xref: /openbmc/linux/include/rdma/ib_verbs.h (revision 66c98360)
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 };
565 
566 enum ib_stat_flag {
567 	IB_STAT_FLAG_OPTIONAL = 1 << 0,
568 };
569 
570 /**
571  * struct rdma_stat_desc
572  * @name - The name of the counter
573  * @flags - Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL
574  * @priv - Driver private information; Core code should not use
575  */
576 struct rdma_stat_desc {
577 	const char *name;
578 	unsigned int flags;
579 	const void *priv;
580 };
581 
582 /**
583  * struct rdma_hw_stats
584  * @lock - Mutex to protect parallel write access to lifespan and values
585  *    of counters, which are 64bits and not guaranteed to be written
586  *    atomicaly on 32bits systems.
587  * @timestamp - Used by the core code to track when the last update was
588  * @lifespan - Used by the core code to determine how old the counters
589  *   should be before being updated again.  Stored in jiffies, defaults
590  *   to 10 milliseconds, drivers can override the default be specifying
591  *   their own value during their allocation routine.
592  * @descs - Array of pointers to static descriptors used for the counters
593  *   in directory.
594  * @is_disabled - A bitmap to indicate each counter is currently disabled
595  *   or not.
596  * @num_counters - How many hardware counters there are.  If name is
597  *   shorter than this number, a kernel oops will result.  Driver authors
598  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
599  *   in their code to prevent this.
600  * @value - Array of u64 counters that are accessed by the sysfs code and
601  *   filled in by the drivers get_stats routine
602  */
603 struct rdma_hw_stats {
604 	struct mutex	lock; /* Protect lifespan and values[] */
605 	unsigned long	timestamp;
606 	unsigned long	lifespan;
607 	const struct rdma_stat_desc *descs;
608 	unsigned long	*is_disabled;
609 	int		num_counters;
610 	u64		value[];
611 };
612 
613 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
614 
615 struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
616 	const struct rdma_stat_desc *descs, int num_counters,
617 	unsigned long lifespan);
618 
619 void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
620 
621 /* Define bits for the various functionality this port needs to be supported by
622  * the core.
623  */
624 /* Management                           0x00000FFF */
625 #define RDMA_CORE_CAP_IB_MAD            0x00000001
626 #define RDMA_CORE_CAP_IB_SMI            0x00000002
627 #define RDMA_CORE_CAP_IB_CM             0x00000004
628 #define RDMA_CORE_CAP_IW_CM             0x00000008
629 #define RDMA_CORE_CAP_IB_SA             0x00000010
630 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
631 
632 /* Address format                       0x000FF000 */
633 #define RDMA_CORE_CAP_AF_IB             0x00001000
634 #define RDMA_CORE_CAP_ETH_AH            0x00002000
635 #define RDMA_CORE_CAP_OPA_AH            0x00004000
636 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
637 
638 /* Protocol                             0xFFF00000 */
639 #define RDMA_CORE_CAP_PROT_IB           0x00100000
640 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
641 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
642 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
643 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
644 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
645 
646 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
647 					| RDMA_CORE_CAP_PROT_ROCE     \
648 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
649 
650 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
651 					| RDMA_CORE_CAP_IB_MAD \
652 					| RDMA_CORE_CAP_IB_SMI \
653 					| RDMA_CORE_CAP_IB_CM  \
654 					| RDMA_CORE_CAP_IB_SA  \
655 					| RDMA_CORE_CAP_AF_IB)
656 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
657 					| RDMA_CORE_CAP_IB_MAD  \
658 					| RDMA_CORE_CAP_IB_CM   \
659 					| RDMA_CORE_CAP_AF_IB   \
660 					| RDMA_CORE_CAP_ETH_AH)
661 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
662 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
663 					| RDMA_CORE_CAP_IB_MAD  \
664 					| RDMA_CORE_CAP_IB_CM   \
665 					| RDMA_CORE_CAP_AF_IB   \
666 					| RDMA_CORE_CAP_ETH_AH)
667 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
668 					| RDMA_CORE_CAP_IW_CM)
669 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
670 					| RDMA_CORE_CAP_OPA_MAD)
671 
672 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
673 
674 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
675 
676 struct ib_port_attr {
677 	u64			subnet_prefix;
678 	enum ib_port_state	state;
679 	enum ib_mtu		max_mtu;
680 	enum ib_mtu		active_mtu;
681 	u32                     phys_mtu;
682 	int			gid_tbl_len;
683 	unsigned int		ip_gids:1;
684 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
685 	u32			port_cap_flags;
686 	u32			max_msg_sz;
687 	u32			bad_pkey_cntr;
688 	u32			qkey_viol_cntr;
689 	u16			pkey_tbl_len;
690 	u32			sm_lid;
691 	u32			lid;
692 	u8			lmc;
693 	u8			max_vl_num;
694 	u8			sm_sl;
695 	u8			subnet_timeout;
696 	u8			init_type_reply;
697 	u8			active_width;
698 	u16			active_speed;
699 	u8                      phys_state;
700 	u16			port_cap_flags2;
701 };
702 
703 enum ib_device_modify_flags {
704 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
705 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
706 };
707 
708 #define IB_DEVICE_NODE_DESC_MAX 64
709 
710 struct ib_device_modify {
711 	u64	sys_image_guid;
712 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
713 };
714 
715 enum ib_port_modify_flags {
716 	IB_PORT_SHUTDOWN		= 1,
717 	IB_PORT_INIT_TYPE		= (1<<2),
718 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
719 	IB_PORT_OPA_MASK_CHG		= (1<<4)
720 };
721 
722 struct ib_port_modify {
723 	u32	set_port_cap_mask;
724 	u32	clr_port_cap_mask;
725 	u8	init_type;
726 };
727 
728 enum ib_event_type {
729 	IB_EVENT_CQ_ERR,
730 	IB_EVENT_QP_FATAL,
731 	IB_EVENT_QP_REQ_ERR,
732 	IB_EVENT_QP_ACCESS_ERR,
733 	IB_EVENT_COMM_EST,
734 	IB_EVENT_SQ_DRAINED,
735 	IB_EVENT_PATH_MIG,
736 	IB_EVENT_PATH_MIG_ERR,
737 	IB_EVENT_DEVICE_FATAL,
738 	IB_EVENT_PORT_ACTIVE,
739 	IB_EVENT_PORT_ERR,
740 	IB_EVENT_LID_CHANGE,
741 	IB_EVENT_PKEY_CHANGE,
742 	IB_EVENT_SM_CHANGE,
743 	IB_EVENT_SRQ_ERR,
744 	IB_EVENT_SRQ_LIMIT_REACHED,
745 	IB_EVENT_QP_LAST_WQE_REACHED,
746 	IB_EVENT_CLIENT_REREGISTER,
747 	IB_EVENT_GID_CHANGE,
748 	IB_EVENT_WQ_FATAL,
749 };
750 
751 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
752 
753 struct ib_event {
754 	struct ib_device	*device;
755 	union {
756 		struct ib_cq	*cq;
757 		struct ib_qp	*qp;
758 		struct ib_srq	*srq;
759 		struct ib_wq	*wq;
760 		u32		port_num;
761 	} element;
762 	enum ib_event_type	event;
763 };
764 
765 struct ib_event_handler {
766 	struct ib_device *device;
767 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
768 	struct list_head  list;
769 };
770 
771 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
772 	do {							\
773 		(_ptr)->device  = _device;			\
774 		(_ptr)->handler = _handler;			\
775 		INIT_LIST_HEAD(&(_ptr)->list);			\
776 	} while (0)
777 
778 struct ib_global_route {
779 	const struct ib_gid_attr *sgid_attr;
780 	union ib_gid	dgid;
781 	u32		flow_label;
782 	u8		sgid_index;
783 	u8		hop_limit;
784 	u8		traffic_class;
785 };
786 
787 struct ib_grh {
788 	__be32		version_tclass_flow;
789 	__be16		paylen;
790 	u8		next_hdr;
791 	u8		hop_limit;
792 	union ib_gid	sgid;
793 	union ib_gid	dgid;
794 };
795 
796 union rdma_network_hdr {
797 	struct ib_grh ibgrh;
798 	struct {
799 		/* The IB spec states that if it's IPv4, the header
800 		 * is located in the last 20 bytes of the header.
801 		 */
802 		u8		reserved[20];
803 		struct iphdr	roce4grh;
804 	};
805 };
806 
807 #define IB_QPN_MASK		0xFFFFFF
808 
809 enum {
810 	IB_MULTICAST_QPN = 0xffffff
811 };
812 
813 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
814 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
815 
816 enum ib_ah_flags {
817 	IB_AH_GRH	= 1
818 };
819 
820 enum ib_rate {
821 	IB_RATE_PORT_CURRENT = 0,
822 	IB_RATE_2_5_GBPS = 2,
823 	IB_RATE_5_GBPS   = 5,
824 	IB_RATE_10_GBPS  = 3,
825 	IB_RATE_20_GBPS  = 6,
826 	IB_RATE_30_GBPS  = 4,
827 	IB_RATE_40_GBPS  = 7,
828 	IB_RATE_60_GBPS  = 8,
829 	IB_RATE_80_GBPS  = 9,
830 	IB_RATE_120_GBPS = 10,
831 	IB_RATE_14_GBPS  = 11,
832 	IB_RATE_56_GBPS  = 12,
833 	IB_RATE_112_GBPS = 13,
834 	IB_RATE_168_GBPS = 14,
835 	IB_RATE_25_GBPS  = 15,
836 	IB_RATE_100_GBPS = 16,
837 	IB_RATE_200_GBPS = 17,
838 	IB_RATE_300_GBPS = 18,
839 	IB_RATE_28_GBPS  = 19,
840 	IB_RATE_50_GBPS  = 20,
841 	IB_RATE_400_GBPS = 21,
842 	IB_RATE_600_GBPS = 22,
843 };
844 
845 /**
846  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
847  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
848  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
849  * @rate: rate to convert.
850  */
851 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
852 
853 /**
854  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
855  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
856  * @rate: rate to convert.
857  */
858 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
859 
860 
861 /**
862  * enum ib_mr_type - memory region type
863  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
864  *                            normal registration
865  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
866  *                            register any arbitrary sg lists (without
867  *                            the normal mr constraints - see
868  *                            ib_map_mr_sg)
869  * @IB_MR_TYPE_DM:            memory region that is used for device
870  *                            memory registration
871  * @IB_MR_TYPE_USER:          memory region that is used for the user-space
872  *                            application
873  * @IB_MR_TYPE_DMA:           memory region that is used for DMA operations
874  *                            without address translations (VA=PA)
875  * @IB_MR_TYPE_INTEGRITY:     memory region that is used for
876  *                            data integrity operations
877  */
878 enum ib_mr_type {
879 	IB_MR_TYPE_MEM_REG,
880 	IB_MR_TYPE_SG_GAPS,
881 	IB_MR_TYPE_DM,
882 	IB_MR_TYPE_USER,
883 	IB_MR_TYPE_DMA,
884 	IB_MR_TYPE_INTEGRITY,
885 };
886 
887 enum ib_mr_status_check {
888 	IB_MR_CHECK_SIG_STATUS = 1,
889 };
890 
891 /**
892  * struct ib_mr_status - Memory region status container
893  *
894  * @fail_status: Bitmask of MR checks status. For each
895  *     failed check a corresponding status bit is set.
896  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
897  *     failure.
898  */
899 struct ib_mr_status {
900 	u32		    fail_status;
901 	struct ib_sig_err   sig_err;
902 };
903 
904 /**
905  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
906  * enum.
907  * @mult: multiple to convert.
908  */
909 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
910 
911 struct rdma_ah_init_attr {
912 	struct rdma_ah_attr *ah_attr;
913 	u32 flags;
914 	struct net_device *xmit_slave;
915 };
916 
917 enum rdma_ah_attr_type {
918 	RDMA_AH_ATTR_TYPE_UNDEFINED,
919 	RDMA_AH_ATTR_TYPE_IB,
920 	RDMA_AH_ATTR_TYPE_ROCE,
921 	RDMA_AH_ATTR_TYPE_OPA,
922 };
923 
924 struct ib_ah_attr {
925 	u16			dlid;
926 	u8			src_path_bits;
927 };
928 
929 struct roce_ah_attr {
930 	u8			dmac[ETH_ALEN];
931 };
932 
933 struct opa_ah_attr {
934 	u32			dlid;
935 	u8			src_path_bits;
936 	bool			make_grd;
937 };
938 
939 struct rdma_ah_attr {
940 	struct ib_global_route	grh;
941 	u8			sl;
942 	u8			static_rate;
943 	u32			port_num;
944 	u8			ah_flags;
945 	enum rdma_ah_attr_type type;
946 	union {
947 		struct ib_ah_attr ib;
948 		struct roce_ah_attr roce;
949 		struct opa_ah_attr opa;
950 	};
951 };
952 
953 enum ib_wc_status {
954 	IB_WC_SUCCESS,
955 	IB_WC_LOC_LEN_ERR,
956 	IB_WC_LOC_QP_OP_ERR,
957 	IB_WC_LOC_EEC_OP_ERR,
958 	IB_WC_LOC_PROT_ERR,
959 	IB_WC_WR_FLUSH_ERR,
960 	IB_WC_MW_BIND_ERR,
961 	IB_WC_BAD_RESP_ERR,
962 	IB_WC_LOC_ACCESS_ERR,
963 	IB_WC_REM_INV_REQ_ERR,
964 	IB_WC_REM_ACCESS_ERR,
965 	IB_WC_REM_OP_ERR,
966 	IB_WC_RETRY_EXC_ERR,
967 	IB_WC_RNR_RETRY_EXC_ERR,
968 	IB_WC_LOC_RDD_VIOL_ERR,
969 	IB_WC_REM_INV_RD_REQ_ERR,
970 	IB_WC_REM_ABORT_ERR,
971 	IB_WC_INV_EECN_ERR,
972 	IB_WC_INV_EEC_STATE_ERR,
973 	IB_WC_FATAL_ERR,
974 	IB_WC_RESP_TIMEOUT_ERR,
975 	IB_WC_GENERAL_ERR
976 };
977 
978 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
979 
980 enum ib_wc_opcode {
981 	IB_WC_SEND = IB_UVERBS_WC_SEND,
982 	IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
983 	IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
984 	IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
985 	IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
986 	IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
987 	IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
988 	IB_WC_LSO = IB_UVERBS_WC_TSO,
989 	IB_WC_ATOMIC_WRITE = IB_UVERBS_WC_ATOMIC_WRITE,
990 	IB_WC_REG_MR,
991 	IB_WC_MASKED_COMP_SWAP,
992 	IB_WC_MASKED_FETCH_ADD,
993 	IB_WC_FLUSH = IB_UVERBS_WC_FLUSH,
994 /*
995  * Set value of IB_WC_RECV so consumers can test if a completion is a
996  * receive by testing (opcode & IB_WC_RECV).
997  */
998 	IB_WC_RECV			= 1 << 7,
999 	IB_WC_RECV_RDMA_WITH_IMM
1000 };
1001 
1002 enum ib_wc_flags {
1003 	IB_WC_GRH		= 1,
1004 	IB_WC_WITH_IMM		= (1<<1),
1005 	IB_WC_WITH_INVALIDATE	= (1<<2),
1006 	IB_WC_IP_CSUM_OK	= (1<<3),
1007 	IB_WC_WITH_SMAC		= (1<<4),
1008 	IB_WC_WITH_VLAN		= (1<<5),
1009 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
1010 };
1011 
1012 struct ib_wc {
1013 	union {
1014 		u64		wr_id;
1015 		struct ib_cqe	*wr_cqe;
1016 	};
1017 	enum ib_wc_status	status;
1018 	enum ib_wc_opcode	opcode;
1019 	u32			vendor_err;
1020 	u32			byte_len;
1021 	struct ib_qp	       *qp;
1022 	union {
1023 		__be32		imm_data;
1024 		u32		invalidate_rkey;
1025 	} ex;
1026 	u32			src_qp;
1027 	u32			slid;
1028 	int			wc_flags;
1029 	u16			pkey_index;
1030 	u8			sl;
1031 	u8			dlid_path_bits;
1032 	u32 port_num; /* valid only for DR SMPs on switches */
1033 	u8			smac[ETH_ALEN];
1034 	u16			vlan_id;
1035 	u8			network_hdr_type;
1036 };
1037 
1038 enum ib_cq_notify_flags {
1039 	IB_CQ_SOLICITED			= 1 << 0,
1040 	IB_CQ_NEXT_COMP			= 1 << 1,
1041 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1042 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
1043 };
1044 
1045 enum ib_srq_type {
1046 	IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1047 	IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1048 	IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1049 };
1050 
1051 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1052 {
1053 	return srq_type == IB_SRQT_XRC ||
1054 	       srq_type == IB_SRQT_TM;
1055 }
1056 
1057 enum ib_srq_attr_mask {
1058 	IB_SRQ_MAX_WR	= 1 << 0,
1059 	IB_SRQ_LIMIT	= 1 << 1,
1060 };
1061 
1062 struct ib_srq_attr {
1063 	u32	max_wr;
1064 	u32	max_sge;
1065 	u32	srq_limit;
1066 };
1067 
1068 struct ib_srq_init_attr {
1069 	void		      (*event_handler)(struct ib_event *, void *);
1070 	void		       *srq_context;
1071 	struct ib_srq_attr	attr;
1072 	enum ib_srq_type	srq_type;
1073 
1074 	struct {
1075 		struct ib_cq   *cq;
1076 		union {
1077 			struct {
1078 				struct ib_xrcd *xrcd;
1079 			} xrc;
1080 
1081 			struct {
1082 				u32		max_num_tags;
1083 			} tag_matching;
1084 		};
1085 	} ext;
1086 };
1087 
1088 struct ib_qp_cap {
1089 	u32	max_send_wr;
1090 	u32	max_recv_wr;
1091 	u32	max_send_sge;
1092 	u32	max_recv_sge;
1093 	u32	max_inline_data;
1094 
1095 	/*
1096 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1097 	 * ib_create_qp() will calculate the right amount of neededed WRs
1098 	 * and MRs based on this.
