * ibdev_to_node - return the NUMA node for a given ib_device
4734  * @dev:	device to get the NUMA node for.
4735  */
4736 static inline int ibdev_to_node(struct ib_device *ibdev)
4737 {
4738 	struct device *parent = ibdev->dev.parent;
4739 
4740 	if (!parent)
4741 		return NUMA_NO_NODE;
4742 	return dev_to_node(parent);
4743 }
4744 
4745 /**
4746  * rdma_device_to_drv_device - Helper macro to reach back to driver's
4747  *			       ib_device holder structure from device pointer.
4748  *
4749  * NOTE: New drivers should not make use of this API; This API is only for
4750  * existing drivers who have exposed sysfs entries using
4751  * ops->device_group.
4752  */
4753 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
4754 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4755 
4756 bool rdma_dev_access_netns(const struct ib_device *device,
4757 			   const struct net *net);
4758 
4759 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4760 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4761 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4762 
4763 /**
4764  * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4765  *                               on the flow_label
4766  *
4767  * This function will convert the 20 bit flow_label input to a valid RoCE v2
4768  * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4769  * convention.
4770  */
4771 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4772 {
4773 	u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4774 
4775 	fl_low ^= fl_high >> 14;
4776 	return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4777 }
4778 
4779 /**
4780  * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4781  *                        local and remote qpn values
4782  *
4783  * This function folded the multiplication results of two qpns, 24 bit each,
4784  * fields, and converts it to a 20 bit results.
4785  *
4786  * This function will create symmetric flow_label value based on the local
4787  * and remote qpn values. this will allow both the requester and responder
4788  * to calculate the same flow_label for a given connection.
4789  *
4790  * This helper function should be used by driver in case the upper layer
4791  * provide a zero flow_label value. This is to improve entropy of RDMA
4792  * traffic in the network.
4793  */
4794 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4795 {
4796 	u64 v = (u64)lqpn * rqpn;
4797 
4798 	v ^= v >> 20;
4799 	v ^= v >> 40;
4800 
4801 	return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4802 }
4803 
4804 /**
4805  * rdma_get_udp_sport - Calculate and set UDP source port based on the flow
4806  *                      label. If flow label is not defined in GRH then
4807  *                      calculate it based on lqpn/rqpn.
4808  *
4809  * @fl:                 flow label from GRH
4810  * @lqpn:               local qp number
4811  * @rqpn:               remote qp number
4812  */
4813 static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
4814 {
4815 	if (!fl)
4816 		fl = rdma_calc_flow_label(lqpn, rqpn);
4817 
4818 	return rdma_flow_label_to_udp_sport(fl);
4819 }
4820 
4821 const struct ib_port_immutable*
4822 ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4823 #endif /* IB_VERBS_H */
4824