1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
2 /*
3  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
4  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
5  * Copyright (c) 2004, 2020 Intel Corporation.  All rights reserved.
6  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
7  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
8  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
9  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
10  */
11 
12 #ifndef IB_VERBS_H
13 #define IB_VERBS_H
14 
15 #include <linux/ethtool.h>
16 #include <linux/types.h>
17 #include <linux/device.h>
18 #include <linux/dma-mapping.h>
19 #include <linux/kref.h>
20 #include <linux/list.h>
21 #include <linux/rwsem.h>
22 #include <linux/workqueue.h>
23 #include <linux/irq_poll.h>
24 #include <uapi/linux/if_ether.h>
25 #include <net/ipv6.h>
26 #include <net/ip.h>
27 #include <linux/string.h>
28 #include <linux/slab.h>
29 #include <linux/netdevice.h>
30 #include <linux/refcount.h>
31 #include <linux/if_link.h>
32 #include <linux/atomic.h>
33 #include <linux/mmu_notifier.h>
34 #include <linux/uaccess.h>
35 #include <linux/cgroup_rdma.h>
36 #include <linux/irqflags.h>
37 #include <linux/preempt.h>
38 #include <linux/dim.h>
39 #include <uapi/rdma/ib_user_verbs.h>
40 #include <rdma/rdma_counter.h>
41 #include <rdma/restrack.h>
42 #include <rdma/signature.h>
43 #include <uapi/rdma/rdma_user_ioctl.h>
44 #include <uapi/rdma/ib_user_ioctl_verbs.h>
45 
46 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
47 
48 struct ib_umem_odp;
49 struct ib_uqp_object;
50 struct ib_usrq_object;
51 struct ib_uwq_object;
52 struct rdma_cm_id;
53 struct ib_port;
54 struct hw_stats_device_data;
55 
56 extern struct workqueue_struct *ib_wq;
57 extern struct workqueue_struct *ib_comp_wq;
58 extern struct workqueue_struct *ib_comp_unbound_wq;
59 
60 struct ib_ucq_object;
61 
62 __printf(3, 4) __cold
63 void ibdev_printk(const char *level, const struct ib_device *ibdev,
64 		  const char *format, ...);
65 __printf(2, 3) __cold
66 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
67 __printf(2, 3) __cold
68 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
69 __printf(2, 3) __cold
70 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
71 __printf(2, 3) __cold
72 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
73 __printf(2, 3) __cold
74 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
75 __printf(2, 3) __cold
76 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
77 __printf(2, 3) __cold
78 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
79 
80 #if defined(CONFIG_DYNAMIC_DEBUG) || \
81 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
82 #define ibdev_dbg(__dev, format, args...)                       \
83 	dynamic_ibdev_dbg(__dev, format, ##args)
84 #else
85 __printf(2, 3) __cold
86 static inline
ibdev_dbg(const struct ib_device * ibdev,const char * format,...)87 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
88 #endif
89 
90 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...)           \
91 do {                                                                    \
92 	static DEFINE_RATELIMIT_STATE(_rs,                              \
93 				      DEFAULT_RATELIMIT_INTERVAL,       \
94 				      DEFAULT_RATELIMIT_BURST);         \
95 	if (__ratelimit(&_rs))                                          \
96 		ibdev_level(ibdev, fmt, ##__VA_ARGS__);                 \
97 } while (0)
98 
99 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
100 	ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
101 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
102 	ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
103 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
104 	ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
105 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
106 	ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
107 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
108 	ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
109 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
110 	ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
111 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
112 	ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
113 
114 #if defined(CONFIG_DYNAMIC_DEBUG) || \
115 	(defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
116 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
117 #define ibdev_dbg_ratelimited(ibdev, fmt, ...)                          \
118 do {                                                                    \
119 	static DEFINE_RATELIMIT_STATE(_rs,                              \
120 				      DEFAULT_RATELIMIT_INTERVAL,       \
121 				      DEFAULT_RATELIMIT_BURST);         \
122 	DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt);                 \
123 	if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs))      \
124 		__dynamic_ibdev_dbg(&descriptor, ibdev, fmt,            \
125 				    ##__VA_ARGS__);                     \
126 } while (0)
127 #else
128 __printf(2, 3) __cold
129 static inline
ibdev_dbg_ratelimited(const struct ib_device * ibdev,const char * format,...)130 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
131 #endif
132 
133 union ib_gid {
134 	u8	raw[16];
135 	struct {
136 		__be64	subnet_prefix;
137 		__be64	interface_id;
138 	} global;
139 };
140 
141 extern union ib_gid zgid;
142 
143 enum ib_gid_type {
144 	IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
145 	IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
146 	IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
147 	IB_GID_TYPE_SIZE
148 };
149 
150 #define ROCE_V2_UDP_DPORT      4791
151 struct ib_gid_attr {
152 	struct net_device __rcu	*ndev;
153 	struct ib_device	*device;
154 	union ib_gid		gid;
155 	enum ib_gid_type	gid_type;
156 	u16			index;
157 	u32			port_num;
158 };
159 
160 enum {
161 	/* set the local administered indication */
162 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
163 };
164 
165 enum rdma_transport_type {
166 	RDMA_TRANSPORT_IB,
167 	RDMA_TRANSPORT_IWARP,
168 	RDMA_TRANSPORT_USNIC,
169 	RDMA_TRANSPORT_USNIC_UDP,
170 	RDMA_TRANSPORT_UNSPECIFIED,
171 };
172 
173 enum rdma_protocol_type {
174 	RDMA_PROTOCOL_IB,
175 	RDMA_PROTOCOL_IBOE,
176 	RDMA_PROTOCOL_IWARP,
177 	RDMA_PROTOCOL_USNIC_UDP
178 };
179 
180 __attribute_const__ enum rdma_transport_type
181 rdma_node_get_transport(unsigned int node_type);
182 
183 enum rdma_network_type {
184 	RDMA_NETWORK_IB,
185 	RDMA_NETWORK_ROCE_V1,
186 	RDMA_NETWORK_IPV4,
187 	RDMA_NETWORK_IPV6
188 };
189 
ib_network_to_gid_type(enum rdma_network_type network_type)190 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
191 {
192 	if (network_type == RDMA_NETWORK_IPV4 ||
193 	    network_type == RDMA_NETWORK_IPV6)
194 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
195 	else if (network_type == RDMA_NETWORK_ROCE_V1)
196 		return IB_GID_TYPE_ROCE;
197 	else
198 		return IB_GID_TYPE_IB;
199 }
200 
201 static inline enum rdma_network_type
rdma_gid_attr_network_type(const struct ib_gid_attr * attr)202 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
203 {
204 	if (attr->gid_type == IB_GID_TYPE_IB)
205 		return RDMA_NETWORK_IB;
206 
207 	if (attr->gid_type == IB_GID_TYPE_ROCE)
208 		return RDMA_NETWORK_ROCE_V1;
209 
210 	if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
211 		return RDMA_NETWORK_IPV4;
212 	else
213 		return RDMA_NETWORK_IPV6;
214 }
215 
216 enum rdma_link_layer {
217 	IB_LINK_LAYER_UNSPECIFIED,
218 	IB_LINK_LAYER_INFINIBAND,
219 	IB_LINK_LAYER_ETHERNET,
220 };
221 
222 enum ib_device_cap_flags {
223 	IB_DEVICE_RESIZE_MAX_WR = IB_UVERBS_DEVICE_RESIZE_MAX_WR,
224 	IB_DEVICE_BAD_PKEY_CNTR = IB_UVERBS_DEVICE_BAD_PKEY_CNTR,
225 	IB_DEVICE_BAD_QKEY_CNTR = IB_UVERBS_DEVICE_BAD_QKEY_CNTR,
226 	IB_DEVICE_RAW_MULTI = IB_UVERBS_DEVICE_RAW_MULTI,
227 	IB_DEVICE_AUTO_PATH_MIG = IB_UVERBS_DEVICE_AUTO_PATH_MIG,
228 	IB_DEVICE_CHANGE_PHY_PORT = IB_UVERBS_DEVICE_CHANGE_PHY_PORT,
229 	IB_DEVICE_UD_AV_PORT_ENFORCE = IB_UVERBS_DEVICE_UD_AV_PORT_ENFORCE,
230 	IB_DEVICE_CURR_QP_STATE_MOD = IB_UVERBS_DEVICE_CURR_QP_STATE_MOD,
231 	IB_DEVICE_SHUTDOWN_PORT = IB_UVERBS_DEVICE_SHUTDOWN_PORT,
232 	/* IB_DEVICE_INIT_TYPE = IB_UVERBS_DEVICE_INIT_TYPE, (not in use) */
233 	IB_DEVICE_PORT_ACTIVE_EVENT = IB_UVERBS_DEVICE_PORT_ACTIVE_EVENT,
234 	IB_DEVICE_SYS_IMAGE_GUID = IB_UVERBS_DEVICE_SYS_IMAGE_GUID,
235 	IB_DEVICE_RC_RNR_NAK_GEN = IB_UVERBS_DEVICE_RC_RNR_NAK_GEN,
236 	IB_DEVICE_SRQ_RESIZE = IB_UVERBS_DEVICE_SRQ_RESIZE,
237 	IB_DEVICE_N_NOTIFY_CQ = IB_UVERBS_DEVICE_N_NOTIFY_CQ,
238 
239 	/* Reserved, old SEND_W_INV = 1 << 16,*/
240 	IB_DEVICE_MEM_WINDOW = IB_UVERBS_DEVICE_MEM_WINDOW,
241 	/*
242 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
243 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
244 	 * messages and can verify the validity of checksum for
245 	 * incoming messages.  Setting this flag implies that the
246 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
247 	 */
248 	IB_DEVICE_UD_IP_CSUM = IB_UVERBS_DEVICE_UD_IP_CSUM,
249 	IB_DEVICE_XRC = IB_UVERBS_DEVICE_XRC,
250 
251 	/*
252 	 * This device supports the IB "base memory management extension",
253 	 * which includes support for fast registrations (IB_WR_REG_MR,
254 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
255 	 * also be set by any iWarp device which must support FRs to comply
256 	 * to the iWarp verbs spec.  iWarp devices also support the
257 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
258 	 * stag.
259 	 */
260 	IB_DEVICE_MEM_MGT_EXTENSIONS = IB_UVERBS_DEVICE_MEM_MGT_EXTENSIONS,
261 	IB_DEVICE_MEM_WINDOW_TYPE_2A = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2A,
262 	IB_DEVICE_MEM_WINDOW_TYPE_2B = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2B,
263 	IB_DEVICE_RC_IP_CSUM = IB_UVERBS_DEVICE_RC_IP_CSUM,
264 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
265 	IB_DEVICE_RAW_IP_CSUM = IB_UVERBS_DEVICE_RAW_IP_CSUM,
266 	IB_DEVICE_MANAGED_FLOW_STEERING =
267 		IB_UVERBS_DEVICE_MANAGED_FLOW_STEERING,
268 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
269 	IB_DEVICE_RAW_SCATTER_FCS = IB_UVERBS_DEVICE_RAW_SCATTER_FCS,
270 	/* The device supports padding incoming writes to cacheline. */
271 	IB_DEVICE_PCI_WRITE_END_PADDING =
272 		IB_UVERBS_DEVICE_PCI_WRITE_END_PADDING,
273 };
274 
275 enum ib_kernel_cap_flags {
276 	/*
277 	 * This device supports a per-device lkey or stag that can be
278 	 * used without performing a memory registration for the local
279 	 * memory.  Note that ULPs should never check this flag, but
280 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
281 	 * which will always contain a usable lkey.
282 	 */
283 	IBK_LOCAL_DMA_LKEY = 1 << 0,
284 	/* IB_QP_CREATE_INTEGRITY_EN is supported to implement T10-PI */
285 	IBK_INTEGRITY_HANDOVER = 1 << 1,
286 	/* IB_ACCESS_ON_DEMAND is supported during reg_user_mr() */
287 	IBK_ON_DEMAND_PAGING = 1 << 2,
288 	/* IB_MR_TYPE_SG_GAPS is supported */
289 	IBK_SG_GAPS_REG = 1 << 3,
290 	/* Driver supports RDMA_NLDEV_CMD_DELLINK */
291 	IBK_ALLOW_USER_UNREG = 1 << 4,
292 
293 	/* ipoib will use IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK */
294 	IBK_BLOCK_MULTICAST_LOOPBACK = 1 << 5,
295 	/* iopib will use IB_QP_CREATE_IPOIB_UD_LSO for its QPs */
296 	IBK_UD_TSO = 1 << 6,
297 	/* iopib will use the device ops:
298 	 *   get_vf_config
299 	 *   get_vf_guid
300 	 *   get_vf_stats
301 	 *   set_vf_guid
302 	 *   set_vf_link_state
303 	 */
304 	IBK_VIRTUAL_FUNCTION = 1 << 7,
305 	/* ipoib will use IB_QP_CREATE_NETDEV_USE for its QPs */
306 	IBK_RDMA_NETDEV_OPA = 1 << 8,
307 };
308 
309 enum ib_atomic_cap {
310 	IB_ATOMIC_NONE,
311 	IB_ATOMIC_HCA,
312 	IB_ATOMIC_GLOB
313 };
314 
315 enum ib_odp_general_cap_bits {
316 	IB_ODP_SUPPORT		= 1 << 0,
317 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
318 };
319 
320 enum ib_odp_transport_cap_bits {
321 	IB_ODP_SUPPORT_SEND	= 1 << 0,
322 	IB_ODP_SUPPORT_RECV	= 1 << 1,
323 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
324 	IB_ODP_SUPPORT_READ	= 1 << 3,
325 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
326 	IB_ODP_SUPPORT_SRQ_RECV	= 1 << 5,
327 };
328 
329 struct ib_odp_caps {
330 	uint64_t general_caps;
331 	struct {
332 		uint32_t  rc_odp_caps;
333 		uint32_t  uc_odp_caps;
334 		uint32_t  ud_odp_caps;
335 		uint32_t  xrc_odp_caps;
336 	} per_transport_caps;
337 };
338 
339 struct ib_rss_caps {
340 	/* Corresponding bit will be set if qp type from
341 	 * 'enum ib_qp_type' is supported, e.g.
342 	 * supported_qpts |= 1 << IB_QPT_UD
343 	 */
344 	u32 supported_qpts;
345 	u32 max_rwq_indirection_tables;
346 	u32 max_rwq_indirection_table_size;
347 };
348 
349 enum ib_tm_cap_flags {
350 	/*  Support tag matching with rendezvous offload for RC transport */
351 	IB_TM_CAP_RNDV_RC = 1 << 0,
352 };
353 
354 struct ib_tm_caps {
355 	/* Max size of RNDV header */
356 	u32 max_rndv_hdr_size;
357 	/* Max number of entries in tag matching list */
358 	u32 max_num_tags;
359 	/* From enum ib_tm_cap_flags */
360 	u32 flags;
361 	/* Max number of outstanding list operations */
362 	u32 max_ops;
363 	/* Max number of SGE in tag matching entry */
364 	u32 max_sge;
365 };
366 
367 struct ib_cq_init_attr {
368 	unsigned int	cqe;
369 	u32		comp_vector;
370 	u32		flags;
371 };
372 
373 enum ib_cq_attr_mask {
374 	IB_CQ_MODERATE = 1 << 0,
375 };
376 
377 struct ib_cq_caps {
378 	u16     max_cq_moderation_count;
379 	u16     max_cq_moderation_period;
380 };
381 
382 struct ib_dm_mr_attr {
383 	u64		length;
384 	u64		offset;
385 	u32		access_flags;
386 };
387 
388 struct ib_dm_alloc_attr {
389 	u64	length;
390 	u32	alignment;
391 	u32	flags;
392 };
393 
394 struct ib_device_attr {
395 	u64			fw_ver;
396 	__be64			sys_image_guid;
397 	u64			max_mr_size;
398 	u64			page_size_cap;
399 	u32			vendor_id;
400 	u32			vendor_part_id;
401 	u32			hw_ver;
402 	int			max_qp;
403 	int			max_qp_wr;
404 	u64			device_cap_flags;
405 	u64			kernel_cap_flags;
406 	int			max_send_sge;
407 	int			max_recv_sge;
408 	int			max_sge_rd;
409 	int			max_cq;
410 	int			max_cqe;
411 	int			max_mr;
412 	int			max_pd;
413 	int			max_qp_rd_atom;
414 	int			max_ee_rd_atom;
415 	int			max_res_rd_atom;
416 	int			max_qp_init_rd_atom;
417 	int			max_ee_init_rd_atom;
418 	enum ib_atomic_cap	atomic_cap;
419 	enum ib_atomic_cap	masked_atomic_cap;
420 	int			max_ee;
421 	int			max_rdd;
422 	int			max_mw;
423 	int			max_raw_ipv6_qp;
424 	int			max_raw_ethy_qp;
425 	int			max_mcast_grp;
426 	int			max_mcast_qp_attach;
427 	int			max_total_mcast_qp_attach;
428 	int			max_ah;
429 	int			max_srq;
430 	int			max_srq_wr;
431 	int			max_srq_sge;
432 	unsigned int		max_fast_reg_page_list_len;
433 	unsigned int		max_pi_fast_reg_page_list_len;
434 	u16			max_pkeys;
435 	u8			local_ca_ack_delay;
436 	int			sig_prot_cap;
437 	int			sig_guard_cap;
438 	struct ib_odp_caps	odp_caps;
439 	uint64_t		timestamp_mask;
440 	uint64_t		hca_core_clock; /* in KHZ */
441 	struct ib_rss_caps	rss_caps;
442 	u32			max_wq_type_rq;
443 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
444 	struct ib_tm_caps	tm_caps;
445 	struct ib_cq_caps       cq_caps;
446 	u64			max_dm_size;
447 	/* Max entries for sgl for optimized performance per READ */
448 	u32			max_sgl_rd;
449 };
450 
451 enum ib_mtu {
452 	IB_MTU_256  = 1,
453 	IB_MTU_512  = 2,
454 	IB_MTU_1024 = 3,
455 	IB_MTU_2048 = 4,
456 	IB_MTU_4096 = 5
457 };
458 
459 enum opa_mtu {
460 	OPA_MTU_8192 = 6,
461 	OPA_MTU_10240 = 7
462 };
463 
ib_mtu_enum_to_int(enum ib_mtu mtu)464 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
465 {
466 	switch (mtu) {
467 	case IB_MTU_256:  return  256;
468 	case IB_MTU_512:  return  512;
469 	case IB_MTU_1024: return 1024;
470 	case IB_MTU_2048: return 2048;
471 	case IB_MTU_4096: return 4096;
472 	default: 	  return -1;
473 	}
474 }
475 
ib_mtu_int_to_enum(int mtu)476 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
477 {
478 	if (mtu >= 4096)
479 		return IB_MTU_4096;
480 	else if (mtu >= 2048)
481 		return IB_MTU_2048;
482 	else if (mtu >= 1024)
483 		return IB_MTU_1024;
484 	else if (mtu >= 512)
485 		return IB_MTU_512;
486 	else
487 		return IB_MTU_256;
488 }
489 
opa_mtu_enum_to_int(enum opa_mtu mtu)490 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
491 {
492 	switch (mtu) {
493 	case OPA_MTU_8192:
494 		return 8192;
495 	case OPA_MTU_10240:
496 		return 10240;
497 	default:
498 		return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
499 	}
500 }
501 
opa_mtu_int_to_enum(int mtu)502 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
503 {
504 	if (mtu >= 10240)
505 		return OPA_MTU_10240;
506 	else if (mtu >= 8192)
507 		return OPA_MTU_8192;
508 	else
509 		return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
510 }
511 
512 enum ib_port_state {
513 	IB_PORT_NOP		= 0,
514 	IB_PORT_DOWN		= 1,
515 	IB_PORT_INIT		= 2,
516 	IB_PORT_ARMED		= 3,
517 	IB_PORT_ACTIVE		= 4,
518 	IB_PORT_ACTIVE_DEFER	= 5
519 };
520 
521 enum ib_port_phys_state {
522 	IB_PORT_PHYS_STATE_SLEEP = 1,
523 	IB_PORT_PHYS_STATE_POLLING = 2,
524 	IB_PORT_PHYS_STATE_DISABLED = 3,
525 	IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
526 	IB_PORT_PHYS_STATE_LINK_UP = 5,
527 	IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
528 	IB_PORT_PHYS_STATE_PHY_TEST = 7,
529 };
530 
531 enum ib_port_width {
532 	IB_WIDTH_1X	= 1,
533 	IB_WIDTH_2X	= 16,
534 	IB_WIDTH_4X	= 2,
535 	IB_WIDTH_8X	= 4,
536 	IB_WIDTH_12X	= 8
537 };
538 
ib_width_enum_to_int(enum ib_port_width width)539 static inline int ib_width_enum_to_int(enum ib_port_width width)
540 {
541 	switch (width) {
542 	case IB_WIDTH_1X:  return  1;
543 	case IB_WIDTH_2X:  return  2;
544 	case IB_WIDTH_4X:  return  4;
545 	case IB_WIDTH_8X:  return  8;
546 	case IB_WIDTH_12X: return 12;
547 	default: 	  return -1;
548 	}
549 }
550 
551 enum ib_port_speed {
552 	IB_SPEED_SDR	= 1,
553 	IB_SPEED_DDR	= 2,
554 	IB_SPEED_QDR	= 4,
555 	IB_SPEED_FDR10	= 8,
556 	IB_SPEED_FDR	= 16,
557 	IB_SPEED_EDR	= 32,
558 	IB_SPEED_HDR	= 64,
559 	IB_SPEED_NDR	= 128,
560 };
561 
562 enum ib_stat_flag {
563 	IB_STAT_FLAG_OPTIONAL = 1 << 0,
564 };
565 
566 /**
567  * struct rdma_stat_desc
568  * @name - The name of the counter
569  * @flags - Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL
570  * @priv - Driver private information; Core code should not use
571  */
572 struct rdma_stat_desc {
573 	const char *name;
574 	unsigned int flags;
575 	const void *priv;
576 };
577 
578 /**
579  * struct rdma_hw_stats
580  * @lock - Mutex to protect parallel write access to lifespan and values
581  *    of counters, which are 64bits and not guaranteed to be written
582  *    atomicaly on 32bits systems.