1099 	 */
1100 	u32	max_rdma_ctxs;
1101 };
1102 
1103 enum ib_sig_type {
1104 	IB_SIGNAL_ALL_WR,
1105 	IB_SIGNAL_REQ_WR
1106 };
1107 
1108 enum ib_qp_type {
1109 	/*
1110 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1111 	 * here (and in that order) since the MAD layer uses them as
1112 	 * indices into a 2-entry table.
1113 	 */
1114 	IB_QPT_SMI,
1115 	IB_QPT_GSI,
1116 
1117 	IB_QPT_RC = IB_UVERBS_QPT_RC,
1118 	IB_QPT_UC = IB_UVERBS_QPT_UC,
1119 	IB_QPT_UD = IB_UVERBS_QPT_UD,
1120 	IB_QPT_RAW_IPV6,
1121 	IB_QPT_RAW_ETHERTYPE,
1122 	IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1123 	IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1124 	IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1125 	IB_QPT_MAX,
1126 	IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1127 	/* Reserve a range for qp types internal to the low level driver.
1128 	 * These qp types will not be visible at the IB core layer, so the
1129 	 * IB_QPT_MAX usages should not be affected in the core layer
1130 	 */
1131 	IB_QPT_RESERVED1 = 0x1000,
1132 	IB_QPT_RESERVED2,
1133 	IB_QPT_RESERVED3,
1134 	IB_QPT_RESERVED4,
1135 	IB_QPT_RESERVED5,
1136 	IB_QPT_RESERVED6,
1137 	IB_QPT_RESERVED7,
1138 	IB_QPT_RESERVED8,
1139 	IB_QPT_RESERVED9,
1140 	IB_QPT_RESERVED10,
1141 };
1142 
1143 enum ib_qp_create_flags {
1144 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1145 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	=
1146 		IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1147 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1148 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1149 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1150 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1151 	IB_QP_CREATE_INTEGRITY_EN		= 1 << 6,
1152 	IB_QP_CREATE_NETDEV_USE			= 1 << 7,
1153 	IB_QP_CREATE_SCATTER_FCS		=
1154 		IB_UVERBS_QP_CREATE_SCATTER_FCS,
1155 	IB_QP_CREATE_CVLAN_STRIPPING		=
1156 		IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1157 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1158 	IB_QP_CREATE_PCI_WRITE_END_PADDING	=
1159 		IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1160 	/* reserve bits 26-31 for low level drivers' internal use */
1161 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1162 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1163 };
1164 
1165 /*
1166  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1167  * callback to destroy the passed in QP.
1168  */
1169 
1170 struct ib_qp_init_attr {
1171 	/* This callback occurs in workqueue context */
1172 	void                  (*event_handler)(struct ib_event *, void *);
1173 
1174 	void		       *qp_context;
1175 	struct ib_cq	       *send_cq;
1176 	struct ib_cq	       *recv_cq;
1177 	struct ib_srq	       *srq;
1178 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1179 	struct ib_qp_cap	cap;
1180 	enum ib_sig_type	sq_sig_type;
1181 	enum ib_qp_type		qp_type;
1182 	u32			create_flags;
1183 
1184 	/*
1185 	 * Only needed for special QP types, or when using the RW API.
1186 	 */
1187 	u32			port_num;
1188 	struct ib_rwq_ind_table *rwq_ind_tbl;
1189 	u32			source_qpn;
1190 };
1191 
1192 struct ib_qp_open_attr {
1193 	void                  (*event_handler)(struct ib_event *, void *);
1194 	void		       *qp_context;
1195 	u32			qp_num;
1196 	enum ib_qp_type		qp_type;
1197 };
1198 
1199 enum ib_rnr_timeout {
1200 	IB_RNR_TIMER_655_36 =  0,
1201 	IB_RNR_TIMER_000_01 =  1,
1202 	IB_RNR_TIMER_000_02 =  2,
1203 	IB_RNR_TIMER_000_03 =  3,
1204 	IB_RNR_TIMER_000_04 =  4,
1205 	IB_RNR_TIMER_000_06 =  5,
1206 	IB_RNR_TIMER_000_08 =  6,
1207 	IB_RNR_TIMER_000_12 =  7,
1208 	IB_RNR_TIMER_000_16 =  8,
1209 	IB_RNR_TIMER_000_24 =  9,
1210 	IB_RNR_TIMER_000_32 = 10,
1211 	IB_RNR_TIMER_000_48 = 11,
1212 	IB_RNR_TIMER_000_64 = 12,
1213 	IB_RNR_TIMER_000_96 = 13,
1214 	IB_RNR_TIMER_001_28 = 14,
1215 	IB_RNR_TIMER_001_92 = 15,
1216 	IB_RNR_TIMER_002_56 = 16,
1217 	IB_RNR_TIMER_003_84 = 17,
1218 	IB_RNR_TIMER_005_12 = 18,
1219 	IB_RNR_TIMER_007_68 = 19,
1220 	IB_RNR_TIMER_010_24 = 20,
1221 	IB_RNR_TIMER_015_36 = 21,
1222 	IB_RNR_TIMER_020_48 = 22,
1223 	IB_RNR_TIMER_030_72 = 23,
1224 	IB_RNR_TIMER_040_96 = 24,
1225 	IB_RNR_TIMER_061_44 = 25,
1226 	IB_RNR_TIMER_081_92 = 26,
1227 	IB_RNR_TIMER_122_88 = 27,
1228 	IB_RNR_TIMER_163_84 = 28,
1229 	IB_RNR_TIMER_245_76 = 29,
1230 	IB_RNR_TIMER_327_68 = 30,
1231 	IB_RNR_TIMER_491_52 = 31
1232 };
1233 
1234 enum ib_qp_attr_mask {
1235 	IB_QP_STATE			= 1,
1236 	IB_QP_CUR_STATE			= (1<<1),
1237 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1238 	IB_QP_ACCESS_FLAGS		= (1<<3),
1239 	IB_QP_PKEY_INDEX		= (1<<4),
1240 	IB_QP_PORT			= (1<<5),
1241 	IB_QP_QKEY			= (1<<6),
1242 	IB_QP_AV			= (1<<7),
1243 	IB_QP_PATH_MTU			= (1<<8),
1244 	IB_QP_TIMEOUT			= (1<<9),
1245 	IB_QP_RETRY_CNT			= (1<<10),
1246 	IB_QP_RNR_RETRY			= (1<<11),
1247 	IB_QP_RQ_PSN			= (1<<12),
1248 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1249 	IB_QP_ALT_PATH			= (1<<14),
1250 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1251 	IB_QP_SQ_PSN			= (1<<16),
1252 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1253 	IB_QP_PATH_MIG_STATE		= (1<<18),
1254 	IB_QP_CAP			= (1<<19),
1255 	IB_QP_DEST_QPN			= (1<<20),
1256 	IB_QP_RESERVED1			= (1<<21),
1257 	IB_QP_RESERVED2			= (1<<22),
1258 	IB_QP_RESERVED3			= (1<<23),
1259 	IB_QP_RESERVED4			= (1<<24),
1260 	IB_QP_RATE_LIMIT		= (1<<25),
1261 
1262 	IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1263 };
1264 
1265 enum ib_qp_state {
1266 	IB_QPS_RESET,
1267 	IB_QPS_INIT,
1268 	IB_QPS_RTR,
1269 	IB_QPS_RTS,
1270 	IB_QPS_SQD,
1271 	IB_QPS_SQE,
1272 	IB_QPS_ERR
1273 };
1274 
1275 enum ib_mig_state {
1276 	IB_MIG_MIGRATED,
1277 	IB_MIG_REARM,
1278 	IB_MIG_ARMED
1279 };
1280 
1281 enum ib_mw_type {
1282 	IB_MW_TYPE_1 = 1,
1283 	IB_MW_TYPE_2 = 2
1284 };
1285 
1286 struct ib_qp_attr {
1287 	enum ib_qp_state	qp_state;
1288 	enum ib_qp_state	cur_qp_state;
1289 	enum ib_mtu		path_mtu;
1290 	enum ib_mig_state	path_mig_state;
1291 	u32			qkey;
1292 	u32			rq_psn;
1293 	u32			sq_psn;
1294 	u32			dest_qp_num;
1295 	int			qp_access_flags;
1296 	struct ib_qp_cap	cap;
1297 	struct rdma_ah_attr	ah_attr;
1298 	struct rdma_ah_attr	alt_ah_attr;
1299 	u16			pkey_index;
1300 	u16			alt_pkey_index;
1301 	u8			en_sqd_async_notify;
1302 	u8			sq_draining;
1303 	u8			max_rd_atomic;
1304 	u8			max_dest_rd_atomic;
1305 	u8			min_rnr_timer;
1306 	u32			port_num;
1307 	u8			timeout;
1308 	u8			retry_cnt;
1309 	u8			rnr_retry;
1310 	u32			alt_port_num;
1311 	u8			alt_timeout;
1312 	u32			rate_limit;
1313 	struct net_device	*xmit_slave;
1314 };
1315 
1316 enum ib_wr_opcode {
1317 	/* These are shared with userspace */
1318 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1319 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1320 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1321 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1322 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1323 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1324 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1325 	IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1326 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1327 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1328 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1329 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1330 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1331 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1332 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1333 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1334 	IB_WR_FLUSH = IB_UVERBS_WR_FLUSH,
1335 	IB_WR_ATOMIC_WRITE = IB_UVERBS_WR_ATOMIC_WRITE,
1336 
1337 	/* These are kernel only and can not be issued by userspace */
1338 	IB_WR_REG_MR = 0x20,
1339 	IB_WR_REG_MR_INTEGRITY,
1340 
1341 	/* reserve values for low level drivers' internal use.
1342 	 * These values will not be used at all in the ib core layer.
1343 	 */
1344 	IB_WR_RESERVED1 = 0xf0,
1345 	IB_WR_RESERVED2,
1346 	IB_WR_RESERVED3,
1347 	IB_WR_RESERVED4,
1348 	IB_WR_RESERVED5,
1349 	IB_WR_RESERVED6,
1350 	IB_WR_RESERVED7,
1351 	IB_WR_RESERVED8,
1352 	IB_WR_RESERVED9,
1353 	IB_WR_RESERVED10,
1354 };
1355 
1356 enum ib_send_flags {
1357 	IB_SEND_FENCE		= 1,
1358 	IB_SEND_SIGNALED	= (1<<1),
1359 	IB_SEND_SOLICITED	= (1<<2),
1360 	IB_SEND_INLINE		= (1<<3),
1361 	IB_SEND_IP_CSUM		= (1<<4),
1362 
1363 	/* reserve bits 26-31 for low level drivers' internal use */
1364 	IB_SEND_RESERVED_START	= (1 << 26),
1365 	IB_SEND_RESERVED_END	= (1 << 31),
1366 };
1367 
1368 struct ib_sge {
1369 	u64	addr;
1370 	u32	length;
1371 	u32	lkey;
1372 };
1373 
1374 struct ib_cqe {
1375 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1376 };
1377 
1378 struct ib_send_wr {
1379 	struct ib_send_wr      *next;
1380 	union {
1381 		u64		wr_id;
1382 		struct ib_cqe	*wr_cqe;
1383 	};
1384 	struct ib_sge	       *sg_list;
1385 	int			num_sge;
1386 	enum ib_wr_opcode	opcode;
1387 	int			send_flags;
1388 	union {
1389 		__be32		imm_data;
1390 		u32		invalidate_rkey;
1391 	} ex;
1392 };
1393 
1394 struct ib_rdma_wr {
1395 	struct ib_send_wr	wr;
1396 	u64			remote_addr;
1397 	u32			rkey;
1398 };
1399 
1400 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1401 {
1402 	return container_of(wr, struct ib_rdma_wr, wr);
1403 }
1404 
1405 struct ib_atomic_wr {
1406 	struct ib_send_wr	wr;
1407 	u64			remote_addr;
1408 	u64			compare_add;
1409 	u64			swap;
1410 	u64			compare_add_mask;
1411 	u64			swap_mask;
1412 	u32			rkey;
1413 };
1414 
1415 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1416 {
1417 	return container_of(wr, struct ib_atomic_wr, wr);
1418 }
1419 
1420 struct ib_ud_wr {
1421 	struct ib_send_wr	wr;
1422 	struct ib_ah		*ah;
1423 	void			*header;
1424 	int			hlen;
1425 	int			mss;
1426 	u32			remote_qpn;
1427 	u32			remote_qkey;
1428 	u16			pkey_index; /* valid for GSI only */
1429 	u32			port_num; /* valid for DR SMPs on switch only */
1430 };
1431 
1432 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1433 {
1434 	return container_of(wr, struct ib_ud_wr, wr);
1435 }
1436 
1437 struct ib_reg_wr {
1438 	struct ib_send_wr	wr;
1439 	struct ib_mr		*mr;
1440 	u32			key;
1441 	int			access;
1442 };
1443 
1444 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1445 {
1446 	return container_of(wr, struct ib_reg_wr, wr);
1447 }
1448 
1449 struct ib_recv_wr {
1450 	struct ib_recv_wr      *next;
1451 	union {
1452 		u64		wr_id;
1453 		struct ib_cqe	*wr_cqe;
1454 	};
1455 	struct ib_sge	       *sg_list;
1456 	int			num_sge;
1457 };
1458 
1459 enum ib_access_flags {
1460 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1461 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1462 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1463 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1464 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1465 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1466 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1467 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1468 	IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1469 	IB_ACCESS_FLUSH_GLOBAL = IB_UVERBS_ACCESS_FLUSH_GLOBAL,
1470 	IB_ACCESS_FLUSH_PERSISTENT = IB_UVERBS_ACCESS_FLUSH_PERSISTENT,
1471 
1472 	IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1473 	IB_ACCESS_SUPPORTED =
1474 		((IB_ACCESS_FLUSH_PERSISTENT << 1) - 1) | IB_ACCESS_OPTIONAL,
1475 };
1476 
1477 /*
1478  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1479  * are hidden here instead of a uapi header!
1480  */
1481 enum ib_mr_rereg_flags {
1482 	IB_MR_REREG_TRANS	= 1,
1483 	IB_MR_REREG_PD		= (1<<1),
1484 	IB_MR_REREG_ACCESS	= (1<<2),
1485 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1486 };
1487 
1488 struct ib_umem;
1489 
1490 enum rdma_remove_reason {
1491 	/*
1492 	 * Userspace requested uobject deletion or initial try
1493 	 * to remove uobject via cleanup. Call could fail
1494 	 */
1495 	RDMA_REMOVE_DESTROY,
1496 	/* Context deletion. This call should delete the actual object itself */
1497 	RDMA_REMOVE_CLOSE,
1498 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1499 	RDMA_REMOVE_DRIVER_REMOVE,
1500 	/* uobj is being cleaned-up before being committed */
1501 	RDMA_REMOVE_ABORT,
1502 	/* The driver failed to destroy the uobject and is being disconnected */
1503 	RDMA_REMOVE_DRIVER_FAILURE,
1504 };
1505 
1506 struct ib_rdmacg_object {
1507 #ifdef CONFIG_CGROUP_RDMA
1508 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1509 #endif
1510 };
1511 
1512 struct ib_ucontext {
1513 	struct ib_device       *device;
1514 	struct ib_uverbs_file  *ufile;
1515 
1516 	struct ib_rdmacg_object	cg_obj;
1517 	/*
1518 	 * Implementation details of the RDMA core, don't use in drivers:
1519 	 */
1520 	struct rdma_restrack_entry res;
1521 	struct xarray mmap_xa;
1522 };
1523 
1524 struct ib_uobject {
1525 	u64			user_handle;	/* handle given to us by userspace */
1526 	/* ufile & ucontext owning this object */
1527 	struct ib_uverbs_file  *ufile;
1528 	/* FIXME, save memory: ufile->context == context */
1529 	struct ib_ucontext     *context;	/* associated user context */
1530 	void		       *object;		/* containing object */
1531 	struct list_head	list;		/* link to context's list */
1532 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1533 	int			id;		/* index into kernel idr */
1534 	struct kref		ref;
1535 	atomic_t		usecnt;		/* protects exclusive access */
1536 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1537 
1538 	const struct uverbs_api_object *uapi_object;
1539 };
1540 
1541 struct ib_udata {
1542 	const void __user *inbuf;
1543 	void __user *outbuf;
1544 	size_t       inlen;
1545 	size_t       outlen;
1546 };
1547 
1548 struct ib_pd {
1549 	u32			local_dma_lkey;
1550 	u32			flags;
1551 	struct ib_device       *device;
1552 	struct ib_uobject      *uobject;
1553 	atomic_t          	usecnt; /* count all resources */
1554 
1555 	u32			unsafe_global_rkey;
1556 
1557 	/*
1558 	 * Implementation details of the RDMA core, don't use in drivers:
1559 	 */
1560 	struct ib_mr	       *__internal_mr;
1561 	struct rdma_restrack_entry res;
1562 };
1563 
1564 struct ib_xrcd {
1565 	struct ib_device       *device;
1566 	atomic_t		usecnt; /* count all exposed resources */
1567 	struct inode	       *inode;
1568 	struct rw_semaphore	tgt_qps_rwsem;
1569 	struct xarray		tgt_qps;
1570 };
1571 
1572 struct ib_ah {
1573 	struct ib_device	*device;
1574 	struct ib_pd		*pd;
1575 	struct ib_uobject	*uobject;
1576 	const struct ib_gid_attr *sgid_attr;
1577 	enum rdma_ah_attr_type	type;
1578 };
1579 
1580 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1581 
1582 enum ib_poll_context {
1583 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1584 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1585 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1586 	IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1587 
1588 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1589 };
1590 
1591 struct ib_cq {
1592 	struct ib_device       *device;
1593 	struct ib_ucq_object   *uobject;
1594 	ib_comp_handler   	comp_handler;
1595 	void                  (*event_handler)(struct ib_event *, void *);
1596 	void                   *cq_context;
1597 	int               	cqe;
1598 	unsigned int		cqe_used;
1599 	atomic_t          	usecnt; /* count number of work queues */
1600 	enum ib_poll_context	poll_ctx;
1601 	struct ib_wc		*wc;
1602 	struct list_head        pool_entry;
1603 	union {
1604 		struct irq_poll		iop;
1605 		struct work_struct	work;
1606 	};
1607 	struct workqueue_struct *comp_wq;
1608 	struct dim *dim;
1609 
1610 	/* updated only by trace points */
1611 	ktime_t timestamp;
1612 	u8 interrupt:1;
1613 	u8 shared:1;
1614 	unsigned int comp_vector;
1615 
1616 	/*
1617 	 * Implementation details of the RDMA core, don't use in drivers:
1618 	 */
1619 	struct rdma_restrack_entry res;
1620 };
1621 
1622 struct ib_srq {
1623 	struct ib_device       *device;
1624 	struct ib_pd	       *pd;
1625 	struct ib_usrq_object  *uobject;
1626 	void		      (*event_handler)(struct ib_event *, void *);
1627 	void		       *srq_context;
1628 	enum ib_srq_type	srq_type;
1629 	atomic_t		usecnt;
1630 
1631 	struct {
1632 		struct ib_cq   *cq;
1633 		union {
1634 			struct {
1635 				struct ib_xrcd *xrcd;
1636 				u32		srq_num;
1637 			} xrc;
1638 		};
1639 	} ext;
1640 
1641 	/*
1642 	 * Implementation details of the RDMA core, don't use in drivers:
1643 	 */
1644 	struct rdma_restrack_entry res;
1645 };
1646 
1647 enum ib_raw_packet_caps {
1648 	/*
1649 	 * Strip cvlan from incoming packet and report it in the matching work
1650 	 * completion is supported.
