?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 5.10.7
[linux/fpc-iii.git] / include / rdma / ib_verbs.h
blob65771bef5e654ac280d45d160315abcf099aa781
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/types.h>
16 #include <linux/device.h>
17 #include <linux/dma-mapping.h>
18 #include <linux/kref.h>
19 #include <linux/list.h>
20 #include <linux/rwsem.h>
21 #include <linux/workqueue.h>
22 #include <linux/irq_poll.h>
23 #include <uapi/linux/if_ether.h>
24 #include <net/ipv6.h>
25 #include <net/ip.h>
26 #include <linux/string.h>
27 #include <linux/slab.h>
28 #include <linux/netdevice.h>
29 #include <linux/refcount.h>
30 #include <linux/if_link.h>
31 #include <linux/atomic.h>
32 #include <linux/mmu_notifier.h>
33 #include <linux/uaccess.h>
34 #include <linux/cgroup_rdma.h>
35 #include <linux/irqflags.h>
36 #include <linux/preempt.h>
37 #include <linux/dim.h>
38 #include <uapi/rdma/ib_user_verbs.h>
39 #include <rdma/rdma_counter.h>
40 #include <rdma/restrack.h>
41 #include <rdma/signature.h>
42 #include <uapi/rdma/rdma_user_ioctl.h>
43 #include <uapi/rdma/ib_user_ioctl_verbs.h>
44
45 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
46
47 struct ib_umem_odp;
48 struct ib_uqp_object;
49 struct ib_usrq_object;
50 struct ib_uwq_object;
51 struct rdma_cm_id;
52
53 extern struct workqueue_struct *ib_wq;
54 extern struct workqueue_struct *ib_comp_wq;
55 extern struct workqueue_struct *ib_comp_unbound_wq;
56
57 struct ib_ucq_object;
58
59 __printf(3, 4) __cold
60 void ibdev_printk(const char *level, const struct ib_device *ibdev,
61 const char *format, ...);
62 __printf(2, 3) __cold
63 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
64 __printf(2, 3) __cold
65 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
66 __printf(2, 3) __cold
67 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
68 __printf(2, 3) __cold
69 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
70 __printf(2, 3) __cold
71 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
72 __printf(2, 3) __cold
73 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
74 __printf(2, 3) __cold
75 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
76
77 #if defined(CONFIG_DYNAMIC_DEBUG) || \
78 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
79 #define ibdev_dbg(__dev, format, args...) \
80 dynamic_ibdev_dbg(__dev, format, ##args)
81 #else
82 __printf(2, 3) __cold
83 static inline
84 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
85 #endif
86
87 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
88 do { \
89 static DEFINE_RATELIMIT_STATE(_rs, \
90 DEFAULT_RATELIMIT_INTERVAL, \
91 DEFAULT_RATELIMIT_BURST); \
92 if (__ratelimit(&_rs)) \
93 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
94 } while (0)
95
96 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
97 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
98 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
99 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
100 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
101 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
102 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
103 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
104 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
105 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
106 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
107 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
108 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
109 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
110
111 #if defined(CONFIG_DYNAMIC_DEBUG) || \
112 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
113 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
114 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
115 do { \
116 static DEFINE_RATELIMIT_STATE(_rs, \
117 DEFAULT_RATELIMIT_INTERVAL, \
118 DEFAULT_RATELIMIT_BURST); \
119 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
120 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
121 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
122 ##__VA_ARGS__); \
123 } while (0)
124 #else
125 __printf(2, 3) __cold
126 static inline
127 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
128 #endif
129
130 union ib_gid {
131 u8 raw[16];
132 struct {
133 __be64 subnet_prefix;
134 __be64 interface_id;
135 } global;
136 };
137
138 extern union ib_gid zgid;
139
140 enum ib_gid_type {
141 IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
142 IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
143 IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
144 IB_GID_TYPE_SIZE
145 };
146
147 #define ROCE_V2_UDP_DPORT 4791
148 struct ib_gid_attr {
149 struct net_device __rcu *ndev;
150 struct ib_device *device;
151 union ib_gid gid;
152 enum ib_gid_type gid_type;
153 u16 index;
154 u8 port_num;
155 };
156
157 enum {
158 /* set the local administered indication */
159 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
160 };
161
162 enum rdma_transport_type {
163 RDMA_TRANSPORT_IB,
164 RDMA_TRANSPORT_IWARP,
165 RDMA_TRANSPORT_USNIC,
166 RDMA_TRANSPORT_USNIC_UDP,
167 RDMA_TRANSPORT_UNSPECIFIED,
168 };
169
170 enum rdma_protocol_type {
171 RDMA_PROTOCOL_IB,
172 RDMA_PROTOCOL_IBOE,
173 RDMA_PROTOCOL_IWARP,
174 RDMA_PROTOCOL_USNIC_UDP
175 };
176
177 __attribute_const__ enum rdma_transport_type
178 rdma_node_get_transport(unsigned int node_type);
179
180 enum rdma_network_type {
181 RDMA_NETWORK_IB,