583  * @timestamp - Used by the core code to track when the last update was
584  * @lifespan - Used by the core code to determine how old the counters
585  *   should be before being updated again.  Stored in jiffies, defaults
586  *   to 10 milliseconds, drivers can override the default be specifying
587  *   their own value during their allocation routine.
588  * @descs - Array of pointers to static descriptors used for the counters
589  *   in directory.
590  * @is_disabled - A bitmap to indicate each counter is currently disabled
591  *   or not.
592  * @num_counters - How many hardware counters there are.  If name is
593  *   shorter than this number, a kernel oops will result.  Driver authors
594  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
595  *   in their code to prevent this.
596  * @value - Array of u64 counters that are accessed by the sysfs code and
597  *   filled in by the drivers get_stats routine
598  */
599 struct rdma_hw_stats {
600 	struct mutex	lock; /* Protect lifespan and values[] */
601 	unsigned long	timestamp;
602 	unsigned long	lifespan;
603 	const struct rdma_stat_desc *descs;
604 	unsigned long	*is_disabled;
605 	int		num_counters;
606 	u64		value[];
607 };
608 
609 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
610 
611 struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
612 	const struct rdma_stat_desc *descs, int num_counters,
613 	unsigned long lifespan);
614 
615 void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
616 
617 /* Define bits for the various functionality this port needs to be supported by
618  * the core.
619  */
620 /* Management                           0x00000FFF */
621 #define RDMA_CORE_CAP_IB_MAD            0x00000001
622 #define RDMA_CORE_CAP_IB_SMI            0x00000002
623 #define RDMA_CORE_CAP_IB_CM             0x00000004
624 #define RDMA_CORE_CAP_IW_CM             0x00000008
625 #define RDMA_CORE_CAP_IB_SA             0x00000010
626 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
627 
628 /* Address format                       0x000FF000 */
629 #define RDMA_CORE_CAP_AF_IB             0x00001000
630 #define RDMA_CORE_CAP_ETH_AH            0x00002000
631 #define RDMA_CORE_CAP_OPA_AH            0x00004000
632 #define RDMA_CORE_CAP_IB_GRH_REQUIRED   0x00008000
633 
634 /* Protocol                             0xFFF00000 */
635 #define RDMA_CORE_CAP_PROT_IB           0x00100000
636 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
637 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
638 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
639 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
640 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
641 
642 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
643 					| RDMA_CORE_CAP_PROT_ROCE     \
644 					| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
645 
646 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
647 					| RDMA_CORE_CAP_IB_MAD \
648 					| RDMA_CORE_CAP_IB_SMI \
649 					| RDMA_CORE_CAP_IB_CM  \
650 					| RDMA_CORE_CAP_IB_SA  \
651 					| RDMA_CORE_CAP_AF_IB)
652 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
653 					| RDMA_CORE_CAP_IB_MAD  \
654 					| RDMA_CORE_CAP_IB_CM   \
655 					| RDMA_CORE_CAP_AF_IB   \
656 					| RDMA_CORE_CAP_ETH_AH)
657 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
658 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
659 					| RDMA_CORE_CAP_IB_MAD  \
660 					| RDMA_CORE_CAP_IB_CM   \
661 					| RDMA_CORE_CAP_AF_IB   \
662 					| RDMA_CORE_CAP_ETH_AH)
663 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
664 					| RDMA_CORE_CAP_IW_CM)
665 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
666 					| RDMA_CORE_CAP_OPA_MAD)
667 
668 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
669 
670 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
671 
672 struct ib_port_attr {
673 	u64			subnet_prefix;
674 	enum ib_port_state	state;
675 	enum ib_mtu		max_mtu;
676 	enum ib_mtu		active_mtu;
677 	u32                     phys_mtu;
678 	int			gid_tbl_len;
679 	unsigned int		ip_gids:1;
680 	/* This is the value from PortInfo CapabilityMask, defined by IBA */
681 	u32			port_cap_flags;
682 	u32			max_msg_sz;
683 	u32			bad_pkey_cntr;
684 	u32			qkey_viol_cntr;
685 	u16			pkey_tbl_len;
686 	u32			sm_lid;
687 	u32			lid;
688 	u8			lmc;
689 	u8			max_vl_num;
690 	u8			sm_sl;
691 	u8			subnet_timeout;
692 	u8			init_type_reply;
693 	u8			active_width;
694 	u16			active_speed;
695 	u8                      phys_state;
696 	u16			port_cap_flags2;
697 };
698 
699 enum ib_device_modify_flags {
700 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
701 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
702 };
703 
704 #define IB_DEVICE_NODE_DESC_MAX 64
705 
706 struct ib_device_modify {
707 	u64	sys_image_guid;
708 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
709 };
710 
711 enum ib_port_modify_flags {
712 	IB_PORT_SHUTDOWN		= 1,
713 	IB_PORT_INIT_TYPE		= (1<<2),
714 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
715 	IB_PORT_OPA_MASK_CHG		= (1<<4)
716 };
717 
718 struct ib_port_modify {
719 	u32	set_port_cap_mask;
720 	u32	clr_port_cap_mask;
721 	u8	init_type;
722 };
723 
724 enum ib_event_type {
725 	IB_EVENT_CQ_ERR,
726 	IB_EVENT_QP_FATAL,
727 	IB_EVENT_QP_REQ_ERR,
728 	IB_EVENT_QP_ACCESS_ERR,
729 	IB_EVENT_COMM_EST,
730 	IB_EVENT_SQ_DRAINED,
731 	IB_EVENT_PATH_MIG,
732 	IB_EVENT_PATH_MIG_ERR,
733 	IB_EVENT_DEVICE_FATAL,
734 	IB_EVENT_PORT_ACTIVE,
735 	IB_EVENT_PORT_ERR,
736 	IB_EVENT_LID_CHANGE,
737 	IB_EVENT_PKEY_CHANGE,
738 	IB_EVENT_SM_CHANGE,
739 	IB_EVENT_SRQ_ERR,
740 	IB_EVENT_SRQ_LIMIT_REACHED,
741 	IB_EVENT_QP_LAST_WQE_REACHED,
742 	IB_EVENT_CLIENT_REREGISTER,
743 	IB_EVENT_GID_CHANGE,
744 	IB_EVENT_WQ_FATAL,
745 };
746 
747 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
748 
749 struct ib_event {
750 	struct ib_device	*device;
751 	union {
752 		struct ib_cq	*cq;
753 		struct ib_qp	*qp;
754 		struct ib_srq	*srq;
755 		struct ib_wq	*wq;
756 		u32		port_num;
757 	} element;
758 	enum ib_event_type	event;
759 };
760 
761 struct ib_event_handler {
762 	struct ib_device *device;
763 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
764 	struct list_head  list;
765 };
766 
767 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
768 	do {							\
769 		(_ptr)->device  = _device;			\
770 		(_ptr)->handler = _handler;			\
771 		INIT_LIST_HEAD(&(_ptr)->list);			\
772 	} while (0)
773 
774 struct ib_global_route {
775 	const struct ib_gid_attr *sgid_attr;
776 	union ib_gid	dgid;
777 	u32		flow_label;
778 	u8		sgid_index;
779 	u8		hop_limit;
780 	u8		traffic_class;
781 };
782 
783 struct ib_grh {
784 	__be32		version_tclass_flow;
785 	__be16		paylen;
786 	u8		next_hdr;
787 	u8		hop_limit;
788 	union ib_gid	sgid;
789 	union ib_gid	dgid;
790 };
791 
792 union rdma_network_hdr {
793 	struct ib_grh ibgrh;
794 	struct {
795 		/* The IB spec states that if it's IPv4, the header
796 		 * is located in the last 20 bytes of the header.
797 		 */
798 		u8		reserved[20];
799 		struct iphdr	roce4grh;
800 	};
801 };
802 
803 #define IB_QPN_MASK		0xFFFFFF
804 
805 enum {
806 	IB_MULTICAST_QPN = 0xffffff
807 };
808 
809 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
810 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
811 
812 enum ib_ah_flags {
813 	IB_AH_GRH	= 1
814 };
815 
816 enum ib_rate {
817 	IB_RATE_PORT_CURRENT = 0,
818 	IB_RATE_2_5_GBPS = 2,
819 	IB_RATE_5_GBPS   = 5,
820 	IB_RATE_10_GBPS  = 3,
821 	IB_RATE_20_GBPS  = 6,
822 	IB_RATE_30_GBPS  = 4,
823 	IB_RATE_40_GBPS  = 7,
824 	IB_RATE_60_GBPS  = 8,
825 	IB_RATE_80_GBPS  = 9,
826 	IB_RATE_120_GBPS = 10,
827 	IB_RATE_14_GBPS  = 11,
828 	IB_RATE_56_GBPS  = 12,
829 	IB_RATE_112_GBPS = 13,
830 	IB_RATE_168_GBPS = 14,
831 	IB_RATE_25_GBPS  = 15,
832 	IB_RATE_100_GBPS = 16,
833 	IB_RATE_200_GBPS = 17,
834 	IB_RATE_300_GBPS = 18,
835 	IB_RATE_28_GBPS  = 19,
836 	IB_RATE_50_GBPS  = 20,
837 	IB_RATE_400_GBPS = 21,
838 	IB_RATE_600_GBPS = 22,
839 };
840 
841 /**
842  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
843  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
844  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
845  * @rate: rate to convert.
846  */
847 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
848 
849 /**
850  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
851  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
852  * @rate: rate to convert.
853  */
854 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
855 
856 
857 /**
858  * enum ib_mr_type - memory region type
859  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
860  *                            normal registration
861  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
862  *                            register any arbitrary sg lists (without
863  *                            the normal mr constraints - see
864  *                            ib_map_mr_sg)
865  * @IB_MR_TYPE_DM:            memory region that is used for device
866  *                            memory registration
867  * @IB_MR_TYPE_USER:          memory region that is used for the user-space
868  *                            application
869  * @IB_MR_TYPE_DMA:           memory region that is used for DMA operations
870  *                            without address translations (VA=PA)
871  * @IB_MR_TYPE_INTEGRITY:     memory region that is used for
872  *                            data integrity operations
873  */
874 enum ib_mr_type {
875 	IB_MR_TYPE_MEM_REG,
876 	IB_MR_TYPE_SG_GAPS,
877 	IB_MR_TYPE_DM,
878 	IB_MR_TYPE_USER,
879 	IB_MR_TYPE_DMA,
880 	IB_MR_TYPE_INTEGRITY,
881 };
882 
883 enum ib_mr_status_check {
884 	IB_MR_CHECK_SIG_STATUS = 1,
885 };
886 
887 /**
888  * struct ib_mr_status - Memory region status container
889  *
890  * @fail_status: Bitmask of MR checks status. For each
891  *     failed check a corresponding status bit is set.
892  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
893  *     failure.
894  */
895 struct ib_mr_status {
896 	u32		    fail_status;
897 	struct ib_sig_err   sig_err;
898 };
899 
900 /**
901  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
902  * enum.
903  * @mult: multiple to convert.
904  */
905 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
906 
907 struct rdma_ah_init_attr {
908 	struct rdma_ah_attr *ah_attr;
909 	u32 flags;
910 	struct net_device *xmit_slave;
911 };
912 
913 enum rdma_ah_attr_type {
914 	RDMA_AH_ATTR_TYPE_UNDEFINED,
915 	RDMA_AH_ATTR_TYPE_IB,
916 	RDMA_AH_ATTR_TYPE_ROCE,
917 	RDMA_AH_ATTR_TYPE_OPA,
918 };
919 
920 struct ib_ah_attr {
921 	u16			dlid;
922 	u8			src_path_bits;
923 };
924 
925 struct roce_ah_attr {
926 	u8			dmac[ETH_ALEN];
927 };
928 
929 struct opa_ah_attr {
930 	u32			dlid;
931 	u8			src_path_bits;
932 	bool			make_grd;
933 };
934 
935 struct rdma_ah_attr {
936 	struct ib_global_route	grh;
937 	u8			sl;
938 	u8			static_rate;
939 	u32			port_num;
940 	u8			ah_flags;
941 	enum rdma_ah_attr_type type;
942 	union {
943 		struct ib_ah_attr ib;
944 		struct roce_ah_attr roce;
945 		struct opa_ah_attr opa;
946 	};
947 };
948 
949 enum ib_wc_status {
950 	IB_WC_SUCCESS,
951 	IB_WC_LOC_LEN_ERR,
952 	IB_WC_LOC_QP_OP_ERR,
953 	IB_WC_LOC_EEC_OP_ERR,
954 	IB_WC_LOC_PROT_ERR,
955 	IB_WC_WR_FLUSH_ERR,
956 	IB_WC_MW_BIND_ERR,
957 	IB_WC_BAD_RESP_ERR,
958 	IB_WC_LOC_ACCESS_ERR,
959 	IB_WC_REM_INV_REQ_ERR,
960 	IB_WC_REM_ACCESS_ERR,
961 	IB_WC_REM_OP_ERR,
962 	IB_WC_RETRY_EXC_ERR,
963 	IB_WC_RNR_RETRY_EXC_ERR,
964 	IB_WC_LOC_RDD_VIOL_ERR,
965 	IB_WC_REM_INV_RD_REQ_ERR,
966 	IB_WC_REM_ABORT_ERR,
967 	IB_WC_INV_EECN_ERR,
968 	IB_WC_INV_EEC_STATE_ERR,
969 	IB_WC_FATAL_ERR,
970 	IB_WC_RESP_TIMEOUT_ERR,
971 	IB_WC_GENERAL_ERR
972 };
973 
974 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
975 
976 enum ib_wc_opcode {
977 	IB_WC_SEND = IB_UVERBS_WC_SEND,
978 	IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
979 	IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
980 	IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
981 	IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
982 	IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
983 	IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
984 	IB_WC_LSO = IB_UVERBS_WC_TSO,
985 	IB_WC_REG_MR,
986 	IB_WC_MASKED_COMP_SWAP,
987 	IB_WC_MASKED_FETCH_ADD,
988 /*
989  * Set value of IB_WC_RECV so consumers can test if a completion is a
990  * receive by testing (opcode & IB_WC_RECV).
991  */
992 	IB_WC_RECV			= 1 << 7,
993 	IB_WC_RECV_RDMA_WITH_IMM
994 };
995 
996 enum ib_wc_flags {
997 	IB_WC_GRH		= 1,
998 	IB_WC_WITH_IMM		= (1<<1),
999 	IB_WC_WITH_INVALIDATE	= (1<<2),
1000 	IB_WC_IP_CSUM_OK	= (1<<3),
1001 	IB_WC_WITH_SMAC		= (1<<4),
1002 	IB_WC_WITH_VLAN		= (1<<5),
1003 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
1004 };
1005 
1006 struct ib_wc {
1007 	union {
1008 		u64		wr_id;
1009 		struct ib_cqe	*wr_cqe;
1010 	};
1011 	enum ib_wc_status	status;
1012 	enum ib_wc_opcode	opcode;
1013 	u32			vendor_err;
1014 	u32			byte_len;
1015 	struct ib_qp	       *qp;
1016 	union {
1017 		__be32		imm_data;
1018 		u32		invalidate_rkey;
1019 	} ex;
1020 	u32			src_qp;
1021 	u32			slid;
1022 	int			wc_flags;
1023 	u16			pkey_index;
1024 	u8			sl;
1025 	u8			dlid_path_bits;
1026 	u32 port_num; /* valid only for DR SMPs on switches */
1027 	u8			smac[ETH_ALEN];
1028 	u16			vlan_id;
1029 	u8			network_hdr_type;
1030 };
1031 
1032 enum ib_cq_notify_flags {
1033 	IB_CQ_SOLICITED			= 1 << 0,
1034 	IB_CQ_NEXT_COMP			= 1 << 1,
1035 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1036 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
1037 };
1038 
1039 enum ib_srq_type {
1040 	IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1041 	IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1042 	IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1043 };
1044 
ib_srq_has_cq(enum ib_srq_type srq_type)1045 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1046 {
1047 	return srq_type == IB_SRQT_XRC ||
1048 	       srq_type == IB_SRQT_TM;
1049 }
1050 
1051 enum ib_srq_attr_mask {
1052 	IB_SRQ_MAX_WR	= 1 << 0,
1053 	IB_SRQ_LIMIT	= 1 << 1,
1054 };
1055 
1056 struct ib_srq_attr {
1057 	u32	max_wr;
1058 	u32	max_sge;
1059 	u32	srq_limit;
1060 };
1061 
1062 struct ib_srq_init_attr {
1063 	void		      (*event_handler)(struct ib_event *, void *);
1064 	void		       *srq_context;
1065 	struct ib_srq_attr	attr;
1066 	enum ib_srq_type	srq_type;
1067 
1068 	struct {
1069 		struct ib_cq   *cq;
1070 		union {
1071 			struct {
1072 				struct ib_xrcd *xrcd;
1073 			} xrc;
1074 
1075 			struct {
1076 				u32		max_num_tags;
1077 			} tag_matching;
1078 		};
1079 	} ext;
1080 };
1081 
1082 struct ib_qp_cap {
1083 	u32	max_send_wr;
1084 	u32	max_recv_wr;
1085 	u32	max_send_sge;
1086 	u32	max_recv_sge;
1087 	u32	max_inline_data;
1088 
1089 	/*
1090 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1091 	 * ib_create_qp() will calculate the right amount of neededed WRs
1092 	 * and MRs based on this.
1093 	 */
1094 	u32	max_rdma_ctxs;
1095 };
1096 
1097 enum ib_sig_type {
1098 	IB_SIGNAL_ALL_WR,
1099 	IB_SIGNAL_REQ_WR
1100 };
1101 
1102 enum ib_qp_type {
1103 	/*
1104 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1105 	 * here (and in that order) since the MAD layer uses them as
1106 	 * indices into a 2-entry table.
1107 	 */
1108 	IB_QPT_SMI,
1109 	IB_QPT_GSI,
1110 
1111 	IB_QPT_RC = IB_UVERBS_QPT_RC,
1112 	IB_QPT_UC = IB_UVERBS_QPT_UC,
1113 	IB_QPT_UD = IB_UVERBS_QPT_UD,
1114 	IB_QPT_RAW_IPV6,
1115 	IB_QPT_RAW_ETHERTYPE,
1116 	IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1117 	IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1118 	IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1119 	IB_QPT_MAX,
1120 	IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1121 	/* Reserve a range for qp types internal to the low level driver.
1122 	 * These qp types will not be visible at the IB core layer, so the
1123 	 * IB_QPT_MAX usages should not be affected in the core layer
1124 	 */
1125 	IB_QPT_RESERVED1 = 0x1000,
1126 	IB_QPT_RESERVED2,
1127 	IB_QPT_RESERVED3,
1128 	IB_QPT_RESERVED4,
1129 	IB_QPT_RESERVED5,
1130 	IB_QPT_RESERVED6,
1131 	IB_QPT_RESERVED7,
1132 	IB_QPT_RESERVED8,
1133 	IB_QPT_RESERVED9,
1134 	IB_QPT_RESERVED10,
1135 };
1136 
1137 enum ib_qp_create_flags {
1138 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1139 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	=
1140 		IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1141 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1142 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1143 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1144 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1145 	IB_QP_CREATE_INTEGRITY_EN		= 1 << 6,
1146 	IB_QP_CREATE_NETDEV_USE			= 1 << 7,
1147 	IB_QP_CREATE_SCATTER_FCS		=
1148 		IB_UVERBS_QP_CREATE_SCATTER_FCS,
1149 	IB_QP_CREATE_CVLAN_STRIPPING		=
1150 		IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1151 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1152 	IB_QP_CREATE_PCI_WRITE_END_PADDING	=
1153 		IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1154 	/* reserve bits 26-31 for low level drivers' internal use */
1155 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1156 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1157 };
1158 
1159 /*
1160  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1161  * callback to destroy the passed in QP.
1162  */
1163 
1164 struct ib_qp_init_attr {
1165 	/* Consumer's event_handler callback must not block */
1166 	void                  (*event_handler)(struct ib_event *, void *);
1167 
1168 	void		       *qp_context;
1169 	struct ib_cq	       *send_cq;
1170 	struct ib_cq	       *recv_cq;
1171 	struct ib_srq	       *srq;
1172 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1173 	struct ib_qp_cap	cap;
1174 	enum ib_sig_type	sq_sig_type;
1175 	enum ib_qp_type		qp_type;
1176 	u32			create_flags;
1177 
1178 	/*
1179 	 * Only needed for special QP types, or when using the RW API.
