?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
LiteX: driver for ICAPBitstream fpga manager
[linux/fpc-iii.git] / include / rdma / ib_verbs.h
blob9fed65bf92792f0745a94ae77d33bff0d960717a
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 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
54 extern struct workqueue_struct *ib_wq;
55 extern struct workqueue_struct *ib_comp_wq;
56 extern struct workqueue_struct *ib_comp_unbound_wq;
57
58 struct ib_ucq_object;
59
60 __printf(3, 4) __cold
61 void ibdev_printk(const char *level, const struct ib_device *ibdev,
62 const char *format, ...);
63 __printf(2, 3) __cold
64 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
65 __printf(2, 3) __cold
66 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
67 __printf(2, 3) __cold
68 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
69 __printf(2, 3) __cold
70 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
71 __printf(2, 3) __cold
72 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
73 __printf(2, 3) __cold
74 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
75 __printf(2, 3) __cold
76 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
77
78 #if defined(CONFIG_DYNAMIC_DEBUG) || \
79 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
80 #define ibdev_dbg(__dev, format, args...) \
81 dynamic_ibdev_dbg(__dev, format, ##args)
82 #else
83 __printf(2, 3) __cold
84 static inline
85 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
86 #endif
87
88 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
89 do { \
90 static DEFINE_RATELIMIT_STATE(_rs, \
91 DEFAULT_RATELIMIT_INTERVAL, \
92 DEFAULT_RATELIMIT_BURST); \
93 if (__ratelimit(&_rs)) \
94 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
95 } while (0)
96
97 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
98 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
99 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
100 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
101 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
102 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
103 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
104 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
105 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
106 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
107 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
108 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
109 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
110 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
111
112 #if defined(CONFIG_DYNAMIC_DEBUG) || \
113 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
114 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
115 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
116 do { \
117 static DEFINE_RATELIMIT_STATE(_rs, \
118 DEFAULT_RATELIMIT_INTERVAL, \
119 DEFAULT_RATELIMIT_BURST); \
120 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
121 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
122 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
123 ##__VA_ARGS__); \
124 } while (0)
125 #else
126 __printf(2, 3) __cold
127 static inline
128 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
129 #endif
130
131 union ib_gid {
132 u8 raw[16];
133 struct {
134 __be64 subnet_prefix;
135 __be64 interface_id;
136 } global;
137 };
138
139 extern union ib_gid zgid;
140
141 enum ib_gid_type {
142 IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
143 IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
144 IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
145 IB_GID_TYPE_SIZE
146 };
147
148 #define ROCE_V2_UDP_DPORT 4791
149 struct ib_gid_attr {
150 struct net_device __rcu *ndev;
151 struct ib_device *device;
152 union ib_gid gid;
153 enum ib_gid_type gid_type;
154 u16 index;
155 u8 port_num;
156 };
157
158 enum {
159 /* set the local administered indication */
160 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
161 };
162
163 enum rdma_transport_type {
164 RDMA_TRANSPORT_IB,
165 RDMA_TRANSPORT_IWARP,
166 RDMA_TRANSPORT_USNIC,
167 RDMA_TRANSPORT_USNIC_UDP,
168 RDMA_TRANSPORT_UNSPECIFIED,
169 };
170
171 enum rdma_protocol_type {
172 RDMA_PROTOCOL_IB,
173 RDMA_PROTOCOL_IBOE,
174 RDMA_PROTOCOL_IWARP,
175 RDMA_PROTOCOL_USNIC_UDP
176 };
177
178 __attribute_const__ enum rdma_transport_type
179 rdma_node_get_transport(unsigned int node_type);
180
181 enum rdma_network_type {
182 RDMA_NETWORK_IB,
183 RDMA_NETWORK_ROCE_V1,
184 RDMA_NETWORK_IPV4,
185 RDMA_NETWORK_IPV6
186 };
187
188 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
189 {
190 if (network_type == RDMA_NETWORK_IPV4 ||
191 network_type == RDMA_NETWORK_IPV6)
192 return IB_GID_TYPE_ROCE_UDP_ENCAP;
193 else if (network_type == RDMA_NETWORK_ROCE_V1)
194 return IB_GID_TYPE_ROCE;
195 else
196 return IB_GID_TYPE_IB;
197 }
198
199 static inline enum rdma_network_type
200 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
201 {
202 if (attr->gid_type == IB_GID_TYPE_IB)
203 return RDMA_NETWORK_IB;
204
205 if (attr->gid_type == IB_GID_TYPE_ROCE)
206 return RDMA_NETWORK_ROCE_V1;
207
208 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
209 return RDMA_NETWORK_IPV4;
210 else
211 return RDMA_NETWORK_IPV6;
212 }
213
214 enum rdma_link_layer {
215 IB_LINK_LAYER_UNSPECIFIED,
216 IB_LINK_LAYER_INFINIBAND,
217 IB_LINK_LAYER_ETHERNET,
218 };
219
220 enum ib_device_cap_flags {
221 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
222 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
223 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
224 IB_DEVICE_RAW_MULTI = (1 << 3),
225 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
226 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
227 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
228 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
229 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
230 /* Not in use, former INIT_TYPE = (1 << 9),*/
231 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
232 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
233 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
234 IB_DEVICE_SRQ_RESIZE = (1 << 13),
235 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
236
237 /*
238 * This device supports a per-device lkey or stag that can be
239 * used without performing a memory registration for the local
240 * memory. Note that ULPs should never check this flag, but
241 * instead of use the local_dma_lkey flag in the ib_pd structure,
242 * which will always contain a usable lkey.
243 */
244 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
245 /* Reserved, old SEND_W_INV = (1 << 16),*/
246 IB_DEVICE_MEM_WINDOW = (1 << 17),
247 /*
248 * Devices should set IB_DEVICE_UD_IP_SUM if they support
249 * insertion of UDP and TCP checksum on outgoing UD IPoIB
250 * messages and can verify the validity of checksum for
251 * incoming messages. Setting this flag implies that the
252 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
253 */
254 IB_DEVICE_UD_IP_CSUM = (1 << 18),
255 IB_DEVICE_UD_TSO = (1 << 19),
256 IB_DEVICE_XRC = (1 << 20),
257
258 /*
259 * This device supports the IB "base memory management extension",
260 * which includes support for fast registrations (IB_WR_REG_MR,
261 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
262 * also be set by any iWarp device which must support FRs to comply
263 * to the iWarp verbs spec. iWarp devices also support the
264 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
265 * stag.
266 */
267 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
268 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
269 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
270 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
271 IB_DEVICE_RC_IP_CSUM = (1 << 25),
272 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
273 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
274 /*
275 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
276 * support execution of WQEs that involve synchronization
277 * of I/O operations with single completion queue managed
278 * by hardware.
279 */
280 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
281 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
282 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
283 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
284 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
285 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
286 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
287 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
288 IB_DEVICE_RDMA_NETDEV_OPA = (1ULL << 35),
289 /* The device supports padding incoming writes to cacheline. */
290 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
291 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
292 };
293
294 enum ib_atomic_cap {
295 IB_ATOMIC_NONE,
296 IB_ATOMIC_HCA,
297 IB_ATOMIC_GLOB
298 };
299
300 enum ib_odp_general_cap_bits {
301 IB_ODP_SUPPORT = 1 << 0,
302 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
303 };
304
305 enum ib_odp_transport_cap_bits {
306 IB_ODP_SUPPORT_SEND = 1 << 0,
307 IB_ODP_SUPPORT_RECV = 1 << 1,
308 IB_ODP_SUPPORT_WRITE = 1 << 2,
309 IB_ODP_SUPPORT_READ = 1 << 3,
310 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
311 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
312 };
313
314 struct ib_odp_caps {
315 uint64_t general_caps;
316 struct {
317 uint32_t rc_odp_caps;
318 uint32_t uc_odp_caps;
319 uint32_t ud_odp_caps;
320 uint32_t xrc_odp_caps;
321 } per_transport_caps;
322 };
323
324 struct ib_rss_caps {
325 /* Corresponding bit will be set if qp type from
326 * 'enum ib_qp_type' is supported, e.g.
327 * supported_qpts |= 1 << IB_QPT_UD
328 */
329 u32 supported_qpts;
330 u32 max_rwq_indirection_tables;
331 u32 max_rwq_indirection_table_size;
332 };
333
334 enum ib_tm_cap_flags {
335 /* Support tag matching with rendezvous offload for RC transport */
336 IB_TM_CAP_RNDV_RC = 1 << 0,
337 };
338
339 struct ib_tm_caps {
340 /* Max size of RNDV header */
341 u32 max_rndv_hdr_size;
342 /* Max number of entries in tag matching list */
343 u32 max_num_tags;
344 /* From enum ib_tm_cap_flags */
345 u32 flags;
346 /* Max number of outstanding list operations */
347 u32 max_ops;
348 /* Max number of SGE in tag matching entry */
349 u32 max_sge;
350 };
351
352 struct ib_cq_init_attr {
353 unsigned int cqe;
354 u32 comp_vector;
355 u32 flags;
356 };
357
358 enum ib_cq_attr_mask {
359 IB_CQ_MODERATE = 1 << 0,
360 };
361
362 struct ib_cq_caps {
363 u16 max_cq_moderation_count;
364 u16 max_cq_moderation_period;
365 };
366
367 struct ib_dm_mr_attr {
368 u64 length;
369 u64 offset;
370 u32 access_flags;
371 };
372
373 struct ib_dm_alloc_attr {
374 u64 length;
375 u32 alignment;
376 u32 flags;
377 };
378
379 struct ib_device_attr {
380 u64 fw_ver;
381 __be64 sys_image_guid;
382 u64 max_mr_size;
383 u64 page_size_cap;
384 u32 vendor_id;
385 u32 vendor_part_id;
386 u32 hw_ver;
387 int max_qp;
388 int max_qp_wr;
389 u64 device_cap_flags;
390 int max_send_sge;
391 int max_recv_sge;
392 int max_sge_rd;
393 int max_cq;
394 int max_cqe;
395 int max_mr;
396 int max_pd;
397 int max_qp_rd_atom;
398 int max_ee_rd_atom;
399 int max_res_rd_atom;
400 int max_qp_init_rd_atom;
401 int max_ee_init_rd_atom;
402 enum ib_atomic_cap atomic_cap;
403 enum ib_atomic_cap masked_atomic_cap;
404 int max_ee;
405 int max_rdd;
406 int max_mw;
407 int max_raw_ipv6_qp;
408 int max_raw_ethy_qp;
409 int max_mcast_grp;
410 int max_mcast_qp_attach;
411 int max_total_mcast_qp_attach;
412 int max_ah;
413 int max_srq;
414 int max_srq_wr;
415 int max_srq_sge;
416 unsigned int max_fast_reg_page_list_len;
417 unsigned int max_pi_fast_reg_page_list_len;
418 u16 max_pkeys;
419 u8 local_ca_ack_delay;
420 int sig_prot_cap;
421 int sig_guard_cap;
422 struct ib_odp_caps odp_caps;
423 uint64_t timestamp_mask;
424 uint64_t hca_core_clock; /* in KHZ */
425 struct ib_rss_caps rss_caps;
426 u32 max_wq_type_rq;
427 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
428 struct ib_tm_caps tm_caps;
429 struct ib_cq_caps cq_caps;
430 u64 max_dm_size;
431 /* Max entries for sgl for optimized performance per READ */
432 u32 max_sgl_rd;
433 };
434
435 enum ib_mtu {
436 IB_MTU_256 = 1,
437 IB_MTU_512 = 2,
438 IB_MTU_1024 = 3,
439 IB_MTU_2048 = 4,
440 IB_MTU_4096 = 5
441 };
442
443 enum opa_mtu {
444 OPA_MTU_8192 = 6,
445 OPA_MTU_10240 = 7
446 };
447
448 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
449 {
450 switch (mtu) {
451 case IB_MTU_256: return 256;
452 case IB_MTU_512: return 512;
453 case IB_MTU_1024: return 1024;
454 case IB_MTU_2048: return 2048;
455 case IB_MTU_4096: return 4096;
456 default: return -1;
457 }
458 }
459
460 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
461 {
462 if (mtu >= 4096)
463 return IB_MTU_4096;
464 else if (mtu >= 2048)
465 return IB_MTU_2048;
466 else if (mtu >= 1024)
467 return IB_MTU_1024;
468 else if (mtu >= 512)
469 return IB_MTU_512;
470 else
471 return IB_MTU_256;
472 }
473
474 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
475 {
476 switch (mtu) {
477 case OPA_MTU_8192:
478 return 8192;
479 case OPA_MTU_10240:
480 return 10240;
481 default:
482 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
483 }
484 }
485
486 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
487 {
488 if (mtu >= 10240)
489 return OPA_MTU_10240;
490 else if (mtu >= 8192)
491 return OPA_MTU_8192;
492 else
493 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
494 }
495
496 enum ib_port_state {
497 IB_PORT_NOP = 0,
498 IB_PORT_DOWN = 1,
499 IB_PORT_INIT = 2,
500 IB_PORT_ARMED = 3,
501 IB_PORT_ACTIVE = 4,
502 IB_PORT_ACTIVE_DEFER = 5
503 };
504
505 enum ib_port_phys_state {
506 IB_PORT_PHYS_STATE_SLEEP = 1,
507 IB_PORT_PHYS_STATE_POLLING = 2,
508 IB_PORT_PHYS_STATE_DISABLED = 3,
509 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
510 IB_PORT_PHYS_STATE_LINK_UP = 5,
511 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
512 IB_PORT_PHYS_STATE_PHY_TEST = 7,
513 };
514
515 enum ib_port_width {
516 IB_WIDTH_1X = 1,
517 IB_WIDTH_2X = 16,
518 IB_WIDTH_4X = 2,
519 IB_WIDTH_8X = 4,
520 IB_WIDTH_12X = 8
521 };
522
523 static inline int ib_width_enum_to_int(enum ib_port_width width)
524 {
525 switch (width) {
526 case IB_WIDTH_1X: return 1;
527 case IB_WIDTH_2X: return 2;
528 case IB_WIDTH_4X: return 4;
529 case IB_WIDTH_8X: return 8;
530 case IB_WIDTH_12X: return 12;
531 default: return -1;
532 }
533 }
534
535 enum ib_port_speed {
536 IB_SPEED_SDR = 1,
537 IB_SPEED_DDR = 2,
538 IB_SPEED_QDR = 4,
539 IB_SPEED_FDR10 = 8,
540 IB_SPEED_FDR = 16,
541 IB_SPEED_EDR = 32,
542 IB_SPEED_HDR = 64,
543 IB_SPEED_NDR = 128,
544 };
545
546 /**
547 * struct rdma_hw_stats
548 * @lock - Mutex to protect parallel write access to lifespan and values
549 * of counters, which are 64bits and not guaranteeed to be written
550 * atomicaly on 32bits systems.
551 * @timestamp - Used by the core code to track when the last update was
552 * @lifespan - Used by the core code to determine how old the counters
553 * should be before being updated again. Stored in jiffies, defaults
554 * to 10 milliseconds, drivers can override the default be specifying
555 * their own value during their allocation routine.
556 * @name - Array of pointers to static names used for the counters in
557 * directory.
558 * @num_counters - How many hardware counters there are. If name is
559 * shorter than this number, a kernel oops will result. Driver authors
560 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
561 * in their code to prevent this.
562 * @value - Array of u64 counters that are accessed by the sysfs code and
563 * filled in by the drivers get_stats routine
564 */
565 struct rdma_hw_stats {
566 struct mutex lock; /* Protect lifespan and values[] */
567 unsigned long timestamp;
568 unsigned long lifespan;
569 const char * const *names;
570 int num_counters;
571 u64 value[];
572 };
573
574 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
575 /**
576 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
577 * for drivers.
578 * @names - Array of static const char *
579 * @num_counters - How many elements in array
580 * @lifespan - How many milliseconds between updates
581 */
582 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
583 const char * const *names, int num_counters,
584 unsigned long lifespan)
585 {
586 struct rdma_hw_stats *stats;
587
588 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
589 GFP_KERNEL);
590 if (!stats)
591 return NULL;
592 stats->names = names;
593 stats->num_counters = num_counters;
594 stats->lifespan = msecs_to_jiffies(lifespan);
595
596 return stats;
597 }
598
599
600 /* Define bits for the various functionality this port needs to be supported by
601 * the core.