1651 	 */
1652 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING =
1653 		IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING,
1654 	/*
1655 	 * Scatter FCS field of an incoming packet to host memory is supported.
1656 	 */
1657 	IB_RAW_PACKET_CAP_SCATTER_FCS = IB_UVERBS_RAW_PACKET_CAP_SCATTER_FCS,
1658 	/* Checksum offloads are supported (for both send and receive). */
1659 	IB_RAW_PACKET_CAP_IP_CSUM = IB_UVERBS_RAW_PACKET_CAP_IP_CSUM,
1660 	/*
1661 	 * When a packet is received for an RQ with no receive WQEs, the
1662 	 * packet processing is delayed.
1663 	 */
1664 	IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP,
1665 };
1666 
1667 enum ib_wq_type {
1668 	IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1669 };
1670 
1671 enum ib_wq_state {
1672 	IB_WQS_RESET,
1673 	IB_WQS_RDY,
1674 	IB_WQS_ERR
1675 };
1676 
1677 struct ib_wq {
1678 	struct ib_device       *device;
1679 	struct ib_uwq_object   *uobject;
1680 	void		    *wq_context;
1681 	void		    (*event_handler)(struct ib_event *, void *);
1682 	struct ib_pd	       *pd;
1683 	struct ib_cq	       *cq;
1684 	u32		wq_num;
1685 	enum ib_wq_state       state;
1686 	enum ib_wq_type	wq_type;
1687 	atomic_t		usecnt;
1688 };
1689 
1690 enum ib_wq_flags {
1691 	IB_WQ_FLAGS_CVLAN_STRIPPING	= IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1692 	IB_WQ_FLAGS_SCATTER_FCS		= IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1693 	IB_WQ_FLAGS_DELAY_DROP		= IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1694 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1695 				IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1696 };
1697 
1698 struct ib_wq_init_attr {
1699 	void		       *wq_context;
1700 	enum ib_wq_type	wq_type;
1701 	u32		max_wr;
1702 	u32		max_sge;
1703 	struct	ib_cq	       *cq;
1704 	void		    (*event_handler)(struct ib_event *, void *);
1705 	u32		create_flags; /* Use enum ib_wq_flags */
1706 };
1707 
1708 enum ib_wq_attr_mask {
1709 	IB_WQ_STATE		= 1 << 0,
1710 	IB_WQ_CUR_STATE		= 1 << 1,
1711 	IB_WQ_FLAGS		= 1 << 2,
1712 };
1713 
1714 struct ib_wq_attr {
1715 	enum	ib_wq_state	wq_state;
1716 	enum	ib_wq_state	curr_wq_state;
1717 	u32			flags; /* Use enum ib_wq_flags */
1718 	u32			flags_mask; /* Use enum ib_wq_flags */
1719 };
1720 
1721 struct ib_rwq_ind_table {
1722 	struct ib_device	*device;
1723 	struct ib_uobject      *uobject;
1724 	atomic_t		usecnt;
1725 	u32		ind_tbl_num;
1726 	u32		log_ind_tbl_size;
1727 	struct ib_wq	**ind_tbl;
1728 };
1729 
1730 struct ib_rwq_ind_table_init_attr {
1731 	u32		log_ind_tbl_size;
1732 	/* Each entry is a pointer to Receive Work Queue */
1733 	struct ib_wq	**ind_tbl;
1734 };
1735 
1736 enum port_pkey_state {
1737 	IB_PORT_PKEY_NOT_VALID = 0,
1738 	IB_PORT_PKEY_VALID = 1,
1739 	IB_PORT_PKEY_LISTED = 2,
1740 };
1741 
1742 struct ib_qp_security;
1743 
1744 struct ib_port_pkey {
1745 	enum port_pkey_state	state;
1746 	u16			pkey_index;
1747 	u32			port_num;
1748 	struct list_head	qp_list;
1749 	struct list_head	to_error_list;
1750 	struct ib_qp_security  *sec;
1751 };
1752 
1753 struct ib_ports_pkeys {
1754 	struct ib_port_pkey	main;
1755 	struct ib_port_pkey	alt;
1756 };
1757 
1758 struct ib_qp_security {
1759 	struct ib_qp	       *qp;
1760 	struct ib_device       *dev;
1761 	/* Hold this mutex when changing port and pkey settings. */
1762 	struct mutex		mutex;
1763 	struct ib_ports_pkeys  *ports_pkeys;
1764 	/* A list of all open shared QP handles.  Required to enforce security
1765 	 * properly for all users of a shared QP.
1766 	 */
1767 	struct list_head        shared_qp_list;
1768 	void                   *security;
1769 	bool			destroying;
1770 	atomic_t		error_list_count;
1771 	struct completion	error_complete;
1772 	int			error_comps_pending;
1773 };
1774 
1775 /*
1776  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1777  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1778  */
1779 struct ib_qp {
1780 	struct ib_device       *device;
1781 	struct ib_pd	       *pd;
1782 	struct ib_cq	       *send_cq;
1783 	struct ib_cq	       *recv_cq;
1784 	spinlock_t		mr_lock;
1785 	int			mrs_used;
1786 	struct list_head	rdma_mrs;
1787 	struct list_head	sig_mrs;
1788 	struct ib_srq	       *srq;
1789 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1790 	struct list_head	xrcd_list;
1791 
1792 	/* count times opened, mcast attaches, flow attaches */
1793 	atomic_t		usecnt;
1794 	struct list_head	open_list;
1795 	struct ib_qp           *real_qp;
1796 	struct ib_uqp_object   *uobject;
1797 	void                  (*event_handler)(struct ib_event *, void *);
1798 	void		       *qp_context;
1799 	/* sgid_attrs associated with the AV's */
1800 	const struct ib_gid_attr *av_sgid_attr;
1801 	const struct ib_gid_attr *alt_path_sgid_attr;
1802 	u32			qp_num;
1803 	u32			max_write_sge;
1804 	u32			max_read_sge;
1805 	enum ib_qp_type		qp_type;
1806 	struct ib_rwq_ind_table *rwq_ind_tbl;
1807 	struct ib_qp_security  *qp_sec;
1808 	u32			port;
1809 
1810 	bool			integrity_en;
1811 	/*
1812 	 * Implementation details of the RDMA core, don't use in drivers:
1813 	 */
1814 	struct rdma_restrack_entry     res;
1815 
1816 	/* The counter the qp is bind to */
1817 	struct rdma_counter    *counter;
1818 };
1819 
1820 struct ib_dm {
1821 	struct ib_device  *device;
1822 	u32		   length;
1823 	u32		   flags;
1824 	struct ib_uobject *uobject;
1825 	atomic_t	   usecnt;
1826 };
1827 
1828 struct ib_mr {
1829 	struct ib_device  *device;
1830 	struct ib_pd	  *pd;
1831 	u32		   lkey;
1832 	u32		   rkey;
1833 	u64		   iova;
1834 	u64		   length;
1835 	unsigned int	   page_size;
1836 	enum ib_mr_type	   type;
1837 	bool		   need_inval;
1838 	union {
1839 		struct ib_uobject	*uobject;	/* user */
1840 		struct list_head	qp_entry;	/* FR */
1841 	};
1842 
1843 	struct ib_dm      *dm;
1844 	struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1845 	/*
1846 	 * Implementation details of the RDMA core, don't use in drivers:
1847 	 */
1848 	struct rdma_restrack_entry res;
1849 };
1850 
1851 struct ib_mw {
1852 	struct ib_device	*device;
1853 	struct ib_pd		*pd;
1854 	struct ib_uobject	*uobject;
1855 	u32			rkey;
1856 	enum ib_mw_type         type;
1857 };
1858 
1859 /* Supported steering options */
1860 enum ib_flow_attr_type {
1861 	/* steering according to rule specifications */
1862 	IB_FLOW_ATTR_NORMAL		= 0x0,
1863 	/* default unicast and multicast rule -
1864 	 * receive all Eth traffic which isn't steered to any QP
1865 	 */
1866 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1867 	/* default multicast rule -
1868 	 * receive all Eth multicast traffic which isn't steered to any QP
1869 	 */
1870 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1871 	/* sniffer rule - receive all port traffic */
1872 	IB_FLOW_ATTR_SNIFFER		= 0x3
1873 };
1874 
1875 /* Supported steering header types */
1876 enum ib_flow_spec_type {
1877 	/* L2 headers*/
1878 	IB_FLOW_SPEC_ETH		= 0x20,
1879 	IB_FLOW_SPEC_IB			= 0x22,
1880 	/* L3 header*/
1881 	IB_FLOW_SPEC_IPV4		= 0x30,
1882 	IB_FLOW_SPEC_IPV6		= 0x31,
1883 	IB_FLOW_SPEC_ESP                = 0x34,
1884 	/* L4 headers*/
1885 	IB_FLOW_SPEC_TCP		= 0x40,
1886 	IB_FLOW_SPEC_UDP		= 0x41,
1887 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1888 	IB_FLOW_SPEC_GRE		= 0x51,
1889 	IB_FLOW_SPEC_MPLS		= 0x60,
1890 	IB_FLOW_SPEC_INNER		= 0x100,
1891 	/* Actions */
1892 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1893 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1894 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1895 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1896 };
1897 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1898 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1899 
1900 enum ib_flow_flags {
1901 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1902 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1903 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1904 };
1905 
1906 struct ib_flow_eth_filter {
1907 	u8	dst_mac[6];
1908 	u8	src_mac[6];
1909 	__be16	ether_type;
1910 	__be16	vlan_tag;
1911 	/* Must be last */
1912 	u8	real_sz[];
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 	/* Must be last */
1926 	u8	real_sz[];
1927 };
1928 
1929 struct ib_flow_spec_ib {
1930 	u32			 type;
1931 	u16			 size;
1932 	struct ib_flow_ib_filter val;
1933 	struct ib_flow_ib_filter mask;
1934 };
1935 
1936 /* IPv4 header flags */
1937 enum ib_ipv4_flags {
1938 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1939 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1940 				    last have this flag set */
1941 };
1942 
1943 struct ib_flow_ipv4_filter {
1944 	__be32	src_ip;
1945 	__be32	dst_ip;
1946 	u8	proto;
1947 	u8	tos;
1948 	u8	ttl;
1949 	u8	flags;
1950 	/* Must be last */
1951 	u8	real_sz[];
1952 };
1953 
1954 struct ib_flow_spec_ipv4 {
1955 	u32			   type;
1956 	u16			   size;
1957 	struct ib_flow_ipv4_filter val;
1958 	struct ib_flow_ipv4_filter mask;
1959 };
1960 
1961 struct ib_flow_ipv6_filter {
1962 	u8	src_ip[16];
1963 	u8	dst_ip[16];
1964 	__be32	flow_label;
1965 	u8	next_hdr;
1966 	u8	traffic_class;
1967 	u8	hop_limit;
1968 	/* Must be last */
1969 	u8	real_sz[];
1970 };
1971 
1972 struct ib_flow_spec_ipv6 {
1973 	u32			   type;
1974 	u16			   size;
1975 	struct ib_flow_ipv6_filter val;
1976 	struct ib_flow_ipv6_filter mask;
1977 };
1978 
1979 struct ib_flow_tcp_udp_filter {
1980 	__be16	dst_port;
1981 	__be16	src_port;
1982 	/* Must be last */
1983 	u8	real_sz[];
1984 };
1985 
1986 struct ib_flow_spec_tcp_udp {
1987 	u32			      type;
1988 	u16			      size;
1989 	struct ib_flow_tcp_udp_filter val;
1990 	struct ib_flow_tcp_udp_filter mask;
1991 };
1992 
1993 struct ib_flow_tunnel_filter {
1994 	__be32	tunnel_id;
1995 	u8	real_sz[];
1996 };
1997 
1998 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1999  * the tunnel_id from val has the vni value
2000  */
2001 struct ib_flow_spec_tunnel {
2002 	u32			      type;
2003 	u16			      size;
2004 	struct ib_flow_tunnel_filter  val;
2005 	struct ib_flow_tunnel_filter  mask;
2006 };
2007 
2008 struct ib_flow_esp_filter {
2009 	__be32	spi;
2010 	__be32  seq;
2011 	/* Must be last */
2012 	u8	real_sz[];
2013 };
2014 
2015 struct ib_flow_spec_esp {
2016 	u32                           type;
2017 	u16			      size;
2018 	struct ib_flow_esp_filter     val;
2019 	struct ib_flow_esp_filter     mask;
2020 };
2021 
2022 struct ib_flow_gre_filter {
2023 	__be16 c_ks_res0_ver;
2024 	__be16 protocol;
2025 	__be32 key;
2026 	/* Must be last */
2027 	u8	real_sz[];
2028 };
2029 
2030 struct ib_flow_spec_gre {
2031 	u32                           type;
2032 	u16			      size;
2033 	struct ib_flow_gre_filter     val;
2034 	struct ib_flow_gre_filter     mask;
2035 };
2036 
2037 struct ib_flow_mpls_filter {
2038 	__be32 tag;
2039 	/* Must be last */
2040 	u8	real_sz[];
2041 };
2042 
2043 struct ib_flow_spec_mpls {
2044 	u32                           type;
2045 	u16			      size;
2046 	struct ib_flow_mpls_filter     val;
2047 	struct ib_flow_mpls_filter     mask;
2048 };
2049 
2050 struct ib_flow_spec_action_tag {
2051 	enum ib_flow_spec_type	      type;
2052 	u16			      size;
2053 	u32                           tag_id;
2054 };
2055 
2056 struct ib_flow_spec_action_drop {
2057 	enum ib_flow_spec_type	      type;
2058 	u16			      size;
2059 };
2060 
2061 struct ib_flow_spec_action_handle {
2062 	enum ib_flow_spec_type	      type;
2063 	u16			      size;
2064 	struct ib_flow_action	     *act;
2065 };
2066 
2067 enum ib_counters_description {
2068 	IB_COUNTER_PACKETS,
2069 	IB_COUNTER_BYTES,
2070 };
2071 
2072 struct ib_flow_spec_action_count {
2073 	enum ib_flow_spec_type type;
2074 	u16 size;
2075 	struct ib_counters *counters;
2076 };
2077 
2078 union ib_flow_spec {
2079 	struct {
2080 		u32			type;
2081 		u16			size;
2082 	};
2083 	struct ib_flow_spec_eth		eth;
2084 	struct ib_flow_spec_ib		ib;
2085 	struct ib_flow_spec_ipv4        ipv4;
2086 	struct ib_flow_spec_tcp_udp	tcp_udp;
2087 	struct ib_flow_spec_ipv6        ipv6;
2088 	struct ib_flow_spec_tunnel      tunnel;
2089 	struct ib_flow_spec_esp		esp;
2090 	struct ib_flow_spec_gre		gre;
2091 	struct ib_flow_spec_mpls	mpls;
2092 	struct ib_flow_spec_action_tag  flow_tag;
2093 	struct ib_flow_spec_action_drop drop;
2094 	struct ib_flow_spec_action_handle action;
2095 	struct ib_flow_spec_action_count flow_count;
2096 };
2097 
2098 struct ib_flow_attr {
2099 	enum ib_flow_attr_type type;
2100 	u16	     size;
2101 	u16	     priority;
2102 	u32	     flags;
2103 	u8	     num_of_specs;
2104 	u32	     port;
2105 	union ib_flow_spec flows[];
2106 };
2107 
2108 struct ib_flow {
2109 	struct ib_qp		*qp;
2110 	struct ib_device	*device;
2111 	struct ib_uobject	*uobject;
2112 };
2113 
2114 enum ib_flow_action_type {
2115 	IB_FLOW_ACTION_UNSPECIFIED,
2116 	IB_FLOW_ACTION_ESP = 1,
2117 };
2118 
2119 struct ib_flow_action_attrs_esp_keymats {
2120 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2121 	union {
2122 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2123 	} keymat;
2124 };
2125 
2126 struct ib_flow_action_attrs_esp_replays {
2127 	enum ib_uverbs_flow_action_esp_replay			protocol;
2128 	union {
2129 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2130 	} replay;
2131 };
2132 
2133 enum ib_flow_action_attrs_esp_flags {
2134 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2135 	 * This is done in order to share the same flags between user-space and
2136 	 * kernel and spare an unnecessary translation.