1180 	 */
1181 	u32			port_num;
1182 	struct ib_rwq_ind_table *rwq_ind_tbl;
1183 	u32			source_qpn;
1184 };
1185 
1186 struct ib_qp_open_attr {
1187 	void                  (*event_handler)(struct ib_event *, void *);
1188 	void		       *qp_context;
1189 	u32			qp_num;
1190 	enum ib_qp_type		qp_type;
1191 };
1192 
1193 enum ib_rnr_timeout {
1194 	IB_RNR_TIMER_655_36 =  0,
1195 	IB_RNR_TIMER_000_01 =  1,
1196 	IB_RNR_TIMER_000_02 =  2,
1197 	IB_RNR_TIMER_000_03 =  3,
1198 	IB_RNR_TIMER_000_04 =  4,
1199 	IB_RNR_TIMER_000_06 =  5,
1200 	IB_RNR_TIMER_000_08 =  6,
1201 	IB_RNR_TIMER_000_12 =  7,
1202 	IB_RNR_TIMER_000_16 =  8,
1203 	IB_RNR_TIMER_000_24 =  9,
1204 	IB_RNR_TIMER_000_32 = 10,
1205 	IB_RNR_TIMER_000_48 = 11,
1206 	IB_RNR_TIMER_000_64 = 12,
1207 	IB_RNR_TIMER_000_96 = 13,
1208 	IB_RNR_TIMER_001_28 = 14,
1209 	IB_RNR_TIMER_001_92 = 15,
1210 	IB_RNR_TIMER_002_56 = 16,
1211 	IB_RNR_TIMER_003_84 = 17,
1212 	IB_RNR_TIMER_005_12 = 18,
1213 	IB_RNR_TIMER_007_68 = 19,
1214 	IB_RNR_TIMER_010_24 = 20,
1215 	IB_RNR_TIMER_015_36 = 21,
1216 	IB_RNR_TIMER_020_48 = 22,
1217 	IB_RNR_TIMER_030_72 = 23,
1218 	IB_RNR_TIMER_040_96 = 24,
1219 	IB_RNR_TIMER_061_44 = 25,
1220 	IB_RNR_TIMER_081_92 = 26,
1221 	IB_RNR_TIMER_122_88 = 27,
1222 	IB_RNR_TIMER_163_84 = 28,
1223 	IB_RNR_TIMER_245_76 = 29,
1224 	IB_RNR_TIMER_327_68 = 30,
1225 	IB_RNR_TIMER_491_52 = 31
1226 };
1227 
1228 enum ib_qp_attr_mask {
1229 	IB_QP_STATE			= 1,
1230 	IB_QP_CUR_STATE			= (1<<1),
1231 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1232 	IB_QP_ACCESS_FLAGS		= (1<<3),
1233 	IB_QP_PKEY_INDEX		= (1<<4),
1234 	IB_QP_PORT			= (1<<5),
1235 	IB_QP_QKEY			= (1<<6),
1236 	IB_QP_AV			= (1<<7),
1237 	IB_QP_PATH_MTU			= (1<<8),
1238 	IB_QP_TIMEOUT			= (1<<9),
1239 	IB_QP_RETRY_CNT			= (1<<10),
1240 	IB_QP_RNR_RETRY			= (1<<11),
1241 	IB_QP_RQ_PSN			= (1<<12),
1242 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1243 	IB_QP_ALT_PATH			= (1<<14),
1244 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1245 	IB_QP_SQ_PSN			= (1<<16),
1246 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1247 	IB_QP_PATH_MIG_STATE		= (1<<18),
1248 	IB_QP_CAP			= (1<<19),
1249 	IB_QP_DEST_QPN			= (1<<20),
1250 	IB_QP_RESERVED1			= (1<<21),
1251 	IB_QP_RESERVED2			= (1<<22),
1252 	IB_QP_RESERVED3			= (1<<23),
1253 	IB_QP_RESERVED4			= (1<<24),
1254 	IB_QP_RATE_LIMIT		= (1<<25),
1255 
1256 	IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1257 };
1258 
1259 enum ib_qp_state {
1260 	IB_QPS_RESET,
1261 	IB_QPS_INIT,
1262 	IB_QPS_RTR,
1263 	IB_QPS_RTS,
1264 	IB_QPS_SQD,
1265 	IB_QPS_SQE,
1266 	IB_QPS_ERR
1267 };
1268 
1269 enum ib_mig_state {
1270 	IB_MIG_MIGRATED,
1271 	IB_MIG_REARM,
1272 	IB_MIG_ARMED
1273 };
1274 
1275 enum ib_mw_type {
1276 	IB_MW_TYPE_1 = 1,
1277 	IB_MW_TYPE_2 = 2
1278 };
1279 
1280 struct ib_qp_attr {
1281 	enum ib_qp_state	qp_state;
1282 	enum ib_qp_state	cur_qp_state;
1283 	enum ib_mtu		path_mtu;
1284 	enum ib_mig_state	path_mig_state;
1285 	u32			qkey;
1286 	u32			rq_psn;
1287 	u32			sq_psn;
1288 	u32			dest_qp_num;
1289 	int			qp_access_flags;
1290 	struct ib_qp_cap	cap;
1291 	struct rdma_ah_attr	ah_attr;
1292 	struct rdma_ah_attr	alt_ah_attr;
1293 	u16			pkey_index;
1294 	u16			alt_pkey_index;
1295 	u8			en_sqd_async_notify;
1296 	u8			sq_draining;
1297 	u8			max_rd_atomic;
1298 	u8			max_dest_rd_atomic;
1299 	u8			min_rnr_timer;
1300 	u32			port_num;
1301 	u8			timeout;
1302 	u8			retry_cnt;
1303 	u8			rnr_retry;
1304 	u32			alt_port_num;
1305 	u8			alt_timeout;
1306 	u32			rate_limit;
1307 	struct net_device	*xmit_slave;
1308 };
1309 
1310 enum ib_wr_opcode {
1311 	/* These are shared with userspace */
1312 	IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1313 	IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1314 	IB_WR_SEND = IB_UVERBS_WR_SEND,
1315 	IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1316 	IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1317 	IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1318 	IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1319 	IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1320 	IB_WR_LSO = IB_UVERBS_WR_TSO,
1321 	IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1322 	IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1323 	IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1324 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1325 		IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1326 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1327 		IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1328 
1329 	/* These are kernel only and can not be issued by userspace */
1330 	IB_WR_REG_MR = 0x20,
1331 	IB_WR_REG_MR_INTEGRITY,
1332 
1333 	/* reserve values for low level drivers' internal use.
1334 	 * These values will not be used at all in the ib core layer.
1335 	 */
1336 	IB_WR_RESERVED1 = 0xf0,
1337 	IB_WR_RESERVED2,
1338 	IB_WR_RESERVED3,
1339 	IB_WR_RESERVED4,
1340 	IB_WR_RESERVED5,
1341 	IB_WR_RESERVED6,
1342 	IB_WR_RESERVED7,
1343 	IB_WR_RESERVED8,
1344 	IB_WR_RESERVED9,
1345 	IB_WR_RESERVED10,
1346 };
1347 
1348 enum ib_send_flags {
1349 	IB_SEND_FENCE		= 1,
1350 	IB_SEND_SIGNALED	= (1<<1),
1351 	IB_SEND_SOLICITED	= (1<<2),
1352 	IB_SEND_INLINE		= (1<<3),
1353 	IB_SEND_IP_CSUM		= (1<<4),
1354 
1355 	/* reserve bits 26-31 for low level drivers' internal use */
1356 	IB_SEND_RESERVED_START	= (1 << 26),
1357 	IB_SEND_RESERVED_END	= (1 << 31),
1358 };
1359 
1360 struct ib_sge {
1361 	u64	addr;
1362 	u32	length;
1363 	u32	lkey;
1364 };
1365 
1366 struct ib_cqe {
1367 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1368 };
1369 
1370 struct ib_send_wr {
1371 	struct ib_send_wr      *next;
1372 	union {
1373 		u64		wr_id;
1374 		struct ib_cqe	*wr_cqe;
1375 	};
1376 	struct ib_sge	       *sg_list;
1377 	int			num_sge;
1378 	enum ib_wr_opcode	opcode;
1379 	int			send_flags;
1380 	union {
1381 		__be32		imm_data;
1382 		u32		invalidate_rkey;
1383 	} ex;
1384 };
1385 
1386 struct ib_rdma_wr {
1387 	struct ib_send_wr	wr;
1388 	u64			remote_addr;
1389 	u32			rkey;
1390 };
1391 
rdma_wr(const struct ib_send_wr * wr)1392 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1393 {
1394 	return container_of(wr, struct ib_rdma_wr, wr);
1395 }
1396 
1397 struct ib_atomic_wr {
1398 	struct ib_send_wr	wr;
1399 	u64			remote_addr;
1400 	u64			compare_add;
1401 	u64			swap;
1402 	u64			compare_add_mask;
1403 	u64			swap_mask;
1404 	u32			rkey;
1405 };
1406 
atomic_wr(const struct ib_send_wr * wr)1407 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1408 {
1409 	return container_of(wr, struct ib_atomic_wr, wr);
1410 }
1411 
1412 struct ib_ud_wr {
1413 	struct ib_send_wr	wr;
1414 	struct ib_ah		*ah;
1415 	void			*header;
1416 	int			hlen;
1417 	int			mss;
1418 	u32			remote_qpn;
1419 	u32			remote_qkey;
1420 	u16			pkey_index; /* valid for GSI only */
1421 	u32			port_num; /* valid for DR SMPs on switch only */
1422 };
1423 
ud_wr(const struct ib_send_wr * wr)1424 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1425 {
1426 	return container_of(wr, struct ib_ud_wr, wr);
1427 }
1428 
1429 struct ib_reg_wr {
1430 	struct ib_send_wr	wr;
1431 	struct ib_mr		*mr;
1432 	u32			key;
1433 	int			access;
1434 };
1435 
reg_wr(const struct ib_send_wr * wr)1436 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1437 {
1438 	return container_of(wr, struct ib_reg_wr, wr);
1439 }
1440 
1441 struct ib_recv_wr {
1442 	struct ib_recv_wr      *next;
1443 	union {
1444 		u64		wr_id;
1445 		struct ib_cqe	*wr_cqe;
1446 	};
1447 	struct ib_sge	       *sg_list;
1448 	int			num_sge;
1449 };
1450 
1451 enum ib_access_flags {
1452 	IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1453 	IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1454 	IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1455 	IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1456 	IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1457 	IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1458 	IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1459 	IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1460 	IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1461 
1462 	IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1463 	IB_ACCESS_SUPPORTED =
1464 		((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1465 };
1466 
1467 /*
1468  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1469  * are hidden here instead of a uapi header!
1470  */
1471 enum ib_mr_rereg_flags {
1472 	IB_MR_REREG_TRANS	= 1,
1473 	IB_MR_REREG_PD		= (1<<1),
1474 	IB_MR_REREG_ACCESS	= (1<<2),
1475 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1476 };
1477 
1478 struct ib_umem;
1479 
1480 enum rdma_remove_reason {
1481 	/*
1482 	 * Userspace requested uobject deletion or initial try
1483 	 * to remove uobject via cleanup. Call could fail
1484 	 */
1485 	RDMA_REMOVE_DESTROY,
1486 	/* Context deletion. This call should delete the actual object itself */
1487 	RDMA_REMOVE_CLOSE,
1488 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1489 	RDMA_REMOVE_DRIVER_REMOVE,
1490 	/* uobj is being cleaned-up before being committed */
1491 	RDMA_REMOVE_ABORT,
1492 	/* The driver failed to destroy the uobject and is being disconnected */
1493 	RDMA_REMOVE_DRIVER_FAILURE,
1494 };
1495 
1496 struct ib_rdmacg_object {
1497 #ifdef CONFIG_CGROUP_RDMA
1498 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1499 #endif
1500 };
1501 
1502 struct ib_ucontext {
1503 	struct ib_device       *device;
1504 	struct ib_uverbs_file  *ufile;
1505 
1506 	struct ib_rdmacg_object	cg_obj;
1507 	/*
1508 	 * Implementation details of the RDMA core, don't use in drivers:
1509 	 */
1510 	struct rdma_restrack_entry res;
1511 	struct xarray mmap_xa;
1512 };
1513 
1514 struct ib_uobject {
1515 	u64			user_handle;	/* handle given to us by userspace */
1516 	/* ufile & ucontext owning this object */
1517 	struct ib_uverbs_file  *ufile;
1518 	/* FIXME, save memory: ufile->context == context */
1519 	struct ib_ucontext     *context;	/* associated user context */
1520 	void		       *object;		/* containing object */
1521 	struct list_head	list;		/* link to context's list */
1522 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1523 	int			id;		/* index into kernel idr */
1524 	struct kref		ref;
1525 	atomic_t		usecnt;		/* protects exclusive access */
1526 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1527 
1528 	const struct uverbs_api_object *uapi_object;
1529 };
1530 
1531 struct ib_udata {
1532 	const void __user *inbuf;
1533 	void __user *outbuf;
1534 	size_t       inlen;
1535 	size_t       outlen;
1536 };
1537 
1538 struct ib_pd {
1539 	u32			local_dma_lkey;
1540 	u32			flags;
1541 	struct ib_device       *device;
1542 	struct ib_uobject      *uobject;
1543 	atomic_t          	usecnt; /* count all resources */
1544 
1545 	u32			unsafe_global_rkey;
1546 
1547 	/*
1548 	 * Implementation details of the RDMA core, don't use in drivers:
1549 	 */
1550 	struct ib_mr	       *__internal_mr;
1551 	struct rdma_restrack_entry res;
1552 };
1553 
1554 struct ib_xrcd {
1555 	struct ib_device       *device;
1556 	atomic_t		usecnt; /* count all exposed resources */
1557 	struct inode	       *inode;
1558 	struct rw_semaphore	tgt_qps_rwsem;
1559 	struct xarray		tgt_qps;
1560 };
1561 
1562 struct ib_ah {
1563 	struct ib_device	*device;
1564 	struct ib_pd		*pd;
1565 	struct ib_uobject	*uobject;
1566 	const struct ib_gid_attr *sgid_attr;
1567 	enum rdma_ah_attr_type	type;
1568 };
1569 
1570 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1571 
1572 enum ib_poll_context {
1573 	IB_POLL_SOFTIRQ,	   /* poll from softirq context */
1574 	IB_POLL_WORKQUEUE,	   /* poll from workqueue */
1575 	IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1576 	IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1577 
1578 	IB_POLL_DIRECT,		   /* caller context, no hw completions */
1579 };
1580 
1581 struct ib_cq {
1582 	struct ib_device       *device;
1583 	struct ib_ucq_object   *uobject;
1584 	ib_comp_handler   	comp_handler;
1585 	void                  (*event_handler)(struct ib_event *, void *);
1586 	void                   *cq_context;
1587 	int               	cqe;
1588 	unsigned int		cqe_used;
1589 	atomic_t          	usecnt; /* count number of work queues */
1590 	enum ib_poll_context	poll_ctx;
1591 	struct ib_wc		*wc;
1592 	struct list_head        pool_entry;
1593 	union {
1594 		struct irq_poll		iop;
1595 		struct work_struct	work;
1596 	};
1597 	struct workqueue_struct *comp_wq;
1598 	struct dim *dim;
1599 
1600 	/* updated only by trace points */
1601 	ktime_t timestamp;
1602 	u8 interrupt:1;
1603 	u8 shared:1;
1604 	unsigned int comp_vector;
1605 
1606 	/*
1607 	 * Implementation details of the RDMA core, don't use in drivers:
1608 	 */
1609 	struct rdma_restrack_entry res;
1610 };
1611 
1612 struct ib_srq {
1613 	struct ib_device       *device;
1614 	struct ib_pd	       *pd;
1615 	struct ib_usrq_object  *uobject;
1616 	void		      (*event_handler)(struct ib_event *, void *);
1617 	void		       *srq_context;
1618 	enum ib_srq_type	srq_type;
1619 	atomic_t		usecnt;
1620 
1621 	struct {
1622 		struct ib_cq   *cq;
1623 		union {
1624 			struct {
1625 				struct ib_xrcd *xrcd;
1626 				u32		srq_num;
1627 			} xrc;
1628 		};
1629 	} ext;
1630 
1631 	/*
1632 	 * Implementation details of the RDMA core, don't use in drivers:
1633 	 */
1634 	struct rdma_restrack_entry res;
1635 };
1636 
1637 enum ib_raw_packet_caps {
1638 	/*
1639 	 * Strip cvlan from incoming packet and report it in the matching work
1640 	 * completion is supported.
1641 	 */
1642 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING =
1643 		IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING,
1644 	/*
1645 	 * Scatter FCS field of an incoming packet to host memory is supported.
1646 	 */
1647 	IB_RAW_PACKET_CAP_SCATTER_FCS = IB_UVERBS_RAW_PACKET_CAP_SCATTER_FCS,
1648 	/* Checksum offloads are supported (for both send and receive). */
1649 	IB_RAW_PACKET_CAP_IP_CSUM = IB_UVERBS_RAW_PACKET_CAP_IP_CSUM,
1650 	/*
1651 	 * When a packet is received for an RQ with no receive WQEs, the
1652 	 * packet processing is delayed.
1653 	 */
1654 	IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP,
1655 };
1656 
1657 enum ib_wq_type {
1658 	IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1659 };
1660 
1661 enum ib_wq_state {
1662 	IB_WQS_RESET,
1663 	IB_WQS_RDY,
1664 	IB_WQS_ERR
1665 };
1666 
1667 struct ib_wq {
1668 	struct ib_device       *device;
1669 	struct ib_uwq_object   *uobject;
1670 	void		    *wq_context;
1671 	void		    (*event_handler)(struct ib_event *, void *);
1672 	struct ib_pd	       *pd;
1673 	struct ib_cq	       *cq;
1674 	u32		wq_num;
1675 	enum ib_wq_state       state;
1676 	enum ib_wq_type	wq_type;
1677 	atomic_t		usecnt;
1678 };
1679 
1680 enum ib_wq_flags {
1681 	IB_WQ_FLAGS_CVLAN_STRIPPING	= IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1682 	IB_WQ_FLAGS_SCATTER_FCS		= IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1683 	IB_WQ_FLAGS_DELAY_DROP		= IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1684 	IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1685 				IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1686 };
1687 
1688 struct ib_wq_init_attr {
1689 	void		       *wq_context;
1690 	enum ib_wq_type	wq_type;
1691 	u32		max_wr;
1692 	u32		max_sge;
1693 	struct	ib_cq	       *cq;
1694 	void		    (*event_handler)(struct ib_event *, void *);
1695 	u32		create_flags; /* Use enum ib_wq_flags */
1696 };
1697 
1698 enum ib_wq_attr_mask {
1699 	IB_WQ_STATE		= 1 << 0,
1700 	IB_WQ_CUR_STATE		= 1 << 1,
1701 	IB_WQ_FLAGS		= 1 << 2,
1702 };
1703 
1704 struct ib_wq_attr {
1705 	enum	ib_wq_state	wq_state;
1706 	enum	ib_wq_state	curr_wq_state;
1707 	u32			flags; /* Use enum ib_wq_flags */
1708 	u32			flags_mask; /* Use enum ib_wq_flags */
1709 };
1710 
1711 struct ib_rwq_ind_table {
1712 	struct ib_device	*device;
1713 	struct ib_uobject      *uobject;
1714 	atomic_t		usecnt;
1715 	u32		ind_tbl_num;
1716 	u32		log_ind_tbl_size;
1717 	struct ib_wq	**ind_tbl;
1718 };
1719 
1720 struct ib_rwq_ind_table_init_attr {
1721 	u32		log_ind_tbl_size;
1722 	/* Each entry is a pointer to Receive Work Queue */
1723 	struct ib_wq	**ind_tbl;
1724 };
1725 
1726 enum port_pkey_state {
1727 	IB_PORT_PKEY_NOT_VALID = 0,
1728 	IB_PORT_PKEY_VALID = 1,
1729 	IB_PORT_PKEY_LISTED = 2,
1730 };
1731 
1732 struct ib_qp_security;
1733 
1734 struct ib_port_pkey {
1735 	enum port_pkey_state	state;
1736 	u16			pkey_index;
1737 	u32			port_num;
1738 	struct list_head	qp_list;
1739 	struct list_head	to_error_list;
1740 	struct ib_qp_security  *sec;
1741 };
1742 
1743 struct ib_ports_pkeys {
1744 	struct ib_port_pkey	main;
1745 	struct ib_port_pkey	alt;
1746 };
1747 
1748 struct ib_qp_security {
1749 	struct ib_qp	       *qp;
1750 	struct ib_device       *dev;
1751 	/* Hold this mutex when changing port and pkey settings. */
1752 	struct mutex		mutex;
1753 	struct ib_ports_pkeys  *ports_pkeys;
1754 	/* A list of all open shared QP handles.  Required to enforce security
1755 	 * properly for all users of a shared QP.