602 */
603 /* Management 0x00000FFF */
604 #define RDMA_CORE_CAP_IB_MAD 0x00000001
605 #define RDMA_CORE_CAP_IB_SMI 0x00000002
606 #define RDMA_CORE_CAP_IB_CM 0x00000004
607 #define RDMA_CORE_CAP_IW_CM 0x00000008
608 #define RDMA_CORE_CAP_IB_SA 0x00000010
609 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
610
611 /* Address format 0x000FF000 */
612 #define RDMA_CORE_CAP_AF_IB 0x00001000
613 #define RDMA_CORE_CAP_ETH_AH 0x00002000
614 #define RDMA_CORE_CAP_OPA_AH 0x00004000
615 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
616
617 /* Protocol 0xFFF00000 */
618 #define RDMA_CORE_CAP_PROT_IB 0x00100000
619 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
620 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
621 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
622 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
623 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
624
625 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
626 | RDMA_CORE_CAP_PROT_ROCE \
627 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
628
629 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
630 | RDMA_CORE_CAP_IB_MAD \
631 | RDMA_CORE_CAP_IB_SMI \
632 | RDMA_CORE_CAP_IB_CM \
633 | RDMA_CORE_CAP_IB_SA \
634 | RDMA_CORE_CAP_AF_IB)
635 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
636 | RDMA_CORE_CAP_IB_MAD \
637 | RDMA_CORE_CAP_IB_CM \
638 | RDMA_CORE_CAP_AF_IB \
639 | RDMA_CORE_CAP_ETH_AH)
640 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
641 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
642 | RDMA_CORE_CAP_IB_MAD \
643 | RDMA_CORE_CAP_IB_CM \
644 | RDMA_CORE_CAP_AF_IB \
645 | RDMA_CORE_CAP_ETH_AH)
646 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
647 | RDMA_CORE_CAP_IW_CM)
648 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
649 | RDMA_CORE_CAP_OPA_MAD)
650
651 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
652
653 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
654
655 struct ib_port_attr {
656 u64 subnet_prefix;
657 enum ib_port_state state;
658 enum ib_mtu max_mtu;
659 enum ib_mtu active_mtu;
660 u32 phys_mtu;
661 int gid_tbl_len;
662 unsigned int ip_gids:1;
663 /* This is the value from PortInfo CapabilityMask, defined by IBA */
664 u32 port_cap_flags;
665 u32 max_msg_sz;
666 u32 bad_pkey_cntr;
667 u32 qkey_viol_cntr;
668 u16 pkey_tbl_len;
669 u32 sm_lid;
670 u32 lid;
671 u8 lmc;
672 u8 max_vl_num;
673 u8 sm_sl;
674 u8 subnet_timeout;
675 u8 init_type_reply;
676 u8 active_width;
677 u16 active_speed;
678 u8 phys_state;
679 u16 port_cap_flags2;
680 };
681
682 enum ib_device_modify_flags {
683 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
684 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
685 };
686
687 #define IB_DEVICE_NODE_DESC_MAX 64
688
689 struct ib_device_modify {
690 u64 sys_image_guid;
691 char node_desc[IB_DEVICE_NODE_DESC_MAX];
692 };
693
694 enum ib_port_modify_flags {
695 IB_PORT_SHUTDOWN = 1,
696 IB_PORT_INIT_TYPE = (1<<2),
697 IB_PORT_RESET_QKEY_CNTR = (1<<3),
698 IB_PORT_OPA_MASK_CHG = (1<<4)
699 };
700
701 struct ib_port_modify {
702 u32 set_port_cap_mask;
703 u32 clr_port_cap_mask;
704 u8 init_type;
705 };
706
707 enum ib_event_type {
708 IB_EVENT_CQ_ERR,
709 IB_EVENT_QP_FATAL,
710 IB_EVENT_QP_REQ_ERR,
711 IB_EVENT_QP_ACCESS_ERR,
712 IB_EVENT_COMM_EST,
713 IB_EVENT_SQ_DRAINED,
714 IB_EVENT_PATH_MIG,
715 IB_EVENT_PATH_MIG_ERR,
716 IB_EVENT_DEVICE_FATAL,
717 IB_EVENT_PORT_ACTIVE,
718 IB_EVENT_PORT_ERR,
719 IB_EVENT_LID_CHANGE,
720 IB_EVENT_PKEY_CHANGE,
721 IB_EVENT_SM_CHANGE,
722 IB_EVENT_SRQ_ERR,
723 IB_EVENT_SRQ_LIMIT_REACHED,
724 IB_EVENT_QP_LAST_WQE_REACHED,
725 IB_EVENT_CLIENT_REREGISTER,
726 IB_EVENT_GID_CHANGE,
727 IB_EVENT_WQ_FATAL,
728 };
729
730 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
731
732 struct ib_event {
733 struct ib_device *device;
734 union {
735 struct ib_cq *cq;
736 struct ib_qp *qp;
737 struct ib_srq *srq;
738 struct ib_wq *wq;
739 u8 port_num;
740 } element;
741 enum ib_event_type event;
742 };
743
744 struct ib_event_handler {
745 struct ib_device *device;
746 void (*handler)(struct ib_event_handler *, struct ib_event *);
747 struct list_head list;
748 };
749
750 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
751 do { \
752 (_ptr)->device = _device; \
753 (_ptr)->handler = _handler; \
754 INIT_LIST_HEAD(&(_ptr)->list); \
755 } while (0)
756
757 struct ib_global_route {
758 const struct ib_gid_attr *sgid_attr;
759 union ib_gid dgid;
760 u32 flow_label;
761 u8 sgid_index;
762 u8 hop_limit;
763 u8 traffic_class;
764 };
765
766 struct ib_grh {
767 __be32 version_tclass_flow;
768 __be16 paylen;
769 u8 next_hdr;
770 u8 hop_limit;
771 union ib_gid sgid;
772 union ib_gid dgid;
773 };
774
775 union rdma_network_hdr {
776 struct ib_grh ibgrh;
777 struct {
778 /* The IB spec states that if it's IPv4, the header
779 * is located in the last 20 bytes of the header.
780 */
781 u8 reserved[20];
782 struct iphdr roce4grh;
783 };
784 };
785
786 #define IB_QPN_MASK 0xFFFFFF
787
788 enum {
789 IB_MULTICAST_QPN = 0xffffff
790 };
791
792 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
793 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
794
795 enum ib_ah_flags {
796 IB_AH_GRH = 1
797 };
798
799 enum ib_rate {
800 IB_RATE_PORT_CURRENT = 0,
801 IB_RATE_2_5_GBPS = 2,
802 IB_RATE_5_GBPS = 5,
803 IB_RATE_10_GBPS = 3,
804 IB_RATE_20_GBPS = 6,
805 IB_RATE_30_GBPS = 4,
806 IB_RATE_40_GBPS = 7,
807 IB_RATE_60_GBPS = 8,
808 IB_RATE_80_GBPS = 9,
809 IB_RATE_120_GBPS = 10,
810 IB_RATE_14_GBPS = 11,
811 IB_RATE_56_GBPS = 12,
812 IB_RATE_112_GBPS = 13,
813 IB_RATE_168_GBPS = 14,
814 IB_RATE_25_GBPS = 15,
815 IB_RATE_100_GBPS = 16,
816 IB_RATE_200_GBPS = 17,
817 IB_RATE_300_GBPS = 18,
818 IB_RATE_28_GBPS = 19,
819 IB_RATE_50_GBPS = 20,
820 IB_RATE_400_GBPS = 21,
821 IB_RATE_600_GBPS = 22,
822 };
823
824 /**
825 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
826 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
827 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
828 * @rate: rate to convert.
829 */
830 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
831
832 /**
833 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
834 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
835 * @rate: rate to convert.
836 */
837 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
838
839
840 /**
841 * enum ib_mr_type - memory region type
842 * @IB_MR_TYPE_MEM_REG: memory region that is used for
843 * normal registration
844 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
845 * register any arbitrary sg lists (without
846 * the normal mr constraints - see
847 * ib_map_mr_sg)
848 * @IB_MR_TYPE_DM: memory region that is used for device
849 * memory registration
850 * @IB_MR_TYPE_USER: memory region that is used for the user-space
851 * application
852 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
853 * without address translations (VA=PA)
854 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
855 * data integrity operations
856 */
857 enum ib_mr_type {
858 IB_MR_TYPE_MEM_REG,
859 IB_MR_TYPE_SG_GAPS,
860 IB_MR_TYPE_DM,
861 IB_MR_TYPE_USER,
862 IB_MR_TYPE_DMA,
863 IB_MR_TYPE_INTEGRITY,
864 };
865
866 enum ib_mr_status_check {
867 IB_MR_CHECK_SIG_STATUS = 1,
868 };
869
870 /**
871 * struct ib_mr_status - Memory region status container
872 *
873 * @fail_status: Bitmask of MR checks status. For each
874 * failed check a corresponding status bit is set.
875 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
876 * failure.
877 */
878 struct ib_mr_status {
879 u32 fail_status;
880 struct ib_sig_err sig_err;
881 };
882
883 /**
884 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
885 * enum.
886 * @mult: multiple to convert.
887 */
888 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
889
890 struct rdma_ah_init_attr {
891 struct rdma_ah_attr *ah_attr;
892 u32 flags;
893 struct net_device *xmit_slave;
894 };
895
896 enum rdma_ah_attr_type {
897 RDMA_AH_ATTR_TYPE_UNDEFINED,
898 RDMA_AH_ATTR_TYPE_IB,
899 RDMA_AH_ATTR_TYPE_ROCE,
900 RDMA_AH_ATTR_TYPE_OPA,
901 };
902
903 struct ib_ah_attr {
904 u16 dlid;
905 u8 src_path_bits;
906 };
907
908 struct roce_ah_attr {
909 u8 dmac[ETH_ALEN];
910 };
911
912 struct opa_ah_attr {
913 u32 dlid;
914 u8 src_path_bits;
915 bool make_grd;
916 };
917
918 struct rdma_ah_attr {
919 struct ib_global_route grh;
920 u8 sl;
921 u8 static_rate;
922 u8 port_num;
923 u8 ah_flags;
924 enum rdma_ah_attr_type type;
925 union {
926 struct ib_ah_attr ib;
927 struct roce_ah_attr roce;
928 struct opa_ah_attr opa;
929 };
930 };
931
932 enum ib_wc_status {
933 IB_WC_SUCCESS,
934 IB_WC_LOC_LEN_ERR,
935 IB_WC_LOC_QP_OP_ERR,
936 IB_WC_LOC_EEC_OP_ERR,
937 IB_WC_LOC_PROT_ERR,
938 IB_WC_WR_FLUSH_ERR,
939 IB_WC_MW_BIND_ERR,
940 IB_WC_BAD_RESP_ERR,
941 IB_WC_LOC_ACCESS_ERR,
942 IB_WC_REM_INV_REQ_ERR,
943 IB_WC_REM_ACCESS_ERR,
944 IB_WC_REM_OP_ERR,
945 IB_WC_RETRY_EXC_ERR,
946 IB_WC_RNR_RETRY_EXC_ERR,
947 IB_WC_LOC_RDD_VIOL_ERR,
948 IB_WC_REM_INV_RD_REQ_ERR,
949 IB_WC_REM_ABORT_ERR,
950 IB_WC_INV_EECN_ERR,
951 IB_WC_INV_EEC_STATE_ERR,
952 IB_WC_FATAL_ERR,
953 IB_WC_RESP_TIMEOUT_ERR,
954 IB_WC_GENERAL_ERR
955 };
956
957 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
958
959 enum ib_wc_opcode {
960 IB_WC_SEND = IB_UVERBS_WC_SEND,
961 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
962 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
963 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
964 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
965 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
966 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
967 IB_WC_LSO = IB_UVERBS_WC_TSO,
968 IB_WC_REG_MR,
969 IB_WC_MASKED_COMP_SWAP,
970 IB_WC_MASKED_FETCH_ADD,
971 /*
972 * Set value of IB_WC_RECV so consumers can test if a completion is a
973 * receive by testing (opcode & IB_WC_RECV).
974 */
975 IB_WC_RECV = 1 << 7,
976 IB_WC_RECV_RDMA_WITH_IMM
977 };
978
979 enum ib_wc_flags {
980 IB_WC_GRH = 1,
981 IB_WC_WITH_IMM = (1<<1),
982 IB_WC_WITH_INVALIDATE = (1<<2),
983 IB_WC_IP_CSUM_OK = (1<<3),
984 IB_WC_WITH_SMAC = (1<<4),
985 IB_WC_WITH_VLAN = (1<<5),
986 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
987 };
988
989 struct ib_wc {
990 union {
991 u64 wr_id;
992 struct ib_cqe *wr_cqe;
993 };
994 enum ib_wc_status status;
995 enum ib_wc_opcode opcode;
996 u32 vendor_err;
997 u32 byte_len;
998 struct ib_qp *qp;
999 union {
1000 __be32 imm_data;
1001 u32 invalidate_rkey;
1002 } ex;
1003 u32 src_qp;
1004 u32 slid;
1005 int wc_flags;
1006 u16 pkey_index;
1007 u8 sl;
1008 u8 dlid_path_bits;
1009 u8 port_num; /* valid only for DR SMPs on switches */
1010 u8 smac[ETH_ALEN];
1011 u16 vlan_id;
1012 u8 network_hdr_type;
1013 };
1014
1015 enum ib_cq_notify_flags {
1016 IB_CQ_SOLICITED = 1 << 0,
1017 IB_CQ_NEXT_COMP = 1 << 1,
1018 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1019 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1020 };
1021
1022 enum ib_srq_type {
1023 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1024 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1025 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1026 };
1027
1028 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1029 {
1030 return srq_type == IB_SRQT_XRC ||
1031 srq_type == IB_SRQT_TM;
1032 }
1033
1034 enum ib_srq_attr_mask {
1035 IB_SRQ_MAX_WR = 1 << 0,
1036 IB_SRQ_LIMIT = 1 << 1,
1037 };
1038
1039 struct ib_srq_attr {
1040 u32 max_wr;
1041 u32 max_sge;
1042 u32 srq_limit;
1043 };
1044
1045 struct ib_srq_init_attr {
1046 void (*event_handler)(struct ib_event *, void *);
1047 void *srq_context;
1048 struct ib_srq_attr attr;
1049 enum ib_srq_type srq_type;
1050
1051 struct {
1052 struct ib_cq *cq;
1053 union {
1054 struct {
1055 struct ib_xrcd *xrcd;
1056 } xrc;
1057
1058 struct {
1059 u32 max_num_tags;
1060 } tag_matching;
1061 };
1062 } ext;
1063 };
1064
1065 struct ib_qp_cap {
1066 u32 max_send_wr;
1067 u32 max_recv_wr;
1068 u32 max_send_sge;
1069 u32 max_recv_sge;
1070 u32 max_inline_data;
1071
1072 /*
1073 * Maximum number of rdma_rw_ctx structures in flight at a time.
1074 * ib_create_qp() will calculate the right amount of neededed WRs
1075 * and MRs based on this.
1076 */
1077 u32 max_rdma_ctxs;
1078 };
1079
1080 enum ib_sig_type {
1081 IB_SIGNAL_ALL_WR,
1082 IB_SIGNAL_REQ_WR
1083 };
1084
1085 enum ib_qp_type {
1086 /*
1087 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1088 * here (and in that order) since the MAD layer uses them as
1089 * indices into a 2-entry table.
1090 */
1091 IB_QPT_SMI,
1092 IB_QPT_GSI,
1093
1094 IB_QPT_RC = IB_UVERBS_QPT_RC,
1095 IB_QPT_UC = IB_UVERBS_QPT_UC,
1096 IB_QPT_UD = IB_UVERBS_QPT_UD,
1097 IB_QPT_RAW_IPV6,
1098 IB_QPT_RAW_ETHERTYPE,
1099 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1100 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1101 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1102 IB_QPT_MAX,
1103 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1104 /* Reserve a range for qp types internal to the low level driver.
1105 * These qp types will not be visible at the IB core layer, so the
1106 * IB_QPT_MAX usages should not be affected in the core layer
1107 */
1108 IB_QPT_RESERVED1 = 0x1000,
1109 IB_QPT_RESERVED2,
1110 IB_QPT_RESERVED3,
1111 IB_QPT_RESERVED4,
1112 IB_QPT_RESERVED5,
1113 IB_QPT_RESERVED6,
1114 IB_QPT_RESERVED7,
1115 IB_QPT_RESERVED8,
1116 IB_QPT_RESERVED9,
1117 IB_QPT_RESERVED10,
1118 };
1119
1120 enum ib_qp_create_flags {
1121 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1122 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1123 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1124 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1125 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1126 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1127 IB_QP_CREATE_NETIF_QP = 1 << 5,
1128 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1129 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1130 IB_QP_CREATE_SCATTER_FCS =
1131 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1132 IB_QP_CREATE_CVLAN_STRIPPING =
1133 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1134 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1135 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1136 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1137 /* reserve bits 26-31 for low level drivers' internal use */
1138 IB_QP_CREATE_RESERVED_START = 1 << 26,
1139 IB_QP_CREATE_RESERVED_END = 1 << 31,
1140 };
1141
1142 /*
1143 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1144 * callback to destroy the passed in QP.
1145 */
1146
1147 struct ib_qp_init_attr {
1148 /* Consumer's event_handler callback must not block */
1149 void (*event_handler)(struct ib_event *, void *);
1150
1151 void *qp_context;
1152 struct ib_cq *send_cq;
1153 struct ib_cq *recv_cq;
1154 struct ib_srq *srq;
1155 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1156 struct ib_qp_cap cap;
1157 enum ib_sig_type sq_sig_type;
1158 enum ib_qp_type qp_type;
1159 u32 create_flags;
1160
1161 /*
1162 * Only needed for special QP types, or when using the RW API.