2137 	 */
2138 
2139 	/* Kernel flags */
2140 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2141 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2142 };
2143 
2144 struct ib_flow_spec_list {
2145 	struct ib_flow_spec_list	*next;
2146 	union ib_flow_spec		spec;
2147 };
2148 
2149 struct ib_flow_action_attrs_esp {
2150 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2151 	struct ib_flow_action_attrs_esp_replays		*replay;
2152 	struct ib_flow_spec_list			*encap;
2153 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2154 	 * Value of 0 is a valid value.
2155 	 */
2156 	u32						esn;
2157 	u32						spi;
2158 	u32						seq;
2159 	u32						tfc_pad;
2160 	/* Use enum ib_flow_action_attrs_esp_flags */
2161 	u64						flags;
2162 	u64						hard_limit_pkts;
2163 };
2164 
2165 struct ib_flow_action {
2166 	struct ib_device		*device;
2167 	struct ib_uobject		*uobject;
2168 	enum ib_flow_action_type	type;
2169 	atomic_t			usecnt;
2170 };
2171 
2172 struct ib_mad;
2173 
2174 enum ib_process_mad_flags {
2175 	IB_MAD_IGNORE_MKEY	= 1,
2176 	IB_MAD_IGNORE_BKEY	= 2,
2177 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2178 };
2179 
2180 enum ib_mad_result {
2181 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2182 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2183 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2184 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2185 };
2186 
2187 struct ib_port_cache {
2188 	u64		      subnet_prefix;
2189 	struct ib_pkey_cache  *pkey;
2190 	struct ib_gid_table   *gid;
2191 	u8                     lmc;
2192 	enum ib_port_state     port_state;
2193 };
2194 
2195 struct ib_port_immutable {
2196 	int                           pkey_tbl_len;
2197 	int                           gid_tbl_len;
2198 	u32                           core_cap_flags;
2199 	u32                           max_mad_size;
2200 };
2201 
2202 struct ib_port_data {
2203 	struct ib_device *ib_dev;
2204 
2205 	struct ib_port_immutable immutable;
2206 
2207 	spinlock_t pkey_list_lock;
2208 
2209 	spinlock_t netdev_lock;
2210 
2211 	struct list_head pkey_list;
2212 
2213 	struct ib_port_cache cache;
2214 
2215 	struct net_device __rcu *netdev;
2216 	netdevice_tracker netdev_tracker;
2217 	struct hlist_node ndev_hash_link;
2218 	struct rdma_port_counter port_counter;
2219 	struct ib_port *sysfs;
2220 };
2221 
2222 /* rdma netdev type - specifies protocol type */
2223 enum rdma_netdev_t {
2224 	RDMA_NETDEV_OPA_VNIC,
2225 	RDMA_NETDEV_IPOIB,
2226 };
2227 
2228 /**
2229  * struct rdma_netdev - rdma netdev
2230  * For cases where netstack interfacing is required.
2231  */
2232 struct rdma_netdev {
2233 	void              *clnt_priv;
2234 	struct ib_device  *hca;
2235 	u32		   port_num;
2236 	int                mtu;
2237 
2238 	/*
2239 	 * cleanup function must be specified.
2240 	 * FIXME: This is only used for OPA_VNIC and that usage should be
2241 	 * removed too.
2242 	 */
2243 	void (*free_rdma_netdev)(struct net_device *netdev);
2244 
2245 	/* control functions */
2246 	void (*set_id)(struct net_device *netdev, int id);
2247 	/* send packet */
2248 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2249 		    struct ib_ah *address, u32 dqpn);
2250 	/* multicast */
2251 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2252 			    union ib_gid *gid, u16 mlid,
2253 			    int set_qkey, u32 qkey);
2254 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2255 			    union ib_gid *gid, u16 mlid);
2256 	/* timeout */
2257 	void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2258 };
2259 
2260 struct rdma_netdev_alloc_params {
2261 	size_t sizeof_priv;
2262 	unsigned int txqs;
2263 	unsigned int rxqs;
2264 	void *param;
2265 
2266 	int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2267 				      struct net_device *netdev, void *param);
2268 };
2269 
2270 struct ib_odp_counters {
2271 	atomic64_t faults;
2272 	atomic64_t invalidations;
2273 	atomic64_t prefetch;
2274 };
2275 
2276 struct ib_counters {
2277 	struct ib_device	*device;
2278 	struct ib_uobject	*uobject;
2279 	/* num of objects attached */
2280 	atomic_t	usecnt;
2281 };
2282 
2283 struct ib_counters_read_attr {
2284 	u64	*counters_buff;
2285 	u32	ncounters;
2286 	u32	flags; /* use enum ib_read_counters_flags */
2287 };
2288 
2289 struct uverbs_attr_bundle;
2290 struct iw_cm_id;
2291 struct iw_cm_conn_param;
2292 
2293 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member)                      \
2294 	.size_##ib_struct =                                                    \
2295 		(sizeof(struct drv_struct) +                                   \
2296 		 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) +      \
2297 		 BUILD_BUG_ON_ZERO(                                            \
2298 			 !__same_type(((struct drv_struct *)NULL)->member,     \
2299 				      struct ib_struct)))
2300 
2301 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp)                          \
2302 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2303 					   gfp, false))
2304 
2305 #define rdma_zalloc_drv_obj_numa(ib_dev, ib_type)                              \
2306 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2307 					   GFP_KERNEL, true))
2308 
2309 #define rdma_zalloc_drv_obj(ib_dev, ib_type)                                   \
2310 	rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2311 
2312 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2313 
2314 struct rdma_user_mmap_entry {
2315 	struct kref ref;
2316 	struct ib_ucontext *ucontext;
2317 	unsigned long start_pgoff;
2318 	size_t npages;
2319 	bool driver_removed;
2320 };
2321 
2322 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2323 static inline u64
2324 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2325 {
2326 	return (u64)entry->start_pgoff << PAGE_SHIFT;
2327 }
2328 
2329 /**
2330  * struct ib_device_ops - InfiniBand device operations
2331  * This structure defines all the InfiniBand device operations, providers will
2332  * need to define the supported operations, otherwise they will be set to null.
2333  */
2334 struct ib_device_ops {
2335 	struct module *owner;
2336 	enum rdma_driver_id driver_id;
2337 	u32 uverbs_abi_ver;
2338 	unsigned int uverbs_no_driver_id_binding:1;
2339 
2340 	/*
2341 	 * NOTE: New drivers should not make use of device_group; instead new
2342 	 * device parameter should be exposed via netlink command. This
2343 	 * mechanism exists only for existing drivers.
2344 	 */
2345 	const struct attribute_group *device_group;
2346 	const struct attribute_group **port_groups;
2347 
2348 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2349 			 const struct ib_send_wr **bad_send_wr);
2350 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2351 			 const struct ib_recv_wr **bad_recv_wr);
2352 	void (*drain_rq)(struct ib_qp *qp);
2353 	void (*drain_sq)(struct ib_qp *qp);
2354 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2355 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2356 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2357 	int (*post_srq_recv)(struct ib_srq *srq,
2358 			     const struct ib_recv_wr *recv_wr,
2359 			     const struct ib_recv_wr **bad_recv_wr);
2360 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2361 			   u32 port_num, const struct ib_wc *in_wc,
2362 			   const struct ib_grh *in_grh,
2363 			   const struct ib_mad *in_mad, struct ib_mad *out_mad,
2364 			   size_t *out_mad_size, u16 *out_mad_pkey_index);
2365 	int (*query_device)(struct ib_device *device,
2366 			    struct ib_device_attr *device_attr,
2367 			    struct ib_udata *udata);
2368 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2369 			     struct ib_device_modify *device_modify);
2370 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2371 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2372 						     int comp_vector);
2373 	int (*query_port)(struct ib_device *device, u32 port_num,
2374 			  struct ib_port_attr *port_attr);
2375 	int (*modify_port)(struct ib_device *device, u32 port_num,
2376 			   int port_modify_mask,
2377 			   struct ib_port_modify *port_modify);
2378 	/**
2379 	 * The following mandatory functions are used only at device
2380 	 * registration.  Keep functions such as these at the end of this
2381 	 * structure to avoid cache line misses when accessing struct ib_device
2382 	 * in fast paths.
2383 	 */
2384 	int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2385 				  struct ib_port_immutable *immutable);
2386 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2387 					       u32 port_num);
2388 	/**
2389 	 * When calling get_netdev, the HW vendor's driver should return the
2390 	 * net device of device @device at port @port_num or NULL if such
2391 	 * a net device doesn't exist. The vendor driver should call dev_hold
2392 	 * on this net device. The HW vendor's device driver must guarantee
2393 	 * that this function returns NULL before the net device has finished
2394 	 * NETDEV_UNREGISTER state.
2395 	 */
2396 	struct net_device *(*get_netdev)(struct ib_device *device,
2397 					 u32 port_num);
2398 	/**
2399 	 * rdma netdev operation
2400 	 *
2401 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2402 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2403 	 */
2404 	struct net_device *(*alloc_rdma_netdev)(
2405 		struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2406 		const char *name, unsigned char name_assign_type,
2407 		void (*setup)(struct net_device *));
2408 
2409 	int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2410 				      enum rdma_netdev_t type,
2411 				      struct rdma_netdev_alloc_params *params);
2412 	/**
2413 	 * query_gid should be return GID value for @device, when @port_num
2414 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2415 	 * is RoCE link layer.
2416 	 */
2417 	int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2418 			 union ib_gid *gid);
2419 	/**
2420 	 * When calling add_gid, the HW vendor's driver should add the gid
2421 	 * of device of port at gid index available at @attr. Meta-info of
2422 	 * that gid (for example, the network device related to this gid) is
2423 	 * available at @attr. @context allows the HW vendor driver to store
2424 	 * extra information together with a GID entry. The HW vendor driver may
2425 	 * allocate memory to contain this information and store it in @context
2426 	 * when a new GID entry is written to. Params are consistent until the
2427 	 * next call of add_gid or delete_gid. The function should return 0 on
2428 	 * success or error otherwise. The function could be called
2429 	 * concurrently for different ports. This function is only called when
2430 	 * roce_gid_table is used.
2431 	 */
2432 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2433 	/**
2434 	 * When calling del_gid, the HW vendor's driver should delete the
2435 	 * gid of device @device at gid index gid_index of port port_num
2436 	 * available in @attr.
2437 	 * Upon the deletion of a GID entry, the HW vendor must free any
2438 	 * allocated memory. The caller will clear @context afterwards.
2439 	 * This function is only called when roce_gid_table is used.
2440 	 */
2441 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2442 	int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2443 			  u16 *pkey);
2444 	int (*alloc_ucontext)(struct ib_ucontext *context,
2445 			      struct ib_udata *udata);
2446 	void (*dealloc_ucontext)(struct ib_ucontext *context);
2447 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2448 	/**
2449 	 * This will be called once refcount of an entry in mmap_xa reaches
2450 	 * zero. The type of the memory that was mapped may differ between
2451 	 * entries and is opaque to the rdma_user_mmap interface.
2452 	 * Therefore needs to be implemented by the driver in mmap_free.
2453 	 */
2454 	void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2455 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2456 	int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2457 	int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2458 	int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2459 			 struct ib_udata *udata);
2460 	int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2461 			      struct ib_udata *udata);
2462 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2463 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2464 	int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2465 	int (*create_srq)(struct ib_srq *srq,
2466 			  struct ib_srq_init_attr *srq_init_attr,
2467 			  struct ib_udata *udata);
2468 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2469 			  enum ib_srq_attr_mask srq_attr_mask,
2470 			  struct ib_udata *udata);
2471 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2472 	int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2473 	int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2474 			 struct ib_udata *udata);
2475 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2476 			 int qp_attr_mask, struct ib_udata *udata);
2477 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2478 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2479 	int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2480 	int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2481 			 struct ib_udata *udata);
2482 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2483 	int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2484 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2485 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2486 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2487 				     u64 virt_addr, int mr_access_flags,
2488 				     struct ib_udata *udata);
2489 	struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2490 					    u64 length, u64 virt_addr, int fd,
2491 					    int mr_access_flags,
2492 					    struct ib_udata *udata);
2493 	struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2494 				       u64 length, u64 virt_addr,
2495 				       int mr_access_flags, struct ib_pd *pd,
2496 				       struct ib_udata *udata);
2497 	int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2498 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2499 				  u32 max_num_sg);
2500 	struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2501 					    u32 max_num_data_sg,
2502 					    u32 max_num_meta_sg);
2503 	int (*advise_mr)(struct ib_pd *pd,
2504 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2505 			 struct ib_sge *sg_list, u32 num_sge,
2506 			 struct uverbs_attr_bundle *attrs);
2507 
2508 	/*
2509 	 * Kernel users should universally support relaxed ordering (RO), as
2510 	 * they are designed to read data only after observing the CQE and use
2511 	 * the DMA API correctly.
2512 	 *
2513 	 * Some drivers implicitly enable RO if platform supports it.
2514 	 */
2515 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2516 			 unsigned int *sg_offset);
2517 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2518 			       struct ib_mr_status *mr_status);
2519 	int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2520 	int (*dealloc_mw)(struct ib_mw *mw);
2521 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2522 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2523 	int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2524 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2525 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2526 				       struct ib_flow_attr *flow_attr,
2527 				       struct ib_udata *udata);
2528 	int (*destroy_flow)(struct ib_flow *flow_id);
2529 	int (*destroy_flow_action)(struct ib_flow_action *action);
2530 	int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2531 				 int state);
2532 	int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2533 			     struct ifla_vf_info *ivf);
2534 	int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2535 			    struct ifla_vf_stats *stats);
2536 	int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2537 			    struct ifla_vf_guid *node_guid,
2538 			    struct ifla_vf_guid *port_guid);
2539 	int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2540 			   int type);
2541 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2542 				   struct ib_wq_init_attr *init_attr,
2543 				   struct ib_udata *udata);
2544 	int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2545 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2546 			 u32 wq_attr_mask, struct ib_udata *udata);
2547 	int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2548 				    struct ib_rwq_ind_table_init_attr *init_attr,
2549 				    struct ib_udata *udata);
2550 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2551 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2552 				  struct ib_ucontext *context,
2553 				  struct ib_dm_alloc_attr *attr,
2554 				  struct uverbs_attr_bundle *attrs);
2555 	int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2556 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2557 				   struct ib_dm_mr_attr *attr,
2558 				   struct uverbs_attr_bundle *attrs);
2559 	int (*create_counters)(struct ib_counters *counters,
2560 			       struct uverbs_attr_bundle *attrs);
2561 	int (*destroy_counters)(struct ib_counters *counters);
2562 	int (*read_counters)(struct ib_counters *counters,
2563 			     struct ib_counters_read_attr *counters_read_attr,
2564 			     struct uverbs_attr_bundle *attrs);
2565 	int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2566 			    int data_sg_nents, unsigned int *data_sg_offset,
2567 			    struct scatterlist *meta_sg, int meta_sg_nents,
2568 			    unsigned int *meta_sg_offset);
2569 
2570 	/**
2571 	 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2572 	 *   fill in the driver initialized data.  The struct is kfree()'ed by
2573 	 *   the sysfs core when the device is removed.  A lifespan of -1 in the
2574 	 *   return struct tells the core to set a default lifespan.
2575 	 */
2576 	struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2577 	struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2578 						     u32 port_num);
2579 	/**
2580 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2581 	 * @index - The index in the value array we wish to have updated, or
2582 	 *   num_counters if we want all stats updated
2583 	 * Return codes -
2584 	 *   < 0 - Error, no counters updated
2585 	 *   index - Updated the single counter pointed to by index
2586 	 *   num_counters - Updated all counters (will reset the timestamp
2587 	 *     and prevent further calls for lifespan milliseconds)
2588 	 * Drivers are allowed to update all counters in leiu of just the
2589 	 *   one given in index at their option
2590 	 */
2591 	int (*get_hw_stats)(struct ib_device *device,
2592 			    struct rdma_hw_stats *stats, u32 port, int index);
2593 
2594 	/**
2595 	 * modify_hw_stat - Modify the counter configuration
2596 	 * @enable: true/false when enable/disable a counter
2597 	 * Return codes - 0 on success or error code otherwise.
2598 	 */
2599 	int (*modify_hw_stat)(struct ib_device *device, u32 port,
2600 			      unsigned int counter_index, bool enable);
2601 	/**
2602 	 * Allows rdma drivers to add their own restrack attributes.