1756 	 */
1757 	struct list_head        shared_qp_list;
1758 	void                   *security;
1759 	bool			destroying;
1760 	atomic_t		error_list_count;
1761 	struct completion	error_complete;
1762 	int			error_comps_pending;
1763 };
1764 
1765 /*
1766  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1767  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1768  */
1769 struct ib_qp {
1770 	struct ib_device       *device;
1771 	struct ib_pd	       *pd;
1772 	struct ib_cq	       *send_cq;
1773 	struct ib_cq	       *recv_cq;
1774 	spinlock_t		mr_lock;
1775 	int			mrs_used;
1776 	struct list_head	rdma_mrs;
1777 	struct list_head	sig_mrs;
1778 	struct ib_srq	       *srq;
1779 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1780 	struct list_head	xrcd_list;
1781 
1782 	/* count times opened, mcast attaches, flow attaches */
1783 	atomic_t		usecnt;
1784 	struct list_head	open_list;
1785 	struct ib_qp           *real_qp;
1786 	struct ib_uqp_object   *uobject;
1787 	void                  (*event_handler)(struct ib_event *, void *);
1788 	void		       *qp_context;
1789 	/* sgid_attrs associated with the AV's */
1790 	const struct ib_gid_attr *av_sgid_attr;
1791 	const struct ib_gid_attr *alt_path_sgid_attr;
1792 	u32			qp_num;
1793 	u32			max_write_sge;
1794 	u32			max_read_sge;
1795 	enum ib_qp_type		qp_type;
1796 	struct ib_rwq_ind_table *rwq_ind_tbl;
1797 	struct ib_qp_security  *qp_sec;
1798 	u32			port;
1799 
1800 	bool			integrity_en;
1801 	/*
1802 	 * Implementation details of the RDMA core, don't use in drivers:
1803 	 */
1804 	struct rdma_restrack_entry     res;
1805 
1806 	/* The counter the qp is bind to */
1807 	struct rdma_counter    *counter;
1808 };
1809 
1810 struct ib_dm {
1811 	struct ib_device  *device;
1812 	u32		   length;
1813 	u32		   flags;
1814 	struct ib_uobject *uobject;
1815 	atomic_t	   usecnt;
1816 };
1817 
1818 struct ib_mr {
1819 	struct ib_device  *device;
1820 	struct ib_pd	  *pd;
1821 	u32		   lkey;
1822 	u32		   rkey;
1823 	u64		   iova;
1824 	u64		   length;
1825 	unsigned int	   page_size;
1826 	enum ib_mr_type	   type;
1827 	bool		   need_inval;
1828 	union {
1829 		struct ib_uobject	*uobject;	/* user */
1830 		struct list_head	qp_entry;	/* FR */
1831 	};
1832 
1833 	struct ib_dm      *dm;
1834 	struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1835 	/*
1836 	 * Implementation details of the RDMA core, don't use in drivers:
1837 	 */
1838 	struct rdma_restrack_entry res;
1839 };
1840 
1841 struct ib_mw {
1842 	struct ib_device	*device;
1843 	struct ib_pd		*pd;
1844 	struct ib_uobject	*uobject;
1845 	u32			rkey;
1846 	enum ib_mw_type         type;
1847 };
1848 
1849 /* Supported steering options */
1850 enum ib_flow_attr_type {
1851 	/* steering according to rule specifications */
1852 	IB_FLOW_ATTR_NORMAL		= 0x0,
1853 	/* default unicast and multicast rule -
1854 	 * receive all Eth traffic which isn't steered to any QP
1855 	 */
1856 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1857 	/* default multicast rule -
1858 	 * receive all Eth multicast traffic which isn't steered to any QP
1859 	 */
1860 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1861 	/* sniffer rule - receive all port traffic */
1862 	IB_FLOW_ATTR_SNIFFER		= 0x3
1863 };
1864 
1865 /* Supported steering header types */
1866 enum ib_flow_spec_type {
1867 	/* L2 headers*/
1868 	IB_FLOW_SPEC_ETH		= 0x20,
1869 	IB_FLOW_SPEC_IB			= 0x22,
1870 	/* L3 header*/
1871 	IB_FLOW_SPEC_IPV4		= 0x30,
1872 	IB_FLOW_SPEC_IPV6		= 0x31,
1873 	IB_FLOW_SPEC_ESP                = 0x34,
1874 	/* L4 headers*/
1875 	IB_FLOW_SPEC_TCP		= 0x40,
1876 	IB_FLOW_SPEC_UDP		= 0x41,
1877 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1878 	IB_FLOW_SPEC_GRE		= 0x51,
1879 	IB_FLOW_SPEC_MPLS		= 0x60,
1880 	IB_FLOW_SPEC_INNER		= 0x100,
1881 	/* Actions */
1882 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1883 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1884 	IB_FLOW_SPEC_ACTION_HANDLE	= 0x1002,
1885 	IB_FLOW_SPEC_ACTION_COUNT       = 0x1003,
1886 };
1887 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1888 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1889 
1890 enum ib_flow_flags {
1891 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1892 	IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1893 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 3  /* Must be last */
1894 };
1895 
1896 struct ib_flow_eth_filter {
1897 	u8	dst_mac[6];
1898 	u8	src_mac[6];
1899 	__be16	ether_type;
1900 	__be16	vlan_tag;
1901 	/* Must be last */
1902 	u8	real_sz[];
1903 };
1904 
1905 struct ib_flow_spec_eth {
1906 	u32			  type;
1907 	u16			  size;
1908 	struct ib_flow_eth_filter val;
1909 	struct ib_flow_eth_filter mask;
1910 };
1911 
1912 struct ib_flow_ib_filter {
1913 	__be16 dlid;
1914 	__u8   sl;
1915 	/* Must be last */
1916 	u8	real_sz[];
1917 };
1918 
1919 struct ib_flow_spec_ib {
1920 	u32			 type;
1921 	u16			 size;
1922 	struct ib_flow_ib_filter val;
1923 	struct ib_flow_ib_filter mask;
1924 };
1925 
1926 /* IPv4 header flags */
1927 enum ib_ipv4_flags {
1928 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1929 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1930 				    last have this flag set */
1931 };
1932 
1933 struct ib_flow_ipv4_filter {
1934 	__be32	src_ip;
1935 	__be32	dst_ip;
1936 	u8	proto;
1937 	u8	tos;
1938 	u8	ttl;
1939 	u8	flags;
1940 	/* Must be last */
1941 	u8	real_sz[];
1942 };
1943 
1944 struct ib_flow_spec_ipv4 {
1945 	u32			   type;
1946 	u16			   size;
1947 	struct ib_flow_ipv4_filter val;
1948 	struct ib_flow_ipv4_filter mask;
1949 };
1950 
1951 struct ib_flow_ipv6_filter {
1952 	u8	src_ip[16];
1953 	u8	dst_ip[16];
1954 	__be32	flow_label;
1955 	u8	next_hdr;
1956 	u8	traffic_class;
1957 	u8	hop_limit;
1958 	/* Must be last */
1959 	u8	real_sz[];
1960 };
1961 
1962 struct ib_flow_spec_ipv6 {
1963 	u32			   type;
1964 	u16			   size;
1965 	struct ib_flow_ipv6_filter val;
1966 	struct ib_flow_ipv6_filter mask;
1967 };
1968 
1969 struct ib_flow_tcp_udp_filter {
1970 	__be16	dst_port;
1971 	__be16	src_port;
1972 	/* Must be last */
1973 	u8	real_sz[];
1974 };
1975 
1976 struct ib_flow_spec_tcp_udp {
1977 	u32			      type;
1978 	u16			      size;
1979 	struct ib_flow_tcp_udp_filter val;
1980 	struct ib_flow_tcp_udp_filter mask;
1981 };
1982 
1983 struct ib_flow_tunnel_filter {
1984 	__be32	tunnel_id;
1985 	u8	real_sz[];
1986 };
1987 
1988 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1989  * the tunnel_id from val has the vni value
1990  */
1991 struct ib_flow_spec_tunnel {
1992 	u32			      type;
1993 	u16			      size;
1994 	struct ib_flow_tunnel_filter  val;
1995 	struct ib_flow_tunnel_filter  mask;
1996 };
1997 
1998 struct ib_flow_esp_filter {
1999 	__be32	spi;
2000 	__be32  seq;
2001 	/* Must be last */
2002 	u8	real_sz[];
2003 };
2004 
2005 struct ib_flow_spec_esp {
2006 	u32                           type;
2007 	u16			      size;
2008 	struct ib_flow_esp_filter     val;
2009 	struct ib_flow_esp_filter     mask;
2010 };
2011 
2012 struct ib_flow_gre_filter {
2013 	__be16 c_ks_res0_ver;
2014 	__be16 protocol;
2015 	__be32 key;
2016 	/* Must be last */
2017 	u8	real_sz[];
2018 };
2019 
2020 struct ib_flow_spec_gre {
2021 	u32                           type;
2022 	u16			      size;
2023 	struct ib_flow_gre_filter     val;
2024 	struct ib_flow_gre_filter     mask;
2025 };
2026 
2027 struct ib_flow_mpls_filter {
2028 	__be32 tag;
2029 	/* Must be last */
2030 	u8	real_sz[];
2031 };
2032 
2033 struct ib_flow_spec_mpls {
2034 	u32                           type;
2035 	u16			      size;
2036 	struct ib_flow_mpls_filter     val;
2037 	struct ib_flow_mpls_filter     mask;
2038 };
2039 
2040 struct ib_flow_spec_action_tag {
2041 	enum ib_flow_spec_type	      type;
2042 	u16			      size;
2043 	u32                           tag_id;
2044 };
2045 
2046 struct ib_flow_spec_action_drop {
2047 	enum ib_flow_spec_type	      type;
2048 	u16			      size;
2049 };
2050 
2051 struct ib_flow_spec_action_handle {
2052 	enum ib_flow_spec_type	      type;
2053 	u16			      size;
2054 	struct ib_flow_action	     *act;
2055 };
2056 
2057 enum ib_counters_description {
2058 	IB_COUNTER_PACKETS,
2059 	IB_COUNTER_BYTES,
2060 };
2061 
2062 struct ib_flow_spec_action_count {
2063 	enum ib_flow_spec_type type;
2064 	u16 size;
2065 	struct ib_counters *counters;
2066 };
2067 
2068 union ib_flow_spec {
2069 	struct {
2070 		u32			type;
2071 		u16			size;
2072 	};
2073 	struct ib_flow_spec_eth		eth;
2074 	struct ib_flow_spec_ib		ib;
2075 	struct ib_flow_spec_ipv4        ipv4;
2076 	struct ib_flow_spec_tcp_udp	tcp_udp;
2077 	struct ib_flow_spec_ipv6        ipv6;
2078 	struct ib_flow_spec_tunnel      tunnel;
2079 	struct ib_flow_spec_esp		esp;
2080 	struct ib_flow_spec_gre		gre;
2081 	struct ib_flow_spec_mpls	mpls;
2082 	struct ib_flow_spec_action_tag  flow_tag;
2083 	struct ib_flow_spec_action_drop drop;
2084 	struct ib_flow_spec_action_handle action;
2085 	struct ib_flow_spec_action_count flow_count;
2086 };
2087 
2088 struct ib_flow_attr {
2089 	enum ib_flow_attr_type type;
2090 	u16	     size;
2091 	u16	     priority;
2092 	u32	     flags;
2093 	u8	     num_of_specs;
2094 	u32	     port;
2095 	union ib_flow_spec flows[];
2096 };
2097 
2098 struct ib_flow {
2099 	struct ib_qp		*qp;
2100 	struct ib_device	*device;
2101 	struct ib_uobject	*uobject;
2102 };
2103 
2104 enum ib_flow_action_type {
2105 	IB_FLOW_ACTION_UNSPECIFIED,
2106 	IB_FLOW_ACTION_ESP = 1,
2107 };
2108 
2109 struct ib_flow_action_attrs_esp_keymats {
2110 	enum ib_uverbs_flow_action_esp_keymat			protocol;
2111 	union {
2112 		struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2113 	} keymat;
2114 };
2115 
2116 struct ib_flow_action_attrs_esp_replays {
2117 	enum ib_uverbs_flow_action_esp_replay			protocol;
2118 	union {
2119 		struct ib_uverbs_flow_action_esp_replay_bmp	bmp;
2120 	} replay;
2121 };
2122 
2123 enum ib_flow_action_attrs_esp_flags {
2124 	/* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2125 	 * This is done in order to share the same flags between user-space and
2126 	 * kernel and spare an unnecessary translation.
2127 	 */
2128 
2129 	/* Kernel flags */
2130 	IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED	= 1ULL << 32,
2131 	IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS	= 1ULL << 33,
2132 };
2133 
2134 struct ib_flow_spec_list {
2135 	struct ib_flow_spec_list	*next;
2136 	union ib_flow_spec		spec;
2137 };
2138 
2139 struct ib_flow_action_attrs_esp {
2140 	struct ib_flow_action_attrs_esp_keymats		*keymat;
2141 	struct ib_flow_action_attrs_esp_replays		*replay;
2142 	struct ib_flow_spec_list			*encap;
2143 	/* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2144 	 * Value of 0 is a valid value.
2145 	 */
2146 	u32						esn;
2147 	u32						spi;
2148 	u32						seq;
2149 	u32						tfc_pad;
2150 	/* Use enum ib_flow_action_attrs_esp_flags */
2151 	u64						flags;
2152 	u64						hard_limit_pkts;
2153 };
2154 
2155 struct ib_flow_action {
2156 	struct ib_device		*device;
2157 	struct ib_uobject		*uobject;
2158 	enum ib_flow_action_type	type;
2159 	atomic_t			usecnt;
2160 };
2161 
2162 struct ib_mad;
2163 
2164 enum ib_process_mad_flags {
2165 	IB_MAD_IGNORE_MKEY	= 1,
2166 	IB_MAD_IGNORE_BKEY	= 2,
2167 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2168 };
2169 
2170 enum ib_mad_result {
2171 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
2172 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
2173 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
2174 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
2175 };
2176 
2177 struct ib_port_cache {
2178 	u64		      subnet_prefix;
2179 	struct ib_pkey_cache  *pkey;
2180 	struct ib_gid_table   *gid;
2181 	u8                     lmc;
2182 	enum ib_port_state     port_state;
2183 };
2184 
2185 struct ib_port_immutable {
2186 	int                           pkey_tbl_len;
2187 	int                           gid_tbl_len;
2188 	u32                           core_cap_flags;
2189 	u32                           max_mad_size;
2190 };
2191 
2192 struct ib_port_data {
2193 	struct ib_device *ib_dev;
2194 
2195 	struct ib_port_immutable immutable;
2196 
2197 	spinlock_t pkey_list_lock;
2198 
2199 	spinlock_t netdev_lock;
2200 
2201 	struct list_head pkey_list;
2202 
2203 	struct ib_port_cache cache;
2204 
2205 	struct net_device __rcu *netdev;
2206 	struct hlist_node ndev_hash_link;
2207 	struct rdma_port_counter port_counter;
2208 	struct ib_port *sysfs;
2209 };
2210 
2211 /* rdma netdev type - specifies protocol type */
2212 enum rdma_netdev_t {
2213 	RDMA_NETDEV_OPA_VNIC,
2214 	RDMA_NETDEV_IPOIB,
2215 };
2216 
2217 /**
2218  * struct rdma_netdev - rdma netdev
2219  * For cases where netstack interfacing is required.
2220  */
2221 struct rdma_netdev {
2222 	void              *clnt_priv;
2223 	struct ib_device  *hca;
2224 	u32		   port_num;
2225 	int                mtu;
2226 
2227 	/*
2228 	 * cleanup function must be specified.
2229 	 * FIXME: This is only used for OPA_VNIC and that usage should be
2230 	 * removed too.
2231 	 */
2232 	void (*free_rdma_netdev)(struct net_device *netdev);
2233 
2234 	/* control functions */
2235 	void (*set_id)(struct net_device *netdev, int id);
2236 	/* send packet */
2237 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2238 		    struct ib_ah *address, u32 dqpn);
2239 	/* multicast */
2240 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2241 			    union ib_gid *gid, u16 mlid,
2242 			    int set_qkey, u32 qkey);
2243 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2244 			    union ib_gid *gid, u16 mlid);
2245 	/* timeout */
2246 	void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2247 };
2248 
2249 struct rdma_netdev_alloc_params {
2250 	size_t sizeof_priv;
2251 	unsigned int txqs;
2252 	unsigned int rxqs;
2253 	void *param;
2254 
2255 	int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2256 				      struct net_device *netdev, void *param);
2257 };
2258 
2259 struct ib_odp_counters {
2260 	atomic64_t faults;
2261 	atomic64_t invalidations;
2262 	atomic64_t prefetch;
2263 };
2264 
2265 struct ib_counters {
2266 	struct ib_device	*device;
2267 	struct ib_uobject	*uobject;
2268 	/* num of objects attached */
2269 	atomic_t	usecnt;
2270 };
2271 
2272 struct ib_counters_read_attr {
2273 	u64	*counters_buff;
2274 	u32	ncounters;
2275 	u32	flags; /* use enum ib_read_counters_flags */
2276 };
2277 
2278 struct uverbs_attr_bundle;
2279 struct iw_cm_id;
2280 struct iw_cm_conn_param;
2281 
2282 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member)                      \
2283 	.size_##ib_struct =                                                    \
2284 		(sizeof(struct drv_struct) +                                   \
2285 		 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) +      \
2286 		 BUILD_BUG_ON_ZERO(                                            \
2287 			 !__same_type(((struct drv_struct *)NULL)->member,     \
2288 				      struct ib_struct)))
2289 
2290 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp)                          \
2291 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2292 					   gfp, false))
2293 
2294 #define rdma_zalloc_drv_obj_numa(ib_dev, ib_type)                              \
2295 	((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2296 					   GFP_KERNEL, true))
2297 
2298 #define rdma_zalloc_drv_obj(ib_dev, ib_type)                                   \
2299 	rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2300 
2301 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2302 
2303 struct rdma_user_mmap_entry {
2304 	struct kref ref;
2305 	struct ib_ucontext *ucontext;
2306 	unsigned long start_pgoff;
2307 	size_t npages;
2308 	bool driver_removed;
2309 };
2310 
2311 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2312 static inline u64
rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry * entry)2313 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2314 {
2315 	return (u64)entry->start_pgoff << PAGE_SHIFT;
2316 }
2317 
2318 /**
2319  * struct ib_device_ops - InfiniBand device operations
2320  * This structure defines all the InfiniBand device operations, providers will
2321  * need to define the supported operations, otherwise they will be set to null.
2322  */
2323 struct ib_device_ops {
2324 	struct module *owner;
2325 	enum rdma_driver_id driver_id;
2326 	u32 uverbs_abi_ver;
2327 	unsigned int uverbs_no_driver_id_binding:1;
2328 
2329 	/*
2330 	 * NOTE: New drivers should not make use of device_group; instead new
2331 	 * device parameter should be exposed via netlink command. This
2332 	 * mechanism exists only for existing drivers.
2333 	 */
2334 	const struct attribute_group *device_group;
2335 	const struct attribute_group **port_groups;
2336 
2337 	int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2338 			 const struct ib_send_wr **bad_send_wr);
2339 	int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2340 			 const struct ib_recv_wr **bad_recv_wr);
2341 	void (*drain_rq)(struct ib_qp *qp);
2342 	void (*drain_sq)(struct ib_qp *qp);
2343 	int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2344 	int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2345 	int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2346 	int (*post_srq_recv)(struct ib_srq *srq,
2347 			     const struct ib_recv_wr *recv_wr,
2348 			     const struct ib_recv_wr **bad_recv_wr);
2349 	int (*process_mad)(struct ib_device *device, int process_mad_flags,
2350 			   u32 port_num, const struct ib_wc *in_wc,
2351 			   const struct ib_grh *in_grh,
2352 			   const struct ib_mad *in_mad, struct ib_mad *out_mad,
2353 			   size_t *out_mad_size, u16 *out_mad_pkey_index);
2354 	int (*query_device)(struct ib_device *device,
2355 			    struct ib_device_attr *device_attr,
2356 			    struct ib_udata *udata);
2357 	int (*modify_device)(struct ib_device *device, int device_modify_mask,
2358 			     struct ib_device_modify *device_modify);
2359 	void (*get_dev_fw_str)(struct ib_device *device, char *str);
2360 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2361 						     int comp_vector);
2362 	int (*query_port)(struct ib_device *device, u32 port_num,
2363 			  struct ib_port_attr *port_attr);
2364 	int (*modify_port)(struct ib_device *device, u32 port_num,
2365 			   int port_modify_mask,
2366 			   struct ib_port_modify *port_modify);
2367 	/**
2368 	 * The following mandatory functions are used only at device
2369 	 * registration.  Keep functions such as these at the end of this
2370 	 * structure to avoid cache line misses when accessing struct ib_device
2371 	 * in fast paths.
2372 	 */
2373 	int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2374 				  struct ib_port_immutable *immutable);
2375 	enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2376 					       u32 port_num);
2377 	/**
2378 	 * When calling get_netdev, the HW vendor's driver should return the
2379 	 * net device of device @device at port @port_num or NULL if such
2380 	 * a net device doesn't exist. The vendor driver should call dev_hold
2381 	 * on this net device. The HW vendor's device driver must guarantee
2382 	 * that this function returns NULL before the net device has finished
2383 	 * NETDEV_UNREGISTER state.
2384 	 */
2385 	struct net_device *(*get_netdev)(struct ib_device *device,
2386 					 u32 port_num);
2387 	/**
2388 	 * rdma netdev operation
2389 	 *
2390 	 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2391 	 * must return -EOPNOTSUPP if it doesn't support the specified type.