1163 */
1164 u8 port_num;
1165 struct ib_rwq_ind_table *rwq_ind_tbl;
1166 u32 source_qpn;
1167 };
1168
1169 struct ib_qp_open_attr {
1170 void (*event_handler)(struct ib_event *, void *);
1171 void *qp_context;
1172 u32 qp_num;
1173 enum ib_qp_type qp_type;
1174 };
1175
1176 enum ib_rnr_timeout {
1177 IB_RNR_TIMER_655_36 = 0,
1178 IB_RNR_TIMER_000_01 = 1,
1179 IB_RNR_TIMER_000_02 = 2,
1180 IB_RNR_TIMER_000_03 = 3,
1181 IB_RNR_TIMER_000_04 = 4,
1182 IB_RNR_TIMER_000_06 = 5,
1183 IB_RNR_TIMER_000_08 = 6,
1184 IB_RNR_TIMER_000_12 = 7,
1185 IB_RNR_TIMER_000_16 = 8,
1186 IB_RNR_TIMER_000_24 = 9,
1187 IB_RNR_TIMER_000_32 = 10,
1188 IB_RNR_TIMER_000_48 = 11,
1189 IB_RNR_TIMER_000_64 = 12,
1190 IB_RNR_TIMER_000_96 = 13,
1191 IB_RNR_TIMER_001_28 = 14,
1192 IB_RNR_TIMER_001_92 = 15,
1193 IB_RNR_TIMER_002_56 = 16,
1194 IB_RNR_TIMER_003_84 = 17,
1195 IB_RNR_TIMER_005_12 = 18,
1196 IB_RNR_TIMER_007_68 = 19,
1197 IB_RNR_TIMER_010_24 = 20,
1198 IB_RNR_TIMER_015_36 = 21,
1199 IB_RNR_TIMER_020_48 = 22,
1200 IB_RNR_TIMER_030_72 = 23,
1201 IB_RNR_TIMER_040_96 = 24,
1202 IB_RNR_TIMER_061_44 = 25,
1203 IB_RNR_TIMER_081_92 = 26,
1204 IB_RNR_TIMER_122_88 = 27,
1205 IB_RNR_TIMER_163_84 = 28,
1206 IB_RNR_TIMER_245_76 = 29,
1207 IB_RNR_TIMER_327_68 = 30,
1208 IB_RNR_TIMER_491_52 = 31
1209 };
1210
1211 enum ib_qp_attr_mask {
1212 IB_QP_STATE = 1,
1213 IB_QP_CUR_STATE = (1<<1),
1214 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1215 IB_QP_ACCESS_FLAGS = (1<<3),
1216 IB_QP_PKEY_INDEX = (1<<4),
1217 IB_QP_PORT = (1<<5),
1218 IB_QP_QKEY = (1<<6),
1219 IB_QP_AV = (1<<7),
1220 IB_QP_PATH_MTU = (1<<8),
1221 IB_QP_TIMEOUT = (1<<9),
1222 IB_QP_RETRY_CNT = (1<<10),
1223 IB_QP_RNR_RETRY = (1<<11),
1224 IB_QP_RQ_PSN = (1<<12),
1225 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1226 IB_QP_ALT_PATH = (1<<14),
1227 IB_QP_MIN_RNR_TIMER = (1<<15),
1228 IB_QP_SQ_PSN = (1<<16),
1229 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1230 IB_QP_PATH_MIG_STATE = (1<<18),
1231 IB_QP_CAP = (1<<19),
1232 IB_QP_DEST_QPN = (1<<20),
1233 IB_QP_RESERVED1 = (1<<21),
1234 IB_QP_RESERVED2 = (1<<22),
1235 IB_QP_RESERVED3 = (1<<23),
1236 IB_QP_RESERVED4 = (1<<24),
1237 IB_QP_RATE_LIMIT = (1<<25),
1238
1239 IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1240 };
1241
1242 enum ib_qp_state {
1243 IB_QPS_RESET,
1244 IB_QPS_INIT,
1245 IB_QPS_RTR,
1246 IB_QPS_RTS,
1247 IB_QPS_SQD,
1248 IB_QPS_SQE,
1249 IB_QPS_ERR
1250 };
1251
1252 enum ib_mig_state {
1253 IB_MIG_MIGRATED,
1254 IB_MIG_REARM,
1255 IB_MIG_ARMED
1256 };
1257
1258 enum ib_mw_type {
1259 IB_MW_TYPE_1 = 1,
1260 IB_MW_TYPE_2 = 2
1261 };
1262
1263 struct ib_qp_attr {
1264 enum ib_qp_state qp_state;
1265 enum ib_qp_state cur_qp_state;
1266 enum ib_mtu path_mtu;
1267 enum ib_mig_state path_mig_state;
1268 u32 qkey;
1269 u32 rq_psn;
1270 u32 sq_psn;
1271 u32 dest_qp_num;
1272 int qp_access_flags;
1273 struct ib_qp_cap cap;
1274 struct rdma_ah_attr ah_attr;
1275 struct rdma_ah_attr alt_ah_attr;
1276 u16 pkey_index;
1277 u16 alt_pkey_index;
1278 u8 en_sqd_async_notify;
1279 u8 sq_draining;
1280 u8 max_rd_atomic;
1281 u8 max_dest_rd_atomic;
1282 u8 min_rnr_timer;
1283 u8 port_num;
1284 u8 timeout;
1285 u8 retry_cnt;
1286 u8 rnr_retry;
1287 u8 alt_port_num;
1288 u8 alt_timeout;
1289 u32 rate_limit;
1290 struct net_device *xmit_slave;
1291 };
1292
1293 enum ib_wr_opcode {
1294 /* These are shared with userspace */
1295 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1296 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1297 IB_WR_SEND = IB_UVERBS_WR_SEND,
1298 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1299 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1300 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1301 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1302 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1303 IB_WR_LSO = IB_UVERBS_WR_TSO,
1304 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1305 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1306 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1307 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1308 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1309 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1310 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1311
1312 /* These are kernel only and can not be issued by userspace */
1313 IB_WR_REG_MR = 0x20,
1314 IB_WR_REG_MR_INTEGRITY,
1315
1316 /* reserve values for low level drivers' internal use.
1317 * These values will not be used at all in the ib core layer.
1318 */
1319 IB_WR_RESERVED1 = 0xf0,
1320 IB_WR_RESERVED2,
1321 IB_WR_RESERVED3,
1322 IB_WR_RESERVED4,
1323 IB_WR_RESERVED5,
1324 IB_WR_RESERVED6,
1325 IB_WR_RESERVED7,
1326 IB_WR_RESERVED8,
1327 IB_WR_RESERVED9,
1328 IB_WR_RESERVED10,
1329 };
1330
1331 enum ib_send_flags {
1332 IB_SEND_FENCE = 1,
1333 IB_SEND_SIGNALED = (1<<1),
1334 IB_SEND_SOLICITED = (1<<2),
1335 IB_SEND_INLINE = (1<<3),
1336 IB_SEND_IP_CSUM = (1<<4),
1337
1338 /* reserve bits 26-31 for low level drivers' internal use */
1339 IB_SEND_RESERVED_START = (1 << 26),
1340 IB_SEND_RESERVED_END = (1 << 31),
1341 };
1342
1343 struct ib_sge {
1344 u64 addr;
1345 u32 length;
1346 u32 lkey;
1347 };
1348
1349 struct ib_cqe {
1350 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1351 };
1352
1353 struct ib_send_wr {
1354 struct ib_send_wr *next;
1355 union {
1356 u64 wr_id;
1357 struct ib_cqe *wr_cqe;
1358 };
1359 struct ib_sge *sg_list;
1360 int num_sge;
1361 enum ib_wr_opcode opcode;
1362 int send_flags;
1363 union {
1364 __be32 imm_data;
1365 u32 invalidate_rkey;
1366 } ex;
1367 };
1368
1369 struct ib_rdma_wr {
1370 struct ib_send_wr wr;
1371 u64 remote_addr;
1372 u32 rkey;
1373 };
1374
1375 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1376 {
1377 return container_of(wr, struct ib_rdma_wr, wr);
1378 }
1379
1380 struct ib_atomic_wr {
1381 struct ib_send_wr wr;
1382 u64 remote_addr;
1383 u64 compare_add;
1384 u64 swap;
1385 u64 compare_add_mask;
1386 u64 swap_mask;
1387 u32 rkey;
1388 };
1389
1390 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1391 {
1392 return container_of(wr, struct ib_atomic_wr, wr);
1393 }
1394
1395 struct ib_ud_wr {
1396 struct ib_send_wr wr;
1397 struct ib_ah *ah;
1398 void *header;
1399 int hlen;
1400 int mss;
1401 u32 remote_qpn;
1402 u32 remote_qkey;
1403 u16 pkey_index; /* valid for GSI only */
1404 u8 port_num; /* valid for DR SMPs on switch only */
1405 };
1406
1407 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1408 {
1409 return container_of(wr, struct ib_ud_wr, wr);
1410 }
1411
1412 struct ib_reg_wr {
1413 struct ib_send_wr wr;
1414 struct ib_mr *mr;
1415 u32 key;
1416 int access;
1417 };
1418
1419 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1420 {
1421 return container_of(wr, struct ib_reg_wr, wr);
1422 }
1423
1424 struct ib_recv_wr {
1425 struct ib_recv_wr *next;
1426 union {
1427 u64 wr_id;
1428 struct ib_cqe *wr_cqe;
1429 };
1430 struct ib_sge *sg_list;
1431 int num_sge;
1432 };
1433
1434 enum ib_access_flags {
1435 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1436 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1437 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1438 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1439 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1440 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1441 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1442 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1443 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1444
1445 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1446 IB_ACCESS_SUPPORTED =
1447 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1448 };
1449
1450 /*
1451 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1452 * are hidden here instead of a uapi header!
1453 */
1454 enum ib_mr_rereg_flags {
1455 IB_MR_REREG_TRANS = 1,
1456 IB_MR_REREG_PD = (1<<1),
1457 IB_MR_REREG_ACCESS = (1<<2),
1458 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1459 };
1460
1461 struct ib_umem;
1462
1463 enum rdma_remove_reason {
1464 /*
1465 * Userspace requested uobject deletion or initial try
1466 * to remove uobject via cleanup. Call could fail
1467 */
1468 RDMA_REMOVE_DESTROY,
1469 /* Context deletion. This call should delete the actual object itself */
1470 RDMA_REMOVE_CLOSE,
1471 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1472 RDMA_REMOVE_DRIVER_REMOVE,
1473 /* uobj is being cleaned-up before being committed */
1474 RDMA_REMOVE_ABORT,
1475 /* The driver failed to destroy the uobject and is being disconnected */
1476 RDMA_REMOVE_DRIVER_FAILURE,
1477 };
1478
1479 struct ib_rdmacg_object {
1480 #ifdef CONFIG_CGROUP_RDMA
1481 struct rdma_cgroup *cg; /* owner rdma cgroup */
1482 #endif
1483 };
1484
1485 struct ib_ucontext {
1486 struct ib_device *device;
1487 struct ib_uverbs_file *ufile;
1488
1489 struct ib_rdmacg_object cg_obj;
1490 /*
1491 * Implementation details of the RDMA core, don't use in drivers:
1492 */
1493 struct rdma_restrack_entry res;
1494 struct xarray mmap_xa;
1495 };
1496
1497 struct ib_uobject {
1498 u64 user_handle; /* handle given to us by userspace */
1499 /* ufile & ucontext owning this object */
1500 struct ib_uverbs_file *ufile;
1501 /* FIXME, save memory: ufile->context == context */
1502 struct ib_ucontext *context; /* associated user context */
1503 void *object; /* containing object */
1504 struct list_head list; /* link to context's list */
1505 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1506 int id; /* index into kernel idr */
1507 struct kref ref;
1508 atomic_t usecnt; /* protects exclusive access */
1509 struct rcu_head rcu; /* kfree_rcu() overhead */
1510
1511 const struct uverbs_api_object *uapi_object;
1512 };
1513
1514 struct ib_udata {
1515 const void __user *inbuf;
1516 void __user *outbuf;
1517 size_t inlen;
1518 size_t outlen;
1519 };
1520
1521 struct ib_pd {
1522 u32 local_dma_lkey;
1523 u32 flags;
1524 struct ib_device *device;
1525 struct ib_uobject *uobject;
1526 atomic_t usecnt; /* count all resources */
1527
1528 u32 unsafe_global_rkey;
1529
1530 /*
1531 * Implementation details of the RDMA core, don't use in drivers:
1532 */
1533 struct ib_mr *__internal_mr;
1534 struct rdma_restrack_entry res;
1535 };
1536
1537 struct ib_xrcd {
1538 struct ib_device *device;
1539 atomic_t usecnt; /* count all exposed resources */
1540 struct inode *inode;
1541 struct rw_semaphore tgt_qps_rwsem;
1542 struct xarray tgt_qps;
1543 };
1544
1545 struct ib_ah {
1546 struct ib_device *device;
1547 struct ib_pd *pd;
1548 struct ib_uobject *uobject;
1549 const struct ib_gid_attr *sgid_attr;
1550 enum rdma_ah_attr_type type;
1551 };
1552
1553 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1554
1555 enum ib_poll_context {
1556 IB_POLL_SOFTIRQ, /* poll from softirq context */
1557 IB_POLL_WORKQUEUE, /* poll from workqueue */
1558 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1559 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1560
1561 IB_POLL_DIRECT, /* caller context, no hw completions */
1562 };
1563
1564 struct ib_cq {
1565 struct ib_device *device;
1566 struct ib_ucq_object *uobject;
1567 ib_comp_handler comp_handler;
1568 void (*event_handler)(struct ib_event *, void *);
1569 void *cq_context;
1570 int cqe;
1571 unsigned int cqe_used;
1572 atomic_t usecnt; /* count number of work queues */
1573 enum ib_poll_context poll_ctx;
1574 struct ib_wc *wc;
1575 struct list_head pool_entry;
1576 union {
1577 struct irq_poll iop;
1578 struct work_struct work;
1579 };
1580 struct workqueue_struct *comp_wq;
1581 struct dim *dim;
1582
1583 /* updated only by trace points */
1584 ktime_t timestamp;
1585 u8 interrupt:1;
1586 u8 shared:1;
1587 unsigned int comp_vector;
1588
1589 /*
1590 * Implementation details of the RDMA core, don't use in drivers:
1591 */
1592 struct rdma_restrack_entry res;
1593 };
1594
1595 struct ib_srq {
1596 struct ib_device *device;
1597 struct ib_pd *pd;
1598 struct ib_usrq_object *uobject;
1599 void (*event_handler)(struct ib_event *, void *);
1600 void *srq_context;
1601 enum ib_srq_type srq_type;
1602 atomic_t usecnt;
1603
1604 struct {
1605 struct ib_cq *cq;
1606 union {
1607 struct {
1608 struct ib_xrcd *xrcd;
1609 u32 srq_num;
1610 } xrc;
1611 };
1612 } ext;
1613 };
1614
1615 enum ib_raw_packet_caps {
1616 /* Strip cvlan from incoming packet and report it in the matching work
1617 * completion is supported.
1618 */
1619 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1620 /* Scatter FCS field of an incoming packet to host memory is supported.
1621 */
1622 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1623 /* Checksum offloads are supported (for both send and receive). */
1624 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1625 /* When a packet is received for an RQ with no receive WQEs, the
1626 * packet processing is delayed.
1627 */
1628 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1629 };
1630
1631 enum ib_wq_type {
1632 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1633 };
1634
1635 enum ib_wq_state {
1636 IB_WQS_RESET,
1637 IB_WQS_RDY,
1638 IB_WQS_ERR
1639 };
1640
1641 struct ib_wq {
1642 struct ib_device *device;
1643 struct ib_uwq_object *uobject;
1644 void *wq_context;
1645 void (*event_handler)(struct ib_event *, void *);
1646 struct ib_pd *pd;
1647 struct ib_cq *cq;
1648 u32 wq_num;
1649 enum ib_wq_state state;
1650 enum ib_wq_type wq_type;
1651 atomic_t usecnt;
1652 };
1653
1654 enum ib_wq_flags {
1655 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1656 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1657 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1658 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1659 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1660 };
1661
1662 struct ib_wq_init_attr {
1663 void *wq_context;
1664 enum ib_wq_type wq_type;
1665 u32 max_wr;
1666 u32 max_sge;
1667 struct ib_cq *cq;
1668 void (*event_handler)(struct ib_event *, void *);
1669 u32 create_flags; /* Use enum ib_wq_flags */
1670 };
1671
1672 enum ib_wq_attr_mask {
1673 IB_WQ_STATE = 1 << 0,
1674 IB_WQ_CUR_STATE = 1 << 1,
1675 IB_WQ_FLAGS = 1 << 2,
1676 };
1677
1678 struct ib_wq_attr {
1679 enum ib_wq_state wq_state;
1680 enum ib_wq_state curr_wq_state;
1681 u32 flags; /* Use enum ib_wq_flags */
1682 u32 flags_mask; /* Use enum ib_wq_flags */
1683 };
1684
1685 struct ib_rwq_ind_table {
1686 struct ib_device *device;
1687 struct ib_uobject *uobject;
1688 atomic_t usecnt;
1689 u32 ind_tbl_num;
1690 u32 log_ind_tbl_size;
1691 struct ib_wq **ind_tbl;
1692 };
1693
1694 struct ib_rwq_ind_table_init_attr {
1695 u32 log_ind_tbl_size;
1696 /* Each entry is a pointer to Receive Work Queue */
1697 struct ib_wq **ind_tbl;
1698 };
1699
1700 enum port_pkey_state {
1701 IB_PORT_PKEY_NOT_VALID = 0,
1702 IB_PORT_PKEY_VALID = 1,
1703 IB_PORT_PKEY_LISTED = 2,
1704 };
1705
1706 struct ib_qp_security;
1707
1708 struct ib_port_pkey {
1709 enum port_pkey_state state;
1710 u16 pkey_index;
1711 u8 port_num;
1712 struct list_head qp_list;
1713 struct list_head to_error_list;
1714 struct ib_qp_security *sec;
1715 };
1716
1717 struct ib_ports_pkeys {
1718 struct ib_port_pkey main;
1719 struct ib_port_pkey alt;
1720 };
1721
1722 struct ib_qp_security {
1723 struct ib_qp *qp;
1724 struct ib_device *dev;
1725 /* Hold this mutex when changing port and pkey settings. */
1726 struct mutex mutex;
1727 struct ib_ports_pkeys *ports_pkeys;
1728 /* A list of all open shared QP handles. Required to enforce security
1729 * properly for all users of a shared QP.