2603 	 */
2604 	int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2605 	int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2606 	int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2607 	int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2608 	int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2609 	int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2610 	int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2611 
2612 	/* Device lifecycle callbacks */
2613 	/*
2614 	 * Called after the device becomes registered, before clients are
2615 	 * attached
2616 	 */
2617 	int (*enable_driver)(struct ib_device *dev);
2618 	/*
2619 	 * This is called as part of ib_dealloc_device().
2620 	 */
2621 	void (*dealloc_driver)(struct ib_device *dev);
2622 
2623 	/* iWarp CM callbacks */
2624 	void (*iw_add_ref)(struct ib_qp *qp);
2625 	void (*iw_rem_ref)(struct ib_qp *qp);
2626 	struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2627 	int (*iw_connect)(struct iw_cm_id *cm_id,
2628 			  struct iw_cm_conn_param *conn_param);
2629 	int (*iw_accept)(struct iw_cm_id *cm_id,
2630 			 struct iw_cm_conn_param *conn_param);
2631 	int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2632 			 u8 pdata_len);
2633 	int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2634 	int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2635 	/**
2636 	 * counter_bind_qp - Bind a QP to a counter.
2637 	 * @counter - The counter to be bound. If counter->id is zero then
2638 	 *   the driver needs to allocate a new counter and set counter->id
2639 	 */
2640 	int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2641 	/**
2642 	 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2643 	 *   counter and bind it onto the default one
2644 	 */
2645 	int (*counter_unbind_qp)(struct ib_qp *qp);
2646 	/**
2647 	 * counter_dealloc -De-allocate the hw counter
2648 	 */
2649 	int (*counter_dealloc)(struct rdma_counter *counter);
2650 	/**
2651 	 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2652 	 * the driver initialized data.
2653 	 */
2654 	struct rdma_hw_stats *(*counter_alloc_stats)(
2655 		struct rdma_counter *counter);
2656 	/**
2657 	 * counter_update_stats - Query the stats value of this counter
2658 	 */
2659 	int (*counter_update_stats)(struct rdma_counter *counter);
2660 
2661 	/**
2662 	 * Allows rdma drivers to add their own restrack attributes
2663 	 * dumped via 'rdma stat' iproute2 command.
2664 	 */
2665 	int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2666 
2667 	/* query driver for its ucontext properties */
2668 	int (*query_ucontext)(struct ib_ucontext *context,
2669 			      struct uverbs_attr_bundle *attrs);
2670 
2671 	/*
2672 	 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2673 	 * Everyone else relies on Linux memory management model.
2674 	 */
2675 	int (*get_numa_node)(struct ib_device *dev);
2676 
2677 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
2678 	DECLARE_RDMA_OBJ_SIZE(ib_counters);
2679 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
2680 	DECLARE_RDMA_OBJ_SIZE(ib_mw);
2681 	DECLARE_RDMA_OBJ_SIZE(ib_pd);
2682 	DECLARE_RDMA_OBJ_SIZE(ib_qp);
2683 	DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2684 	DECLARE_RDMA_OBJ_SIZE(ib_srq);
2685 	DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2686 	DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2687 };
2688 
2689 struct ib_core_device {
2690 	/* device must be the first element in structure until,
2691 	 * union of ib_core_device and device exists in ib_device.
2692 	 */
2693 	struct device dev;
2694 	possible_net_t rdma_net;
2695 	struct kobject *ports_kobj;
2696 	struct list_head port_list;
2697 	struct ib_device *owner; /* reach back to owner ib_device */
2698 };
2699 
2700 struct rdma_restrack_root;
2701 struct ib_device {
2702 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2703 	struct device                *dma_device;
2704 	struct ib_device_ops	     ops;
2705 	char                          name[IB_DEVICE_NAME_MAX];
2706 	struct rcu_head rcu_head;
2707 
2708 	struct list_head              event_handler_list;
2709 	/* Protects event_handler_list */
2710 	struct rw_semaphore event_handler_rwsem;
2711 
2712 	/* Protects QP's event_handler calls and open_qp list */
2713 	spinlock_t qp_open_list_lock;
2714 
2715 	struct rw_semaphore	      client_data_rwsem;
2716 	struct xarray                 client_data;
2717 	struct mutex                  unregistration_lock;
2718 
2719 	/* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2720 	rwlock_t cache_lock;
2721 	/**
2722 	 * port_data is indexed by port number
2723 	 */
2724 	struct ib_port_data *port_data;
2725 
2726 	int			      num_comp_vectors;
2727 
2728 	union {
2729 		struct device		dev;
2730 		struct ib_core_device	coredev;
2731 	};
2732 
2733 	/* First group is for device attributes,
2734 	 * Second group is for driver provided attributes (optional).
2735 	 * Third group is for the hw_stats
2736 	 * It is a NULL terminated array.
2737 	 */
2738 	const struct attribute_group	*groups[4];
2739 
2740 	u64			     uverbs_cmd_mask;
2741 
2742 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2743 	__be64			     node_guid;
2744 	u32			     local_dma_lkey;
2745 	u16                          is_switch:1;
2746 	/* Indicates kernel verbs support, should not be used in drivers */
2747 	u16                          kverbs_provider:1;
2748 	/* CQ adaptive moderation (RDMA DIM) */
2749 	u16                          use_cq_dim:1;
2750 	u8                           node_type;
2751 	u32			     phys_port_cnt;
2752 	struct ib_device_attr        attrs;
2753 	struct hw_stats_device_data *hw_stats_data;
2754 
2755 #ifdef CONFIG_CGROUP_RDMA
2756 	struct rdmacg_device         cg_device;
2757 #endif
2758 
2759 	u32                          index;
2760 
2761 	spinlock_t                   cq_pools_lock;
2762 	struct list_head             cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2763 
2764 	struct rdma_restrack_root *res;
2765 
2766 	const struct uapi_definition   *driver_def;
2767 
2768 	/*
2769 	 * Positive refcount indicates that the device is currently
2770 	 * registered and cannot be unregistered.
2771 	 */
2772 	refcount_t refcount;
2773 	struct completion unreg_completion;
2774 	struct work_struct unregistration_work;
2775 
2776 	const struct rdma_link_ops *link_ops;
2777 
2778 	/* Protects compat_devs xarray modifications */
2779 	struct mutex compat_devs_mutex;
2780 	/* Maintains compat devices for each net namespace */
2781 	struct xarray compat_devs;
2782 
2783 	/* Used by iWarp CM */
2784 	char iw_ifname[IFNAMSIZ];
2785 	u32 iw_driver_flags;
2786 	u32 lag_flags;
2787 };
2788 
2789 static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
2790 				    gfp_t gfp, bool is_numa_aware)
2791 {
2792 	if (is_numa_aware && dev->ops.get_numa_node)
2793 		return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
2794 
2795 	return kzalloc(size, gfp);
2796 }
2797 
2798 struct ib_client_nl_info;
2799 struct ib_client {
2800 	const char *name;
2801 	int (*add)(struct ib_device *ibdev);
2802 	void (*remove)(struct ib_device *, void *client_data);
2803 	void (*rename)(struct ib_device *dev, void *client_data);
2804 	int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2805 			   struct ib_client_nl_info *res);
2806 	int (*get_global_nl_info)(struct ib_client_nl_info *res);
2807 
2808 	/* Returns the net_dev belonging to this ib_client and matching the
2809 	 * given parameters.
2810 	 * @dev:	 An RDMA device that the net_dev use for communication.
2811 	 * @port:	 A physical port number on the RDMA device.
2812 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2813 	 * @gid:	 A GID that the net_dev uses to communicate.
2814 	 * @addr:	 An IP address the net_dev is configured with.
2815 	 * @client_data: The device's client data set by ib_set_client_data().
2816 	 *
2817 	 * An ib_client that implements a net_dev on top of RDMA devices
2818 	 * (such as IP over IB) should implement this callback, allowing the
2819 	 * rdma_cm module to find the right net_dev for a given request.
2820 	 *
2821 	 * The caller is responsible for calling dev_put on the returned
2822 	 * netdev. */
2823 	struct net_device *(*get_net_dev_by_params)(
2824 			struct ib_device *dev,
2825 			u32 port,
2826 			u16 pkey,
2827 			const union ib_gid *gid,
2828 			const struct sockaddr *addr,
2829 			void *client_data);
2830 
2831 	refcount_t uses;
2832 	struct completion uses_zero;
2833 	u32 client_id;
2834 
2835 	/* kverbs are not required by the client */
2836 	u8 no_kverbs_req:1;
2837 };
2838 
2839 /*
2840  * IB block DMA iterator
2841  *
2842  * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2843  * to a HW supported page size.
2844  */
2845 struct ib_block_iter {
2846 	/* internal states */
2847 	struct scatterlist *__sg;	/* sg holding the current aligned block */
2848 	dma_addr_t __dma_addr;		/* unaligned DMA address of this block */
2849 	unsigned int __sg_nents;	/* number of SG entries */
2850 	unsigned int __sg_advance;	/* number of bytes to advance in sg in next step */
2851 	unsigned int __pg_bit;		/* alignment of current block */
2852 };
2853 
2854 struct ib_device *_ib_alloc_device(size_t size);
2855 #define ib_alloc_device(drv_struct, member)                                    \
2856 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2857 				      BUILD_BUG_ON_ZERO(offsetof(              \
2858 					      struct drv_struct, member))),    \
2859 		     struct drv_struct, member)
2860 
2861 void ib_dealloc_device(struct ib_device *device);
2862 
2863 void ib_get_device_fw_str(struct ib_device *device, char *str);
2864 
2865 int ib_register_device(struct ib_device *device, const char *name,
2866 		       struct device *dma_device);
2867 void ib_unregister_device(struct ib_device *device);
2868 void ib_unregister_driver(enum rdma_driver_id driver_id);
2869 void ib_unregister_device_and_put(struct ib_device *device);
2870 void ib_unregister_device_queued(struct ib_device *ib_dev);
2871 
2872 int ib_register_client   (struct ib_client *client);
2873 void ib_unregister_client(struct ib_client *client);
2874 
2875 void __rdma_block_iter_start(struct ib_block_iter *biter,
2876 			     struct scatterlist *sglist,
2877 			     unsigned int nents,
2878 			     unsigned long pgsz);
2879 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2880 
2881 /**
2882  * rdma_block_iter_dma_address - get the aligned dma address of the current
2883  * block held by the block iterator.
2884  * @biter: block iterator holding the memory block
2885  */
2886 static inline dma_addr_t
2887 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2888 {
2889 	return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2890 }
2891 
2892 /**
2893  * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2894  * @sglist: sglist to iterate over
2895  * @biter: block iterator holding the memory block
2896  * @nents: maximum number of sg entries to iterate over
2897  * @pgsz: best HW supported page size to use
2898  *
2899  * Callers may use rdma_block_iter_dma_address() to get each
2900  * blocks aligned DMA address.
2901  */
2902 #define rdma_for_each_block(sglist, biter, nents, pgsz)		\
2903 	for (__rdma_block_iter_start(biter, sglist, nents,	\
2904 				     pgsz);			\
2905 	     __rdma_block_iter_next(biter);)
2906 
2907 /**
2908  * ib_get_client_data - Get IB client context
2909  * @device:Device to get context for
2910  * @client:Client to get context for
2911  *
2912  * ib_get_client_data() returns the client context data set with
2913  * ib_set_client_data(). This can only be called while the client is
2914  * registered to the device, once the ib_client remove() callback returns this
2915  * cannot be called.
2916  */
2917 static inline void *ib_get_client_data(struct ib_device *device,
2918 				       struct ib_client *client)
2919 {
2920 	return xa_load(&device->client_data, client->client_id);
2921 }
2922 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2923 			 void *data);
2924 void ib_set_device_ops(struct ib_device *device,
2925 		       const struct ib_device_ops *ops);
2926 
2927 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2928 		      unsigned long pfn, unsigned long size, pgprot_t prot,
2929 		      struct rdma_user_mmap_entry *entry);
2930 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2931 				struct rdma_user_mmap_entry *entry,
2932 				size_t length);
2933 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2934 				      struct rdma_user_mmap_entry *entry,
2935 				      size_t length, u32 min_pgoff,
2936 				      u32 max_pgoff);
2937 
2938 static inline int
2939 rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
2940 				  struct rdma_user_mmap_entry *entry,
2941 				  size_t length, u32 pgoff)
2942 {
2943 	return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff,
2944 						 pgoff);
2945 }
2946 
2947 struct rdma_user_mmap_entry *
2948 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2949 			       unsigned long pgoff);
2950 struct rdma_user_mmap_entry *
2951 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2952 			 struct vm_area_struct *vma);
2953 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2954 
2955 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2956 
2957 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2958 {
2959 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2960 }
2961 
2962 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2963 {
2964 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2965 }
2966 
2967 static inline bool ib_is_buffer_cleared(const void __user *p,
2968 					size_t len)
2969 {
2970 	bool ret;
2971 	u8 *buf;
2972 
2973 	if (len > USHRT_MAX)
2974 		return false;
2975 
2976 	buf = memdup_user(p, len);
2977 	if (IS_ERR(buf))
2978 		return false;
2979 
2980 	ret = !memchr_inv(buf, 0, len);
2981 	kfree(buf);
2982 	return ret;
2983 }
2984 
2985 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2986 				       size_t offset,
2987 				       size_t len)
2988 {
2989 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
2990 }
2991 
2992 /**
2993  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2994  * contains all required attributes and no attributes not allowed for
2995  * the given QP state transition.
2996  * @cur_state: Current QP state
2997  * @next_state: Next QP state
2998  * @type: QP type
2999  * @mask: Mask of supplied QP attributes
3000  *
3001  * This function is a helper function that a low-level driver's
3002  * modify_qp method can use to validate the consumer's input.  It
3003  * checks that cur_state and next_state are valid QP states, that a
3004  * transition from cur_state to next_state is allowed by the IB spec,
3005  * and that the attribute mask supplied is allowed for the transition.
3006  */
3007 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
3008 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
3009 
3010 void ib_register_event_handler(struct ib_event_handler *event_handler);
3011 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
3012 void ib_dispatch_event(const struct ib_event *event);
3013 
3014 int ib_query_port(struct ib_device *device,
3015 		  u32 port_num, struct ib_port_attr *port_attr);
3016 
3017 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3018 					       u32 port_num);
3019 
3020 /**
3021  * rdma_cap_ib_switch - Check if the device is IB switch
3022  * @device: Device to check
3023  *
3024  * Device driver is responsible for setting is_switch bit on
3025  * in ib_device structure at init time.
3026  *
3027  * Return: true if the device is IB switch.
3028  */
3029 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3030 {
3031 	return device->is_switch;
3032 }
3033 
3034 /**
3035  * rdma_start_port - Return the first valid port number for the device
3036  * specified
3037  *
3038  * @device: Device to be checked
3039  *
3040  * Return start port number
3041  */
3042 static inline u32 rdma_start_port(const struct ib_device *device)
3043 {
3044 	return rdma_cap_ib_switch(device) ? 0 : 1;
3045 }
3046 
3047 /**
3048  * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3049  * @device - The struct ib_device * to iterate over
3050  * @iter - The unsigned int to store the port number
3051  */
3052 #define rdma_for_each_port(device, iter)                                       \
3053 	for (iter = rdma_start_port(device +				       \
3054 				    BUILD_BUG_ON_ZERO(!__same_type(u32,	       \
3055 								   iter)));    \
3056 	     iter <= rdma_end_port(device); iter++)
3057 
3058 /**
3059  * rdma_end_port - Return the last valid port number for the device
3060  * specified
3061  *
3062  * @device: Device to be checked
3063  *
3064  * Return last port number
3065  */
3066 static inline u32 rdma_end_port(const struct ib_device *device)
3067 {
3068 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3069 }
3070 
3071 static inline int rdma_is_port_valid(const struct ib_device *device,
3072 				     unsigned int port)
3073 {
3074 	return (port >= rdma_start_port(device) &&
3075 		port <= rdma_end_port(device));
3076 }
3077 
3078 static inline bool rdma_is_grh_required(const struct ib_device *device,
3079 					u32 port_num)
3080 {
3081 	return device->port_data[port_num].immutable.core_cap_flags &
3082 	       RDMA_CORE_PORT_IB_GRH_REQUIRED;
3083 }
3084 
3085 static inline bool rdma_protocol_ib(const struct ib_device *device,
3086 				    u32 port_num)
3087 {
3088 	return device->port_data[port_num].immutable.core_cap_flags &
3089 	       RDMA_CORE_CAP_PROT_IB;
3090 }
3091 
3092 static inline bool rdma_protocol_roce(const struct ib_device *device,
3093 				      u32 port_num)
3094 {
3095 	return device->port_data[port_num].immutable.core_cap_flags &
3096 	       (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3097 }
3098 
3099 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3100 						u32 port_num)
3101 {
3102 	return device->port_data[port_num].immutable.core_cap_flags &
3103 	       RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3104 }
3105 
3106 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3107 						u32 port_num)
3108 {
3109 	return device->port_data[port_num].immutable.core_cap_flags &
3110 	       RDMA_CORE_CAP_PROT_ROCE;
3111 }
3112 
3113 static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3114 				       u32 port_num)
3115 {
3116 	return device->port_data[port_num].immutable.core_cap_flags &
3117 	       RDMA_CORE_CAP_PROT_IWARP;
3118 }
3119 
3120 static inline bool rdma_ib_or_roce(const struct ib_device *device,
3121 				   u32 port_num)
3122 {
3123 	return rdma_protocol_ib(device, port_num) ||
3124 		rdma_protocol_roce(device, port_num);
3125 }
3126 
3127 static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3128 					    u32 port_num)
3129 {
3130 	return device->port_data[port_num].immutable.core_cap_flags &
3131 	       RDMA_CORE_CAP_PROT_RAW_PACKET;
3132 }
3133 
3134 static inline bool rdma_protocol_usnic(const struct ib_device *device,
3135 				       u32 port_num)
3136 {
3137 	return device->port_data[port_num].immutable.core_cap_flags &
3138 	       RDMA_CORE_CAP_PROT_USNIC;
3139 }
3140 
3141 /**
3142  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3143  * Management Datagrams.