2392 	 */
2393 	struct net_device *(*alloc_rdma_netdev)(
2394 		struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2395 		const char *name, unsigned char name_assign_type,
2396 		void (*setup)(struct net_device *));
2397 
2398 	int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2399 				      enum rdma_netdev_t type,
2400 				      struct rdma_netdev_alloc_params *params);
2401 	/**
2402 	 * query_gid should be return GID value for @device, when @port_num
2403 	 * link layer is either IB or iWarp. It is no-op if @port_num port
2404 	 * is RoCE link layer.
2405 	 */
2406 	int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2407 			 union ib_gid *gid);
2408 	/**
2409 	 * When calling add_gid, the HW vendor's driver should add the gid
2410 	 * of device of port at gid index available at @attr. Meta-info of
2411 	 * that gid (for example, the network device related to this gid) is
2412 	 * available at @attr. @context allows the HW vendor driver to store
2413 	 * extra information together with a GID entry. The HW vendor driver may
2414 	 * allocate memory to contain this information and store it in @context
2415 	 * when a new GID entry is written to. Params are consistent until the
2416 	 * next call of add_gid or delete_gid. The function should return 0 on
2417 	 * success or error otherwise. The function could be called
2418 	 * concurrently for different ports. This function is only called when
2419 	 * roce_gid_table is used.
2420 	 */
2421 	int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2422 	/**
2423 	 * When calling del_gid, the HW vendor's driver should delete the
2424 	 * gid of device @device at gid index gid_index of port port_num
2425 	 * available in @attr.
2426 	 * Upon the deletion of a GID entry, the HW vendor must free any
2427 	 * allocated memory. The caller will clear @context afterwards.
2428 	 * This function is only called when roce_gid_table is used.
2429 	 */
2430 	int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2431 	int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2432 			  u16 *pkey);
2433 	int (*alloc_ucontext)(struct ib_ucontext *context,
2434 			      struct ib_udata *udata);
2435 	void (*dealloc_ucontext)(struct ib_ucontext *context);
2436 	int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2437 	/**
2438 	 * This will be called once refcount of an entry in mmap_xa reaches
2439 	 * zero. The type of the memory that was mapped may differ between
2440 	 * entries and is opaque to the rdma_user_mmap interface.
2441 	 * Therefore needs to be implemented by the driver in mmap_free.
2442 	 */
2443 	void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2444 	void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2445 	int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2446 	int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2447 	int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2448 			 struct ib_udata *udata);
2449 	int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2450 			      struct ib_udata *udata);
2451 	int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2452 	int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2453 	int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2454 	int (*create_srq)(struct ib_srq *srq,
2455 			  struct ib_srq_init_attr *srq_init_attr,
2456 			  struct ib_udata *udata);
2457 	int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2458 			  enum ib_srq_attr_mask srq_attr_mask,
2459 			  struct ib_udata *udata);
2460 	int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2461 	int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2462 	int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2463 			 struct ib_udata *udata);
2464 	int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2465 			 int qp_attr_mask, struct ib_udata *udata);
2466 	int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2467 			int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2468 	int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2469 	int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2470 			 struct ib_udata *udata);
2471 	int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2472 	int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2473 	int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2474 	struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2475 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2476 				     u64 virt_addr, int mr_access_flags,
2477 				     struct ib_udata *udata);
2478 	struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2479 					    u64 length, u64 virt_addr, int fd,
2480 					    int mr_access_flags,
2481 					    struct ib_udata *udata);
2482 	struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2483 				       u64 length, u64 virt_addr,
2484 				       int mr_access_flags, struct ib_pd *pd,
2485 				       struct ib_udata *udata);
2486 	int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2487 	struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2488 				  u32 max_num_sg);
2489 	struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2490 					    u32 max_num_data_sg,
2491 					    u32 max_num_meta_sg);
2492 	int (*advise_mr)(struct ib_pd *pd,
2493 			 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2494 			 struct ib_sge *sg_list, u32 num_sge,
2495 			 struct uverbs_attr_bundle *attrs);
2496 
2497 	/*
2498 	 * Kernel users should universally support relaxed ordering (RO), as
2499 	 * they are designed to read data only after observing the CQE and use
2500 	 * the DMA API correctly.
2501 	 *
2502 	 * Some drivers implicitly enable RO if platform supports it.
2503 	 */
2504 	int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2505 			 unsigned int *sg_offset);
2506 	int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2507 			       struct ib_mr_status *mr_status);
2508 	int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2509 	int (*dealloc_mw)(struct ib_mw *mw);
2510 	int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2511 	int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2512 	int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2513 	int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2514 	struct ib_flow *(*create_flow)(struct ib_qp *qp,
2515 				       struct ib_flow_attr *flow_attr,
2516 				       struct ib_udata *udata);
2517 	int (*destroy_flow)(struct ib_flow *flow_id);
2518 	int (*destroy_flow_action)(struct ib_flow_action *action);
2519 	int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2520 				 int state);
2521 	int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2522 			     struct ifla_vf_info *ivf);
2523 	int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2524 			    struct ifla_vf_stats *stats);
2525 	int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2526 			    struct ifla_vf_guid *node_guid,
2527 			    struct ifla_vf_guid *port_guid);
2528 	int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2529 			   int type);
2530 	struct ib_wq *(*create_wq)(struct ib_pd *pd,
2531 				   struct ib_wq_init_attr *init_attr,
2532 				   struct ib_udata *udata);
2533 	int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2534 	int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2535 			 u32 wq_attr_mask, struct ib_udata *udata);
2536 	int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2537 				    struct ib_rwq_ind_table_init_attr *init_attr,
2538 				    struct ib_udata *udata);
2539 	int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2540 	struct ib_dm *(*alloc_dm)(struct ib_device *device,
2541 				  struct ib_ucontext *context,
2542 				  struct ib_dm_alloc_attr *attr,
2543 				  struct uverbs_attr_bundle *attrs);
2544 	int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2545 	struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2546 				   struct ib_dm_mr_attr *attr,
2547 				   struct uverbs_attr_bundle *attrs);
2548 	int (*create_counters)(struct ib_counters *counters,
2549 			       struct uverbs_attr_bundle *attrs);
2550 	int (*destroy_counters)(struct ib_counters *counters);
2551 	int (*read_counters)(struct ib_counters *counters,
2552 			     struct ib_counters_read_attr *counters_read_attr,
2553 			     struct uverbs_attr_bundle *attrs);
2554 	int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2555 			    int data_sg_nents, unsigned int *data_sg_offset,
2556 			    struct scatterlist *meta_sg, int meta_sg_nents,
2557 			    unsigned int *meta_sg_offset);
2558 
2559 	/**
2560 	 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2561 	 *   fill in the driver initialized data.  The struct is kfree()'ed by
2562 	 *   the sysfs core when the device is removed.  A lifespan of -1 in the
2563 	 *   return struct tells the core to set a default lifespan.
2564 	 */
2565 	struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2566 	struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2567 						     u32 port_num);
2568 	/**
2569 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2570 	 * @index - The index in the value array we wish to have updated, or
2571 	 *   num_counters if we want all stats updated
2572 	 * Return codes -
2573 	 *   < 0 - Error, no counters updated
2574 	 *   index - Updated the single counter pointed to by index
2575 	 *   num_counters - Updated all counters (will reset the timestamp
2576 	 *     and prevent further calls for lifespan milliseconds)
2577 	 * Drivers are allowed to update all counters in leiu of just the
2578 	 *   one given in index at their option
2579 	 */
2580 	int (*get_hw_stats)(struct ib_device *device,
2581 			    struct rdma_hw_stats *stats, u32 port, int index);
2582 
2583 	/**
2584 	 * modify_hw_stat - Modify the counter configuration
2585 	 * @enable: true/false when enable/disable a counter
2586 	 * Return codes - 0 on success or error code otherwise.
2587 	 */
2588 	int (*modify_hw_stat)(struct ib_device *device, u32 port,
2589 			      unsigned int counter_index, bool enable);
2590 	/**
2591 	 * Allows rdma drivers to add their own restrack attributes.
2592 	 */
2593 	int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2594 	int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2595 	int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2596 	int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2597 	int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2598 	int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2599 	int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2600 
2601 	/* Device lifecycle callbacks */
2602 	/*
2603 	 * Called after the device becomes registered, before clients are
2604 	 * attached
2605 	 */
2606 	int (*enable_driver)(struct ib_device *dev);
2607 	/*
2608 	 * This is called as part of ib_dealloc_device().
2609 	 */
2610 	void (*dealloc_driver)(struct ib_device *dev);
2611 
2612 	/* iWarp CM callbacks */
2613 	void (*iw_add_ref)(struct ib_qp *qp);
2614 	void (*iw_rem_ref)(struct ib_qp *qp);
2615 	struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2616 	int (*iw_connect)(struct iw_cm_id *cm_id,
2617 			  struct iw_cm_conn_param *conn_param);
2618 	int (*iw_accept)(struct iw_cm_id *cm_id,
2619 			 struct iw_cm_conn_param *conn_param);
2620 	int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2621 			 u8 pdata_len);
2622 	int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2623 	int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2624 	/**
2625 	 * counter_bind_qp - Bind a QP to a counter.
2626 	 * @counter - The counter to be bound. If counter->id is zero then
2627 	 *   the driver needs to allocate a new counter and set counter->id
2628 	 */
2629 	int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2630 	/**
2631 	 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2632 	 *   counter and bind it onto the default one
2633 	 */
2634 	int (*counter_unbind_qp)(struct ib_qp *qp);
2635 	/**
2636 	 * counter_dealloc -De-allocate the hw counter
2637 	 */
2638 	int (*counter_dealloc)(struct rdma_counter *counter);
2639 	/**
2640 	 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2641 	 * the driver initialized data.
2642 	 */
2643 	struct rdma_hw_stats *(*counter_alloc_stats)(
2644 		struct rdma_counter *counter);
2645 	/**
2646 	 * counter_update_stats - Query the stats value of this counter
2647 	 */
2648 	int (*counter_update_stats)(struct rdma_counter *counter);
2649 
2650 	/**
2651 	 * Allows rdma drivers to add their own restrack attributes
2652 	 * dumped via 'rdma stat' iproute2 command.
2653 	 */
2654 	int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2655 
2656 	/* query driver for its ucontext properties */
2657 	int (*query_ucontext)(struct ib_ucontext *context,
2658 			      struct uverbs_attr_bundle *attrs);
2659 
2660 	/*
2661 	 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2662 	 * Everyone else relies on Linux memory management model.
2663 	 */
2664 	int (*get_numa_node)(struct ib_device *dev);
2665 
2666 	DECLARE_RDMA_OBJ_SIZE(ib_ah);
2667 	DECLARE_RDMA_OBJ_SIZE(ib_counters);
2668 	DECLARE_RDMA_OBJ_SIZE(ib_cq);
2669 	DECLARE_RDMA_OBJ_SIZE(ib_mw);
2670 	DECLARE_RDMA_OBJ_SIZE(ib_pd);
2671 	DECLARE_RDMA_OBJ_SIZE(ib_qp);
2672 	DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2673 	DECLARE_RDMA_OBJ_SIZE(ib_srq);
2674 	DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2675 	DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2676 };
2677 
2678 struct ib_core_device {
2679 	/* device must be the first element in structure until,
2680 	 * union of ib_core_device and device exists in ib_device.
2681 	 */
2682 	struct device dev;
2683 	possible_net_t rdma_net;
2684 	struct kobject *ports_kobj;
2685 	struct list_head port_list;
2686 	struct ib_device *owner; /* reach back to owner ib_device */
2687 };
2688 
2689 struct rdma_restrack_root;
2690 struct ib_device {
2691 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2692 	struct device                *dma_device;
2693 	struct ib_device_ops	     ops;
2694 	char                          name[IB_DEVICE_NAME_MAX];
2695 	struct rcu_head rcu_head;
2696 
2697 	struct list_head              event_handler_list;
2698 	/* Protects event_handler_list */
2699 	struct rw_semaphore event_handler_rwsem;
2700 
2701 	/* Protects QP's event_handler calls and open_qp list */
2702 	spinlock_t qp_open_list_lock;
2703 
2704 	struct rw_semaphore	      client_data_rwsem;
2705 	struct xarray                 client_data;
2706 	struct mutex                  unregistration_lock;
2707 
2708 	/* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2709 	rwlock_t cache_lock;
2710 	/**
2711 	 * port_data is indexed by port number
2712 	 */
2713 	struct ib_port_data *port_data;
2714 
2715 	int			      num_comp_vectors;
2716 
2717 	union {
2718 		struct device		dev;
2719 		struct ib_core_device	coredev;
2720 	};
2721 
2722 	/* First group is for device attributes,
2723 	 * Second group is for driver provided attributes (optional).
2724 	 * Third group is for the hw_stats
2725 	 * It is a NULL terminated array.
2726 	 */
2727 	const struct attribute_group	*groups[4];
2728 
2729 	u64			     uverbs_cmd_mask;
2730 
2731 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2732 	__be64			     node_guid;
2733 	u32			     local_dma_lkey;
2734 	u16                          is_switch:1;
2735 	/* Indicates kernel verbs support, should not be used in drivers */
2736 	u16                          kverbs_provider:1;
2737 	/* CQ adaptive moderation (RDMA DIM) */
2738 	u16                          use_cq_dim:1;
2739 	u8                           node_type;
2740 	u32			     phys_port_cnt;
2741 	struct ib_device_attr        attrs;
2742 	struct hw_stats_device_data *hw_stats_data;
2743 
2744 #ifdef CONFIG_CGROUP_RDMA
2745 	struct rdmacg_device         cg_device;
2746 #endif
2747 
2748 	u32                          index;
2749 
2750 	spinlock_t                   cq_pools_lock;
2751 	struct list_head             cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2752 
2753 	struct rdma_restrack_root *res;
2754 
2755 	const struct uapi_definition   *driver_def;
2756 
2757 	/*
2758 	 * Positive refcount indicates that the device is currently
2759 	 * registered and cannot be unregistered.
2760 	 */
2761 	refcount_t refcount;
2762 	struct completion unreg_completion;
2763 	struct work_struct unregistration_work;
2764 
2765 	const struct rdma_link_ops *link_ops;
2766 
2767 	/* Protects compat_devs xarray modifications */
2768 	struct mutex compat_devs_mutex;
2769 	/* Maintains compat devices for each net namespace */
2770 	struct xarray compat_devs;
2771 
2772 	/* Used by iWarp CM */
2773 	char iw_ifname[IFNAMSIZ];
2774 	u32 iw_driver_flags;
2775 	u32 lag_flags;
2776 };
2777 
rdma_zalloc_obj(struct ib_device * dev,size_t size,gfp_t gfp,bool is_numa_aware)2778 static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
2779 				    gfp_t gfp, bool is_numa_aware)
2780 {
2781 	if (is_numa_aware && dev->ops.get_numa_node)
2782 		return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
2783 
2784 	return kzalloc(size, gfp);
2785 }
2786 
2787 struct ib_client_nl_info;
2788 struct ib_client {
2789 	const char *name;
2790 	int (*add)(struct ib_device *ibdev);
2791 	void (*remove)(struct ib_device *, void *client_data);
2792 	void (*rename)(struct ib_device *dev, void *client_data);
2793 	int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2794 			   struct ib_client_nl_info *res);
2795 	int (*get_global_nl_info)(struct ib_client_nl_info *res);
2796 
2797 	/* Returns the net_dev belonging to this ib_client and matching the
2798 	 * given parameters.
2799 	 * @dev:	 An RDMA device that the net_dev use for communication.
2800 	 * @port:	 A physical port number on the RDMA device.
2801 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2802 	 * @gid:	 A GID that the net_dev uses to communicate.
2803 	 * @addr:	 An IP address the net_dev is configured with.
2804 	 * @client_data: The device's client data set by ib_set_client_data().
2805 	 *
2806 	 * An ib_client that implements a net_dev on top of RDMA devices
2807 	 * (such as IP over IB) should implement this callback, allowing the
2808 	 * rdma_cm module to find the right net_dev for a given request.
2809 	 *
2810 	 * The caller is responsible for calling dev_put on the returned
2811 	 * netdev. */
2812 	struct net_device *(*get_net_dev_by_params)(
2813 			struct ib_device *dev,
2814 			u32 port,
2815 			u16 pkey,
2816 			const union ib_gid *gid,
2817 			const struct sockaddr *addr,
2818 			void *client_data);
2819 
2820 	refcount_t uses;
2821 	struct completion uses_zero;
2822 	u32 client_id;
2823 
2824 	/* kverbs are not required by the client */
2825 	u8 no_kverbs_req:1;
2826 };
2827 
2828 /*
2829  * IB block DMA iterator
2830  *
2831  * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2832  * to a HW supported page size.
2833  */
2834 struct ib_block_iter {
2835 	/* internal states */
2836 	struct scatterlist *__sg;	/* sg holding the current aligned block */
2837 	dma_addr_t __dma_addr;		/* unaligned DMA address of this block */
2838 	unsigned int __sg_nents;	/* number of SG entries */
2839 	unsigned int __sg_advance;	/* number of bytes to advance in sg in next step */
2840 	unsigned int __pg_bit;		/* alignment of current block */
2841 };
2842 
2843 struct ib_device *_ib_alloc_device(size_t size);
2844 #define ib_alloc_device(drv_struct, member)                                    \
2845 	container_of(_ib_alloc_device(sizeof(struct drv_struct) +              \
2846 				      BUILD_BUG_ON_ZERO(offsetof(              \
2847 					      struct drv_struct, member))),    \
2848 		     struct drv_struct, member)
2849 
2850 void ib_dealloc_device(struct ib_device *device);
2851 
2852 void ib_get_device_fw_str(struct ib_device *device, char *str);
2853 
2854 int ib_register_device(struct ib_device *device, const char *name,
2855 		       struct device *dma_device);
2856 void ib_unregister_device(struct ib_device *device);
2857 void ib_unregister_driver(enum rdma_driver_id driver_id);
2858 void ib_unregister_device_and_put(struct ib_device *device);
2859 void ib_unregister_device_queued(struct ib_device *ib_dev);
2860 
2861 int ib_register_client   (struct ib_client *client);
2862 void ib_unregister_client(struct ib_client *client);
2863 
2864 void __rdma_block_iter_start(struct ib_block_iter *biter,
2865 			     struct scatterlist *sglist,
2866 			     unsigned int nents,
2867 			     unsigned long pgsz);
2868 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2869 
2870 /**
2871  * rdma_block_iter_dma_address - get the aligned dma address of the current
2872  * block held by the block iterator.
2873  * @biter: block iterator holding the memory block
2874  */
2875 static inline dma_addr_t
rdma_block_iter_dma_address(struct ib_block_iter * biter)2876 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2877 {
2878 	return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2879 }
2880 
2881 /**
2882  * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2883  * @sglist: sglist to iterate over
2884  * @biter: block iterator holding the memory block
2885  * @nents: maximum number of sg entries to iterate over
2886  * @pgsz: best HW supported page size to use
2887  *
2888  * Callers may use rdma_block_iter_dma_address() to get each
2889  * blocks aligned DMA address.
2890  */
2891 #define rdma_for_each_block(sglist, biter, nents, pgsz)		\
2892 	for (__rdma_block_iter_start(biter, sglist, nents,	\
2893 				     pgsz);			\
2894 	     __rdma_block_iter_next(biter);)
2895 
2896 /**
2897  * ib_get_client_data - Get IB client context
2898  * @device:Device to get context for
2899  * @client:Client to get context for
2900  *
2901  * ib_get_client_data() returns the client context data set with
2902  * ib_set_client_data(). This can only be called while the client is
2903  * registered to the device, once the ib_client remove() callback returns this
2904  * cannot be called.
2905  */
ib_get_client_data(struct ib_device * device,struct ib_client * client)2906 static inline void *ib_get_client_data(struct ib_device *device,
2907 				       struct ib_client *client)
2908 {
2909 	return xa_load(&device->client_data, client->client_id);
2910 }
2911 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2912 			 void *data);
2913 void ib_set_device_ops(struct ib_device *device,
2914 		       const struct ib_device_ops *ops);
2915 
2916 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2917 		      unsigned long pfn, unsigned long size, pgprot_t prot,
2918 		      struct rdma_user_mmap_entry *entry);
2919 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2920 				struct rdma_user_mmap_entry *entry,
2921 				size_t length);
2922 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2923 				      struct rdma_user_mmap_entry *entry,
2924 				      size_t length, u32 min_pgoff,
2925 				      u32 max_pgoff);
2926 
2927 static inline int
rdma_user_mmap_entry_insert_exact(struct ib_ucontext * ucontext,struct rdma_user_mmap_entry * entry,size_t length,u32 pgoff)2928 rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
2929 				  struct rdma_user_mmap_entry *entry,
2930 				  size_t length, u32 pgoff)
2931 {
2932 	return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff,
2933 						 pgoff);
2934 }
2935 
2936 struct rdma_user_mmap_entry *
2937 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2938 			       unsigned long pgoff);
2939 struct rdma_user_mmap_entry *
2940 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2941 			 struct vm_area_struct *vma);
2942 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2943 
2944 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2945 
ib_copy_from_udata(void * dest,struct ib_udata * udata,size_t len)2946 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2947 {
2948 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2949 }
2950 
ib_copy_to_udata(struct ib_udata * udata,void * src,size_t len)2951 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2952 {
2953 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2954 }
2955 
ib_is_buffer_cleared(const void __user * p,size_t len)2956 static inline bool ib_is_buffer_cleared(const void __user *p,
2957 					size_t len)
2958 {
2959 	bool ret;
2960 	u8 *buf;
2961 
2962 	if (len > USHRT_MAX)
2963 		return false;
2964 
2965 	buf = memdup_user(p, len);
2966 	if (IS_ERR(buf))
2967 		return false;
2968 
2969 	ret = !memchr_inv(buf, 0, len);
2970 	kfree(buf);
2971 	return ret;
2972 }
2973 
ib_is_udata_cleared(struct ib_udata * udata,size_t offset,size_t len)2974 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2975 				       size_t offset,
2976 				       size_t len)
2977 {
2978 	return ib_is_buffer_cleared(udata->inbuf + offset, len);
2979 }
2980 
2981 /**
2982  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2983  * contains all required attributes and no attributes not allowed for
2984  * the given QP state transition.