1730 */
1731 struct list_head shared_qp_list;
1732 void *security;
1733 bool destroying;
1734 atomic_t error_list_count;
1735 struct completion error_complete;
1736 int error_comps_pending;
1737 };
1738
1739 /*
1740 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1741 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1742 */
1743 struct ib_qp {
1744 struct ib_device *device;
1745 struct ib_pd *pd;
1746 struct ib_cq *send_cq;
1747 struct ib_cq *recv_cq;
1748 spinlock_t mr_lock;
1749 int mrs_used;
1750 struct list_head rdma_mrs;
1751 struct list_head sig_mrs;
1752 struct ib_srq *srq;
1753 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1754 struct list_head xrcd_list;
1755
1756 /* count times opened, mcast attaches, flow attaches */
1757 atomic_t usecnt;
1758 struct list_head open_list;
1759 struct ib_qp *real_qp;
1760 struct ib_uqp_object *uobject;
1761 void (*event_handler)(struct ib_event *, void *);
1762 void *qp_context;
1763 /* sgid_attrs associated with the AV's */
1764 const struct ib_gid_attr *av_sgid_attr;
1765 const struct ib_gid_attr *alt_path_sgid_attr;
1766 u32 qp_num;
1767 u32 max_write_sge;
1768 u32 max_read_sge;
1769 enum ib_qp_type qp_type;
1770 struct ib_rwq_ind_table *rwq_ind_tbl;
1771 struct ib_qp_security *qp_sec;
1772 u8 port;
1773
1774 bool integrity_en;
1775 /*
1776 * Implementation details of the RDMA core, don't use in drivers:
1777 */
1778 struct rdma_restrack_entry res;
1779
1780 /* The counter the qp is bind to */
1781 struct rdma_counter *counter;
1782 };
1783
1784 struct ib_dm {
1785 struct ib_device *device;
1786 u32 length;
1787 u32 flags;
1788 struct ib_uobject *uobject;
1789 atomic_t usecnt;
1790 };
1791
1792 struct ib_mr {
1793 struct ib_device *device;
1794 struct ib_pd *pd;
1795 u32 lkey;
1796 u32 rkey;
1797 u64 iova;
1798 u64 length;
1799 unsigned int page_size;
1800 enum ib_mr_type type;
1801 bool need_inval;
1802 union {
1803 struct ib_uobject *uobject; /* user */
1804 struct list_head qp_entry; /* FR */
1805 };
1806
1807 struct ib_dm *dm;
1808 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1809 /*
1810 * Implementation details of the RDMA core, don't use in drivers:
1811 */
1812 struct rdma_restrack_entry res;
1813 };
1814
1815 struct ib_mw {
1816 struct ib_device *device;
1817 struct ib_pd *pd;
1818 struct ib_uobject *uobject;
1819 u32 rkey;
1820 enum ib_mw_type type;
1821 };
1822
1823 /* Supported steering options */
1824 enum ib_flow_attr_type {
1825 /* steering according to rule specifications */
1826 IB_FLOW_ATTR_NORMAL = 0x0,
1827 /* default unicast and multicast rule -
1828 * receive all Eth traffic which isn't steered to any QP
1829 */
1830 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1831 /* default multicast rule -
1832 * receive all Eth multicast traffic which isn't steered to any QP
1833 */
1834 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1835 /* sniffer rule - receive all port traffic */
1836 IB_FLOW_ATTR_SNIFFER = 0x3
1837 };
1838
1839 /* Supported steering header types */
1840 enum ib_flow_spec_type {
1841 /* L2 headers*/
1842 IB_FLOW_SPEC_ETH = 0x20,
1843 IB_FLOW_SPEC_IB = 0x22,
1844 /* L3 header*/
1845 IB_FLOW_SPEC_IPV4 = 0x30,
1846 IB_FLOW_SPEC_IPV6 = 0x31,
1847 IB_FLOW_SPEC_ESP = 0x34,
1848 /* L4 headers*/
1849 IB_FLOW_SPEC_TCP = 0x40,
1850 IB_FLOW_SPEC_UDP = 0x41,
1851 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1852 IB_FLOW_SPEC_GRE = 0x51,
1853 IB_FLOW_SPEC_MPLS = 0x60,
1854 IB_FLOW_SPEC_INNER = 0x100,
1855 /* Actions */
1856 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1857 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1858 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1859 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1860 };
1861 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1862 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1863
1864 enum ib_flow_flags {
1865 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1866 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1867 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1868 };
1869
1870 struct ib_flow_eth_filter {
1871 u8 dst_mac[6];
1872 u8 src_mac[6];
1873 __be16 ether_type;
1874 __be16 vlan_tag;
1875 /* Must be last */
1876 u8 real_sz[];
1877 };
1878
1879 struct ib_flow_spec_eth {
1880 u32 type;
1881 u16 size;
1882 struct ib_flow_eth_filter val;
1883 struct ib_flow_eth_filter mask;
1884 };
1885
1886 struct ib_flow_ib_filter {
1887 __be16 dlid;
1888 __u8 sl;
1889 /* Must be last */
1890 u8 real_sz[];
1891 };
1892
1893 struct ib_flow_spec_ib {
1894 u32 type;
1895 u16 size;
1896 struct ib_flow_ib_filter val;
1897 struct ib_flow_ib_filter mask;
1898 };
1899
1900 /* IPv4 header flags */
1901 enum ib_ipv4_flags {
1902 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1903 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1904 last have this flag set */
1905 };
1906
1907 struct ib_flow_ipv4_filter {
1908 __be32 src_ip;
1909 __be32 dst_ip;
1910 u8 proto;
1911 u8 tos;
1912 u8 ttl;
1913 u8 flags;
1914 /* Must be last */
1915 u8 real_sz[];
1916 };
1917
1918 struct ib_flow_spec_ipv4 {
1919 u32 type;
1920 u16 size;
1921 struct ib_flow_ipv4_filter val;
1922 struct ib_flow_ipv4_filter mask;
1923 };
1924
1925 struct ib_flow_ipv6_filter {
1926 u8 src_ip[16];
1927 u8 dst_ip[16];
1928 __be32 flow_label;
1929 u8 next_hdr;
1930 u8 traffic_class;
1931 u8 hop_limit;
1932 /* Must be last */
1933 u8 real_sz[];
1934 };
1935
1936 struct ib_flow_spec_ipv6 {
1937 u32 type;
1938 u16 size;
1939 struct ib_flow_ipv6_filter val;
1940 struct ib_flow_ipv6_filter mask;
1941 };
1942
1943 struct ib_flow_tcp_udp_filter {
1944 __be16 dst_port;
1945 __be16 src_port;
1946 /* Must be last */
1947 u8 real_sz[];
1948 };
1949
1950 struct ib_flow_spec_tcp_udp {
1951 u32 type;
1952 u16 size;
1953 struct ib_flow_tcp_udp_filter val;
1954 struct ib_flow_tcp_udp_filter mask;
1955 };
1956
1957 struct ib_flow_tunnel_filter {
1958 __be32 tunnel_id;
1959 u8 real_sz[];
1960 };
1961
1962 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1963 * the tunnel_id from val has the vni value
1964 */
1965 struct ib_flow_spec_tunnel {
1966 u32 type;
1967 u16 size;
1968 struct ib_flow_tunnel_filter val;
1969 struct ib_flow_tunnel_filter mask;
1970 };
1971
1972 struct ib_flow_esp_filter {
1973 __be32 spi;
1974 __be32 seq;
1975 /* Must be last */
1976 u8 real_sz[];
1977 };
1978
1979 struct ib_flow_spec_esp {
1980 u32 type;
1981 u16 size;
1982 struct ib_flow_esp_filter val;
1983 struct ib_flow_esp_filter mask;
1984 };
1985
1986 struct ib_flow_gre_filter {
1987 __be16 c_ks_res0_ver;
1988 __be16 protocol;
1989 __be32 key;
1990 /* Must be last */
1991 u8 real_sz[];
1992 };
1993
1994 struct ib_flow_spec_gre {
1995 u32 type;
1996 u16 size;
1997 struct ib_flow_gre_filter val;
1998 struct ib_flow_gre_filter mask;
1999 };
2000
2001 struct ib_flow_mpls_filter {
2002 __be32 tag;
2003 /* Must be last */
2004 u8 real_sz[];
2005 };
2006
2007 struct ib_flow_spec_mpls {
2008 u32 type;
2009 u16 size;
2010 struct ib_flow_mpls_filter val;
2011 struct ib_flow_mpls_filter mask;
2012 };
2013
2014 struct ib_flow_spec_action_tag {
2015 enum ib_flow_spec_type type;
2016 u16 size;
2017 u32 tag_id;
2018 };
2019
2020 struct ib_flow_spec_action_drop {
2021 enum ib_flow_spec_type type;
2022 u16 size;
2023 };
2024
2025 struct ib_flow_spec_action_handle {
2026 enum ib_flow_spec_type type;
2027 u16 size;
2028 struct ib_flow_action *act;
2029 };
2030
2031 enum ib_counters_description {
2032 IB_COUNTER_PACKETS,
2033 IB_COUNTER_BYTES,
2034 };
2035
2036 struct ib_flow_spec_action_count {
2037 enum ib_flow_spec_type type;
2038 u16 size;
2039 struct ib_counters *counters;
2040 };
2041
2042 union ib_flow_spec {
2043 struct {
2044 u32 type;
2045 u16 size;
2046 };
2047 struct ib_flow_spec_eth eth;
2048 struct ib_flow_spec_ib ib;
2049 struct ib_flow_spec_ipv4 ipv4;
2050 struct ib_flow_spec_tcp_udp tcp_udp;
2051 struct ib_flow_spec_ipv6 ipv6;
2052 struct ib_flow_spec_tunnel tunnel;
2053 struct ib_flow_spec_esp esp;
2054 struct ib_flow_spec_gre gre;
2055 struct ib_flow_spec_mpls mpls;
2056 struct ib_flow_spec_action_tag flow_tag;
2057 struct ib_flow_spec_action_drop drop;
2058 struct ib_flow_spec_action_handle action;
2059 struct ib_flow_spec_action_count flow_count;
2060 };
2061
2062 struct ib_flow_attr {
2063 enum ib_flow_attr_type type;
2064 u16 size;
2065 u16 priority;
2066 u32 flags;
2067 u8 num_of_specs;
2068 u8 port;
2069 union ib_flow_spec flows[];
2070 };
2071
2072 struct ib_flow {
2073 struct ib_qp *qp;
2074 struct ib_device *device;
2075 struct ib_uobject *uobject;
2076 };
2077
2078 enum ib_flow_action_type {
2079 IB_FLOW_ACTION_UNSPECIFIED,
2080 IB_FLOW_ACTION_ESP = 1,
2081 };
2082
2083 struct ib_flow_action_attrs_esp_keymats {
2084 enum ib_uverbs_flow_action_esp_keymat protocol;
2085 union {
2086 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2087 } keymat;
2088 };
2089
2090 struct ib_flow_action_attrs_esp_replays {
2091 enum ib_uverbs_flow_action_esp_replay protocol;
2092 union {
2093 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2094 } replay;
2095 };
2096
2097 enum ib_flow_action_attrs_esp_flags {
2098 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2099 * This is done in order to share the same flags between user-space and
2100 * kernel and spare an unnecessary translation.
2101 */
2102
2103 /* Kernel flags */
2104 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2105 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2106 };
2107
2108 struct ib_flow_spec_list {
2109 struct ib_flow_spec_list *next;
2110 union ib_flow_spec spec;
2111 };
2112
2113 struct ib_flow_action_attrs_esp {
2114 struct ib_flow_action_attrs_esp_keymats *keymat;
2115 struct ib_flow_action_attrs_esp_replays *replay;
2116 struct ib_flow_spec_list *encap;
2117 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2118 * Value of 0 is a valid value.
2119 */
2120 u32 esn;
2121 u32 spi;
2122 u32 seq;
2123 u32 tfc_pad;
2124 /* Use enum ib_flow_action_attrs_esp_flags */
2125 u64 flags;
2126 u64 hard_limit_pkts;
2127 };
2128
2129 struct ib_flow_action {
2130 struct ib_device *device;
2131 struct ib_uobject *uobject;
2132 enum ib_flow_action_type type;
2133 atomic_t usecnt;
2134 };
2135
2136 struct ib_mad;
2137 struct ib_grh;
2138
2139 enum ib_process_mad_flags {
2140 IB_MAD_IGNORE_MKEY = 1,
2141 IB_MAD_IGNORE_BKEY = 2,
2142 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2143 };
2144
2145 enum ib_mad_result {
2146 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2147 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2148 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2149 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2150 };
2151
2152 struct ib_port_cache {
2153 u64 subnet_prefix;
2154 struct ib_pkey_cache *pkey;
2155 struct ib_gid_table *gid;
2156 u8 lmc;
2157 enum ib_port_state port_state;
2158 };
2159
2160 struct ib_port_immutable {
2161 int pkey_tbl_len;
2162 int gid_tbl_len;
2163 u32 core_cap_flags;
2164 u32 max_mad_size;
2165 };
2166
2167 struct ib_port_data {
2168 struct ib_device *ib_dev;
2169
2170 struct ib_port_immutable immutable;
2171
2172 spinlock_t pkey_list_lock;
2173 struct list_head pkey_list;
2174
2175 struct ib_port_cache cache;
2176
2177 spinlock_t netdev_lock;
2178 struct net_device __rcu *netdev;
2179 struct hlist_node ndev_hash_link;
2180 struct rdma_port_counter port_counter;
2181 struct rdma_hw_stats *hw_stats;
2182 };
2183
2184 /* rdma netdev type - specifies protocol type */
2185 enum rdma_netdev_t {
2186 RDMA_NETDEV_OPA_VNIC,
2187 RDMA_NETDEV_IPOIB,
2188 };
2189
2190 /**
2191 * struct rdma_netdev - rdma netdev
2192 * For cases where netstack interfacing is required.
2193 */
2194 struct rdma_netdev {
2195 void *clnt_priv;
2196 struct ib_device *hca;
2197 u8 port_num;
2198 int mtu;
2199
2200 /*
2201 * cleanup function must be specified.
2202 * FIXME: This is only used for OPA_VNIC and that usage should be
2203 * removed too.
2204 */
2205 void (*free_rdma_netdev)(struct net_device *netdev);
2206
2207 /* control functions */
2208 void (*set_id)(struct net_device *netdev, int id);
2209 /* send packet */
2210 int (*send)(struct net_device *dev, struct sk_buff *skb,
2211 struct ib_ah *address, u32 dqpn);
2212 /* multicast */
2213 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2214 union ib_gid *gid, u16 mlid,
2215 int set_qkey, u32 qkey);
2216 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2217 union ib_gid *gid, u16 mlid);
2218 };
2219
2220 struct rdma_netdev_alloc_params {
2221 size_t sizeof_priv;
2222 unsigned int txqs;
2223 unsigned int rxqs;
2224 void *param;
2225
2226 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2227 struct net_device *netdev, void *param);
2228 };
2229
2230 struct ib_odp_counters {
2231 atomic64_t faults;
2232 atomic64_t invalidations;
2233 atomic64_t prefetch;
2234 };
2235
2236 struct ib_counters {
2237 struct ib_device *device;
2238 struct ib_uobject *uobject;
2239 /* num of objects attached */
2240 atomic_t usecnt;
2241 };
2242
2243 struct ib_counters_read_attr {
2244 u64 *counters_buff;
2245 u32 ncounters;
2246 u32 flags; /* use enum ib_read_counters_flags */
2247 };
2248
2249 struct uverbs_attr_bundle;
2250 struct iw_cm_id;
2251 struct iw_cm_conn_param;
2252
2253 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2254 .size_##ib_struct = \
2255 (sizeof(struct drv_struct) + \
2256 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2257 BUILD_BUG_ON_ZERO( \
2258 !__same_type(((struct drv_struct *)NULL)->member, \
2259 struct ib_struct)))
2260
2261 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2262 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2263
2264 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2265 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2266
2267 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2268
2269 struct rdma_user_mmap_entry {
2270 struct kref ref;
2271 struct ib_ucontext *ucontext;
2272 unsigned long start_pgoff;
2273 size_t npages;
2274 bool driver_removed;
2275 };
2276
2277 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2278 static inline u64
2279 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2280 {
2281 return (u64)entry->start_pgoff << PAGE_SHIFT;
2282 }
2283
2284 /**
2285 * struct ib_device_ops - InfiniBand device operations
2286 * This structure defines all the InfiniBand device operations, providers will
2287 * need to define the supported operations, otherwise they will be set to null.
2288 */
2289 struct ib_device_ops {
2290 struct module *owner;
2291 enum rdma_driver_id driver_id;
2292 u32 uverbs_abi_ver;
2293 unsigned int uverbs_no_driver_id_binding:1;
2294
2295 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2296 const struct ib_send_wr **bad_send_wr);
2297 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2298 const struct ib_recv_wr **bad_recv_wr);
2299 void (*drain_rq)(struct ib_qp *qp);
2300 void (*drain_sq)(struct ib_qp *qp);
2301 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2302 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2303 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2304 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2305 int (*post_srq_recv)(struct ib_srq *srq,
2306 const struct ib_recv_wr *recv_wr,
2307 const struct ib_recv_wr **bad_recv_wr);
2308 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2309 u8 port_num, const struct ib_wc *in_wc,
2310 const struct ib_grh *in_grh,
2311 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2312 size_t *out_mad_size, u16 *out_mad_pkey_index);
2313 int (*query_device)(struct ib_device *device,
2314 struct ib_device_attr *device_attr,
2315 struct ib_udata *udata);
2316 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2317 struct ib_device_modify *device_modify);
2318 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2319 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2320 int comp_vector);
2321 int (*query_port)(struct ib_device *device, u8 port_num,
2322 struct ib_port_attr *port_attr);
2323 int (*modify_port)(struct ib_device *device, u8 port_num,
2324 int port_modify_mask,
2325 struct ib_port_modify *port_modify);
2326 /**
2327 * The following mandatory functions are used only at device
2328 * registration. Keep functions such as these at the end of this
2329 * structure to avoid cache line misses when accessing struct ib_device
2330 * in fast paths.