3144  * @device: Device to check
3145  * @port_num: Port number to check
3146  *
3147  * Management Datagrams (MAD) are a required part of the InfiniBand
3148  * specification and are supported on all InfiniBand devices.  A slightly
3149  * extended version are also supported on OPA interfaces.
3150  *
3151  * Return: true if the port supports sending/receiving of MAD packets.
3152  */
3153 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3154 {
3155 	return device->port_data[port_num].immutable.core_cap_flags &
3156 	       RDMA_CORE_CAP_IB_MAD;
3157 }
3158 
3159 /**
3160  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3161  * Management Datagrams.
3162  * @device: Device to check
3163  * @port_num: Port number to check
3164  *
3165  * Intel OmniPath devices extend and/or replace the InfiniBand Management
3166  * datagrams with their own versions.  These OPA MADs share many but not all of
3167  * the characteristics of InfiniBand MADs.
3168  *
3169  * OPA MADs differ in the following ways:
3170  *
3171  *    1) MADs are variable size up to 2K
3172  *       IBTA defined MADs remain fixed at 256 bytes
3173  *    2) OPA SMPs must carry valid PKeys
3174  *    3) OPA SMP packets are a different format
3175  *
3176  * Return: true if the port supports OPA MAD packet formats.
3177  */
3178 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3179 {
3180 	return device->port_data[port_num].immutable.core_cap_flags &
3181 		RDMA_CORE_CAP_OPA_MAD;
3182 }
3183 
3184 /**
3185  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3186  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3187  * @device: Device to check
3188  * @port_num: Port number to check
3189  *
3190  * Each InfiniBand node is required to provide a Subnet Management Agent
3191  * that the subnet manager can access.  Prior to the fabric being fully
3192  * configured by the subnet manager, the SMA is accessed via a well known
3193  * interface called the Subnet Management Interface (SMI).  This interface
3194  * uses directed route packets to communicate with the SM to get around the
3195  * chicken and egg problem of the SM needing to know what's on the fabric
3196  * in order to configure the fabric, and needing to configure the fabric in
3197  * order to send packets to the devices on the fabric.  These directed
3198  * route packets do not need the fabric fully configured in order to reach
3199  * their destination.  The SMI is the only method allowed to send
3200  * directed route packets on an InfiniBand fabric.
3201  *
3202  * Return: true if the port provides an SMI.
3203  */
3204 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3205 {
3206 	return device->port_data[port_num].immutable.core_cap_flags &
3207 	       RDMA_CORE_CAP_IB_SMI;
3208 }
3209 
3210 /**
3211  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3212  * Communication Manager.
3213  * @device: Device to check
3214  * @port_num: Port number to check
3215  *
3216  * The InfiniBand Communication Manager is one of many pre-defined General
3217  * Service Agents (GSA) that are accessed via the General Service
3218  * Interface (GSI).  It's role is to facilitate establishment of connections
3219  * between nodes as well as other management related tasks for established
3220  * connections.
3221  *
3222  * Return: true if the port supports an IB CM (this does not guarantee that
3223  * a CM is actually running however).
3224  */
3225 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3226 {
3227 	return device->port_data[port_num].immutable.core_cap_flags &
3228 	       RDMA_CORE_CAP_IB_CM;
3229 }
3230 
3231 /**
3232  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3233  * Communication Manager.
3234  * @device: Device to check
3235  * @port_num: Port number to check
3236  *
3237  * Similar to above, but specific to iWARP connections which have a different
3238  * managment protocol than InfiniBand.
3239  *
3240  * Return: true if the port supports an iWARP CM (this does not guarantee that
3241  * a CM is actually running however).
3242  */
3243 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3244 {
3245 	return device->port_data[port_num].immutable.core_cap_flags &
3246 	       RDMA_CORE_CAP_IW_CM;
3247 }
3248 
3249 /**
3250  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3251  * Subnet Administration.
3252  * @device: Device to check
3253  * @port_num: Port number to check
3254  *
3255  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3256  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
3257  * fabrics, devices should resolve routes to other hosts by contacting the
3258  * SA to query the proper route.
3259  *
3260  * Return: true if the port should act as a client to the fabric Subnet
3261  * Administration interface.  This does not imply that the SA service is
3262  * running locally.
3263  */
3264 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3265 {
3266 	return device->port_data[port_num].immutable.core_cap_flags &
3267 	       RDMA_CORE_CAP_IB_SA;
3268 }
3269 
3270 /**
3271  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3272  * Multicast.
3273  * @device: Device to check
3274  * @port_num: Port number to check
3275  *
3276  * InfiniBand multicast registration is more complex than normal IPv4 or
3277  * IPv6 multicast registration.  Each Host Channel Adapter must register
3278  * with the Subnet Manager when it wishes to join a multicast group.  It
3279  * should do so only once regardless of how many queue pairs it subscribes
3280  * to this group.  And it should leave the group only after all queue pairs
3281  * attached to the group have been detached.
3282  *
3283  * Return: true if the port must undertake the additional adminstrative
3284  * overhead of registering/unregistering with the SM and tracking of the
3285  * total number of queue pairs attached to the multicast group.
3286  */
3287 static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3288 				     u32 port_num)
3289 {
3290 	return rdma_cap_ib_sa(device, port_num);
3291 }
3292 
3293 /**
3294  * rdma_cap_af_ib - Check if the port of device has the capability
3295  * Native Infiniband Address.
3296  * @device: Device to check
3297  * @port_num: Port number to check
3298  *
3299  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3300  * GID.  RoCE uses a different mechanism, but still generates a GID via
3301  * a prescribed mechanism and port specific data.
3302  *
3303  * Return: true if the port uses a GID address to identify devices on the
3304  * network.
3305  */
3306 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3307 {
3308 	return device->port_data[port_num].immutable.core_cap_flags &
3309 	       RDMA_CORE_CAP_AF_IB;
3310 }
3311 
3312 /**
3313  * rdma_cap_eth_ah - Check if the port of device has the capability
3314  * Ethernet Address Handle.
3315  * @device: Device to check
3316  * @port_num: Port number to check
3317  *
3318  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3319  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3320  * port.  Normally, packet headers are generated by the sending host
3321  * adapter, but when sending connectionless datagrams, we must manually
3322  * inject the proper headers for the fabric we are communicating over.
3323  *
3324  * Return: true if we are running as a RoCE port and must force the
3325  * addition of a Global Route Header built from our Ethernet Address
3326  * Handle into our header list for connectionless packets.
3327  */
3328 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3329 {
3330 	return device->port_data[port_num].immutable.core_cap_flags &
3331 	       RDMA_CORE_CAP_ETH_AH;
3332 }
3333 
3334 /**
3335  * rdma_cap_opa_ah - Check if the port of device supports
3336  * OPA Address handles
3337  * @device: Device to check
3338  * @port_num: Port number to check
3339  *
3340  * Return: true if we are running on an OPA device which supports
3341  * the extended OPA addressing.
3342  */
3343 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3344 {
3345 	return (device->port_data[port_num].immutable.core_cap_flags &
3346 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3347 }
3348 
3349 /**
3350  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3351  *
3352  * @device: Device
3353  * @port_num: Port number
3354  *
3355  * This MAD size includes the MAD headers and MAD payload.  No other headers
3356  * are included.
3357  *
3358  * Return the max MAD size required by the Port.  Will return 0 if the port
3359  * does not support MADs
3360  */
3361 static inline size_t rdma_max_mad_size(const struct ib_device *device,
3362 				       u32 port_num)
3363 {
3364 	return device->port_data[port_num].immutable.max_mad_size;
3365 }
3366 
3367 /**
3368  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3369  * @device: Device to check
3370  * @port_num: Port number to check
3371  *
3372  * RoCE GID table mechanism manages the various GIDs for a device.
3373  *
3374  * NOTE: if allocating the port's GID table has failed, this call will still
3375  * return true, but any RoCE GID table API will fail.
3376  *
3377  * Return: true if the port uses RoCE GID table mechanism in order to manage
3378  * its GIDs.
3379  */
3380 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3381 					   u32 port_num)
3382 {
3383 	return rdma_protocol_roce(device, port_num) &&
3384 		device->ops.add_gid && device->ops.del_gid;
3385 }
3386 
3387 /*
3388  * Check if the device supports READ W/ INVALIDATE.
3389  */
3390 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3391 {
3392 	/*
3393 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3394 	 * has support for it yet.
3395 	 */
3396 	return rdma_protocol_iwarp(dev, port_num);
3397 }
3398 
3399 /**
3400  * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3401  * @device: Device
3402  * @port_num: 1 based Port number
3403  *
3404  * Return true if port is an Intel OPA port , false if not
3405  */
3406 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3407 					  u32 port_num)
3408 {
3409 	return (device->port_data[port_num].immutable.core_cap_flags &
3410 		RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3411 }
3412 
3413 /**
3414  * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3415  * @device: Device
3416  * @port_num: Port number
3417  * @mtu: enum value of MTU
3418  *
3419  * Return the MTU size supported by the port as an integer value. Will return
3420  * -1 if enum value of mtu is not supported.
3421  */
3422 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3423 				       int mtu)
3424 {
3425 	if (rdma_core_cap_opa_port(device, port))
3426 		return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3427 	else
3428 		return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3429 }
3430 
3431 /**
3432  * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3433  * @device: Device
3434  * @port_num: Port number
3435  * @attr: port attribute
3436  *
3437  * Return the MTU size supported by the port as an integer value.
3438  */
3439 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3440 				     struct ib_port_attr *attr)
3441 {
3442 	if (rdma_core_cap_opa_port(device, port))
3443 		return attr->phys_mtu;
3444 	else
3445 		return ib_mtu_enum_to_int(attr->max_mtu);
3446 }
3447 
3448 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3449 			 int state);
3450 int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3451 		     struct ifla_vf_info *info);
3452 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3453 		    struct ifla_vf_stats *stats);
3454 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3455 		    struct ifla_vf_guid *node_guid,
3456 		    struct ifla_vf_guid *port_guid);
3457 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3458 		   int type);
3459 
3460 int ib_query_pkey(struct ib_device *device,
3461 		  u32 port_num, u16 index, u16 *pkey);
3462 
3463 int ib_modify_device(struct ib_device *device,
3464 		     int device_modify_mask,
3465 		     struct ib_device_modify *device_modify);
3466 
3467 int ib_modify_port(struct ib_device *device,
3468 		   u32 port_num, int port_modify_mask,
3469 		   struct ib_port_modify *port_modify);
3470 
3471 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3472 		u32 *port_num, u16 *index);
3473 
3474 int ib_find_pkey(struct ib_device *device,
3475 		 u32 port_num, u16 pkey, u16 *index);
3476 
3477 enum ib_pd_flags {
3478 	/*
3479 	 * Create a memory registration for all memory in the system and place
3480 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3481 	 * ULPs to avoid the overhead of dynamic MRs.
3482 	 *
3483 	 * This flag is generally considered unsafe and must only be used in
3484 	 * extremly trusted environments.  Every use of it will log a warning
3485 	 * in the kernel log.
3486 	 */
3487 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3488 };
3489 
3490 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3491 		const char *caller);
3492 
3493 /**
3494  * ib_alloc_pd - Allocates an unused protection domain.
3495  * @device: The device on which to allocate the protection domain.
3496  * @flags: protection domain flags
3497  *
3498  * A protection domain object provides an association between QPs, shared
3499  * receive queues, address handles, memory regions, and memory windows.
3500  *
3501  * Every PD has a local_dma_lkey which can be used as the lkey value for local
3502  * memory operations.
3503  */
3504 #define ib_alloc_pd(device, flags) \
3505 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3506 
3507 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3508 
3509 /**
3510  * ib_dealloc_pd - Deallocate kernel PD
3511  * @pd: The protection domain
3512  *
3513  * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3514  */
3515 static inline void ib_dealloc_pd(struct ib_pd *pd)
3516 {
3517 	int ret = ib_dealloc_pd_user(pd, NULL);
3518 
3519 	WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3520 }
3521 
3522 enum rdma_create_ah_flags {
3523 	/* In a sleepable context */
3524 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3525 };
3526 
3527 /**
3528  * rdma_create_ah - Creates an address handle for the given address vector.
3529  * @pd: The protection domain associated with the address handle.
3530  * @ah_attr: The attributes of the address vector.
3531  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3532  *
3533  * The address handle is used to reference a local or global destination
3534  * in all UD QP post sends.
3535  */
3536 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3537 			     u32 flags);
3538 
3539 /**
3540  * rdma_create_user_ah - Creates an address handle for the given address vector.
3541  * It resolves destination mac address for ah attribute of RoCE type.
3542  * @pd: The protection domain associated with the address handle.
3543  * @ah_attr: The attributes of the address vector.
3544  * @udata: pointer to user's input output buffer information need by
3545  *         provider driver.
3546  *
3547  * It returns 0 on success and returns appropriate error code on error.
3548  * The address handle is used to reference a local or global destination
3549  * in all UD QP post sends.
3550  */
3551 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3552 				  struct rdma_ah_attr *ah_attr,
3553 				  struct ib_udata *udata);
3554 /**
3555  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3556  *   work completion.
3557  * @hdr: the L3 header to parse
3558  * @net_type: type of header to parse
3559  * @sgid: place to store source gid
3560  * @dgid: place to store destination gid
3561  */
3562 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3563 			      enum rdma_network_type net_type,
3564 			      union ib_gid *sgid, union ib_gid *dgid);
3565 
3566 /**
3567  * ib_get_rdma_header_version - Get the header version
3568  * @hdr: the L3 header to parse
3569  */
3570 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3571 
3572 /**
3573  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3574  *   work completion.
3575  * @device: Device on which the received message arrived.
3576  * @port_num: Port on which the received message arrived.
3577  * @wc: Work completion associated with the received message.
3578  * @grh: References the received global route header.  This parameter is
3579  *   ignored unless the work completion indicates that the GRH is valid.
3580  * @ah_attr: Returned attributes that can be used when creating an address
3581  *   handle for replying to the message.
3582  * When ib_init_ah_attr_from_wc() returns success,
3583  * (a) for IB link layer it optionally contains a reference to SGID attribute
3584  * when GRH is present for IB link layer.
3585  * (b) for RoCE link layer it contains a reference to SGID attribute.
3586  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3587  * attributes which are initialized using ib_init_ah_attr_from_wc().
3588  *
3589  */
3590 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3591 			    const struct ib_wc *wc, const struct ib_grh *grh,
3592 			    struct rdma_ah_attr *ah_attr);
3593 
3594 /**
3595  * ib_create_ah_from_wc - Creates an address handle associated with the
3596  *   sender of the specified work completion.
3597  * @pd: The protection domain associated with the address handle.
3598  * @wc: Work completion information associated with a received message.
3599  * @grh: References the received global route header.  This parameter is
3600  *   ignored unless the work completion indicates that the GRH is valid.
3601  * @port_num: The outbound port number to associate with the address.
3602  *
3603  * The address handle is used to reference a local or global destination
3604  * in all UD QP post sends.
3605  */
3606 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3607 				   const struct ib_grh *grh, u32 port_num);
3608 
3609 /**
3610  * rdma_modify_ah - Modifies the address vector associated with an address
3611  *   handle.
3612  * @ah: The address handle to modify.
3613  * @ah_attr: The new address vector attributes to associate with the
3614  *   address handle.
3615  */
3616 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3617 
3618 /**
3619  * rdma_query_ah - Queries the address vector associated with an address
3620  *   handle.
3621  * @ah: The address handle to query.
3622  * @ah_attr: The address vector attributes associated with the address
3623  *   handle.
3624  */
3625 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3626 
3627 enum rdma_destroy_ah_flags {
3628 	/* In a sleepable context */
3629 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3630 };
3631 
3632 /**
3633  * rdma_destroy_ah_user - Destroys an address handle.
3634  * @ah: The address handle to destroy.
3635  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3636  * @udata: Valid user data or NULL for kernel objects
3637  */
3638 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3639 
3640 /**
3641  * rdma_destroy_ah - Destroys an kernel address handle.
3642  * @ah: The address handle to destroy.
3643  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3644  *
3645  * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3646  */
3647 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3648 {
3649 	int ret = rdma_destroy_ah_user(ah, flags, NULL);
3650 
3651 	WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3652 }
3653 
3654 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3655 				  struct ib_srq_init_attr *srq_init_attr,
3656 				  struct ib_usrq_object *uobject,
3657 				  struct ib_udata *udata);
3658 static inline struct ib_srq *
3659 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3660 {
3661 	if (!pd->device->ops.create_srq)
3662 		return ERR_PTR(-EOPNOTSUPP);
3663 
3664 	return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3665 }
3666 
3667 /**
3668  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3669  * @srq: The SRQ to modify.
3670  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3671  *   the current values of selected SRQ attributes are returned.
3672  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3673  *   are being modified.
3674  *
3675  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3676  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3677  * the number of receives queued drops below the limit.
3678  */
3679 int ib_modify_srq(struct ib_srq *srq,
3680 		  struct ib_srq_attr *srq_attr,
3681 		  enum ib_srq_attr_mask srq_attr_mask);
3682 
3683 /**
3684  * ib_query_srq - Returns the attribute list and current values for the
3685  *   specified SRQ.
3686  * @srq: The SRQ to query.
3687  * @srq_attr: The attributes of the specified SRQ.