2985  * @cur_state: Current QP state
2986  * @next_state: Next QP state
2987  * @type: QP type
2988  * @mask: Mask of supplied QP attributes
2989  *
2990  * This function is a helper function that a low-level driver's
2991  * modify_qp method can use to validate the consumer's input.  It
2992  * checks that cur_state and next_state are valid QP states, that a
2993  * transition from cur_state to next_state is allowed by the IB spec,
2994  * and that the attribute mask supplied is allowed for the transition.
2995  */
2996 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2997 			enum ib_qp_type type, enum ib_qp_attr_mask mask);
2998 
2999 void ib_register_event_handler(struct ib_event_handler *event_handler);
3000 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
3001 void ib_dispatch_event(const struct ib_event *event);
3002 
3003 int ib_query_port(struct ib_device *device,
3004 		  u32 port_num, struct ib_port_attr *port_attr);
3005 
3006 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3007 					       u32 port_num);
3008 
3009 /**
3010  * rdma_cap_ib_switch - Check if the device is IB switch
3011  * @device: Device to check
3012  *
3013  * Device driver is responsible for setting is_switch bit on
3014  * in ib_device structure at init time.
3015  *
3016  * Return: true if the device is IB switch.
3017  */
rdma_cap_ib_switch(const struct ib_device * device)3018 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3019 {
3020 	return device->is_switch;
3021 }
3022 
3023 /**
3024  * rdma_start_port - Return the first valid port number for the device
3025  * specified
3026  *
3027  * @device: Device to be checked
3028  *
3029  * Return start port number
3030  */
rdma_start_port(const struct ib_device * device)3031 static inline u32 rdma_start_port(const struct ib_device *device)
3032 {
3033 	return rdma_cap_ib_switch(device) ? 0 : 1;
3034 }
3035 
3036 /**
3037  * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3038  * @device - The struct ib_device * to iterate over
3039  * @iter - The unsigned int to store the port number
3040  */
3041 #define rdma_for_each_port(device, iter)                                       \
3042 	for (iter = rdma_start_port(device +				       \
3043 				    BUILD_BUG_ON_ZERO(!__same_type(u32,	       \
3044 								   iter)));    \
3045 	     iter <= rdma_end_port(device); iter++)
3046 
3047 /**
3048  * rdma_end_port - Return the last valid port number for the device
3049  * specified
3050  *
3051  * @device: Device to be checked
3052  *
3053  * Return last port number
3054  */
rdma_end_port(const struct ib_device * device)3055 static inline u32 rdma_end_port(const struct ib_device *device)
3056 {
3057 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3058 }
3059 
rdma_is_port_valid(const struct ib_device * device,unsigned int port)3060 static inline int rdma_is_port_valid(const struct ib_device *device,
3061 				     unsigned int port)
3062 {
3063 	return (port >= rdma_start_port(device) &&
3064 		port <= rdma_end_port(device));
3065 }
3066 
rdma_is_grh_required(const struct ib_device * device,u32 port_num)3067 static inline bool rdma_is_grh_required(const struct ib_device *device,
3068 					u32 port_num)
3069 {
3070 	return device->port_data[port_num].immutable.core_cap_flags &
3071 	       RDMA_CORE_PORT_IB_GRH_REQUIRED;
3072 }
3073 
rdma_protocol_ib(const struct ib_device * device,u32 port_num)3074 static inline bool rdma_protocol_ib(const struct ib_device *device,
3075 				    u32 port_num)
3076 {
3077 	return device->port_data[port_num].immutable.core_cap_flags &
3078 	       RDMA_CORE_CAP_PROT_IB;
3079 }
3080 
rdma_protocol_roce(const struct ib_device * device,u32 port_num)3081 static inline bool rdma_protocol_roce(const struct ib_device *device,
3082 				      u32 port_num)
3083 {
3084 	return device->port_data[port_num].immutable.core_cap_flags &
3085 	       (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3086 }
3087 
rdma_protocol_roce_udp_encap(const struct ib_device * device,u32 port_num)3088 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3089 						u32 port_num)
3090 {
3091 	return device->port_data[port_num].immutable.core_cap_flags &
3092 	       RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3093 }
3094 
rdma_protocol_roce_eth_encap(const struct ib_device * device,u32 port_num)3095 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3096 						u32 port_num)
3097 {
3098 	return device->port_data[port_num].immutable.core_cap_flags &
3099 	       RDMA_CORE_CAP_PROT_ROCE;
3100 }
3101 
rdma_protocol_iwarp(const struct ib_device * device,u32 port_num)3102 static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3103 				       u32 port_num)
3104 {
3105 	return device->port_data[port_num].immutable.core_cap_flags &
3106 	       RDMA_CORE_CAP_PROT_IWARP;
3107 }
3108 
rdma_ib_or_roce(const struct ib_device * device,u32 port_num)3109 static inline bool rdma_ib_or_roce(const struct ib_device *device,
3110 				   u32 port_num)
3111 {
3112 	return rdma_protocol_ib(device, port_num) ||
3113 		rdma_protocol_roce(device, port_num);
3114 }
3115 
rdma_protocol_raw_packet(const struct ib_device * device,u32 port_num)3116 static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3117 					    u32 port_num)
3118 {
3119 	return device->port_data[port_num].immutable.core_cap_flags &
3120 	       RDMA_CORE_CAP_PROT_RAW_PACKET;
3121 }
3122 
rdma_protocol_usnic(const struct ib_device * device,u32 port_num)3123 static inline bool rdma_protocol_usnic(const struct ib_device *device,
3124 				       u32 port_num)
3125 {
3126 	return device->port_data[port_num].immutable.core_cap_flags &
3127 	       RDMA_CORE_CAP_PROT_USNIC;
3128 }
3129 
3130 /**
3131  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3132  * Management Datagrams.
3133  * @device: Device to check
3134  * @port_num: Port number to check
3135  *
3136  * Management Datagrams (MAD) are a required part of the InfiniBand
3137  * specification and are supported on all InfiniBand devices.  A slightly
3138  * extended version are also supported on OPA interfaces.
3139  *
3140  * Return: true if the port supports sending/receiving of MAD packets.
3141  */
rdma_cap_ib_mad(const struct ib_device * device,u32 port_num)3142 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3143 {
3144 	return device->port_data[port_num].immutable.core_cap_flags &
3145 	       RDMA_CORE_CAP_IB_MAD;
3146 }
3147 
3148 /**
3149  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3150  * Management Datagrams.
3151  * @device: Device to check
3152  * @port_num: Port number to check
3153  *
3154  * Intel OmniPath devices extend and/or replace the InfiniBand Management
3155  * datagrams with their own versions.  These OPA MADs share many but not all of
3156  * the characteristics of InfiniBand MADs.
3157  *
3158  * OPA MADs differ in the following ways:
3159  *
3160  *    1) MADs are variable size up to 2K
3161  *       IBTA defined MADs remain fixed at 256 bytes
3162  *    2) OPA SMPs must carry valid PKeys
3163  *    3) OPA SMP packets are a different format
3164  *
3165  * Return: true if the port supports OPA MAD packet formats.
3166  */
rdma_cap_opa_mad(struct ib_device * device,u32 port_num)3167 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3168 {
3169 	return device->port_data[port_num].immutable.core_cap_flags &
3170 		RDMA_CORE_CAP_OPA_MAD;
3171 }
3172 
3173 /**
3174  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3175  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3176  * @device: Device to check
3177  * @port_num: Port number to check
3178  *
3179  * Each InfiniBand node is required to provide a Subnet Management Agent
3180  * that the subnet manager can access.  Prior to the fabric being fully
3181  * configured by the subnet manager, the SMA is accessed via a well known
3182  * interface called the Subnet Management Interface (SMI).  This interface
3183  * uses directed route packets to communicate with the SM to get around the
3184  * chicken and egg problem of the SM needing to know what's on the fabric
3185  * in order to configure the fabric, and needing to configure the fabric in
3186  * order to send packets to the devices on the fabric.  These directed
3187  * route packets do not need the fabric fully configured in order to reach
3188  * their destination.  The SMI is the only method allowed to send
3189  * directed route packets on an InfiniBand fabric.
3190  *
3191  * Return: true if the port provides an SMI.
3192  */
rdma_cap_ib_smi(const struct ib_device * device,u32 port_num)3193 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3194 {
3195 	return device->port_data[port_num].immutable.core_cap_flags &
3196 	       RDMA_CORE_CAP_IB_SMI;
3197 }
3198 
3199 /**
3200  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3201  * Communication Manager.
3202  * @device: Device to check
3203  * @port_num: Port number to check
3204  *
3205  * The InfiniBand Communication Manager is one of many pre-defined General
3206  * Service Agents (GSA) that are accessed via the General Service
3207  * Interface (GSI).  It's role is to facilitate establishment of connections
3208  * between nodes as well as other management related tasks for established
3209  * connections.
3210  *
3211  * Return: true if the port supports an IB CM (this does not guarantee that
3212  * a CM is actually running however).
3213  */
rdma_cap_ib_cm(const struct ib_device * device,u32 port_num)3214 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3215 {
3216 	return device->port_data[port_num].immutable.core_cap_flags &
3217 	       RDMA_CORE_CAP_IB_CM;
3218 }
3219 
3220 /**
3221  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3222  * Communication Manager.
3223  * @device: Device to check
3224  * @port_num: Port number to check
3225  *
3226  * Similar to above, but specific to iWARP connections which have a different
3227  * managment protocol than InfiniBand.
3228  *
3229  * Return: true if the port supports an iWARP CM (this does not guarantee that
3230  * a CM is actually running however).
3231  */
rdma_cap_iw_cm(const struct ib_device * device,u32 port_num)3232 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3233 {
3234 	return device->port_data[port_num].immutable.core_cap_flags &
3235 	       RDMA_CORE_CAP_IW_CM;
3236 }
3237 
3238 /**
3239  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3240  * Subnet Administration.
3241  * @device: Device to check
3242  * @port_num: Port number to check
3243  *
3244  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3245  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
3246  * fabrics, devices should resolve routes to other hosts by contacting the
3247  * SA to query the proper route.
3248  *
3249  * Return: true if the port should act as a client to the fabric Subnet
3250  * Administration interface.  This does not imply that the SA service is
3251  * running locally.
3252  */
rdma_cap_ib_sa(const struct ib_device * device,u32 port_num)3253 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3254 {
3255 	return device->port_data[port_num].immutable.core_cap_flags &
3256 	       RDMA_CORE_CAP_IB_SA;
3257 }
3258 
3259 /**
3260  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3261  * Multicast.
3262  * @device: Device to check
3263  * @port_num: Port number to check
3264  *
3265  * InfiniBand multicast registration is more complex than normal IPv4 or
3266  * IPv6 multicast registration.  Each Host Channel Adapter must register
3267  * with the Subnet Manager when it wishes to join a multicast group.  It
3268  * should do so only once regardless of how many queue pairs it subscribes
3269  * to this group.  And it should leave the group only after all queue pairs
3270  * attached to the group have been detached.
3271  *
3272  * Return: true if the port must undertake the additional adminstrative
3273  * overhead of registering/unregistering with the SM and tracking of the
3274  * total number of queue pairs attached to the multicast group.
3275  */
rdma_cap_ib_mcast(const struct ib_device * device,u32 port_num)3276 static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3277 				     u32 port_num)
3278 {
3279 	return rdma_cap_ib_sa(device, port_num);
3280 }
3281 
3282 /**
3283  * rdma_cap_af_ib - Check if the port of device has the capability
3284  * Native Infiniband Address.
3285  * @device: Device to check
3286  * @port_num: Port number to check
3287  *
3288  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3289  * GID.  RoCE uses a different mechanism, but still generates a GID via
3290  * a prescribed mechanism and port specific data.
3291  *
3292  * Return: true if the port uses a GID address to identify devices on the
3293  * network.
3294  */
rdma_cap_af_ib(const struct ib_device * device,u32 port_num)3295 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3296 {
3297 	return device->port_data[port_num].immutable.core_cap_flags &
3298 	       RDMA_CORE_CAP_AF_IB;
3299 }
3300 
3301 /**
3302  * rdma_cap_eth_ah - Check if the port of device has the capability
3303  * Ethernet Address Handle.
3304  * @device: Device to check
3305  * @port_num: Port number to check
3306  *
3307  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3308  * to fabricate GIDs over Ethernet/IP specific addresses native to the
3309  * port.  Normally, packet headers are generated by the sending host
3310  * adapter, but when sending connectionless datagrams, we must manually
3311  * inject the proper headers for the fabric we are communicating over.
3312  *
3313  * Return: true if we are running as a RoCE port and must force the
3314  * addition of a Global Route Header built from our Ethernet Address
3315  * Handle into our header list for connectionless packets.
3316  */
rdma_cap_eth_ah(const struct ib_device * device,u32 port_num)3317 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3318 {
3319 	return device->port_data[port_num].immutable.core_cap_flags &
3320 	       RDMA_CORE_CAP_ETH_AH;
3321 }
3322 
3323 /**
3324  * rdma_cap_opa_ah - Check if the port of device supports
3325  * OPA Address handles
3326  * @device: Device to check
3327  * @port_num: Port number to check
3328  *
3329  * Return: true if we are running on an OPA device which supports
3330  * the extended OPA addressing.
3331  */
rdma_cap_opa_ah(struct ib_device * device,u32 port_num)3332 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3333 {
3334 	return (device->port_data[port_num].immutable.core_cap_flags &
3335 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3336 }
3337 
3338 /**
3339  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3340  *
3341  * @device: Device
3342  * @port_num: Port number
3343  *
3344  * This MAD size includes the MAD headers and MAD payload.  No other headers
3345  * are included.
3346  *
3347  * Return the max MAD size required by the Port.  Will return 0 if the port
3348  * does not support MADs
3349  */
rdma_max_mad_size(const struct ib_device * device,u32 port_num)3350 static inline size_t rdma_max_mad_size(const struct ib_device *device,
3351 				       u32 port_num)
3352 {
3353 	return device->port_data[port_num].immutable.max_mad_size;
3354 }
3355 
3356 /**
3357  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3358  * @device: Device to check
3359  * @port_num: Port number to check
3360  *
3361  * RoCE GID table mechanism manages the various GIDs for a device.
3362  *
3363  * NOTE: if allocating the port's GID table has failed, this call will still
3364  * return true, but any RoCE GID table API will fail.
3365  *
3366  * Return: true if the port uses RoCE GID table mechanism in order to manage
3367  * its GIDs.
3368  */
rdma_cap_roce_gid_table(const struct ib_device * device,u32 port_num)3369 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3370 					   u32 port_num)
3371 {
3372 	return rdma_protocol_roce(device, port_num) &&
3373 		device->ops.add_gid && device->ops.del_gid;
3374 }
3375 
3376 /*
3377  * Check if the device supports READ W/ INVALIDATE.
3378  */
rdma_cap_read_inv(struct ib_device * dev,u32 port_num)3379 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3380 {
3381 	/*
3382 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
3383 	 * has support for it yet.
3384 	 */
3385 	return rdma_protocol_iwarp(dev, port_num);
3386 }
3387 
3388 /**
3389  * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3390  * @device: Device
3391  * @port_num: 1 based Port number
3392  *
3393  * Return true if port is an Intel OPA port , false if not
3394  */
rdma_core_cap_opa_port(struct ib_device * device,u32 port_num)3395 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3396 					  u32 port_num)
3397 {
3398 	return (device->port_data[port_num].immutable.core_cap_flags &
3399 		RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3400 }
3401 
3402 /**
3403  * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3404  * @device: Device
3405  * @port_num: Port number
3406  * @mtu: enum value of MTU
3407  *
3408  * Return the MTU size supported by the port as an integer value. Will return
3409  * -1 if enum value of mtu is not supported.
3410  */
rdma_mtu_enum_to_int(struct ib_device * device,u32 port,int mtu)3411 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3412 				       int mtu)
3413 {
3414 	if (rdma_core_cap_opa_port(device, port))
3415 		return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3416 	else
3417 		return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3418 }
3419 
3420 /**
3421  * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3422  * @device: Device
3423  * @port_num: Port number
3424  * @attr: port attribute
3425  *
3426  * Return the MTU size supported by the port as an integer value.
3427  */
rdma_mtu_from_attr(struct ib_device * device,u32 port,struct ib_port_attr * attr)3428 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3429 				     struct ib_port_attr *attr)
3430 {
3431 	if (rdma_core_cap_opa_port(device, port))
3432 		return attr->phys_mtu;
3433 	else
3434 		return ib_mtu_enum_to_int(attr->max_mtu);
3435 }
3436 
3437 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3438 			 int state);
3439 int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3440 		     struct ifla_vf_info *info);
3441 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3442 		    struct ifla_vf_stats *stats);
3443 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3444 		    struct ifla_vf_guid *node_guid,
3445 		    struct ifla_vf_guid *port_guid);
3446 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3447 		   int type);
3448 
3449 int ib_query_pkey(struct ib_device *device,
3450 		  u32 port_num, u16 index, u16 *pkey);
3451 
3452 int ib_modify_device(struct ib_device *device,
3453 		     int device_modify_mask,
3454 		     struct ib_device_modify *device_modify);
3455 
3456 int ib_modify_port(struct ib_device *device,
3457 		   u32 port_num, int port_modify_mask,
3458 		   struct ib_port_modify *port_modify);
3459 
3460 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3461 		u32 *port_num, u16 *index);
3462 
3463 int ib_find_pkey(struct ib_device *device,
3464 		 u32 port_num, u16 pkey, u16 *index);
3465 
3466 enum ib_pd_flags {
3467 	/*
3468 	 * Create a memory registration for all memory in the system and place
3469 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
3470 	 * ULPs to avoid the overhead of dynamic MRs.
3471 	 *
3472 	 * This flag is generally considered unsafe and must only be used in
3473 	 * extremly trusted environments.  Every use of it will log a warning
3474 	 * in the kernel log.
3475 	 */
3476 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
3477 };
3478 
3479 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3480 		const char *caller);
3481 
3482 /**
3483  * ib_alloc_pd - Allocates an unused protection domain.
3484  * @device: The device on which to allocate the protection domain.
3485  * @flags: protection domain flags
3486  *
3487  * A protection domain object provides an association between QPs, shared
3488  * receive queues, address handles, memory regions, and memory windows.
3489  *
3490  * Every PD has a local_dma_lkey which can be used as the lkey value for local
3491  * memory operations.
3492  */
3493 #define ib_alloc_pd(device, flags) \
3494 	__ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3495 
3496 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3497 
3498 /**
3499  * ib_dealloc_pd - Deallocate kernel PD
3500  * @pd: The protection domain
3501  *
3502  * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3503  */
ib_dealloc_pd(struct ib_pd * pd)3504 static inline void ib_dealloc_pd(struct ib_pd *pd)
3505 {
3506 	int ret = ib_dealloc_pd_user(pd, NULL);
3507 
3508 	WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3509 }
3510 
3511 enum rdma_create_ah_flags {
3512 	/* In a sleepable context */
3513 	RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3514 };
3515 
3516 /**
3517  * rdma_create_ah - Creates an address handle for the given address vector.
3518  * @pd: The protection domain associated with the address handle.
3519  * @ah_attr: The attributes of the address vector.
3520  * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3521  *
3522  * The address handle is used to reference a local or global destination
3523  * in all UD QP post sends.
3524  */
3525 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3526 			     u32 flags);
3527 
3528 /**
3529  * rdma_create_user_ah - Creates an address handle for the given address vector.
3530  * It resolves destination mac address for ah attribute of RoCE type.
3531  * @pd: The protection domain associated with the address handle.
3532  * @ah_attr: The attributes of the address vector.
3533  * @udata: pointer to user's input output buffer information need by
3534  *         provider driver.
3535  *
3536  * It returns 0 on success and returns appropriate error code on error.
3537  * The address handle is used to reference a local or global destination
3538  * in all UD QP post sends.
3539  */
3540 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3541 				  struct rdma_ah_attr *ah_attr,
3542 				  struct ib_udata *udata);
3543 /**
3544  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3545  *   work completion.
3546  * @hdr: the L3 header to parse
3547  * @net_type: type of header to parse
3548  * @sgid: place to store source gid
3549  * @dgid: place to store destination gid
3550  */
3551 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3552 			      enum rdma_network_type net_type,
3553 			      union ib_gid *sgid, union ib_gid *dgid);
3554 
3555 /**
3556  * ib_get_rdma_header_version - Get the header version
3557  * @hdr: the L3 header to parse
3558  */
3559 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3560 
3561 /**
3562  * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3563  *   work completion.
3564  * @device: Device on which the received message arrived.
3565  * @port_num: Port on which the received message arrived.