2331 */
2332 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2333 struct ib_port_immutable *immutable);
2334 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2335 u8 port_num);
2336 /**
2337 * When calling get_netdev, the HW vendor's driver should return the
2338 * net device of device @device at port @port_num or NULL if such
2339 * a net device doesn't exist. The vendor driver should call dev_hold
2340 * on this net device. The HW vendor's device driver must guarantee
2341 * that this function returns NULL before the net device has finished
2342 * NETDEV_UNREGISTER state.
2343 */
2344 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2345 /**
2346 * rdma netdev operation
2347 *
2348 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2349 * must return -EOPNOTSUPP if it doesn't support the specified type.
2350 */
2351 struct net_device *(*alloc_rdma_netdev)(
2352 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2353 const char *name, unsigned char name_assign_type,
2354 void (*setup)(struct net_device *));
2355
2356 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2357 enum rdma_netdev_t type,
2358 struct rdma_netdev_alloc_params *params);
2359 /**
2360 * query_gid should be return GID value for @device, when @port_num
2361 * link layer is either IB or iWarp. It is no-op if @port_num port
2362 * is RoCE link layer.
2363 */
2364 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2365 union ib_gid *gid);
2366 /**
2367 * When calling add_gid, the HW vendor's driver should add the gid
2368 * of device of port at gid index available at @attr. Meta-info of
2369 * that gid (for example, the network device related to this gid) is
2370 * available at @attr. @context allows the HW vendor driver to store
2371 * extra information together with a GID entry. The HW vendor driver may
2372 * allocate memory to contain this information and store it in @context
2373 * when a new GID entry is written to. Params are consistent until the
2374 * next call of add_gid or delete_gid. The function should return 0 on
2375 * success or error otherwise. The function could be called
2376 * concurrently for different ports. This function is only called when
2377 * roce_gid_table is used.
2378 */
2379 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2380 /**
2381 * When calling del_gid, the HW vendor's driver should delete the
2382 * gid of device @device at gid index gid_index of port port_num
2383 * available in @attr.
2384 * Upon the deletion of a GID entry, the HW vendor must free any
2385 * allocated memory. The caller will clear @context afterwards.
2386 * This function is only called when roce_gid_table is used.
2387 */
2388 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2389 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2390 u16 *pkey);
2391 int (*alloc_ucontext)(struct ib_ucontext *context,
2392 struct ib_udata *udata);
2393 void (*dealloc_ucontext)(struct ib_ucontext *context);
2394 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2395 /**
2396 * This will be called once refcount of an entry in mmap_xa reaches
2397 * zero. The type of the memory that was mapped may differ between
2398 * entries and is opaque to the rdma_user_mmap interface.
2399 * Therefore needs to be implemented by the driver in mmap_free.
2400 */
2401 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2402 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2403 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2404 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2405 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2406 struct ib_udata *udata);
2407 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2408 struct ib_udata *udata);
2409 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2410 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2411 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2412 int (*create_srq)(struct ib_srq *srq,
2413 struct ib_srq_init_attr *srq_init_attr,
2414 struct ib_udata *udata);
2415 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2416 enum ib_srq_attr_mask srq_attr_mask,
2417 struct ib_udata *udata);
2418 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2419 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2420 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2421 struct ib_qp_init_attr *qp_init_attr,
2422 struct ib_udata *udata);
2423 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2424 int qp_attr_mask, struct ib_udata *udata);
2425 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2426 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2427 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2428 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2429 struct ib_udata *udata);
2430 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2431 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2432 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2433 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2434 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2435 u64 virt_addr, int mr_access_flags,
2436 struct ib_udata *udata);
2437 struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2438 u64 length, u64 virt_addr,
2439 int mr_access_flags, struct ib_pd *pd,
2440 struct ib_udata *udata);
2441 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2442 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2443 u32 max_num_sg);
2444 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2445 u32 max_num_data_sg,
2446 u32 max_num_meta_sg);
2447 int (*advise_mr)(struct ib_pd *pd,
2448 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2449 struct ib_sge *sg_list, u32 num_sge,
2450 struct uverbs_attr_bundle *attrs);
2451 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2452 unsigned int *sg_offset);
2453 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2454 struct ib_mr_status *mr_status);
2455 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2456 int (*dealloc_mw)(struct ib_mw *mw);
2457 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2458 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2459 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2460 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2461 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2462 struct ib_flow_attr *flow_attr,
2463 struct ib_udata *udata);
2464 int (*destroy_flow)(struct ib_flow *flow_id);
2465 struct ib_flow_action *(*create_flow_action_esp)(
2466 struct ib_device *device,
2467 const struct ib_flow_action_attrs_esp *attr,
2468 struct uverbs_attr_bundle *attrs);
2469 int (*destroy_flow_action)(struct ib_flow_action *action);
2470 int (*modify_flow_action_esp)(
2471 struct ib_flow_action *action,
2472 const struct ib_flow_action_attrs_esp *attr,
2473 struct uverbs_attr_bundle *attrs);
2474 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2475 int state);
2476 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2477 struct ifla_vf_info *ivf);
2478 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2479 struct ifla_vf_stats *stats);
2480 int (*get_vf_guid)(struct ib_device *device, int vf, u8 port,
2481 struct ifla_vf_guid *node_guid,
2482 struct ifla_vf_guid *port_guid);
2483 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2484 int type);
2485 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2486 struct ib_wq_init_attr *init_attr,
2487 struct ib_udata *udata);
2488 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2489 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2490 u32 wq_attr_mask, struct ib_udata *udata);
2491 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2492 struct ib_rwq_ind_table_init_attr *init_attr,
2493 struct ib_udata *udata);
2494 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2495 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2496 struct ib_ucontext *context,
2497 struct ib_dm_alloc_attr *attr,
2498 struct uverbs_attr_bundle *attrs);
2499 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2500 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2501 struct ib_dm_mr_attr *attr,
2502 struct uverbs_attr_bundle *attrs);
2503 int (*create_counters)(struct ib_counters *counters,
2504 struct uverbs_attr_bundle *attrs);
2505 int (*destroy_counters)(struct ib_counters *counters);
2506 int (*read_counters)(struct ib_counters *counters,
2507 struct ib_counters_read_attr *counters_read_attr,
2508 struct uverbs_attr_bundle *attrs);
2509 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2510 int data_sg_nents, unsigned int *data_sg_offset,
2511 struct scatterlist *meta_sg, int meta_sg_nents,
2512 unsigned int *meta_sg_offset);
2513
2514 /**
2515 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2516 * driver initialized data. The struct is kfree()'ed by the sysfs
2517 * core when the device is removed. A lifespan of -1 in the return
2518 * struct tells the core to set a default lifespan.
2519 */
2520 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2521 u8 port_num);
2522 /**
2523 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2524 * @index - The index in the value array we wish to have updated, or
2525 * num_counters if we want all stats updated
2526 * Return codes -
2527 * < 0 - Error, no counters updated
2528 * index - Updated the single counter pointed to by index
2529 * num_counters - Updated all counters (will reset the timestamp
2530 * and prevent further calls for lifespan milliseconds)
2531 * Drivers are allowed to update all counters in leiu of just the
2532 * one given in index at their option
2533 */
2534 int (*get_hw_stats)(struct ib_device *device,
2535 struct rdma_hw_stats *stats, u8 port, int index);
2536 /*
2537 * This function is called once for each port when a ib device is
2538 * registered.
2539 */
2540 int (*init_port)(struct ib_device *device, u8 port_num,
2541 struct kobject *port_sysfs);
2542 /**
2543 * Allows rdma drivers to add their own restrack attributes.
2544 */
2545 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2546 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2547 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2548 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2549 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2550 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2551 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2552
2553 /* Device lifecycle callbacks */
2554 /*
2555 * Called after the device becomes registered, before clients are
2556 * attached
2557 */
2558 int (*enable_driver)(struct ib_device *dev);
2559 /*
2560 * This is called as part of ib_dealloc_device().
2561 */
2562 void (*dealloc_driver)(struct ib_device *dev);
2563
2564 /* iWarp CM callbacks */
2565 void (*iw_add_ref)(struct ib_qp *qp);
2566 void (*iw_rem_ref)(struct ib_qp *qp);
2567 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2568 int (*iw_connect)(struct iw_cm_id *cm_id,
2569 struct iw_cm_conn_param *conn_param);
2570 int (*iw_accept)(struct iw_cm_id *cm_id,
2571 struct iw_cm_conn_param *conn_param);
2572 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2573 u8 pdata_len);
2574 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2575 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2576 /**
2577 * counter_bind_qp - Bind a QP to a counter.
2578 * @counter - The counter to be bound. If counter->id is zero then
2579 * the driver needs to allocate a new counter and set counter->id
2580 */
2581 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2582 /**
2583 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2584 * counter and bind it onto the default one
2585 */
2586 int (*counter_unbind_qp)(struct ib_qp *qp);
2587 /**
2588 * counter_dealloc -De-allocate the hw counter
2589 */
2590 int (*counter_dealloc)(struct rdma_counter *counter);
2591 /**
2592 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2593 * the driver initialized data.
2594 */
2595 struct rdma_hw_stats *(*counter_alloc_stats)(
2596 struct rdma_counter *counter);
2597 /**
2598 * counter_update_stats - Query the stats value of this counter
2599 */
2600 int (*counter_update_stats)(struct rdma_counter *counter);
2601
2602 /**
2603 * Allows rdma drivers to add their own restrack attributes
2604 * dumped via 'rdma stat' iproute2 command.
2605 */
2606 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2607
2608 /* query driver for its ucontext properties */
2609 int (*query_ucontext)(struct ib_ucontext *context,
2610 struct uverbs_attr_bundle *attrs);
2611
2612 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2613 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2614 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2615 DECLARE_RDMA_OBJ_SIZE(ib_mw);
2616 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2617 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2618 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2619 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2620 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2621 };
2622
2623 struct ib_core_device {
2624 /* device must be the first element in structure until,
2625 * union of ib_core_device and device exists in ib_device.
2626 */
2627 struct device dev;
2628 possible_net_t rdma_net;
2629 struct kobject *ports_kobj;
2630 struct list_head port_list;
2631 struct ib_device *owner; /* reach back to owner ib_device */
2632 };
2633
2634 struct rdma_restrack_root;
2635 struct ib_device {
2636 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2637 struct device *dma_device;
2638 struct ib_device_ops ops;
2639 char name[IB_DEVICE_NAME_MAX];
2640 struct rcu_head rcu_head;
2641
2642 struct list_head event_handler_list;
2643 /* Protects event_handler_list */
2644 struct rw_semaphore event_handler_rwsem;
2645
2646 /* Protects QP's event_handler calls and open_qp list */
2647 spinlock_t qp_open_list_lock;
2648
2649 struct rw_semaphore client_data_rwsem;
2650 struct xarray client_data;
2651 struct mutex unregistration_lock;
2652
2653 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2654 rwlock_t cache_lock;
2655 /**
2656 * port_data is indexed by port number
2657 */
2658 struct ib_port_data *port_data;
2659
2660 int num_comp_vectors;
2661
2662 union {
2663 struct device dev;
2664 struct ib_core_device coredev;
2665 };
2666
2667 /* First group for device attributes,
2668 * Second group for driver provided attributes (optional).
2669 * It is NULL terminated array.
2670 */
2671 const struct attribute_group *groups[3];
2672
2673 u64 uverbs_cmd_mask;
2674
2675 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2676 __be64 node_guid;
2677 u32 local_dma_lkey;
2678 u16 is_switch:1;
2679 /* Indicates kernel verbs support, should not be used in drivers */
2680 u16 kverbs_provider:1;
2681 /* CQ adaptive moderation (RDMA DIM) */
2682 u16 use_cq_dim:1;
2683 u8 node_type;
2684 u8 phys_port_cnt;
2685 struct ib_device_attr attrs;
2686 struct attribute_group *hw_stats_ag;
2687 struct rdma_hw_stats *hw_stats;
2688
2689 #ifdef CONFIG_CGROUP_RDMA
2690 struct rdmacg_device cg_device;
2691 #endif
2692
2693 u32 index;
2694
2695 spinlock_t cq_pools_lock;
2696 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2697
2698 struct rdma_restrack_root *res;
2699
2700 const struct uapi_definition *driver_def;
2701
2702 /*
2703 * Positive refcount indicates that the device is currently
2704 * registered and cannot be unregistered.
2705 */
2706 refcount_t refcount;
2707 struct completion unreg_completion;
2708 struct work_struct unregistration_work;
2709
2710 const struct rdma_link_ops *link_ops;
2711
2712 /* Protects compat_devs xarray modifications */
2713 struct mutex compat_devs_mutex;
2714 /* Maintains compat devices for each net namespace */
2715 struct xarray compat_devs;
2716
2717 /* Used by iWarp CM */
2718 char iw_ifname[IFNAMSIZ];
2719 u32 iw_driver_flags;
2720 u32 lag_flags;
2721 };
2722
2723 struct ib_client_nl_info;
2724 struct ib_client {
2725 const char *name;
2726 int (*add)(struct ib_device *ibdev);
2727 void (*remove)(struct ib_device *, void *client_data);
2728 void (*rename)(struct ib_device *dev, void *client_data);
2729 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2730 struct ib_client_nl_info *res);
2731 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2732
2733 /* Returns the net_dev belonging to this ib_client and matching the
2734 * given parameters.
2735 * @dev: An RDMA device that the net_dev use for communication.
2736 * @port: A physical port number on the RDMA device.
2737 * @pkey: P_Key that the net_dev uses if applicable.
2738 * @gid: A GID that the net_dev uses to communicate.
2739 * @addr: An IP address the net_dev is configured with.
2740 * @client_data: The device's client data set by ib_set_client_data().
2741 *
2742 * An ib_client that implements a net_dev on top of RDMA devices
2743 * (such as IP over IB) should implement this callback, allowing the
2744 * rdma_cm module to find the right net_dev for a given request.
2745 *
2746 * The caller is responsible for calling dev_put on the returned
2747 * netdev. */
2748 struct net_device *(*get_net_dev_by_params)(
2749 struct ib_device *dev,
2750 u8 port,
2751 u16 pkey,
2752 const union ib_gid *gid,
2753 const struct sockaddr *addr,
2754 void *client_data);
2755
2756 refcount_t uses;
2757 struct completion uses_zero;
2758 u32 client_id;
2759
2760 /* kverbs are not required by the client */
2761 u8 no_kverbs_req:1;
2762 };
2763
2764 /*
2765 * IB block DMA iterator
2766 *
2767 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2768 * to a HW supported page size.
2769 */
2770 struct ib_block_iter {
2771 /* internal states */
2772 struct scatterlist *__sg; /* sg holding the current aligned block */
2773 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2774 unsigned int __sg_nents; /* number of SG entries */
2775 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2776 unsigned int __pg_bit; /* alignment of current block */
2777 };
2778
2779 struct ib_device *_ib_alloc_device(size_t size);
2780 #define ib_alloc_device(drv_struct, member) \
2781 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2782 BUILD_BUG_ON_ZERO(offsetof( \
2783 struct drv_struct, member))), \
2784 struct drv_struct, member)
2785
2786 void ib_dealloc_device(struct ib_device *device);
2787
2788 void ib_get_device_fw_str(struct ib_device *device, char *str);
2789
2790 int ib_register_device(struct ib_device *device, const char *name,
2791 struct device *dma_device);
2792 void ib_unregister_device(struct ib_device *device);
2793 void ib_unregister_driver(enum rdma_driver_id driver_id);
2794 void ib_unregister_device_and_put(struct ib_device *device);
2795 void ib_unregister_device_queued(struct ib_device *ib_dev);
2796
2797 int ib_register_client (struct ib_client *client);
2798 void ib_unregister_client(struct ib_client *client);
2799
2800 void __rdma_block_iter_start(struct ib_block_iter *biter,
2801 struct scatterlist *sglist,
2802 unsigned int nents,
2803 unsigned long pgsz);
2804 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2805
2806 /**
2807 * rdma_block_iter_dma_address - get the aligned dma address of the current
2808 * block held by the block iterator.
2809 * @biter: block iterator holding the memory block
2810 */
2811 static inline dma_addr_t
2812 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2813 {
2814 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2815 }
2816
2817 /**
2818 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2819 * @sglist: sglist to iterate over
2820 * @biter: block iterator holding the memory block
2821 * @nents: maximum number of sg entries to iterate over
2822 * @pgsz: best HW supported page size to use
2823 *
2824 * Callers may use rdma_block_iter_dma_address() to get each
2825 * blocks aligned DMA address.
2826 */
2827 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2828 for (__rdma_block_iter_start(biter, sglist, nents, \
2829 pgsz); \
2830 __rdma_block_iter_next(biter);)
2831
2832 /**
2833 * ib_get_client_data - Get IB client context
2834 * @device:Device to get context for
2835 * @client:Client to get context for
2836 *
2837 * ib_get_client_data() returns the client context data set with
2838 * ib_set_client_data(). This can only be called while the client is
2839 * registered to the device, once the ib_client remove() callback returns this
2840 * cannot be called.