3688  */
3689 int ib_query_srq(struct ib_srq *srq,
3690 		 struct ib_srq_attr *srq_attr);
3691 
3692 /**
3693  * ib_destroy_srq_user - Destroys the specified SRQ.
3694  * @srq: The SRQ to destroy.
3695  * @udata: Valid user data or NULL for kernel objects
3696  */
3697 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3698 
3699 /**
3700  * ib_destroy_srq - Destroys the specified kernel SRQ.
3701  * @srq: The SRQ to destroy.
3702  *
3703  * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3704  */
3705 static inline void ib_destroy_srq(struct ib_srq *srq)
3706 {
3707 	int ret = ib_destroy_srq_user(srq, NULL);
3708 
3709 	WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3710 }
3711 
3712 /**
3713  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3714  * @srq: The SRQ to post the work request on.
3715  * @recv_wr: A list of work requests to post on the receive queue.
3716  * @bad_recv_wr: On an immediate failure, this parameter will reference
3717  *   the work request that failed to be posted on the QP.
3718  */
3719 static inline int ib_post_srq_recv(struct ib_srq *srq,
3720 				   const struct ib_recv_wr *recv_wr,
3721 				   const struct ib_recv_wr **bad_recv_wr)
3722 {
3723 	const struct ib_recv_wr *dummy;
3724 
3725 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3726 					      bad_recv_wr ? : &dummy);
3727 }
3728 
3729 struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3730 				  struct ib_qp_init_attr *qp_init_attr,
3731 				  const char *caller);
3732 /**
3733  * ib_create_qp - Creates a kernel QP associated with the specific protection
3734  * domain.
3735  * @pd: The protection domain associated with the QP.
3736  * @init_attr: A list of initial attributes required to create the
3737  *   QP.  If QP creation succeeds, then the attributes are updated to
3738  *   the actual capabilities of the created QP.
3739  */
3740 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3741 					 struct ib_qp_init_attr *init_attr)
3742 {
3743 	return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
3744 }
3745 
3746 /**
3747  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3748  * @qp: The QP to modify.
3749  * @attr: On input, specifies the QP attributes to modify.  On output,
3750  *   the current values of selected QP attributes are returned.
3751  * @attr_mask: A bit-mask used to specify which attributes of the QP
3752  *   are being modified.
3753  * @udata: pointer to user's input output buffer information
3754  *   are being modified.
3755  * It returns 0 on success and returns appropriate error code on error.
3756  */
3757 int ib_modify_qp_with_udata(struct ib_qp *qp,
3758 			    struct ib_qp_attr *attr,
3759 			    int attr_mask,
3760 			    struct ib_udata *udata);
3761 
3762 /**
3763  * ib_modify_qp - Modifies the attributes for the specified QP and then
3764  *   transitions the QP to the given state.
3765  * @qp: The QP to modify.
3766  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3767  *   the current values of selected QP attributes are returned.
3768  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3769  *   are being modified.
3770  */
3771 int ib_modify_qp(struct ib_qp *qp,
3772 		 struct ib_qp_attr *qp_attr,
3773 		 int qp_attr_mask);
3774 
3775 /**
3776  * ib_query_qp - Returns the attribute list and current values for the
3777  *   specified QP.
3778  * @qp: The QP to query.
3779  * @qp_attr: The attributes of the specified QP.
3780  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3781  * @qp_init_attr: Additional attributes of the selected QP.
3782  *
3783  * The qp_attr_mask may be used to limit the query to gathering only the
3784  * selected attributes.
3785  */
3786 int ib_query_qp(struct ib_qp *qp,
3787 		struct ib_qp_attr *qp_attr,
3788 		int qp_attr_mask,
3789 		struct ib_qp_init_attr *qp_init_attr);
3790 
3791 /**
3792  * ib_destroy_qp - Destroys the specified QP.
3793  * @qp: The QP to destroy.
3794  * @udata: Valid udata or NULL for kernel objects
3795  */
3796 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3797 
3798 /**
3799  * ib_destroy_qp - Destroys the specified kernel QP.
3800  * @qp: The QP to destroy.
3801  *
3802  * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3803  */
3804 static inline int ib_destroy_qp(struct ib_qp *qp)
3805 {
3806 	return ib_destroy_qp_user(qp, NULL);
3807 }
3808 
3809 /**
3810  * ib_open_qp - Obtain a reference to an existing sharable QP.
3811  * @xrcd - XRC domain
3812  * @qp_open_attr: Attributes identifying the QP to open.
3813  *
3814  * Returns a reference to a sharable QP.
3815  */
3816 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3817 			 struct ib_qp_open_attr *qp_open_attr);
3818 
3819 /**
3820  * ib_close_qp - Release an external reference to a QP.
3821  * @qp: The QP handle to release
3822  *
3823  * The opened QP handle is released by the caller.  The underlying
3824  * shared QP is not destroyed until all internal references are released.
3825  */
3826 int ib_close_qp(struct ib_qp *qp);
3827 
3828 /**
3829  * ib_post_send - Posts a list of work requests to the send queue of
3830  *   the specified QP.
3831  * @qp: The QP to post the work request on.
3832  * @send_wr: A list of work requests to post on the send queue.
3833  * @bad_send_wr: On an immediate failure, this parameter will reference
3834  *   the work request that failed to be posted on the QP.
3835  *
3836  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3837  * error is returned, the QP state shall not be affected,
3838  * ib_post_send() will return an immediate error after queueing any
3839  * earlier work requests in the list.
3840  */
3841 static inline int ib_post_send(struct ib_qp *qp,
3842 			       const struct ib_send_wr *send_wr,
3843 			       const struct ib_send_wr **bad_send_wr)
3844 {
3845 	const struct ib_send_wr *dummy;
3846 
3847 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3848 }
3849 
3850 /**
3851  * ib_post_recv - Posts a list of work requests to the receive queue of
3852  *   the specified QP.
3853  * @qp: The QP to post the work request on.
3854  * @recv_wr: A list of work requests to post on the receive queue.
3855  * @bad_recv_wr: On an immediate failure, this parameter will reference
3856  *   the work request that failed to be posted on the QP.
3857  */
3858 static inline int ib_post_recv(struct ib_qp *qp,
3859 			       const struct ib_recv_wr *recv_wr,
3860 			       const struct ib_recv_wr **bad_recv_wr)
3861 {
3862 	const struct ib_recv_wr *dummy;
3863 
3864 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3865 }
3866 
3867 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3868 			    int comp_vector, enum ib_poll_context poll_ctx,
3869 			    const char *caller);
3870 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3871 					int nr_cqe, int comp_vector,
3872 					enum ib_poll_context poll_ctx)
3873 {
3874 	return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3875 			     KBUILD_MODNAME);
3876 }
3877 
3878 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3879 				int nr_cqe, enum ib_poll_context poll_ctx,
3880 				const char *caller);
3881 
3882 /**
3883  * ib_alloc_cq_any: Allocate kernel CQ
3884  * @dev: The IB device
3885  * @private: Private data attached to the CQE
3886  * @nr_cqe: Number of CQEs in the CQ
3887  * @poll_ctx: Context used for polling the CQ
3888  */
3889 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3890 					    void *private, int nr_cqe,
3891 					    enum ib_poll_context poll_ctx)
3892 {
3893 	return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3894 				 KBUILD_MODNAME);
3895 }
3896 
3897 void ib_free_cq(struct ib_cq *cq);
3898 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3899 
3900 /**
3901  * ib_create_cq - Creates a CQ on the specified device.
3902  * @device: The device on which to create the CQ.
3903  * @comp_handler: A user-specified callback that is invoked when a
3904  *   completion event occurs on the CQ.
3905  * @event_handler: A user-specified callback that is invoked when an
3906  *   asynchronous event not associated with a completion occurs on the CQ.
3907  * @cq_context: Context associated with the CQ returned to the user via
3908  *   the associated completion and event handlers.
3909  * @cq_attr: The attributes the CQ should be created upon.
3910  *
3911  * Users can examine the cq structure to determine the actual CQ size.
3912  */
3913 struct ib_cq *__ib_create_cq(struct ib_device *device,
3914 			     ib_comp_handler comp_handler,
3915 			     void (*event_handler)(struct ib_event *, void *),
3916 			     void *cq_context,
3917 			     const struct ib_cq_init_attr *cq_attr,
3918 			     const char *caller);
3919 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3920 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3921 
3922 /**
3923  * ib_resize_cq - Modifies the capacity of the CQ.
3924  * @cq: The CQ to resize.
3925  * @cqe: The minimum size of the CQ.
3926  *
3927  * Users can examine the cq structure to determine the actual CQ size.
3928  */
3929 int ib_resize_cq(struct ib_cq *cq, int cqe);
3930 
3931 /**
3932  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3933  * @cq: The CQ to modify.
3934  * @cq_count: number of CQEs that will trigger an event
3935  * @cq_period: max period of time in usec before triggering an event
3936  *
3937  */
3938 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3939 
3940 /**
3941  * ib_destroy_cq_user - Destroys the specified CQ.
3942  * @cq: The CQ to destroy.
3943  * @udata: Valid user data or NULL for kernel objects
3944  */
3945 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3946 
3947 /**
3948  * ib_destroy_cq - Destroys the specified kernel CQ.
3949  * @cq: The CQ to destroy.
3950  *
3951  * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3952  */
3953 static inline void ib_destroy_cq(struct ib_cq *cq)
3954 {
3955 	int ret = ib_destroy_cq_user(cq, NULL);
3956 
3957 	WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3958 }
3959 
3960 /**
3961  * ib_poll_cq - poll a CQ for completion(s)
3962  * @cq:the CQ being polled
3963  * @num_entries:maximum number of completions to return
3964  * @wc:array of at least @num_entries &struct ib_wc where completions
3965  *   will be returned
3966  *
3967  * Poll a CQ for (possibly multiple) completions.  If the return value
3968  * is < 0, an error occurred.  If the return value is >= 0, it is the
3969  * number of completions returned.  If the return value is
3970  * non-negative and < num_entries, then the CQ was emptied.
3971  */
3972 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3973 			     struct ib_wc *wc)
3974 {
3975 	return cq->device->ops.poll_cq(cq, num_entries, wc);
3976 }
3977 
3978 /**
3979  * ib_req_notify_cq - Request completion notification on a CQ.
3980  * @cq: The CQ to generate an event for.
3981  * @flags:
3982  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3983  *   to request an event on the next solicited event or next work
3984  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3985  *   may also be |ed in to request a hint about missed events, as
3986  *   described below.
3987  *
3988  * Return Value:
3989  *    < 0 means an error occurred while requesting notification
3990  *   == 0 means notification was requested successfully, and if
3991  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3992  *        were missed and it is safe to wait for another event.  In
3993  *        this case is it guaranteed that any work completions added
3994  *        to the CQ since the last CQ poll will trigger a completion
3995  *        notification event.
3996  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3997  *        in.  It means that the consumer must poll the CQ again to
3998  *        make sure it is empty to avoid missing an event because of a
3999  *        race between requesting notification and an entry being
4000  *        added to the CQ.  This return value means it is possible
4001  *        (but not guaranteed) that a work completion has been added
4002  *        to the CQ since the last poll without triggering a
4003  *        completion notification event.
4004  */
4005 static inline int ib_req_notify_cq(struct ib_cq *cq,
4006 				   enum ib_cq_notify_flags flags)
4007 {
4008 	return cq->device->ops.req_notify_cq(cq, flags);
4009 }
4010 
4011 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4012 			     int comp_vector_hint,
4013 			     enum ib_poll_context poll_ctx);
4014 
4015 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4016 
4017 /*
4018  * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
4019  * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
4020  * address into the dma address.
4021  */
4022 static inline bool ib_uses_virt_dma(struct ib_device *dev)
4023 {
4024 	return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
4025 }
4026 
4027 /*
4028  * Check if a IB device's underlying DMA mapping supports P2PDMA transfers.
4029  */
4030 static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev)
4031 {
4032 	if (ib_uses_virt_dma(dev))
4033 		return false;
4034 
4035 	return dma_pci_p2pdma_supported(dev->dma_device);
4036 }
4037 
4038 /**
4039  * ib_virt_dma_to_ptr - Convert a dma_addr to a kernel pointer
4040  * @dma_addr: The DMA address
4041  *
4042  * Used by ib_uses_virt_dma() devices to get back to the kernel pointer after
4043  * going through the dma_addr marshalling.
4044  */
4045 static inline void *ib_virt_dma_to_ptr(u64 dma_addr)
4046 {
4047 	/* virt_dma mode maps the kvs's directly into the dma addr */
4048 	return (void *)(uintptr_t)dma_addr;
4049 }
4050 
4051 /**
4052  * ib_virt_dma_to_page - Convert a dma_addr to a struct page
4053  * @dma_addr: The DMA address
4054  *
4055  * Used by ib_uses_virt_dma() device to get back to the struct page after going
4056  * through the dma_addr marshalling.
4057  */
4058 static inline struct page *ib_virt_dma_to_page(u64 dma_addr)
4059 {
4060 	return virt_to_page(ib_virt_dma_to_ptr(dma_addr));
4061 }
4062 
4063 /**
4064  * ib_dma_mapping_error - check a DMA addr for error
4065  * @dev: The device for which the dma_addr was created
4066  * @dma_addr: The DMA address to check
4067  */
4068 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4069 {
4070 	if (ib_uses_virt_dma(dev))
4071 		return 0;
4072 	return dma_mapping_error(dev->dma_device, dma_addr);
4073 }
4074 
4075 /**
4076  * ib_dma_map_single - Map a kernel virtual address to DMA address
4077  * @dev: The device for which the dma_addr is to be created
4078  * @cpu_addr: The kernel virtual address
4079  * @size: The size of the region in bytes
4080  * @direction: The direction of the DMA
4081  */
4082 static inline u64 ib_dma_map_single(struct ib_device *dev,
4083 				    void *cpu_addr, size_t size,
4084 				    enum dma_data_direction direction)
4085 {
4086 	if (ib_uses_virt_dma(dev))
4087 		return (uintptr_t)cpu_addr;
4088 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4089 }
4090 
4091 /**
4092  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4093  * @dev: The device for which the DMA address was created
4094  * @addr: The DMA address
4095  * @size: The size of the region in bytes
4096  * @direction: The direction of the DMA
4097  */
4098 static inline void ib_dma_unmap_single(struct ib_device *dev,
4099 				       u64 addr, size_t size,
4100 				       enum dma_data_direction direction)
4101 {
4102 	if (!ib_uses_virt_dma(dev))
4103 		dma_unmap_single(dev->dma_device, addr, size, direction);
4104 }
4105 
4106 /**
4107  * ib_dma_map_page - Map a physical page to DMA address
4108  * @dev: The device for which the dma_addr is to be created
4109  * @page: The page to be mapped
4110  * @offset: The offset within the page
4111  * @size: The size of the region in bytes
4112  * @direction: The direction of the DMA
4113  */
4114 static inline u64 ib_dma_map_page(struct ib_device *dev,
4115 				  struct page *page,
4116 				  unsigned long offset,
4117 				  size_t size,
4118 					 enum dma_data_direction direction)
4119 {
4120 	if (ib_uses_virt_dma(dev))
4121 		return (uintptr_t)(page_address(page) + offset);
4122 	return dma_map_page(dev->dma_device, page, offset, size, direction);
4123 }
4124 
4125 /**
4126  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4127  * @dev: The device for which the DMA address was created
4128  * @addr: The DMA address
4129  * @size: The size of the region in bytes
4130  * @direction: The direction of the DMA
4131  */
4132 static inline void ib_dma_unmap_page(struct ib_device *dev,
4133 				     u64 addr, size_t size,
4134 				     enum dma_data_direction direction)
4135 {
4136 	if (!ib_uses_virt_dma(dev))
4137 		dma_unmap_page(dev->dma_device, addr, size, direction);
4138 }
4139 
4140 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4141 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4142 				      struct scatterlist *sg, int nents,
4143 				      enum dma_data_direction direction,
4144 				      unsigned long dma_attrs)
4145 {
4146 	if (ib_uses_virt_dma(dev))
4147 		return ib_dma_virt_map_sg(dev, sg, nents);
4148 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4149 				dma_attrs);
4150 }
4151 
4152 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4153 					 struct scatterlist *sg, int nents,
4154 					 enum dma_data_direction direction,
4155 					 unsigned long dma_attrs)
4156 {
4157 	if (!ib_uses_virt_dma(dev))
4158 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4159 				   dma_attrs);
4160 }
4161 
4162 /**
4163  * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4164  * @dev: The device for which the DMA addresses are to be created
4165  * @sg: The sg_table object describing the buffer
4166  * @direction: The direction of the DMA
4167  * @attrs: Optional DMA attributes for the map operation
4168  */
4169 static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4170 					   struct sg_table *sgt,
4171 					   enum dma_data_direction direction,
4172 					   unsigned long dma_attrs)
4173 {
4174 	int nents;
4175 
4176 	if (ib_uses_virt_dma(dev)) {
4177 		nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4178 		if (!nents)
4179 			return -EIO;
4180 		sgt->nents = nents;
4181 		return 0;
4182 	}
4183 	return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4184 }
4185 
4186 static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4187 					      struct sg_table *sgt,
4188 					      enum dma_data_direction direction,
4189 					      unsigned long dma_attrs)
4190 {
4191 	if (!ib_uses_virt_dma(dev))
4192 		dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4193 }
4194 
4195 /**
4196  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4197  * @dev: The device for which the DMA addresses are to be created
4198  * @sg: The array of scatter/gather entries
4199  * @nents: The number of scatter/gather entries
4200  * @direction: The direction of the DMA
4201  */
4202 static inline int ib_dma_map_sg(struct ib_device *dev,
4203 				struct scatterlist *sg, int nents,
4204 				enum dma_data_direction direction)
4205 {
4206 	return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4207 }
4208 
4209 /**
4210  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4211  * @dev: The device for which the DMA addresses were created
4212  * @sg: The array of scatter/gather entries
4213  * @nents: The number of scatter/gather entries
4214  * @direction: The direction of the DMA
4215  */
4216 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4217 				   struct scatterlist *sg, int nents,
4218 				   enum dma_data_direction direction)
4219 {
4220 	ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4221 }
4222 
4223 /**
4224  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4225  * @dev: The device to query
4226  *
4227  * The returned value represents a size in bytes.