3566  * @wc: Work completion associated with the received message.
3567  * @grh: References the received global route header.  This parameter is
3568  *   ignored unless the work completion indicates that the GRH is valid.
3569  * @ah_attr: Returned attributes that can be used when creating an address
3570  *   handle for replying to the message.
3571  * When ib_init_ah_attr_from_wc() returns success,
3572  * (a) for IB link layer it optionally contains a reference to SGID attribute
3573  * when GRH is present for IB link layer.
3574  * (b) for RoCE link layer it contains a reference to SGID attribute.
3575  * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3576  * attributes which are initialized using ib_init_ah_attr_from_wc().
3577  *
3578  */
3579 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3580 			    const struct ib_wc *wc, const struct ib_grh *grh,
3581 			    struct rdma_ah_attr *ah_attr);
3582 
3583 /**
3584  * ib_create_ah_from_wc - Creates an address handle associated with the
3585  *   sender of the specified work completion.
3586  * @pd: The protection domain associated with the address handle.
3587  * @wc: Work completion information associated with a received message.
3588  * @grh: References the received global route header.  This parameter is
3589  *   ignored unless the work completion indicates that the GRH is valid.
3590  * @port_num: The outbound port number to associate with the address.
3591  *
3592  * The address handle is used to reference a local or global destination
3593  * in all UD QP post sends.
3594  */
3595 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3596 				   const struct ib_grh *grh, u32 port_num);
3597 
3598 /**
3599  * rdma_modify_ah - Modifies the address vector associated with an address
3600  *   handle.
3601  * @ah: The address handle to modify.
3602  * @ah_attr: The new address vector attributes to associate with the
3603  *   address handle.
3604  */
3605 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3606 
3607 /**
3608  * rdma_query_ah - Queries the address vector associated with an address
3609  *   handle.
3610  * @ah: The address handle to query.
3611  * @ah_attr: The address vector attributes associated with the address
3612  *   handle.
3613  */
3614 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3615 
3616 enum rdma_destroy_ah_flags {
3617 	/* In a sleepable context */
3618 	RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3619 };
3620 
3621 /**
3622  * rdma_destroy_ah_user - Destroys an address handle.
3623  * @ah: The address handle to destroy.
3624  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3625  * @udata: Valid user data or NULL for kernel objects
3626  */
3627 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3628 
3629 /**
3630  * rdma_destroy_ah - Destroys an kernel address handle.
3631  * @ah: The address handle to destroy.
3632  * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3633  *
3634  * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3635  */
rdma_destroy_ah(struct ib_ah * ah,u32 flags)3636 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3637 {
3638 	int ret = rdma_destroy_ah_user(ah, flags, NULL);
3639 
3640 	WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3641 }
3642 
3643 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3644 				  struct ib_srq_init_attr *srq_init_attr,
3645 				  struct ib_usrq_object *uobject,
3646 				  struct ib_udata *udata);
3647 static inline struct ib_srq *
ib_create_srq(struct ib_pd * pd,struct ib_srq_init_attr * srq_init_attr)3648 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3649 {
3650 	if (!pd->device->ops.create_srq)
3651 		return ERR_PTR(-EOPNOTSUPP);
3652 
3653 	return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3654 }
3655 
3656 /**
3657  * ib_modify_srq - Modifies the attributes for the specified SRQ.
3658  * @srq: The SRQ to modify.
3659  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
3660  *   the current values of selected SRQ attributes are returned.
3661  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3662  *   are being modified.
3663  *
3664  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3665  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3666  * the number of receives queued drops below the limit.
3667  */
3668 int ib_modify_srq(struct ib_srq *srq,
3669 		  struct ib_srq_attr *srq_attr,
3670 		  enum ib_srq_attr_mask srq_attr_mask);
3671 
3672 /**
3673  * ib_query_srq - Returns the attribute list and current values for the
3674  *   specified SRQ.
3675  * @srq: The SRQ to query.
3676  * @srq_attr: The attributes of the specified SRQ.
3677  */
3678 int ib_query_srq(struct ib_srq *srq,
3679 		 struct ib_srq_attr *srq_attr);
3680 
3681 /**
3682  * ib_destroy_srq_user - Destroys the specified SRQ.
3683  * @srq: The SRQ to destroy.
3684  * @udata: Valid user data or NULL for kernel objects
3685  */
3686 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3687 
3688 /**
3689  * ib_destroy_srq - Destroys the specified kernel SRQ.
3690  * @srq: The SRQ to destroy.
3691  *
3692  * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3693  */
ib_destroy_srq(struct ib_srq * srq)3694 static inline void ib_destroy_srq(struct ib_srq *srq)
3695 {
3696 	int ret = ib_destroy_srq_user(srq, NULL);
3697 
3698 	WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3699 }
3700 
3701 /**
3702  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3703  * @srq: The SRQ to post the work request on.
3704  * @recv_wr: A list of work requests to post on the receive queue.
3705  * @bad_recv_wr: On an immediate failure, this parameter will reference
3706  *   the work request that failed to be posted on the QP.
3707  */
ib_post_srq_recv(struct ib_srq * srq,const struct ib_recv_wr * recv_wr,const struct ib_recv_wr ** bad_recv_wr)3708 static inline int ib_post_srq_recv(struct ib_srq *srq,
3709 				   const struct ib_recv_wr *recv_wr,
3710 				   const struct ib_recv_wr **bad_recv_wr)
3711 {
3712 	const struct ib_recv_wr *dummy;
3713 
3714 	return srq->device->ops.post_srq_recv(srq, recv_wr,
3715 					      bad_recv_wr ? : &dummy);
3716 }
3717 
3718 struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3719 				  struct ib_qp_init_attr *qp_init_attr,
3720 				  const char *caller);
3721 /**
3722  * ib_create_qp - Creates a kernel QP associated with the specific protection
3723  * domain.
3724  * @pd: The protection domain associated with the QP.
3725  * @init_attr: A list of initial attributes required to create the
3726  *   QP.  If QP creation succeeds, then the attributes are updated to
3727  *   the actual capabilities of the created QP.
3728  */
ib_create_qp(struct ib_pd * pd,struct ib_qp_init_attr * init_attr)3729 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3730 					 struct ib_qp_init_attr *init_attr)
3731 {
3732 	return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
3733 }
3734 
3735 /**
3736  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3737  * @qp: The QP to modify.
3738  * @attr: On input, specifies the QP attributes to modify.  On output,
3739  *   the current values of selected QP attributes are returned.
3740  * @attr_mask: A bit-mask used to specify which attributes of the QP
3741  *   are being modified.
3742  * @udata: pointer to user's input output buffer information
3743  *   are being modified.
3744  * It returns 0 on success and returns appropriate error code on error.
3745  */
3746 int ib_modify_qp_with_udata(struct ib_qp *qp,
3747 			    struct ib_qp_attr *attr,
3748 			    int attr_mask,
3749 			    struct ib_udata *udata);
3750 
3751 /**
3752  * ib_modify_qp - Modifies the attributes for the specified QP and then
3753  *   transitions the QP to the given state.
3754  * @qp: The QP to modify.
3755  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3756  *   the current values of selected QP attributes are returned.
3757  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3758  *   are being modified.
3759  */
3760 int ib_modify_qp(struct ib_qp *qp,
3761 		 struct ib_qp_attr *qp_attr,
3762 		 int qp_attr_mask);
3763 
3764 /**
3765  * ib_query_qp - Returns the attribute list and current values for the
3766  *   specified QP.
3767  * @qp: The QP to query.
3768  * @qp_attr: The attributes of the specified QP.
3769  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3770  * @qp_init_attr: Additional attributes of the selected QP.
3771  *
3772  * The qp_attr_mask may be used to limit the query to gathering only the
3773  * selected attributes.
3774  */
3775 int ib_query_qp(struct ib_qp *qp,
3776 		struct ib_qp_attr *qp_attr,
3777 		int qp_attr_mask,
3778 		struct ib_qp_init_attr *qp_init_attr);
3779 
3780 /**
3781  * ib_destroy_qp - Destroys the specified QP.
3782  * @qp: The QP to destroy.
3783  * @udata: Valid udata or NULL for kernel objects
3784  */
3785 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3786 
3787 /**
3788  * ib_destroy_qp - Destroys the specified kernel QP.
3789  * @qp: The QP to destroy.
3790  *
3791  * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3792  */
ib_destroy_qp(struct ib_qp * qp)3793 static inline int ib_destroy_qp(struct ib_qp *qp)
3794 {
3795 	return ib_destroy_qp_user(qp, NULL);
3796 }
3797 
3798 /**
3799  * ib_open_qp - Obtain a reference to an existing sharable QP.
3800  * @xrcd - XRC domain
3801  * @qp_open_attr: Attributes identifying the QP to open.
3802  *
3803  * Returns a reference to a sharable QP.
3804  */
3805 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3806 			 struct ib_qp_open_attr *qp_open_attr);
3807 
3808 /**
3809  * ib_close_qp - Release an external reference to a QP.
3810  * @qp: The QP handle to release
3811  *
3812  * The opened QP handle is released by the caller.  The underlying
3813  * shared QP is not destroyed until all internal references are released.
3814  */
3815 int ib_close_qp(struct ib_qp *qp);
3816 
3817 /**
3818  * ib_post_send - Posts a list of work requests to the send queue of
3819  *   the specified QP.
3820  * @qp: The QP to post the work request on.
3821  * @send_wr: A list of work requests to post on the send queue.
3822  * @bad_send_wr: On an immediate failure, this parameter will reference
3823  *   the work request that failed to be posted on the QP.
3824  *
3825  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3826  * error is returned, the QP state shall not be affected,
3827  * ib_post_send() will return an immediate error after queueing any
3828  * earlier work requests in the list.
3829  */
ib_post_send(struct ib_qp * qp,const struct ib_send_wr * send_wr,const struct ib_send_wr ** bad_send_wr)3830 static inline int ib_post_send(struct ib_qp *qp,
3831 			       const struct ib_send_wr *send_wr,
3832 			       const struct ib_send_wr **bad_send_wr)
3833 {
3834 	const struct ib_send_wr *dummy;
3835 
3836 	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3837 }
3838 
3839 /**
3840  * ib_post_recv - Posts a list of work requests to the receive queue of
3841  *   the specified QP.
3842  * @qp: The QP to post the work request on.
3843  * @recv_wr: A list of work requests to post on the receive queue.
3844  * @bad_recv_wr: On an immediate failure, this parameter will reference
3845  *   the work request that failed to be posted on the QP.
3846  */
ib_post_recv(struct ib_qp * qp,const struct ib_recv_wr * recv_wr,const struct ib_recv_wr ** bad_recv_wr)3847 static inline int ib_post_recv(struct ib_qp *qp,
3848 			       const struct ib_recv_wr *recv_wr,
3849 			       const struct ib_recv_wr **bad_recv_wr)
3850 {
3851 	const struct ib_recv_wr *dummy;
3852 
3853 	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3854 }
3855 
3856 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3857 			    int comp_vector, enum ib_poll_context poll_ctx,
3858 			    const char *caller);
ib_alloc_cq(struct ib_device * dev,void * private,int nr_cqe,int comp_vector,enum ib_poll_context poll_ctx)3859 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3860 					int nr_cqe, int comp_vector,
3861 					enum ib_poll_context poll_ctx)
3862 {
3863 	return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3864 			     KBUILD_MODNAME);
3865 }
3866 
3867 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3868 				int nr_cqe, enum ib_poll_context poll_ctx,
3869 				const char *caller);
3870 
3871 /**
3872  * ib_alloc_cq_any: Allocate kernel CQ
3873  * @dev: The IB device
3874  * @private: Private data attached to the CQE
3875  * @nr_cqe: Number of CQEs in the CQ
3876  * @poll_ctx: Context used for polling the CQ
3877  */
ib_alloc_cq_any(struct ib_device * dev,void * private,int nr_cqe,enum ib_poll_context poll_ctx)3878 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3879 					    void *private, int nr_cqe,
3880 					    enum ib_poll_context poll_ctx)
3881 {
3882 	return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3883 				 KBUILD_MODNAME);
3884 }
3885 
3886 void ib_free_cq(struct ib_cq *cq);
3887 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3888 
3889 /**
3890  * ib_create_cq - Creates a CQ on the specified device.
3891  * @device: The device on which to create the CQ.
3892  * @comp_handler: A user-specified callback that is invoked when a
3893  *   completion event occurs on the CQ.
3894  * @event_handler: A user-specified callback that is invoked when an
3895  *   asynchronous event not associated with a completion occurs on the CQ.
3896  * @cq_context: Context associated with the CQ returned to the user via
3897  *   the associated completion and event handlers.
3898  * @cq_attr: The attributes the CQ should be created upon.
3899  *
3900  * Users can examine the cq structure to determine the actual CQ size.
3901  */
3902 struct ib_cq *__ib_create_cq(struct ib_device *device,
3903 			     ib_comp_handler comp_handler,
3904 			     void (*event_handler)(struct ib_event *, void *),
3905 			     void *cq_context,
3906 			     const struct ib_cq_init_attr *cq_attr,
3907 			     const char *caller);
3908 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3909 	__ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3910 
3911 /**
3912  * ib_resize_cq - Modifies the capacity of the CQ.
3913  * @cq: The CQ to resize.
3914  * @cqe: The minimum size of the CQ.
3915  *
3916  * Users can examine the cq structure to determine the actual CQ size.
3917  */
3918 int ib_resize_cq(struct ib_cq *cq, int cqe);
3919 
3920 /**
3921  * rdma_set_cq_moderation - Modifies moderation params of the CQ
3922  * @cq: The CQ to modify.
3923  * @cq_count: number of CQEs that will trigger an event
3924  * @cq_period: max period of time in usec before triggering an event
3925  *
3926  */
3927 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3928 
3929 /**
3930  * ib_destroy_cq_user - Destroys the specified CQ.
3931  * @cq: The CQ to destroy.
3932  * @udata: Valid user data or NULL for kernel objects
3933  */
3934 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3935 
3936 /**
3937  * ib_destroy_cq - Destroys the specified kernel CQ.
3938  * @cq: The CQ to destroy.
3939  *
3940  * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3941  */
ib_destroy_cq(struct ib_cq * cq)3942 static inline void ib_destroy_cq(struct ib_cq *cq)
3943 {
3944 	int ret = ib_destroy_cq_user(cq, NULL);
3945 
3946 	WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3947 }
3948 
3949 /**
3950  * ib_poll_cq - poll a CQ for completion(s)
3951  * @cq:the CQ being polled
3952  * @num_entries:maximum number of completions to return
3953  * @wc:array of at least @num_entries &struct ib_wc where completions
3954  *   will be returned
3955  *
3956  * Poll a CQ for (possibly multiple) completions.  If the return value
3957  * is < 0, an error occurred.  If the return value is >= 0, it is the
3958  * number of completions returned.  If the return value is
3959  * non-negative and < num_entries, then the CQ was emptied.
3960  */
ib_poll_cq(struct ib_cq * cq,int num_entries,struct ib_wc * wc)3961 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3962 			     struct ib_wc *wc)
3963 {
3964 	return cq->device->ops.poll_cq(cq, num_entries, wc);
3965 }
3966 
3967 /**
3968  * ib_req_notify_cq - Request completion notification on a CQ.
3969  * @cq: The CQ to generate an event for.
3970  * @flags:
3971  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3972  *   to request an event on the next solicited event or next work
3973  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3974  *   may also be |ed in to request a hint about missed events, as
3975  *   described below.
3976  *
3977  * Return Value:
3978  *    < 0 means an error occurred while requesting notification
3979  *   == 0 means notification was requested successfully, and if
3980  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3981  *        were missed and it is safe to wait for another event.  In
3982  *        this case is it guaranteed that any work completions added
3983  *        to the CQ since the last CQ poll will trigger a completion
3984  *        notification event.
3985  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3986  *        in.  It means that the consumer must poll the CQ again to
3987  *        make sure it is empty to avoid missing an event because of a
3988  *        race between requesting notification and an entry being
3989  *        added to the CQ.  This return value means it is possible
3990  *        (but not guaranteed) that a work completion has been added
3991  *        to the CQ since the last poll without triggering a
3992  *        completion notification event.
3993  */
ib_req_notify_cq(struct ib_cq * cq,enum ib_cq_notify_flags flags)3994 static inline int ib_req_notify_cq(struct ib_cq *cq,
3995 				   enum ib_cq_notify_flags flags)
3996 {
3997 	return cq->device->ops.req_notify_cq(cq, flags);
3998 }
3999 
4000 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4001 			     int comp_vector_hint,
4002 			     enum ib_poll_context poll_ctx);
4003 
4004 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4005 
4006 /*
4007  * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
4008  * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
4009  * address into the dma address.
4010  */
ib_uses_virt_dma(struct ib_device * dev)4011 static inline bool ib_uses_virt_dma(struct ib_device *dev)
4012 {
4013 	return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
4014 }
4015 
4016 /*
4017  * Check if a IB device's underlying DMA mapping supports P2PDMA transfers.
4018  */
ib_dma_pci_p2p_dma_supported(struct ib_device * dev)4019 static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev)
4020 {
4021 	if (ib_uses_virt_dma(dev))
4022 		return false;
4023 
4024 	return dma_pci_p2pdma_supported(dev->dma_device);
4025 }
4026 
4027 /**
4028  * ib_dma_mapping_error - check a DMA addr for error
4029  * @dev: The device for which the dma_addr was created
4030  * @dma_addr: The DMA address to check
4031  */
ib_dma_mapping_error(struct ib_device * dev,u64 dma_addr)4032 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4033 {
4034 	if (ib_uses_virt_dma(dev))
4035 		return 0;
4036 	return dma_mapping_error(dev->dma_device, dma_addr);
4037 }
4038 
4039 /**
4040  * ib_dma_map_single - Map a kernel virtual address to DMA address
4041  * @dev: The device for which the dma_addr is to be created
4042  * @cpu_addr: The kernel virtual address
4043  * @size: The size of the region in bytes
4044  * @direction: The direction of the DMA
4045  */
ib_dma_map_single(struct ib_device * dev,void * cpu_addr,size_t size,enum dma_data_direction direction)4046 static inline u64 ib_dma_map_single(struct ib_device *dev,
4047 				    void *cpu_addr, size_t size,
4048 				    enum dma_data_direction direction)
4049 {
4050 	if (ib_uses_virt_dma(dev))
4051 		return (uintptr_t)cpu_addr;
4052 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4053 }
4054 
4055 /**
4056  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4057  * @dev: The device for which the DMA address was created
4058  * @addr: The DMA address
4059  * @size: The size of the region in bytes
4060  * @direction: The direction of the DMA
4061  */
ib_dma_unmap_single(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction direction)4062 static inline void ib_dma_unmap_single(struct ib_device *dev,
4063 				       u64 addr, size_t size,
4064 				       enum dma_data_direction direction)
4065 {
4066 	if (!ib_uses_virt_dma(dev))
4067 		dma_unmap_single(dev->dma_device, addr, size, direction);
4068 }
4069 
4070 /**
4071  * ib_dma_map_page - Map a physical page to DMA address
4072  * @dev: The device for which the dma_addr is to be created
4073  * @page: The page to be mapped
4074  * @offset: The offset within the page
4075  * @size: The size of the region in bytes
4076  * @direction: The direction of the DMA
4077  */
ib_dma_map_page(struct ib_device * dev,struct page * page,unsigned long offset,size_t size,enum dma_data_direction direction)4078 static inline u64 ib_dma_map_page(struct ib_device *dev,
4079 				  struct page *page,
4080 				  unsigned long offset,
4081 				  size_t size,
4082 					 enum dma_data_direction direction)
4083 {
4084 	if (ib_uses_virt_dma(dev))
4085 		return (uintptr_t)(page_address(page) + offset);
4086 	return dma_map_page(dev->dma_device, page, offset, size, direction);
4087 }
4088 
4089 /**
4090  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4091  * @dev: The device for which the DMA address was created
4092  * @addr: The DMA address
4093  * @size: The size of the region in bytes
4094  * @direction: The direction of the DMA
4095  */
ib_dma_unmap_page(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction direction)4096 static inline void ib_dma_unmap_page(struct ib_device *dev,
4097 				     u64 addr, size_t size,
4098 				     enum dma_data_direction direction)
4099 {
4100 	if (!ib_uses_virt_dma(dev))
4101 		dma_unmap_page(dev->dma_device, addr, size, direction);
4102 }
4103 
4104 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)4105 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4106 				      struct scatterlist *sg, int nents,
4107 				      enum dma_data_direction direction,
4108 				      unsigned long dma_attrs)
4109 {
4110 	if (ib_uses_virt_dma(dev))
4111 		return ib_dma_virt_map_sg(dev, sg, nents);
4112 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4113 				dma_attrs);
4114 }
4115 
ib_dma_unmap_sg_attrs(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction,unsigned long dma_attrs)4116 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4117 					 struct scatterlist *sg, int nents,
4118 					 enum dma_data_direction direction,
4119 					 unsigned long dma_attrs)
4120 {
4121 	if (!ib_uses_virt_dma(dev))
4122 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4123 				   dma_attrs);
4124 }
4125 
4126 /**
4127  * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4128  * @dev: The device for which the DMA addresses are to be created
4129  * @sg: The sg_table object describing the buffer
4130  * @direction: The direction of the DMA
4131  * @attrs: Optional DMA attributes for the map operation
4132  */
ib_dma_map_sgtable_attrs(struct ib_device * dev,struct sg_table * sgt,enum dma_data_direction direction,unsigned long dma_attrs)4133 static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4134 					   struct sg_table *sgt,
4135 					   enum dma_data_direction direction,
4136 					   unsigned long dma_attrs)
4137 {
4138 	int nents;
4139 
4140 	if (ib_uses_virt_dma(dev)) {
4141 		nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4142 		if (!nents)
4143 			return -EIO;
4144 		sgt->nents = nents;
4145 		return 0;
4146 	}
4147 	return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4148 }
4149 
ib_dma_unmap_sgtable_attrs(struct ib_device * dev,struct sg_table * sgt,enum dma_data_direction direction,unsigned long dma_attrs)4150 static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4151 					      struct sg_table *sgt,
4152 					      enum dma_data_direction direction,
4153 					      unsigned long dma_attrs)
4154 {
4155 	if (!ib_uses_virt_dma(dev))
4156 		dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4157 }
4158 
4159 /**
4160  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4161  * @dev: The device for which the DMA addresses are to be created
4162  * @sg: The array of scatter/gather entries
4163  * @nents: The number of scatter/gather entries
4164  * @direction: The direction of the DMA
4165  */
ib_dma_map_sg(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction)4166 static inline int ib_dma_map_sg(struct ib_device *dev,
4167 				struct scatterlist *sg, int nents,
4168 				enum dma_data_direction direction)
4169 {
4170 	return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4171 }
4172 
4173 /**
4174  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4175  * @dev: The device for which the DMA addresses were created
4176  * @sg: The array of scatter/gather entries
4177  * @nents: The number of scatter/gather entries
4178  * @direction: The direction of the DMA
4179  */
ib_dma_unmap_sg(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction)4180 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4181 				   struct scatterlist *sg, int nents,
4182 				   enum dma_data_direction direction)
4183 {
4184 	ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4185 }
4186 
4187 /**
4188  * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4189  * @dev: The device to query
4190  *
4191  * The returned value represents a size in bytes.