2841 */
2842 static inline void *ib_get_client_data(struct ib_device *device,
2843 struct ib_client *client)
2844 {
2845 return xa_load(&device->client_data, client->client_id);
2846 }
2847 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2848 void *data);
2849 void ib_set_device_ops(struct ib_device *device,
2850 const struct ib_device_ops *ops);
2851
2852 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2853 unsigned long pfn, unsigned long size, pgprot_t prot,
2854 struct rdma_user_mmap_entry *entry);
2855 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2856 struct rdma_user_mmap_entry *entry,
2857 size_t length);
2858 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2859 struct rdma_user_mmap_entry *entry,
2860 size_t length, u32 min_pgoff,
2861 u32 max_pgoff);
2862
2863 struct rdma_user_mmap_entry *
2864 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2865 unsigned long pgoff);
2866 struct rdma_user_mmap_entry *
2867 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2868 struct vm_area_struct *vma);
2869 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2870
2871 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2872
2873 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2874 {
2875 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2876 }
2877
2878 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2879 {
2880 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2881 }
2882
2883 static inline bool ib_is_buffer_cleared(const void __user *p,
2884 size_t len)
2885 {
2886 bool ret;
2887 u8 *buf;
2888
2889 if (len > USHRT_MAX)
2890 return false;
2891
2892 buf = memdup_user(p, len);
2893 if (IS_ERR(buf))
2894 return false;
2895
2896 ret = !memchr_inv(buf, 0, len);
2897 kfree(buf);
2898 return ret;
2899 }
2900
2901 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2902 size_t offset,
2903 size_t len)
2904 {
2905 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2906 }
2907
2908 /**
2909 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2910 * contains all required attributes and no attributes not allowed for
2911 * the given QP state transition.
2912 * @cur_state: Current QP state
2913 * @next_state: Next QP state
2914 * @type: QP type
2915 * @mask: Mask of supplied QP attributes
2916 *
2917 * This function is a helper function that a low-level driver's
2918 * modify_qp method can use to validate the consumer's input. It
2919 * checks that cur_state and next_state are valid QP states, that a
2920 * transition from cur_state to next_state is allowed by the IB spec,
2921 * and that the attribute mask supplied is allowed for the transition.
2922 */
2923 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2924 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2925
2926 void ib_register_event_handler(struct ib_event_handler *event_handler);
2927 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2928 void ib_dispatch_event(const struct ib_event *event);
2929
2930 int ib_query_port(struct ib_device *device,
2931 u8 port_num, struct ib_port_attr *port_attr);
2932
2933 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2934 u8 port_num);
2935
2936 /**
2937 * rdma_cap_ib_switch - Check if the device is IB switch
2938 * @device: Device to check
2939 *
2940 * Device driver is responsible for setting is_switch bit on
2941 * in ib_device structure at init time.
2942 *
2943 * Return: true if the device is IB switch.
2944 */
2945 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2946 {
2947 return device->is_switch;
2948 }
2949
2950 /**
2951 * rdma_start_port - Return the first valid port number for the device
2952 * specified
2953 *
2954 * @device: Device to be checked
2955 *
2956 * Return start port number
2957 */
2958 static inline u8 rdma_start_port(const struct ib_device *device)
2959 {
2960 return rdma_cap_ib_switch(device) ? 0 : 1;
2961 }
2962
2963 /**
2964 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2965 * @device - The struct ib_device * to iterate over
2966 * @iter - The unsigned int to store the port number
2967 */
2968 #define rdma_for_each_port(device, iter) \
2969 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
2970 unsigned int, iter))); \
2971 iter <= rdma_end_port(device); (iter)++)
2972
2973 /**
2974 * rdma_end_port - Return the last valid port number for the device
2975 * specified
2976 *
2977 * @device: Device to be checked
2978 *
2979 * Return last port number
2980 */
2981 static inline u8 rdma_end_port(const struct ib_device *device)
2982 {
2983 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2984 }
2985
2986 static inline int rdma_is_port_valid(const struct ib_device *device,
2987 unsigned int port)
2988 {
2989 return (port >= rdma_start_port(device) &&
2990 port <= rdma_end_port(device));
2991 }
2992
2993 static inline bool rdma_is_grh_required(const struct ib_device *device,
2994 u8 port_num)
2995 {
2996 return device->port_data[port_num].immutable.core_cap_flags &
2997 RDMA_CORE_PORT_IB_GRH_REQUIRED;
2998 }
2999
3000 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
3001 {
3002 return device->port_data[port_num].immutable.core_cap_flags &
3003 RDMA_CORE_CAP_PROT_IB;
3004 }
3005
3006 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
3007 {
3008 return device->port_data[port_num].immutable.core_cap_flags &
3009 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3010 }
3011
3012 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
3013 {
3014 return device->port_data[port_num].immutable.core_cap_flags &
3015 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3016 }
3017
3018 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
3019 {
3020 return device->port_data[port_num].immutable.core_cap_flags &
3021 RDMA_CORE_CAP_PROT_ROCE;
3022 }
3023
3024 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
3025 {
3026 return device->port_data[port_num].immutable.core_cap_flags &
3027 RDMA_CORE_CAP_PROT_IWARP;
3028 }
3029
3030 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
3031 {
3032 return rdma_protocol_ib(device, port_num) ||
3033 rdma_protocol_roce(device, port_num);
3034 }
3035
3036 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
3037 {
3038 return device->port_data[port_num].immutable.core_cap_flags &
3039 RDMA_CORE_CAP_PROT_RAW_PACKET;
3040 }
3041
3042 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
3043 {
3044 return device->port_data[port_num].immutable.core_cap_flags &
3045 RDMA_CORE_CAP_PROT_USNIC;
3046 }
3047
3048 /**
3049 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3050 * Management Datagrams.
3051 * @device: Device to check
3052 * @port_num: Port number to check
3053 *
3054 * Management Datagrams (MAD) are a required part of the InfiniBand
3055 * specification and are supported on all InfiniBand devices. A slightly
3056 * extended version are also supported on OPA interfaces.
3057 *
3058 * Return: true if the port supports sending/receiving of MAD packets.
3059 */
3060 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
3061 {
3062 return device->port_data[port_num].immutable.core_cap_flags &
3063 RDMA_CORE_CAP_IB_MAD;
3064 }
3065
3066 /**
3067 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3068 * Management Datagrams.
3069 * @device: Device to check
3070 * @port_num: Port number to check
3071 *
3072 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3073 * datagrams with their own versions. These OPA MADs share many but not all of
3074 * the characteristics of InfiniBand MADs.
3075 *
3076 * OPA MADs differ in the following ways:
3077 *
3078 * 1) MADs are variable size up to 2K
3079 * IBTA defined MADs remain fixed at 256 bytes
3080 * 2) OPA SMPs must carry valid PKeys
3081 * 3) OPA SMP packets are a different format
3082 *
3083 * Return: true if the port supports OPA MAD packet formats.
3084 */
3085 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3086 {
3087 return device->port_data[port_num].immutable.core_cap_flags &
3088 RDMA_CORE_CAP_OPA_MAD;
3089 }
3090
3091 /**
3092 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3093 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3094 * @device: Device to check
3095 * @port_num: Port number to check
3096 *
3097 * Each InfiniBand node is required to provide a Subnet Management Agent
3098 * that the subnet manager can access. Prior to the fabric being fully
3099 * configured by the subnet manager, the SMA is accessed via a well known
3100 * interface called the Subnet Management Interface (SMI). This interface
3101 * uses directed route packets to communicate with the SM to get around the
3102 * chicken and egg problem of the SM needing to know what's on the fabric
3103 * in order to configure the fabric, and needing to configure the fabric in
3104 * order to send packets to the devices on the fabric. These directed
3105 * route packets do not need the fabric fully configured in order to reach
3106 * their destination. The SMI is the only method allowed to send
3107 * directed route packets on an InfiniBand fabric.
3108 *
3109 * Return: true if the port provides an SMI.
3110 */
3111 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3112 {
3113 return device->port_data[port_num].immutable.core_cap_flags &
3114 RDMA_CORE_CAP_IB_SMI;
3115 }
3116
3117 /**
3118 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3119 * Communication Manager.
3120 * @device: Device to check
3121 * @port_num: Port number to check
3122 *
3123 * The InfiniBand Communication Manager is one of many pre-defined General
3124 * Service Agents (GSA) that are accessed via the General Service
3125 * Interface (GSI). It's role is to facilitate establishment of connections
3126 * between nodes as well as other management related tasks for established
3127 * connections.
3128 *
3129 * Return: true if the port supports an IB CM (this does not guarantee that
3130 * a CM is actually running however).
3131 */
3132 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3133 {
3134 return device->port_data[port_num].immutable.core_cap_flags &
3135 RDMA_CORE_CAP_IB_CM;
3136 }
3137
3138 /**
3139 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3140 * Communication Manager.
3141 * @device: Device to check
3142 * @port_num: Port number to check
3143 *
3144 * Similar to above, but specific to iWARP connections which have a different
3145 * managment protocol than InfiniBand.
3146 *
3147 * Return: true if the port supports an iWARP CM (this does not guarantee that
3148 * a CM is actually running however).
3149 */
3150 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3151 {
3152 return device->port_data[port_num].immutable.core_cap_flags &
3153 RDMA_CORE_CAP_IW_CM;
3154 }
3155
3156 /**
3157 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3158 * Subnet Administration.
3159 * @device: Device to check
3160 * @port_num: Port number to check
3161 *
3162 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3163 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3164 * fabrics, devices should resolve routes to other hosts by contacting the
3165 * SA to query the proper route.
3166 *
3167 * Return: true if the port should act as a client to the fabric Subnet
3168 * Administration interface. This does not imply that the SA service is
3169 * running locally.
3170 */
3171 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3172 {
3173 return device->port_data[port_num].immutable.core_cap_flags &
3174 RDMA_CORE_CAP_IB_SA;
3175 }
3176
3177 /**
3178 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3179 * Multicast.
3180 * @device: Device to check
3181 * @port_num: Port number to check
3182 *
3183 * InfiniBand multicast registration is more complex than normal IPv4 or
3184 * IPv6 multicast registration. Each Host Channel Adapter must register
3185 * with the Subnet Manager when it wishes to join a multicast group. It
3186 * should do so only once regardless of how many queue pairs it subscribes
3187 * to this group. And it should leave the group only after all queue pairs
3188 * attached to the group have been detached.
3189 *
3190 * Return: true if the port must undertake the additional adminstrative
3191 * overhead of registering/unregistering with the SM and tracking of the
3192 * total number of queue pairs attached to the multicast group.
3193 */
3194 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3195 {
3196 return rdma_cap_ib_sa(device, port_num);
3197 }
3198
3199 /**
3200 * rdma_cap_af_ib - Check if the port of device has the capability
3201 * Native Infiniband Address.
3202 * @device: Device to check
3203 * @port_num: Port number to check
3204 *
3205 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3206 * GID. RoCE uses a different mechanism, but still generates a GID via
3207 * a prescribed mechanism and port specific data.
3208 *
3209 * Return: true if the port uses a GID address to identify devices on the
3210 * network.
3211 */
3212 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3213 {
3214 return device->port_data[port_num].immutable.core_cap_flags &
3215 RDMA_CORE_CAP_AF_IB;
3216 }
3217
3218 /**
3219 * rdma_cap_eth_ah - Check if the port of device has the capability
3220 * Ethernet Address Handle.
3221 * @device: Device to check
3222 * @port_num: Port number to check
3223 *
3224 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3225 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3226 * port. Normally, packet headers are generated by the sending host
3227 * adapter, but when sending connectionless datagrams, we must manually
3228 * inject the proper headers for the fabric we are communicating over.
3229 *
3230 * Return: true if we are running as a RoCE port and must force the
3231 * addition of a Global Route Header built from our Ethernet Address
3232 * Handle into our header list for connectionless packets.
3233 */
3234 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3235 {
3236 return device->port_data[port_num].immutable.core_cap_flags &
3237 RDMA_CORE_CAP_ETH_AH;
3238 }
3239
3240 /**
3241 * rdma_cap_opa_ah - Check if the port of device supports
3242 * OPA Address handles
3243 * @device: Device to check
3244 * @port_num: Port number to check
3245 *
3246 * Return: true if we are running on an OPA device which supports
3247 * the extended OPA addressing.
3248 */
3249 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3250 {
3251 return (device->port_data[port_num].immutable.core_cap_flags &
3252 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3253 }
3254
3255 /**
3256 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3257 *
3258 * @device: Device
3259 * @port_num: Port number
3260 *
3261 * This MAD size includes the MAD headers and MAD payload. No other headers
3262 * are included.
3263 *
3264 * Return the max MAD size required by the Port. Will return 0 if the port
3265 * does not support MADs
3266 */
3267 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3268 {
3269 return device->port_data[port_num].immutable.max_mad_size;
3270 }
3271
3272 /**
3273 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3274 * @device: Device to check
3275 * @port_num: Port number to check
3276 *
3277 * RoCE GID table mechanism manages the various GIDs for a device.
3278 *
3279 * NOTE: if allocating the port's GID table has failed, this call will still
3280 * return true, but any RoCE GID table API will fail.
3281 *
3282 * Return: true if the port uses RoCE GID table mechanism in order to manage
3283 * its GIDs.
3284 */
3285 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3286 u8 port_num)
3287 {
3288 return rdma_protocol_roce(device, port_num) &&
3289 device->ops.add_gid && device->ops.del_gid;
3290 }
3291
3292 /*
3293 * Check if the device supports READ W/ INVALIDATE.
3294 */
3295 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3296 {
3297 /*
3298 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3299 * has support for it yet.
3300 */
3301 return rdma_protocol_iwarp(dev, port_num);
3302 }
3303
3304 /**
3305 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3306 * @device: Device
3307 * @port_num: 1 based Port number
3308 *
3309 * Return true if port is an Intel OPA port , false if not
3310 */
3311 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3312 u32 port_num)
3313 {
3314 return (device->port_data[port_num].immutable.core_cap_flags &
3315 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3316 }
3317
3318 /**
3319 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3320 * @device: Device
3321 * @port_num: Port number
3322 * @mtu: enum value of MTU
3323 *
3324 * Return the MTU size supported by the port as an integer value. Will return
3325 * -1 if enum value of mtu is not supported.
3326 */
3327 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u8 port,
3328 int mtu)
3329 {
3330 if (rdma_core_cap_opa_port(device, port))
3331 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3332 else
3333 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3334 }
3335
3336 /**
3337 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3338 * @device: Device
3339 * @port_num: Port number
3340 * @attr: port attribute
3341 *
3342 * Return the MTU size supported by the port as an integer value.
3343 */
3344 static inline int rdma_mtu_from_attr(struct ib_device *device, u8 port,
3345 struct ib_port_attr *attr)
3346 {
3347 if (rdma_core_cap_opa_port(device, port))
3348 return attr->phys_mtu;
3349 else
3350 return ib_mtu_enum_to_int(attr->max_mtu);
3351 }
3352
3353 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3354 int state);
3355 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3356 struct ifla_vf_info *info);
3357 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3358 struct ifla_vf_stats *stats);
3359 int ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
3360 struct ifla_vf_guid *node_guid,
3361 struct ifla_vf_guid *port_guid);
3362 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3363 int type);
3364
3365 int ib_query_pkey(struct ib_device *device,
3366 u8 port_num, u16 index, u16 *pkey);
3367
3368 int ib_modify_device(struct ib_device *device,
3369 int device_modify_mask,
3370 struct ib_device_modify *device_modify);
3371
3372 int ib_modify_port(struct ib_device *device,
3373 u8 port_num, int port_modify_mask,
3374 struct ib_port_modify *port_modify);
3375
3376 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3377 u8 *port_num, u16 *index);
3378
3379 int ib_find_pkey(struct ib_device *device,
3380 u8 port_num, u16 pkey, u16 *index);
3381
3382 enum ib_pd_flags {
3383 /*
3384 * Create a memory registration for all memory in the system and place
3385 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3386 * ULPs to avoid the overhead of dynamic MRs.
3387 *
3388 * This flag is generally considered unsafe and must only be used in
3389 * extremly trusted environments. Every use of it will log a warning
3390 * in the kernel log.
3391 */
3392 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3393 };
3394
3395 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3396 const char *caller);
3397
3398 /**
3399 * ib_alloc_pd - Allocates an unused protection domain.
3400 * @device: The device on which to allocate the protection domain.
3401 * @flags: protection domain flags
3402 *
3403 * A protection domain object provides an association between QPs, shared
3404 * receive queues, address handles, memory regions, and memory windows.
3405 *
3406 * Every PD has a local_dma_lkey which can be used as the lkey value for local
3407 * memory operations.
3408 */
3409 #define ib_alloc_pd(device, flags) \
3410 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3411
3412 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3413
3414 /**
3415 * ib_dealloc_pd - Deallocate kernel PD
3416 * @pd: The protection domain
3417 *
3418 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3419 */
3420 static inline void ib_dealloc_pd(struct ib_pd *pd)
3421 {
3422 int ret = ib_dealloc_pd_user(pd, NULL);
3423
3424 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3425 }
3426
3427 enum rdma_create_ah_flags {
3428 /* In a sleepable context */
3429 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3430 };
3431
3432 /**
3433 * rdma_create_ah - Creates an address handle for the given address vector.
3434 * @pd: The protection domain associated with the address handle.
3435 * @ah_attr: The attributes of the address vector.
3436 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3437 *
3438 * The address handle is used to reference a local or global destination
3439 * in all UD QP post sends.
3440 */
3441 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3442 u32 flags);
3443
3444 /**
3445 * rdma_create_user_ah - Creates an address handle for the given address vector.
3446 * It resolves destination mac address for ah attribute of RoCE type.
3447 * @pd: The protection domain associated with the address handle.
3448 * @ah_attr: The attributes of the address vector.
3449 * @udata: pointer to user's input output buffer information need by
3450 * provider driver.
3451 *
3452 * It returns 0 on success and returns appropriate error code on error.
3453 * The address handle is used to reference a local or global destination
3454 * in all UD QP post sends.
3455 */
3456 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3457 struct rdma_ah_attr *ah_attr,
3458 struct ib_udata *udata);
3459 /**
3460 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3461 * work completion.
