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