4192  */
ib_dma_max_seg_size(struct ib_device * dev)4193 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4194 {
4195 	if (ib_uses_virt_dma(dev))
4196 		return UINT_MAX;
4197 	return dma_get_max_seg_size(dev->dma_device);
4198 }
4199 
4200 /**
4201  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4202  * @dev: The device for which the DMA address was created
4203  * @addr: The DMA address
4204  * @size: The size of the region in bytes
4205  * @dir: The direction of the DMA
4206  */
ib_dma_sync_single_for_cpu(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction dir)4207 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4208 					      u64 addr,
4209 					      size_t size,
4210 					      enum dma_data_direction dir)
4211 {
4212 	if (!ib_uses_virt_dma(dev))
4213 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4214 }
4215 
4216 /**
4217  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4218  * @dev: The device for which the DMA address was created
4219  * @addr: The DMA address
4220  * @size: The size of the region in bytes
4221  * @dir: The direction of the DMA
4222  */
ib_dma_sync_single_for_device(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction dir)4223 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4224 						 u64 addr,
4225 						 size_t size,
4226 						 enum dma_data_direction dir)
4227 {
4228 	if (!ib_uses_virt_dma(dev))
4229 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4230 }
4231 
4232 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4233  * space. This function should be called when 'current' is the owning MM.
4234  */
4235 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4236 			     u64 virt_addr, int mr_access_flags);
4237 
4238 /* ib_advise_mr -  give an advice about an address range in a memory region */
4239 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4240 		 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4241 /**
4242  * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4243  *   HCA translation table.
4244  * @mr: The memory region to deregister.
4245  * @udata: Valid user data or NULL for kernel object
4246  *
4247  * This function can fail, if the memory region has memory windows bound to it.
4248  */
4249 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4250 
4251 /**
4252  * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4253  *   HCA translation table.
4254  * @mr: The memory region to deregister.
4255  *
4256  * This function can fail, if the memory region has memory windows bound to it.
4257  *
4258  * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4259  */
ib_dereg_mr(struct ib_mr * mr)4260 static inline int ib_dereg_mr(struct ib_mr *mr)
4261 {
4262 	return ib_dereg_mr_user(mr, NULL);
4263 }
4264 
4265 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4266 			  u32 max_num_sg);
4267 
4268 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4269 				    u32 max_num_data_sg,
4270 				    u32 max_num_meta_sg);
4271 
4272 /**
4273  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4274  *   R_Key and L_Key.
4275  * @mr - struct ib_mr pointer to be updated.
4276  * @newkey - new key to be used.
4277  */
ib_update_fast_reg_key(struct ib_mr * mr,u8 newkey)4278 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4279 {
4280 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4281 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4282 }
4283 
4284 /**
4285  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4286  * for calculating a new rkey for type 2 memory windows.
4287  * @rkey - the rkey to increment.
4288  */
ib_inc_rkey(u32 rkey)4289 static inline u32 ib_inc_rkey(u32 rkey)
4290 {
4291 	const u32 mask = 0x000000ff;
4292 	return ((rkey + 1) & mask) | (rkey & ~mask);
4293 }
4294 
4295 /**
4296  * ib_attach_mcast - Attaches the specified QP to a multicast group.
4297  * @qp: QP to attach to the multicast group.  The QP must be type
4298  *   IB_QPT_UD.
4299  * @gid: Multicast group GID.
4300  * @lid: Multicast group LID in host byte order.
4301  *
4302  * In order to send and receive multicast packets, subnet
4303  * administration must have created the multicast group and configured
4304  * the fabric appropriately.  The port associated with the specified
4305  * QP must also be a member of the multicast group.
4306  */
4307 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4308 
4309 /**
4310  * ib_detach_mcast - Detaches the specified QP from a multicast group.
4311  * @qp: QP to detach from the multicast group.
4312  * @gid: Multicast group GID.
4313  * @lid: Multicast group LID in host byte order.
4314  */
4315 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4316 
4317 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4318 				   struct inode *inode, struct ib_udata *udata);
4319 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4320 
ib_check_mr_access(struct ib_device * ib_dev,unsigned int flags)4321 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4322 				     unsigned int flags)
4323 {
4324 	/*
4325 	 * Local write permission is required if remote write or
4326 	 * remote atomic permission is also requested.
4327 	 */
4328 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4329 	    !(flags & IB_ACCESS_LOCAL_WRITE))
4330 		return -EINVAL;
4331 
4332 	if (flags & ~IB_ACCESS_SUPPORTED)
4333 		return -EINVAL;
4334 
4335 	if (flags & IB_ACCESS_ON_DEMAND &&
4336 	    !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING))
4337 		return -EINVAL;
4338 	return 0;
4339 }
4340 
ib_access_writable(int access_flags)4341 static inline bool ib_access_writable(int access_flags)
4342 {
4343 	/*
4344 	 * We have writable memory backing the MR if any of the following
4345 	 * access flags are set.  "Local write" and "remote write" obviously
4346 	 * require write access.  "Remote atomic" can do things like fetch and
4347 	 * add, which will modify memory, and "MW bind" can change permissions
4348 	 * by binding a window.
4349 	 */
4350 	return access_flags &
4351 		(IB_ACCESS_LOCAL_WRITE   | IB_ACCESS_REMOTE_WRITE |
4352 		 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4353 }
4354 
4355 /**
4356  * ib_check_mr_status: lightweight check of MR status.
4357  *     This routine may provide status checks on a selected
4358  *     ib_mr. first use is for signature status check.
4359  *
4360  * @mr: A memory region.
4361  * @check_mask: Bitmask of which checks to perform from
4362  *     ib_mr_status_check enumeration.
4363  * @mr_status: The container of relevant status checks.
4364  *     failed checks will be indicated in the status bitmask
4365  *     and the relevant info shall be in the error item.
4366  */
4367 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4368 		       struct ib_mr_status *mr_status);
4369 
4370 /**
4371  * ib_device_try_get: Hold a registration lock
4372  * device: The device to lock
4373  *
4374  * A device under an active registration lock cannot become unregistered. It
4375  * is only possible to obtain a registration lock on a device that is fully
4376  * registered, otherwise this function returns false.
4377  *
4378  * The registration lock is only necessary for actions which require the
4379  * device to still be registered. Uses that only require the device pointer to
4380  * be valid should use get_device(&ibdev->dev) to hold the memory.
4381  *
4382  */
ib_device_try_get(struct ib_device * dev)4383 static inline bool ib_device_try_get(struct ib_device *dev)
4384 {
4385 	return refcount_inc_not_zero(&dev->refcount);
4386 }
4387 
4388 void ib_device_put(struct ib_device *device);
4389 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4390 					  enum rdma_driver_id driver_id);
4391 struct ib_device *ib_device_get_by_name(const char *name,
4392 					enum rdma_driver_id driver_id);
4393 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4394 					    u16 pkey, const union ib_gid *gid,
4395 					    const struct sockaddr *addr);
4396 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4397 			 unsigned int port);
4398 struct net_device *ib_device_netdev(struct ib_device *dev, u32 port);
4399 
4400 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4401 			   struct ib_wq_init_attr *init_attr);
4402 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4403 
4404 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4405 		 unsigned int *sg_offset, unsigned int page_size);
4406 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4407 		    int data_sg_nents, unsigned int *data_sg_offset,
4408 		    struct scatterlist *meta_sg, int meta_sg_nents,
4409 		    unsigned int *meta_sg_offset, unsigned int page_size);
4410 
4411 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)4412 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4413 		  unsigned int *sg_offset, unsigned int page_size)
4414 {
4415 	int n;
4416 
4417 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4418 	mr->iova = 0;
4419 
4420 	return n;
4421 }
4422 
4423 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4424 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4425 
4426 void ib_drain_rq(struct ib_qp *qp);
4427 void ib_drain_sq(struct ib_qp *qp);
4428 void ib_drain_qp(struct ib_qp *qp);
4429 
4430 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4431 		     u8 *width);
4432 
rdma_ah_retrieve_dmac(struct rdma_ah_attr * attr)4433 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4434 {
4435 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4436 		return attr->roce.dmac;
4437 	return NULL;
4438 }
4439 
rdma_ah_set_dlid(struct rdma_ah_attr * attr,u32 dlid)4440 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4441 {
4442 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4443 		attr->ib.dlid = (u16)dlid;
4444 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4445 		attr->opa.dlid = dlid;
4446 }
4447 
rdma_ah_get_dlid(const struct rdma_ah_attr * attr)4448 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4449 {
4450 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4451 		return attr->ib.dlid;
4452 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4453 		return attr->opa.dlid;
4454 	return 0;
4455 }
4456 
rdma_ah_set_sl(struct rdma_ah_attr * attr,u8 sl)4457 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4458 {
4459 	attr->sl = sl;
4460 }
4461 
rdma_ah_get_sl(const struct rdma_ah_attr * attr)4462 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4463 {
4464 	return attr->sl;
4465 }
4466 
rdma_ah_set_path_bits(struct rdma_ah_attr * attr,u8 src_path_bits)4467 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4468 					 u8 src_path_bits)
4469 {
4470 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4471 		attr->ib.src_path_bits = src_path_bits;
4472 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4473 		attr->opa.src_path_bits = src_path_bits;
4474 }
4475 
rdma_ah_get_path_bits(const struct rdma_ah_attr * attr)4476 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4477 {
4478 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4479 		return attr->ib.src_path_bits;
4480 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4481 		return attr->opa.src_path_bits;
4482 	return 0;
4483 }
4484 
rdma_ah_set_make_grd(struct rdma_ah_attr * attr,bool make_grd)4485 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4486 					bool make_grd)
4487 {
4488 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4489 		attr->opa.make_grd = make_grd;
4490 }
4491 
rdma_ah_get_make_grd(const struct rdma_ah_attr * attr)4492 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4493 {
4494 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4495 		return attr->opa.make_grd;
4496 	return false;
4497 }
4498 
rdma_ah_set_port_num(struct rdma_ah_attr * attr,u32 port_num)4499 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4500 {
4501 	attr->port_num = port_num;
4502 }
4503 
rdma_ah_get_port_num(const struct rdma_ah_attr * attr)4504 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4505 {
4506 	return attr->port_num;
4507 }
4508 
rdma_ah_set_static_rate(struct rdma_ah_attr * attr,u8 static_rate)4509 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4510 					   u8 static_rate)
4511 {
4512 	attr->static_rate = static_rate;
4513 }
4514 
rdma_ah_get_static_rate(const struct rdma_ah_attr * attr)4515 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4516 {
4517 	return attr->static_rate;
4518 }
4519 
rdma_ah_set_ah_flags(struct rdma_ah_attr * attr,enum ib_ah_flags flag)4520 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4521 					enum ib_ah_flags flag)
4522 {
4523 	attr->ah_flags = flag;
4524 }
4525 
4526 static inline enum ib_ah_flags
rdma_ah_get_ah_flags(const struct rdma_ah_attr * attr)4527 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4528 {
4529 	return attr->ah_flags;
4530 }
4531 
4532 static inline const struct ib_global_route
rdma_ah_read_grh(const struct rdma_ah_attr * attr)4533 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4534 {
4535 	return &attr->grh;
4536 }
4537 
4538 /*To retrieve and modify the grh */
4539 static inline struct ib_global_route
rdma_ah_retrieve_grh(struct rdma_ah_attr * attr)4540 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4541 {
4542 	return &attr->grh;
4543 }
4544 
rdma_ah_set_dgid_raw(struct rdma_ah_attr * attr,void * dgid)4545 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4546 {
4547 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4548 
4549 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4550 }
4551 
rdma_ah_set_subnet_prefix(struct rdma_ah_attr * attr,__be64 prefix)4552 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4553 					     __be64 prefix)
4554 {
4555 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4556 
4557 	grh->dgid.global.subnet_prefix = prefix;
4558 }
4559 
rdma_ah_set_interface_id(struct rdma_ah_attr * attr,__be64 if_id)4560 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4561 					    __be64 if_id)
4562 {
4563 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4564 
4565 	grh->dgid.global.interface_id = if_id;
4566 }
4567 
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)4568 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4569 				   union ib_gid *dgid, u32 flow_label,
4570 				   u8 sgid_index, u8 hop_limit,
4571 				   u8 traffic_class)
4572 {
4573 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4574 
4575 	attr->ah_flags = IB_AH_GRH;
4576 	if (dgid)
4577 		grh->dgid = *dgid;
4578 	grh->flow_label = flow_label;
4579 	grh->sgid_index = sgid_index;
4580 	grh->hop_limit = hop_limit;
4581 	grh->traffic_class = traffic_class;
4582 	grh->sgid_attr = NULL;
4583 }
4584 
4585 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4586 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4587 			     u32 flow_label, u8 hop_limit, u8 traffic_class,
4588 			     const struct ib_gid_attr *sgid_attr);
4589 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4590 		       const struct rdma_ah_attr *src);
4591 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4592 			  const struct rdma_ah_attr *new);
4593 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4594 
4595 /**
4596  * rdma_ah_find_type - Return address handle type.
4597  *
4598  * @dev: Device to be checked
4599  * @port_num: Port number
4600  */
rdma_ah_find_type(struct ib_device * dev,u32 port_num)4601 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4602 						       u32 port_num)
4603 {
4604 	if (rdma_protocol_roce(dev, port_num))
4605 		return RDMA_AH_ATTR_TYPE_ROCE;
4606 	if (rdma_protocol_ib(dev, port_num)) {
4607 		if (rdma_cap_opa_ah(dev, port_num))
4608 			return RDMA_AH_ATTR_TYPE_OPA;
4609 		return RDMA_AH_ATTR_TYPE_IB;
4610 	}
4611 
4612 	return RDMA_AH_ATTR_TYPE_UNDEFINED;
4613 }
4614 
4615 /**
4616  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4617  *     In the current implementation the only way to
4618  *     get the 32bit lid is from other sources for OPA.
4619  *     For IB, lids will always be 16bits so cast the
4620  *     value accordingly.
4621  *
4622  * @lid: A 32bit LID
4623  */
ib_lid_cpu16(u32 lid)4624 static inline u16 ib_lid_cpu16(u32 lid)
4625 {
4626 	WARN_ON_ONCE(lid & 0xFFFF0000);
4627 	return (u16)lid;
4628 }
4629 
4630 /**
4631  * ib_lid_be16 - Return lid in 16bit BE encoding.
4632  *
4633  * @lid: A 32bit LID
4634  */
ib_lid_be16(u32 lid)4635 static inline __be16 ib_lid_be16(u32 lid)
4636 {
4637 	WARN_ON_ONCE(lid & 0xFFFF0000);
4638 	return cpu_to_be16((u16)lid);
4639 }
4640 
4641 /**
4642  * ib_get_vector_affinity - Get the affinity mappings of a given completion
4643  *   vector
4644  * @device:         the rdma device
4645  * @comp_vector:    index of completion vector
4646  *
4647  * Returns NULL on failure, otherwise a corresponding cpu map of the
4648  * completion vector (returns all-cpus map if the device driver doesn't
4649  * implement get_vector_affinity).
4650  */
4651 static inline const struct cpumask *
ib_get_vector_affinity(struct ib_device * device,int comp_vector)4652 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4653 {
4654 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4655 	    !device->ops.get_vector_affinity)
4656 		return NULL;
4657 
4658 	return device->ops.get_vector_affinity(device, comp_vector);
4659 
4660 }
4661 
4662 /**
4663  * rdma_roce_rescan_device - Rescan all of the network devices in the system
4664  * and add their gids, as needed, to the relevant RoCE devices.
4665  *
4666  * @device:         the rdma device
4667  */
4668 void rdma_roce_rescan_device(struct ib_device *ibdev);
4669 
4670 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4671 
4672 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4673 
4674 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4675 				     enum rdma_netdev_t type, const char *name,
4676 				     unsigned char name_assign_type,
4677 				     void (*setup)(struct net_device *));
4678 
4679 int rdma_init_netdev(struct ib_device *device, u32 port_num,
4680 		     enum rdma_netdev_t type, const char *name,
4681 		     unsigned char name_assign_type,
4682 		     void (*setup)(struct net_device *),
4683 		     struct net_device *netdev);
4684 
4685 /**
4686  * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4687  *
4688  * @device:	device pointer for which ib_device pointer to retrieve
4689  *
4690  * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4691  *
4692  */
rdma_device_to_ibdev(struct device * device)4693 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4694 {
4695 	struct ib_core_device *coredev =
4696 		container_of(device, struct ib_core_device, dev);
4697 
4698 	return coredev->owner;
4699 }
4700 
4701 /**
4702  * ibdev_to_node - return the NUMA node for a given ib_device
4703  * @dev:	device to get the NUMA node for.
4704  */
ibdev_to_node(struct ib_device * ibdev)4705 static inline int ibdev_to_node(struct ib_device *ibdev)
4706 {
4707 	struct device *parent = ibdev->dev.parent;
4708 
4709 	if (!parent)
4710 		return NUMA_NO_NODE;
4711 	return dev_to_node(parent);
4712 }
4713 
4714 /**
4715  * rdma_device_to_drv_device - Helper macro to reach back to driver's
4716  *			       ib_device holder structure from device pointer.
4717  *
4718  * NOTE: New drivers should not make use of this API; This API is only for
4719  * existing drivers who have exposed sysfs entries using
4720  * ops->device_group.
4721  */
4722 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member)           \
4723 	container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4724 
4725 bool rdma_dev_access_netns(const struct ib_device *device,
4726 			   const struct net *net);
4727 
4728 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4729 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4730 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4731 
4732 /**
4733  * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4734  *                               on the flow_label
4735  *
4736  * This function will convert the 20 bit flow_label input to a valid RoCE v2
4737  * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4738  * convention.
4739  */
rdma_flow_label_to_udp_sport(u32 fl)4740 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4741 {
4742 	u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4743 
4744 	fl_low ^= fl_high >> 14;
4745 	return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4746 }
4747 
4748 /**
4749  * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4750  *                        local and remote qpn values
4751  *
4752  * This function folded the multiplication results of two qpns, 24 bit each,
4753  * fields, and converts it to a 20 bit results.
4754  *
4755  * This function will create symmetric flow_label value based on the local
4756  * and remote qpn values. this will allow both the requester and responder
4757  * to calculate the same flow_label for a given connection.
4758  *
4759  * This helper function should be used by driver in case the upper layer
4760  * provide a zero flow_label value. This is to improve entropy of RDMA
4761  * traffic in the network.
4762  */
rdma_calc_flow_label(u32 lqpn,u32 rqpn)4763 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4764 {
4765 	u64 v = (u64)lqpn * rqpn;
4766 
4767 	v ^= v >> 20;
4768 	v ^= v >> 40;
4769 
4770 	return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4771 }
4772 
4773 /**
4774  * rdma_get_udp_sport - Calculate and set UDP source port based on the flow
4775  *                      label. If flow label is not defined in GRH then
4776  *                      calculate it based on lqpn/rqpn.
4777  *
4778  * @fl:                 flow label from GRH
4779  * @lqpn:               local qp number
4780  * @rqpn:               remote qp number
4781  */
rdma_get_udp_sport(u32 fl,u32 lqpn,u32 rqpn)4782 static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
4783 {
4784 	if (!fl)
4785 		fl = rdma_calc_flow_label(lqpn, rqpn);
4786 
4787 	return rdma_flow_label_to_udp_sport(fl);
4788 }
4789 
4790 const struct ib_port_immutable*
4791 ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4792 #endif /* IB_VERBS_H */
4793