3462 * @hdr: the L3 header to parse
3463 * @net_type: type of header to parse
3464 * @sgid: place to store source gid
3465 * @dgid: place to store destination gid
3466 */
3467 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3468 enum rdma_network_type net_type,
3469 union ib_gid *sgid, union ib_gid *dgid);
3470
3471 /**
3472 * ib_get_rdma_header_version - Get the header version
3473 * @hdr: the L3 header to parse
3474 */
3475 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3476
3477 /**
3478 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3479 * work completion.
3480 * @device: Device on which the received message arrived.
3481 * @port_num: Port on which the received message arrived.
3482 * @wc: Work completion associated with the received message.
3483 * @grh: References the received global route header. This parameter is
3484 * ignored unless the work completion indicates that the GRH is valid.
3485 * @ah_attr: Returned attributes that can be used when creating an address
3486 * handle for replying to the message.
3487 * When ib_init_ah_attr_from_wc() returns success,
3488 * (a) for IB link layer it optionally contains a reference to SGID attribute
3489 * when GRH is present for IB link layer.
3490 * (b) for RoCE link layer it contains a reference to SGID attribute.
3491 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3492 * attributes which are initialized using ib_init_ah_attr_from_wc().
3493 *
3494 */
3495 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3496 const struct ib_wc *wc, const struct ib_grh *grh,
3497 struct rdma_ah_attr *ah_attr);
3498
3499 /**
3500 * ib_create_ah_from_wc - Creates an address handle associated with the
3501 * sender of the specified work completion.
3502 * @pd: The protection domain associated with the address handle.
3503 * @wc: Work completion information associated with a received message.
3504 * @grh: References the received global route header. This parameter is
3505 * ignored unless the work completion indicates that the GRH is valid.
3506 * @port_num: The outbound port number to associate with the address.
3507 *
3508 * The address handle is used to reference a local or global destination
3509 * in all UD QP post sends.
3510 */
3511 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3512 const struct ib_grh *grh, u8 port_num);
3513
3514 /**
3515 * rdma_modify_ah - Modifies the address vector associated with an address
3516 * handle.
3517 * @ah: The address handle to modify.
3518 * @ah_attr: The new address vector attributes to associate with the
3519 * address handle.
3520 */
3521 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3522
3523 /**
3524 * rdma_query_ah - Queries the address vector associated with an address
3525 * handle.
3526 * @ah: The address handle to query.
3527 * @ah_attr: The address vector attributes associated with the address
3528 * handle.
3529 */
3530 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3531
3532 enum rdma_destroy_ah_flags {
3533 /* In a sleepable context */
3534 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3535 };
3536
3537 /**
3538 * rdma_destroy_ah_user - Destroys an address handle.
3539 * @ah: The address handle to destroy.
3540 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3541 * @udata: Valid user data or NULL for kernel objects
3542 */
3543 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3544
3545 /**
3546 * rdma_destroy_ah - Destroys an kernel address handle.
3547 * @ah: The address handle to destroy.
3548 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3549 *
3550 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3551 */
3552 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3553 {
3554 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3555
3556 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3557 }
3558
3559 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3560 struct ib_srq_init_attr *srq_init_attr,
3561 struct ib_usrq_object *uobject,
3562 struct ib_udata *udata);
3563 static inline struct ib_srq *
3564 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3565 {
3566 if (!pd->device->ops.create_srq)
3567 return ERR_PTR(-EOPNOTSUPP);
3568
3569 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3570 }
3571
3572 /**
3573 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3574 * @srq: The SRQ to modify.
3575 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3576 * the current values of selected SRQ attributes are returned.
3577 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3578 * are being modified.
3579 *
3580 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3581 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3582 * the number of receives queued drops below the limit.
3583 */
3584 int ib_modify_srq(struct ib_srq *srq,
3585 struct ib_srq_attr *srq_attr,
3586 enum ib_srq_attr_mask srq_attr_mask);
3587
3588 /**
3589 * ib_query_srq - Returns the attribute list and current values for the
3590 * specified SRQ.
3591 * @srq: The SRQ to query.
3592 * @srq_attr: The attributes of the specified SRQ.
3593 */
3594 int ib_query_srq(struct ib_srq *srq,
3595 struct ib_srq_attr *srq_attr);
3596
3597 /**
3598 * ib_destroy_srq_user - Destroys the specified SRQ.
3599 * @srq: The SRQ to destroy.
3600 * @udata: Valid user data or NULL for kernel objects
3601 */
3602 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3603
3604 /**
3605 * ib_destroy_srq - Destroys the specified kernel SRQ.
3606 * @srq: The SRQ to destroy.
3607 *
3608 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3609 */
3610 static inline void ib_destroy_srq(struct ib_srq *srq)
3611 {
3612 int ret = ib_destroy_srq_user(srq, NULL);
3613
3614 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3615 }
3616
3617 /**
3618 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3619 * @srq: The SRQ to post the work request on.
3620 * @recv_wr: A list of work requests to post on the receive queue.
3621 * @bad_recv_wr: On an immediate failure, this parameter will reference
3622 * the work request that failed to be posted on the QP.
3623 */
3624 static inline int ib_post_srq_recv(struct ib_srq *srq,
3625 const struct ib_recv_wr *recv_wr,
3626 const struct ib_recv_wr **bad_recv_wr)
3627 {
3628 const struct ib_recv_wr *dummy;
3629
3630 return srq->device->ops.post_srq_recv(srq, recv_wr,
3631 bad_recv_wr ? : &dummy);
3632 }
3633
3634 struct ib_qp *ib_create_named_qp(struct ib_pd *pd,
3635 struct ib_qp_init_attr *qp_init_attr,
3636 const char *caller);
3637 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3638 struct ib_qp_init_attr *init_attr)
3639 {
3640 return ib_create_named_qp(pd, init_attr, KBUILD_MODNAME);
3641 }
3642
3643 /**
3644 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3645 * @qp: The QP to modify.
3646 * @attr: On input, specifies the QP attributes to modify. On output,
3647 * the current values of selected QP attributes are returned.
3648 * @attr_mask: A bit-mask used to specify which attributes of the QP
3649 * are being modified.
3650 * @udata: pointer to user's input output buffer information
3651 * are being modified.
3652 * It returns 0 on success and returns appropriate error code on error.
3653 */
3654 int ib_modify_qp_with_udata(struct ib_qp *qp,
3655 struct ib_qp_attr *attr,
3656 int attr_mask,
3657 struct ib_udata *udata);
3658
3659 /**
3660 * ib_modify_qp - Modifies the attributes for the specified QP and then
3661 * transitions the QP to the given state.
3662 * @qp: The QP to modify.
3663 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3664 * the current values of selected QP attributes are returned.
3665 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3666 * are being modified.
3667 */
3668 int ib_modify_qp(struct ib_qp *qp,
3669 struct ib_qp_attr *qp_attr,
3670 int qp_attr_mask);
3671
3672 /**
3673 * ib_query_qp - Returns the attribute list and current values for the
3674 * specified QP.
3675 * @qp: The QP to query.
3676 * @qp_attr: The attributes of the specified QP.
3677 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3678 * @qp_init_attr: Additional attributes of the selected QP.
3679 *
3680 * The qp_attr_mask may be used to limit the query to gathering only the
3681 * selected attributes.
3682 */
3683 int ib_query_qp(struct ib_qp *qp,
3684 struct ib_qp_attr *qp_attr,
3685 int qp_attr_mask,
3686 struct ib_qp_init_attr *qp_init_attr);
3687
3688 /**
3689 * ib_destroy_qp - Destroys the specified QP.
3690 * @qp: The QP to destroy.
3691 * @udata: Valid udata or NULL for kernel objects
3692 */
3693 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3694
3695 /**
3696 * ib_destroy_qp - Destroys the specified kernel QP.
3697 * @qp: The QP to destroy.
3698 *
3699 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3700 */
3701 static inline int ib_destroy_qp(struct ib_qp *qp)
3702 {
3703 return ib_destroy_qp_user(qp, NULL);
3704 }
3705
3706 /**
3707 * ib_open_qp - Obtain a reference to an existing sharable QP.
3708 * @xrcd - XRC domain
3709 * @qp_open_attr: Attributes identifying the QP to open.
3710 *
3711 * Returns a reference to a sharable QP.
3712 */
3713 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3714 struct ib_qp_open_attr *qp_open_attr);
3715
3716 /**
3717 * ib_close_qp - Release an external reference to a QP.
3718 * @qp: The QP handle to release
3719 *
3720 * The opened QP handle is released by the caller. The underlying
3721 * shared QP is not destroyed until all internal references are released.
3722 */
3723 int ib_close_qp(struct ib_qp *qp);
3724
3725 /**
3726 * ib_post_send - Posts a list of work requests to the send queue of
3727 * the specified QP.
3728 * @qp: The QP to post the work request on.
3729 * @send_wr: A list of work requests to post on the send queue.
3730 * @bad_send_wr: On an immediate failure, this parameter will reference
3731 * the work request that failed to be posted on the QP.
3732 *
3733 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3734 * error is returned, the QP state shall not be affected,
3735 * ib_post_send() will return an immediate error after queueing any
3736 * earlier work requests in the list.
3737 */
3738 static inline int ib_post_send(struct ib_qp *qp,
3739 const struct ib_send_wr *send_wr,
3740 const struct ib_send_wr **bad_send_wr)
3741 {
3742 const struct ib_send_wr *dummy;
3743
3744 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3745 }
3746
3747 /**
3748 * ib_post_recv - Posts a list of work requests to the receive queue of
3749 * the specified QP.
3750 * @qp: The QP to post the work request on.
3751 * @recv_wr: A list of work requests to post on the receive queue.
3752 * @bad_recv_wr: On an immediate failure, this parameter will reference
3753 * the work request that failed to be posted on the QP.
3754 */
3755 static inline int ib_post_recv(struct ib_qp *qp,
3756 const struct ib_recv_wr *recv_wr,
3757 const struct ib_recv_wr **bad_recv_wr)
3758 {
3759 const struct ib_recv_wr *dummy;
3760
3761 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3762 }
3763
3764 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3765 int comp_vector, enum ib_poll_context poll_ctx,
3766 const char *caller);
3767 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3768 int nr_cqe, int comp_vector,
3769 enum ib_poll_context poll_ctx)
3770 {
3771 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3772 KBUILD_MODNAME);
3773 }
3774
3775 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3776 int nr_cqe, enum ib_poll_context poll_ctx,
3777 const char *caller);
3778
3779 /**
3780 * ib_alloc_cq_any: Allocate kernel CQ
3781 * @dev: The IB device
3782 * @private: Private data attached to the CQE
3783 * @nr_cqe: Number of CQEs in the CQ
3784 * @poll_ctx: Context used for polling the CQ
3785 */
3786 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3787 void *private, int nr_cqe,
3788 enum ib_poll_context poll_ctx)
3789 {
3790 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3791 KBUILD_MODNAME);
3792 }
3793
3794 void ib_free_cq(struct ib_cq *cq);
3795 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3796
3797 /**
3798 * ib_create_cq - Creates a CQ on the specified device.
3799 * @device: The device on which to create the CQ.
3800 * @comp_handler: A user-specified callback that is invoked when a
3801 * completion event occurs on the CQ.
3802 * @event_handler: A user-specified callback that is invoked when an
3803 * asynchronous event not associated with a completion occurs on the CQ.
3804 * @cq_context: Context associated with the CQ returned to the user via
3805 * the associated completion and event handlers.
3806 * @cq_attr: The attributes the CQ should be created upon.
3807 *
3808 * Users can examine the cq structure to determine the actual CQ size.
3809 */
3810 struct ib_cq *__ib_create_cq(struct ib_device *device,
3811 ib_comp_handler comp_handler,
3812 void (*event_handler)(struct ib_event *, void *),
3813 void *cq_context,
3814 const struct ib_cq_init_attr *cq_attr,
3815 const char *caller);
3816 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3817 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3818
3819 /**
3820 * ib_resize_cq - Modifies the capacity of the CQ.
3821 * @cq: The CQ to resize.
3822 * @cqe: The minimum size of the CQ.
3823 *
3824 * Users can examine the cq structure to determine the actual CQ size.
3825 */
3826 int ib_resize_cq(struct ib_cq *cq, int cqe);
3827
3828 /**
3829 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3830 * @cq: The CQ to modify.
3831 * @cq_count: number of CQEs that will trigger an event
3832 * @cq_period: max period of time in usec before triggering an event
3833 *
3834 */
3835 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3836
3837 /**
3838 * ib_destroy_cq_user - Destroys the specified CQ.
3839 * @cq: The CQ to destroy.
3840 * @udata: Valid user data or NULL for kernel objects
3841 */
3842 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3843
3844 /**
3845 * ib_destroy_cq - Destroys the specified kernel CQ.
3846 * @cq: The CQ to destroy.
3847 *
3848 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3849 */
3850 static inline void ib_destroy_cq(struct ib_cq *cq)
3851 {
3852 int ret = ib_destroy_cq_user(cq, NULL);
3853
3854 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3855 }
3856
3857 /**
3858 * ib_poll_cq - poll a CQ for completion(s)
3859 * @cq:the CQ being polled
3860 * @num_entries:maximum number of completions to return
3861 * @wc:array of at least @num_entries &struct ib_wc where completions
3862 * will be returned
3863 *
3864 * Poll a CQ for (possibly multiple) completions. If the return value
3865 * is < 0, an error occurred. If the return value is >= 0, it is the
3866 * number of completions returned. If the return value is
3867 * non-negative and < num_entries, then the CQ was emptied.
3868 */
3869 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3870 struct ib_wc *wc)
3871 {
3872 return cq->device->ops.poll_cq(cq, num_entries, wc);
3873 }
3874
3875 /**
3876 * ib_req_notify_cq - Request completion notification on a CQ.
3877 * @cq: The CQ to generate an event for.
3878 * @flags:
3879 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3880 * to request an event on the next solicited event or next work
3881 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3882 * may also be |ed in to request a hint about missed events, as
3883 * described below.
3884 *
3885 * Return Value:
3886 * < 0 means an error occurred while requesting notification
3887 * == 0 means notification was requested successfully, and if
3888 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3889 * were missed and it is safe to wait for another event. In
3890 * this case is it guaranteed that any work completions added
3891 * to the CQ since the last CQ poll will trigger a completion
3892 * notification event.
3893 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3894 * in. It means that the consumer must poll the CQ again to
3895 * make sure it is empty to avoid missing an event because of a
3896 * race between requesting notification and an entry being
3897 * added to the CQ. This return value means it is possible
3898 * (but not guaranteed) that a work completion has been added
3899 * to the CQ since the last poll without triggering a
3900 * completion notification event.
3901 */
3902 static inline int ib_req_notify_cq(struct ib_cq *cq,
3903 enum ib_cq_notify_flags flags)
3904 {
3905 return cq->device->ops.req_notify_cq(cq, flags);
3906 }
3907
3908 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
3909 int comp_vector_hint,
3910 enum ib_poll_context poll_ctx);
3911
3912 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
3913
3914 /**
3915 * ib_req_ncomp_notif - Request completion notification when there are
3916 * at least the specified number of unreaped completions on the CQ.
3917 * @cq: The CQ to generate an event for.
3918 * @wc_cnt: The number of unreaped completions that should be on the
3919 * CQ before an event is generated.
3920 */
3921 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3922 {
3923 return cq->device->ops.req_ncomp_notif ?
3924 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3925 -ENOSYS;
3926 }
3927
3928 /*
3929 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
3930 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
3931 * address into the dma address.
3932 */
3933 static inline bool ib_uses_virt_dma(struct ib_device *dev)
3934 {
3935 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
3936 }
3937
3938 /**
3939 * ib_dma_mapping_error - check a DMA addr for error
3940 * @dev: The device for which the dma_addr was created
3941 * @dma_addr: The DMA address to check
3942 */
3943 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3944 {
3945 if (ib_uses_virt_dma(dev))
3946 return 0;
3947 return dma_mapping_error(dev->dma_device, dma_addr);
3948 }
3949
3950 /**
3951 * ib_dma_map_single - Map a kernel virtual address to DMA address
3952 * @dev: The device for which the dma_addr is to be created
3953 * @cpu_addr: The kernel virtual address
3954 * @size: The size of the region in bytes
3955 * @direction: The direction of the DMA
3956 */
3957 static inline u64 ib_dma_map_single(struct ib_device *dev,
3958 void *cpu_addr, size_t size,
3959 enum dma_data_direction direction)
3960 {
3961 if (ib_uses_virt_dma(dev))
3962 return (uintptr_t)cpu_addr;
3963 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3964 }
3965
3966 /**
3967 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3968 * @dev: The device for which the DMA address was created
3969 * @addr: The DMA address
3970 * @size: The size of the region in bytes
3971 * @direction: The direction of the DMA
3972 */
3973 static inline void ib_dma_unmap_single(struct ib_device *dev,
3974 u64 addr, size_t size,
3975 enum dma_data_direction direction)
3976 {
3977 if (!ib_uses_virt_dma(dev))
3978 dma_unmap_single(dev->dma_device, addr, size, direction);
3979 }
3980
3981 /**
3982 * ib_dma_map_page - Map a physical page to DMA address
3983 * @dev: The device for which the dma_addr is to be created
3984 * @page: The page to be mapped
3985 * @offset: The offset within the page
3986 * @size: The size of the region in bytes
3987 * @direction: The direction of the DMA
3988 */
3989 static inline u64 ib_dma_map_page(struct ib_device *dev,
3990 struct page *page,
3991 unsigned long offset,
3992 size_t size,
3993 enum dma_data_direction direction)
3994 {
3995 if (ib_uses_virt_dma(dev))
3996 return (uintptr_t)(page_address(page) + offset);
3997 return dma_map_page(dev->dma_device, page, offset, size, direction);
3998 }
3999
4000 /**
4001 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4002 * @dev: The device for which the DMA address was created
4003 * @addr: The DMA address
4004 * @size: The size of the region in bytes
4005 * @direction: The direction of the DMA
4006 */
4007 static inline void ib_dma_unmap_page(struct ib_device *dev,
4008 u64 addr, size_t size,
4009 enum dma_data_direction direction)
4010 {
4011 if (!ib_uses_virt_dma(dev))
4012 dma_unmap_page(dev->dma_device, addr, size, direction);
4013 }
4014
4015 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4016 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4017 struct scatterlist *sg, int nents,
4018 enum dma_data_direction direction,
4019 unsigned long dma_attrs)
4020 {
4021 if (ib_uses_virt_dma(dev))
4022 return ib_dma_virt_map_sg(dev, sg, nents);
4023 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4024 dma_attrs);
4025 }
4026
4027 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4028 struct scatterlist *sg, int nents,
4029 enum dma_data_direction direction,
4030 unsigned long dma_attrs)
4031 {
4032 if (!ib_uses_virt_dma(dev))
4033 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4034 dma_attrs);
4035 }
4036
4037 /**
4038 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4039 * @dev: The device for which the DMA addresses are to be created
4040 * @sg: The array of scatter/gather entries
4041 * @nents: The number of scatter/gather entries
4042 * @direction: The direction of the DMA
4043 */
4044 static inline int ib_dma_map_sg(struct ib_device *dev,
4045 struct scatterlist *sg, int nents,
4046 enum dma_data_direction direction)
4047 {
4048 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4049 }
4050
4051 /**
4052 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4053 * @dev: The device for which the DMA addresses were created
4054 * @sg: The array of scatter/gather entries
4055 * @nents: The number of scatter/gather entries
4056 * @direction: The direction of the DMA
4057 */
4058 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4059 struct scatterlist *sg, int nents,
4060 enum dma_data_direction direction)
4061 {
4062 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4063 }
4064
4065 /**
4066 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4067 * @dev: The device to query
4068 *
4069 * The returned value represents a size in bytes.