4228  */
4229 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4230 {
4231 	if (ib_uses_virt_dma(dev))
4232 		return UINT_MAX;
4233 	return dma_get_max_seg_size(dev->dma_device);
4234 }
4235 
4236 /**
4237  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4238  * @dev: The device for which the DMA address was created
4239  * @addr: The DMA address
4240  * @size: The size of the region in bytes
4241  * @dir: The direction of the DMA
4242  */
4243 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4244 					      u64 addr,
4245 					      size_t size,
4246 					      enum dma_data_direction dir)
4247 {
4248 	if (!ib_uses_virt_dma(dev))
4249 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4250 }
4251 
4252 /**
4253  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4254  * @dev: The device for which the DMA address was created
4255  * @addr: The DMA address
4256  * @size: The size of the region in bytes
4257  * @dir: The direction of the DMA
4258  */
4259 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4260 						 u64 addr,
4261 						 size_t size,
4262 						 enum dma_data_direction dir)
4263 {
4264 	if (!ib_uses_virt_dma(dev))
4265 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4266 }
4267 
4268 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4269  * space. This function should be called when 'current' is the owning MM.
4270  */
4271 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4272 			     u64 virt_addr, int mr_access_flags);
4273 
4274 /* ib_advise_mr -  give an advice about an address range in a memory region */
4275 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4276 		 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4277 /**
4278  * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4279  *   HCA translation table.
4280  * @mr: The memory region to deregister.
4281  * @udata: Valid user data or NULL for kernel object
4282  *
4283  * This function can fail, if the memory region has memory windows bound to it.
4284  */
4285 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4286 
4287 /**
4288  * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4289  *   HCA translation table.
4290  * @mr: The memory region to deregister.
4291  *
4292  * This function can fail, if the memory region has memory windows bound to it.
4293  *
4294  * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4295  */
4296 static inline int ib_dereg_mr(struct ib_mr *mr)
4297 {
4298 	return ib_dereg_mr_user(mr, NULL);
4299 }
4300 
4301 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4302 			  u32 max_num_sg);
4303 
4304 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4305 				    u32 max_num_data_sg,
4306 				    u32 max_num_meta_sg);
4307 
4308 /**
4309  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4310  *   R_Key and L_Key.
4311  * @mr - struct ib_mr pointer to be updated.
4312  * @newkey - new key to be used.
4313  */
4314 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4315 {
4316 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4317 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4318 }
4319 
4320 /**
4321  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4322  * for calculating a new rkey for type 2 memory windows.
4323  * @rkey - the rkey to increment.
4324  */
4325 static inline u32 ib_inc_rkey(u32 rkey)
4326 {
4327 	const u32 mask = 0x000000ff;
4328 	return ((rkey + 1) & mask) | (rkey & ~mask);
4329 }
4330 
4331 /**
4332  * ib_attach_mcast - Attaches the specified QP to a multicast group.
4333  * @qp: QP to attach to the multicast group.  The QP must be type
4334  *   IB_QPT_UD.
4335  * @gid: Multicast group GID.
4336  * @lid: Multicast group LID in host byte order.
4337  *
4338  * In order to send and receive multicast packets, subnet
4339  * administration must have created the multicast group and configured
4340  * the fabric appropriately.  The port associated with the specified
4341  * QP must also be a member of the multicast group.
4342  */
4343 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4344 
4345 /**
4346  * ib_detach_mcast - Detaches the specified QP from a multicast group.
4347  * @qp: QP to detach from the multicast group.
4348  * @gid: Multicast group GID.
4349  * @lid: Multicast group LID in host byte order.
4350  */
4351 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4352 
4353 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4354 				   struct inode *inode, struct ib_udata *udata);
4355 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4356 
4357 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4358 				     unsigned int flags)
4359 {
4360 	u64 device_cap = ib_dev->attrs.device_cap_flags;
4361 
4362 	/*
4363 	 * Local write permission is required if remote write or
4364 	 * remote atomic permission is also requested.
4365 	 */
4366 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4367 	    !(flags & IB_ACCESS_LOCAL_WRITE))
4368 		return -EINVAL;
4369 
4370 	if (flags & ~IB_ACCESS_SUPPORTED)
4371 		return -EINVAL;
4372 
4373 	if (flags & IB_ACCESS_ON_DEMAND &&
4374 	    !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING))
4375 		return -EOPNOTSUPP;
4376 
4377 	if ((flags & IB_ACCESS_FLUSH_GLOBAL &&
4378 	    !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) ||
4379 	    (flags & IB_ACCESS_FLUSH_PERSISTENT &&
4380 	    !(device_cap & IB_DEVICE_FLUSH_PERSISTENT)))
4381 		return -EOPNOTSUPP;
4382 
4383 	return 0;
4384 }
4385 
4386 static inline bool ib_access_writable(int access_flags)
4387 {
4388 	/*
4389 	 * We have writable memory backing the MR if any of the following
4390 	 * access flags are set.  "Local write" and "remote write" obviously
4391 	 * require write access.  "Remote atomic" can do things like fetch and
4392 	 * add, which will modify memory, and "MW bind" can change permissions
4393 	 * by binding a window.
4394 	 */
4395 	return access_flags &
4396 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
4397 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4398 }
4399 
4400 /**
4401  * ib_check_mr_status: lightweight check of MR status.
4402  *     This routine may provide status checks on a selected
4403  *     ib_mr. first use is for signature status check.
4404  *
4405  * @mr: A memory region.
4406  * @check_mask: Bitmask of which checks to perform from
4407  *     ib_mr_status_check enumeration.
4408  * @mr_status: The container of relevant status checks.
4409  *     failed checks will be indicated in the status bitmask
4410  *     and the relevant info shall be in the error item.
4411  */
4412 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4413 		       struct ib_mr_status *mr_status);
4414 
4415 /**
4416  * ib_device_try_get: Hold a registration lock
4417  * device: The device to lock
4418  *
4419  * A device under an active registration lock cannot become unregistered. It
4420  * is only possible to obtain a registration lock on a device that is fully
4421  * registered, otherwise this function returns false.
4422  *
4423  * The registration lock is only necessary for actions which require the
4424  * device to still be registered. Uses that only require the device pointer to
4425  * be valid should use get_device(&ibdev->dev) to hold the memory.
4426  *
4427  */
4428 static inline bool ib_device_try_get(struct ib_device *dev)
4429 {
4430 	return refcount_inc_not_zero(&dev->refcount);
4431 }
4432 
4433 void ib_device_put(struct ib_device *device);
4434 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4435 					  enum rdma_driver_id driver_id);
4436 struct ib_device *ib_device_get_by_name(const char *name,
4437 					enum rdma_driver_id driver_id);
4438 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4439 					    u16 pkey, const union ib_gid *gid,
4440 					    const struct sockaddr *addr);
4441 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4442 			 unsigned int port);
4443 struct net_device *ib_device_netdev(struct ib_device *dev, u32 port);
4444 
4445 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4446 			   struct ib_wq_init_attr *init_attr);
4447 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4448 
4449 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4450 		 unsigned int *sg_offset, unsigned int page_size);
4451 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4452 		    int data_sg_nents, unsigned int *data_sg_offset,
4453 		    struct scatterlist *meta_sg, int meta_sg_nents,
4454 		    unsigned int *meta_sg_offset, unsigned int page_size);
4455 
4456 static inline int
4457 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4458 		  unsigned int *sg_offset, unsigned int page_size)
4459 {
4460 	int n;
4461 
4462 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4463 	mr->iova = 0;
4464 
4465 	return n;
4466 }
4467 
4468 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4469 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4470 
4471 void ib_drain_rq(struct ib_qp *qp);
4472 void ib_drain_sq(struct ib_qp *qp);
4473 void ib_drain_qp(struct ib_qp *qp);
4474 
4475 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4476 		     u8 *width);
4477 
4478 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4479 {
4480 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4481 		return attr->roce.dmac;
4482 	return NULL;
4483 }
4484 
4485 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4486 {
4487 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4488 		attr->ib.dlid = (u16)dlid;
4489 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4490 		attr->opa.dlid = dlid;
4491 }
4492 
4493 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4494 {
4495 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4496 		return attr->ib.dlid;
4497 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4498 		return attr->opa.dlid;
4499 	return 0;
4500 }
4501 
4502 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4503 {
4504 	attr->sl = sl;
4505 }
4506 
4507 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4508 {
4509 	return attr->sl;
4510 }
4511 
4512 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4513 					 u8 src_path_bits)
4514 {
4515 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4516 		attr->ib.src_path_bits = src_path_bits;
4517 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4518 		attr->opa.src_path_bits = src_path_bits;
4519 }
4520 
4521 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4522 {
4523 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4524 		return attr->ib.src_path_bits;
4525 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4526 		return attr->opa.src_path_bits;
4527 	return 0;
4528 }
4529 
4530 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4531 					bool make_grd)
4532 {
4533 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4534 		attr->opa.make_grd = make_grd;
4535 }
4536 
4537 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4538 {
4539 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4540 		return attr->opa.make_grd;
4541 	return false;
4542 }
4543 
4544 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4545 {
4546 	attr->port_num = port_num;
4547 }
4548 
4549 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4550 {
4551 	return attr->port_num;
4552 }
4553 
4554 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4555 					   u8 static_rate)
4556 {
4557 	attr->static_rate = static_rate;
4558 }
4559 
4560 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4561 {
4562 	return attr->static_rate;
4563 }
4564 
4565 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4566 					enum ib_ah_flags flag)
4567 {
4568 	attr->ah_flags = flag;
4569 }
4570 
4571 static inline enum ib_ah_flags
4572 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4573 {
4574 	return attr->ah_flags;
4575 }
4576 
4577 static inline const struct ib_global_route
4578 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4579 {
4580 	return &attr->grh;
4581 }
4582 
4583 /*To retrieve and modify the grh */
4584 static inline struct ib_global_route
4585 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4586 {
4587 	return &attr->grh;
4588 }
4589 
4590 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4591 {
4592 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4593 
4594 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4595 }
4596 
4597 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4598 					     __be64 prefix)
4599 {
4600 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4601 
4602 	grh->dgid.global.subnet_prefix = prefix;
4603 }
4604 
4605 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4606 					    __be64 if_id)
4607 {
4608 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4609 
4610 	grh->dgid.global.interface_id = if_id;
4611 }
4612 
4613 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4614 				   union ib_gid *dgid, u32 flow_label,
4615 				   u8 sgid_index, u8 hop_limit,
4616 				   u8 traffic_class)
4617 {
4618 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4619 
4620 	attr->ah_flags = IB_AH_GRH;
4621 	if (dgid)
4622 		grh->dgid = *dgid;
4623 	grh->flow_label = flow_label;
4624 	grh->sgid_index = sgid_index;
4625 	grh->hop_limit = hop_limit;
4626 	grh->traffic_class = traffic_class;
4627 	grh->sgid_attr = NULL;
4628 }
4629 
4630 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4631 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4632 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4633 			     const struct ib_gid_attr *sgid_attr);
4634 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4635 		       const struct rdma_ah_attr *src);
4636 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4637 			  const struct rdma_ah_attr *new);
4638 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4639 
4640 /**
4641  * rdma_ah_find_type - Return address handle type.
4642  *
4643  * @dev: Device to be checked
4644  * @port_num: Port number
4645  */
4646 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4647 						       u32 port_num)
4648 {
4649 	if (rdma_protocol_roce(dev, port_num))
4650 		return RDMA_AH_ATTR_TYPE_ROCE;
4651 	if (rdma_protocol_ib(dev, port_num)) {
4652 		if (rdma_cap_opa_ah(dev, port_num))
4653 			return RDMA_AH_ATTR_TYPE_OPA;
4654 		return RDMA_AH_ATTR_TYPE_IB;
4655 	}
4656 
4657 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4658 }
4659 
4660 /**
4661  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4662  *     In the current implementation the only way to
4663  *     get the 32bit lid is from other sources for OPA.
4664  *     For IB, lids will always be 16bits so cast the
4665  *     value accordingly.
4666  *
4667  * @lid: A 32bit LID
4668  */
4669 static inline u16 ib_lid_cpu16(u32 lid)
4670 {
4671 	WARN_ON_ONCE(lid & 0xFFFF0000);
4672 	return (u16)lid;
4673 }
4674 
4675 /**
4676  * ib_lid_be16 - Return lid in 16bit BE encoding.
4677  *
4678  * @lid: A 32bit LID
4679  */
4680 static inline __be16 ib_lid_be16(u32 lid)
4681 {
4682 	WARN_ON_ONCE(lid & 0xFFFF0000);
4683 	return cpu_to_be16((u16)lid);
4684 }
4685 
4686 /**
4687  * ib_get_vector_affinity - Get the affinity mappings of a given completion
4688  *   vector
4689  * @device:         the rdma device
4690  * @comp_vector:    index of completion vector
4691  *
4692  * Returns NULL on failure, otherwise a corresponding cpu map of the
4693  * completion vector (returns all-cpus map if the device driver doesn't
4694  * implement get_vector_affinity).
4695  */
4696 static inline const struct cpumask *
4697 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4698 {
4699 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4700 	    !device->ops.get_vector_affinity)
4701 		return NULL;
4702 
4703 	return device->ops.get_vector_affinity(device, comp_vector);
4704 
4705 }
4706 
4707 /**
4708  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4709  * and add their gids, as needed, to the relevant RoCE devices.
4710  *
4711  * @device:         the rdma device
4712  */
4713 void rdma_roce_rescan_device(struct ib_device *ibdev);
4714 
4715 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4716 
4717 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4718 
4719 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4720 				     enum rdma_netdev_t type, const char *name,
4721 				     unsigned char name_assign_type,
4722 				     void (*setup)(struct net_device *));
4723 
4724 int rdma_init_netdev(struct ib_device *device, u32 port_num,
4725 		     enum rdma_netdev_t type, const char *name,
4726 		     unsigned char name_assign_type,
4727 		     void (*setup)(struct net_device *),
4728 		     struct net_device *netdev);
4729 
4730 /**
4731  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4732  *
4733  * @device:	device pointer for which ib_device pointer to retrieve
4734  *
4735  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4736  *
4737  */
4738 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4739 {
4740 	struct ib_core_device *coredev =
4741 		container_of(device, struct ib_core_device, dev);
4742 
4743 	return coredev->owner;
4744 }
4745 
4746 /**
4747  * ibdev_to_node - return the NUMA node for a given ib_device
4748  * @dev:	device to get the NUMA node for.
4749  */
4750 static inline int ibdev_to_node(struct ib_device *ibdev)
4751 {
4752 	struct device *parent = ibdev->dev.parent;
4753 
4754 	if (!parent)
4755 		return NUMA_NO_NODE;
4756 	return dev_to_node(parent);
4757 }
4758 
4759 /**
4760  * rdma_device_to_drv_device - Helper macro to reach back to driver's
4761  *			       ib_device holder structure from device pointer.
4762  *
4763  * NOTE: New drivers should not make use of this API; This API is only for
4764  * existing drivers who have exposed sysfs entries using
4765  * ops->device_group.
4766  */
4767 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
4768 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4769 
4770 bool rdma_dev_access_netns(const struct ib_device *device,
4771 			   const struct net *net);
4772 
4773 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4774 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4775 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4776 
4777 /**
4778  * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4779  *                               on the flow_label
4780  *
4781  * This function will convert the 20 bit flow_label input to a valid RoCE v2
4782  * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4783  * convention.
4784  */
4785 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4786 {
4787 	u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4788 
4789 	fl_low ^= fl_high >> 14;
4790 	return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4791 }
4792 
4793 /**
4794  * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4795  *                        local and remote qpn values
4796  *
4797  * This function folded the multiplication results of two qpns, 24 bit each,
4798  * fields, and converts it to a 20 bit results.
4799  *
4800  * This function will create symmetric flow_label value based on the local
4801  * and remote qpn values. this will allow both the requester and responder
4802  * to calculate the same flow_label for a given connection.
4803  *
4804  * This helper function should be used by driver in case the upper layer
4805  * provide a zero flow_label value. This is to improve entropy of RDMA
4806  * traffic in the network.
4807  */
4808 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4809 {
4810 	u64 v = (u64)lqpn * rqpn;
4811 
4812 	v ^= v >> 20;
4813 	v ^= v >> 40;
4814 
4815 	return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4816 }
4817 
4818 /**
4819  * rdma_get_udp_sport - Calculate and set UDP source port based on the flow
4820  *                      label. If flow label is not defined in GRH then
4821  *                      calculate it based on lqpn/rqpn.
4822  *
4823  * @fl:                 flow label from GRH
4824  * @lqpn:               local qp number
4825  * @rqpn:               remote qp number
4826  */
4827 static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
4828 {
4829 	if (!fl)
4830 		fl = rdma_calc_flow_label(lqpn, rqpn);
4831 
4832 	return rdma_flow_label_to_udp_sport(fl);
4833 }
4834 
4835 const struct ib_port_immutable*
4836 ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4837 #endif /* IB_VERBS_H */
4838