4070 */
4071 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4072 {
4073 if (ib_uses_virt_dma(dev))
4074 return UINT_MAX;
4075 return dma_get_max_seg_size(dev->dma_device);
4076 }
4077
4078 /**
4079 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4080 * @dev: The device for which the DMA address was created
4081 * @addr: The DMA address
4082 * @size: The size of the region in bytes
4083 * @dir: The direction of the DMA
4084 */
4085 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4086 u64 addr,
4087 size_t size,
4088 enum dma_data_direction dir)
4089 {
4090 if (!ib_uses_virt_dma(dev))
4091 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4092 }
4093
4094 /**
4095 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4096 * @dev: The device for which the DMA address was created
4097 * @addr: The DMA address
4098 * @size: The size of the region in bytes
4099 * @dir: The direction of the DMA
4100 */
4101 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4102 u64 addr,
4103 size_t size,
4104 enum dma_data_direction dir)
4105 {
4106 if (!ib_uses_virt_dma(dev))
4107 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4108 }
4109
4110 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4111 * space. This function should be called when 'current' is the owning MM.
4112 */
4113 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4114 u64 virt_addr, int mr_access_flags);
4115
4116 /* ib_advise_mr - give an advice about an address range in a memory region */
4117 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4118 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4119 /**
4120 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4121 * HCA translation table.
4122 * @mr: The memory region to deregister.
4123 * @udata: Valid user data or NULL for kernel object
4124 *
4125 * This function can fail, if the memory region has memory windows bound to it.
4126 */
4127 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4128
4129 /**
4130 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4131 * HCA translation table.
4132 * @mr: The memory region to deregister.
4133 *
4134 * This function can fail, if the memory region has memory windows bound to it.
4135 *
4136 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4137 */
4138 static inline int ib_dereg_mr(struct ib_mr *mr)
4139 {
4140 return ib_dereg_mr_user(mr, NULL);
4141 }
4142
4143 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4144 u32 max_num_sg);
4145
4146 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4147 u32 max_num_data_sg,
4148 u32 max_num_meta_sg);
4149
4150 /**
4151 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4152 * R_Key and L_Key.
4153 * @mr - struct ib_mr pointer to be updated.
4154 * @newkey - new key to be used.
4155 */
4156 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4157 {
4158 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4159 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4160 }
4161
4162 /**
4163 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4164 * for calculating a new rkey for type 2 memory windows.
4165 * @rkey - the rkey to increment.
4166 */
4167 static inline u32 ib_inc_rkey(u32 rkey)
4168 {
4169 const u32 mask = 0x000000ff;
4170 return ((rkey + 1) & mask) | (rkey & ~mask);
4171 }
4172
4173 /**
4174 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4175 * @qp: QP to attach to the multicast group. The QP must be type
4176 * IB_QPT_UD.
4177 * @gid: Multicast group GID.
4178 * @lid: Multicast group LID in host byte order.
4179 *
4180 * In order to send and receive multicast packets, subnet
4181 * administration must have created the multicast group and configured
4182 * the fabric appropriately. The port associated with the specified
4183 * QP must also be a member of the multicast group.
4184 */
4185 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4186
4187 /**
4188 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4189 * @qp: QP to detach from the multicast group.
4190 * @gid: Multicast group GID.
4191 * @lid: Multicast group LID in host byte order.
4192 */
4193 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4194
4195 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4196 struct inode *inode, struct ib_udata *udata);
4197 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4198
4199 static inline int ib_check_mr_access(struct ib_device *ib_dev,
4200 unsigned int flags)
4201 {
4202 /*
4203 * Local write permission is required if remote write or
4204 * remote atomic permission is also requested.
4205 */
4206 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4207 !(flags & IB_ACCESS_LOCAL_WRITE))
4208 return -EINVAL;
4209
4210 if (flags & ~IB_ACCESS_SUPPORTED)
4211 return -EINVAL;
4212
4213 if (flags & IB_ACCESS_ON_DEMAND &&
4214 !(ib_dev->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING))
4215 return -EINVAL;
4216 return 0;
4217 }
4218
4219 static inline bool ib_access_writable(int access_flags)
4220 {
4221 /*
4222 * We have writable memory backing the MR if any of the following
4223 * access flags are set. "Local write" and "remote write" obviously
4224 * require write access. "Remote atomic" can do things like fetch and
4225 * add, which will modify memory, and "MW bind" can change permissions
4226 * by binding a window.
4227 */
4228 return access_flags &
4229 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4230 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4231 }
4232
4233 /**
4234 * ib_check_mr_status: lightweight check of MR status.
4235 * This routine may provide status checks on a selected
4236 * ib_mr. first use is for signature status check.
4237 *
4238 * @mr: A memory region.
4239 * @check_mask: Bitmask of which checks to perform from
4240 * ib_mr_status_check enumeration.
4241 * @mr_status: The container of relevant status checks.
4242 * failed checks will be indicated in the status bitmask
4243 * and the relevant info shall be in the error item.
4244 */
4245 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4246 struct ib_mr_status *mr_status);
4247
4248 /**
4249 * ib_device_try_get: Hold a registration lock
4250 * device: The device to lock
4251 *
4252 * A device under an active registration lock cannot become unregistered. It
4253 * is only possible to obtain a registration lock on a device that is fully
4254 * registered, otherwise this function returns false.
4255 *
4256 * The registration lock is only necessary for actions which require the
4257 * device to still be registered. Uses that only require the device pointer to
4258 * be valid should use get_device(&ibdev->dev) to hold the memory.
4259 *
4260 */
4261 static inline bool ib_device_try_get(struct ib_device *dev)
4262 {
4263 return refcount_inc_not_zero(&dev->refcount);
4264 }
4265
4266 void ib_device_put(struct ib_device *device);
4267 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4268 enum rdma_driver_id driver_id);
4269 struct ib_device *ib_device_get_by_name(const char *name,
4270 enum rdma_driver_id driver_id);
4271 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4272 u16 pkey, const union ib_gid *gid,
4273 const struct sockaddr *addr);
4274 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4275 unsigned int port);
4276 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4277
4278 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4279 struct ib_wq_init_attr *init_attr);
4280 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4281 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4282 u32 wq_attr_mask);
4283
4284 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4285 unsigned int *sg_offset, unsigned int page_size);
4286 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4287 int data_sg_nents, unsigned int *data_sg_offset,
4288 struct scatterlist *meta_sg, int meta_sg_nents,
4289 unsigned int *meta_sg_offset, unsigned int page_size);
4290
4291 static inline int
4292 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4293 unsigned int *sg_offset, unsigned int page_size)
4294 {
4295 int n;
4296
4297 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4298 mr->iova = 0;
4299
4300 return n;
4301 }
4302
4303 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4304 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4305
4306 void ib_drain_rq(struct ib_qp *qp);
4307 void ib_drain_sq(struct ib_qp *qp);
4308 void ib_drain_qp(struct ib_qp *qp);
4309
4310 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u16 *speed, u8 *width);
4311
4312 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4313 {
4314 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4315 return attr->roce.dmac;
4316 return NULL;
4317 }
4318
4319 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4320 {
4321 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4322 attr->ib.dlid = (u16)dlid;
4323 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4324 attr->opa.dlid = dlid;
4325 }
4326
4327 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4328 {
4329 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4330 return attr->ib.dlid;
4331 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4332 return attr->opa.dlid;
4333 return 0;
4334 }
4335
4336 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4337 {
4338 attr->sl = sl;
4339 }
4340
4341 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4342 {
4343 return attr->sl;
4344 }
4345
4346 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4347 u8 src_path_bits)
4348 {
4349 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4350 attr->ib.src_path_bits = src_path_bits;
4351 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4352 attr->opa.src_path_bits = src_path_bits;
4353 }
4354
4355 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4356 {
4357 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4358 return attr->ib.src_path_bits;
4359 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4360 return attr->opa.src_path_bits;
4361 return 0;
4362 }
4363
4364 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4365 bool make_grd)
4366 {
4367 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4368 attr->opa.make_grd = make_grd;
4369 }
4370
4371 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4372 {
4373 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4374 return attr->opa.make_grd;
4375 return false;
4376 }
4377
4378 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4379 {
4380 attr->port_num = port_num;
4381 }
4382
4383 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4384 {
4385 return attr->port_num;
4386 }
4387
4388 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4389 u8 static_rate)
4390 {
4391 attr->static_rate = static_rate;
4392 }
4393
4394 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4395 {
4396 return attr->static_rate;
4397 }
4398
4399 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4400 enum ib_ah_flags flag)
4401 {
4402 attr->ah_flags = flag;
4403 }
4404
4405 static inline enum ib_ah_flags
4406 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4407 {
4408 return attr->ah_flags;
4409 }
4410
4411 static inline const struct ib_global_route
4412 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4413 {
4414 return &attr->grh;
4415 }
4416
4417 /*To retrieve and modify the grh */
4418 static inline struct ib_global_route
4419 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4420 {
4421 return &attr->grh;
4422 }
4423
4424 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4425 {
4426 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4427
4428 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4429 }
4430
4431 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4432 __be64 prefix)
4433 {
4434 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4435
4436 grh->dgid.global.subnet_prefix = prefix;
4437 }
4438
4439 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4440 __be64 if_id)
4441 {
4442 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4443
4444 grh->dgid.global.interface_id = if_id;
4445 }
4446
4447 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4448 union ib_gid *dgid, u32 flow_label,
4449 u8 sgid_index, u8 hop_limit,
4450 u8 traffic_class)
4451 {
4452 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4453
4454 attr->ah_flags = IB_AH_GRH;
4455 if (dgid)
4456 grh->dgid = *dgid;
4457 grh->flow_label = flow_label;
4458 grh->sgid_index = sgid_index;
4459 grh->hop_limit = hop_limit;
4460 grh->traffic_class = traffic_class;
4461 grh->sgid_attr = NULL;
4462 }
4463
4464 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4465 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4466 u32 flow_label, u8 hop_limit, u8 traffic_class,
4467 const struct ib_gid_attr *sgid_attr);
4468 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4469 const struct rdma_ah_attr *src);
4470 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4471 const struct rdma_ah_attr *new);
4472 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4473
4474 /**
4475 * rdma_ah_find_type - Return address handle type.
4476 *
4477 * @dev: Device to be checked
4478 * @port_num: Port number
4479 */
4480 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4481 u8 port_num)
4482 {
4483 if (rdma_protocol_roce(dev, port_num))
4484 return RDMA_AH_ATTR_TYPE_ROCE;
4485 if (rdma_protocol_ib(dev, port_num)) {
4486 if (rdma_cap_opa_ah(dev, port_num))
4487 return RDMA_AH_ATTR_TYPE_OPA;
4488 return RDMA_AH_ATTR_TYPE_IB;
4489 }
4490
4491 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4492 }
4493
4494 /**
4495 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4496 * In the current implementation the only way to get
4497 * get the 32bit lid is from other sources for OPA.
4498 * For IB, lids will always be 16bits so cast the
4499 * value accordingly.
4500 *
4501 * @lid: A 32bit LID
4502 */
4503 static inline u16 ib_lid_cpu16(u32 lid)
4504 {
4505 WARN_ON_ONCE(lid & 0xFFFF0000);
4506 return (u16)lid;
4507 }
4508
4509 /**
4510 * ib_lid_be16 - Return lid in 16bit BE encoding.
4511 *
4512 * @lid: A 32bit LID
4513 */
4514 static inline __be16 ib_lid_be16(u32 lid)
4515 {
4516 WARN_ON_ONCE(lid & 0xFFFF0000);
4517 return cpu_to_be16((u16)lid);
4518 }
4519
4520 /**
4521 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4522 * vector
4523 * @device: the rdma device
4524 * @comp_vector: index of completion vector
4525 *
4526 * Returns NULL on failure, otherwise a corresponding cpu map of the
4527 * completion vector (returns all-cpus map if the device driver doesn't
4528 * implement get_vector_affinity).
4529 */
4530 static inline const struct cpumask *
4531 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4532 {
4533 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4534 !device->ops.get_vector_affinity)
4535 return NULL;
4536
4537 return device->ops.get_vector_affinity(device, comp_vector);
4538
4539 }
4540
4541 /**
4542 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4543 * and add their gids, as needed, to the relevant RoCE devices.
4544 *
4545 * @device: the rdma device
4546 */
4547 void rdma_roce_rescan_device(struct ib_device *ibdev);
4548
4549 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4550
4551 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4552
4553 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4554 enum rdma_netdev_t type, const char *name,
4555 unsigned char name_assign_type,
4556 void (*setup)(struct net_device *));
4557
4558 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4559 enum rdma_netdev_t type, const char *name,
4560 unsigned char name_assign_type,
4561 void (*setup)(struct net_device *),
4562 struct net_device *netdev);
4563
4564 /**
4565 * rdma_set_device_sysfs_group - Set device attributes group to have
4566 * driver specific sysfs entries at
4567 * for infiniband class.
4568 *
4569 * @device: device pointer for which attributes to be created
4570 * @group: Pointer to group which should be added when device
4571 * is registered with sysfs.
4572 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4573 * group per device to have sysfs attributes.
4574 *
4575 * NOTE: New drivers should not make use of this API; instead new device
4576 * parameter should be exposed via netlink command. This API and mechanism
4577 * exist only for existing drivers.
4578 */
4579 static inline void
4580 rdma_set_device_sysfs_group(struct ib_device *dev,
4581 const struct attribute_group *group)
4582 {
4583 dev->groups[1] = group;
4584 }
4585
4586 /**
4587 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4588 *
4589 * @device: device pointer for which ib_device pointer to retrieve
4590 *
4591 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4592 *
4593 */
4594 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4595 {
4596 struct ib_core_device *coredev =
4597 container_of(device, struct ib_core_device, dev);
4598
4599 return coredev->owner;
4600 }
4601
4602 /**
4603 * ibdev_to_node - return the NUMA node for a given ib_device
4604 * @dev: device to get the NUMA node for.
4605 */
4606 static inline int ibdev_to_node(struct ib_device *ibdev)
4607 {
4608 struct device *parent = ibdev->dev.parent;
4609
4610 if (!parent)
4611 return NUMA_NO_NODE;
4612 return dev_to_node(parent);
4613 }
4614
4615 /**
4616 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4617 * ib_device holder structure from device pointer.
4618 *
4619 * NOTE: New drivers should not make use of this API; This API is only for
4620 * existing drivers who have exposed sysfs entries using
4621 * rdma_set_device_sysfs_group().
4622 */
4623 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4624 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4625
4626 bool rdma_dev_access_netns(const struct ib_device *device,
4627 const struct net *net);
4628
4629 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4630 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4631 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4632
4633 /**
4634 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4635 * on the flow_label
4636 *
4637 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4638 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4639 * convention.
4640 */
4641 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4642 {
4643 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4644
4645 fl_low ^= fl_high >> 14;
4646 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4647 }
4648
4649 /**
4650 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4651 * local and remote qpn values
4652 *
4653 * This function folded the multiplication results of two qpns, 24 bit each,
4654 * fields, and converts it to a 20 bit results.
4655 *
4656 * This function will create symmetric flow_label value based on the local
4657 * and remote qpn values. this will allow both the requester and responder
4658 * to calculate the same flow_label for a given connection.
4659 *
4660 * This helper function should be used by driver in case the upper layer
4661 * provide a zero flow_label value. This is to improve entropy of RDMA
4662 * traffic in the network.
4663 */
4664 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4665 {
4666 u64 v = (u64)lqpn * rqpn;
4667
4668 v ^= v >> 20;
4669 v ^= v >> 40;
4670
4671 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4672 }
4673 #endif /* IB_VERBS_H */
Linux with added FPC-III support