182 RDMA_NETWORK_ROCE_V1,
183 RDMA_NETWORK_IPV4,
184 RDMA_NETWORK_IPV6
185 };
186
187 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
188 {
189 if (network_type == RDMA_NETWORK_IPV4 ||
190 network_type == RDMA_NETWORK_IPV6)
191 return IB_GID_TYPE_ROCE_UDP_ENCAP;
192 else if (network_type == RDMA_NETWORK_ROCE_V1)
193 return IB_GID_TYPE_ROCE;
194 else
195 return IB_GID_TYPE_IB;
196 }
197
198 static inline enum rdma_network_type
199 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
200 {
201 if (attr->gid_type == IB_GID_TYPE_IB)
202 return RDMA_NETWORK_IB;
203
204 if (attr->gid_type == IB_GID_TYPE_ROCE)
205 return RDMA_NETWORK_ROCE_V1;
206
207 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
208 return RDMA_NETWORK_IPV4;
209 else
210 return RDMA_NETWORK_IPV6;
211 }
212
213 enum rdma_link_layer {
214 IB_LINK_LAYER_UNSPECIFIED,
215 IB_LINK_LAYER_INFINIBAND,
216 IB_LINK_LAYER_ETHERNET,
217 };
218
219 enum ib_device_cap_flags {
220 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
221 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
222 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
223 IB_DEVICE_RAW_MULTI = (1 << 3),
224 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
225 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
226 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
227 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
228 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
229 /* Not in use, former INIT_TYPE = (1 << 9),*/
230 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
231 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
232 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
233 IB_DEVICE_SRQ_RESIZE = (1 << 13),
234 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
235
236 /*
237 * This device supports a per-device lkey or stag that can be
238 * used without performing a memory registration for the local
239 * memory. Note that ULPs should never check this flag, but
240 * instead of use the local_dma_lkey flag in the ib_pd structure,
241 * which will always contain a usable lkey.
242 */
243 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
244 /* Reserved, old SEND_W_INV = (1 << 16),*/
245 IB_DEVICE_MEM_WINDOW = (1 << 17),
246 /*
247 * Devices should set IB_DEVICE_UD_IP_SUM if they support
248 * insertion of UDP and TCP checksum on outgoing UD IPoIB
249 * messages and can verify the validity of checksum for
250 * incoming messages. Setting this flag implies that the
251 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
252 */
253 IB_DEVICE_UD_IP_CSUM = (1 << 18),
254 IB_DEVICE_UD_TSO = (1 << 19),
255 IB_DEVICE_XRC = (1 << 20),
256
257 /*
258 * This device supports the IB "base memory management extension",
259 * which includes support for fast registrations (IB_WR_REG_MR,
260 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
261 * also be set by any iWarp device which must support FRs to comply
262 * to the iWarp verbs spec. iWarp devices also support the
263 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
264 * stag.
265 */
266 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
267 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
268 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
269 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
270 IB_DEVICE_RC_IP_CSUM = (1 << 25),
271 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
272 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
273 /*
274 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
275 * support execution of WQEs that involve synchronization
276 * of I/O operations with single completion queue managed
277 * by hardware.
278 */
279 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
280 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
281 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
282 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
283 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
284 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
285 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
286 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
287 IB_DEVICE_RDMA_NETDEV_OPA = (1ULL << 35),
288 /* The device supports padding incoming writes to cacheline. */
289 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
290 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
291 };
292
293 enum ib_atomic_cap {
294 IB_ATOMIC_NONE,
295 IB_ATOMIC_HCA,
296 IB_ATOMIC_GLOB
297 };
298
299 enum ib_odp_general_cap_bits {
300 IB_ODP_SUPPORT = 1 << 0,
301 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
302 };
303
304 enum ib_odp_transport_cap_bits {
305 IB_ODP_SUPPORT_SEND = 1 << 0,
306 IB_ODP_SUPPORT_RECV = 1 << 1,
307 IB_ODP_SUPPORT_WRITE = 1 << 2,
308 IB_ODP_SUPPORT_READ = 1 << 3,
309 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
310 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
311 };
312
313 struct ib_odp_caps {
314 uint64_t general_caps;
315 struct {
316 uint32_t rc_odp_caps;
317 uint32_t uc_odp_caps;
318 uint32_t ud_odp_caps;
319 uint32_t xrc_odp_caps;
320 } per_transport_caps;
321 };
322
323 struct ib_rss_caps {
324 /* Corresponding bit will be set if qp type from
325 * 'enum ib_qp_type' is supported, e.g.
326 * supported_qpts |= 1 << IB_QPT_UD
327 */
328 u32 supported_qpts;
329 u32 max_rwq_indirection_tables;
330 u32 max_rwq_indirection_table_size;
331 };
332
333 enum ib_tm_cap_flags {
334 /* Support tag matching with rendezvous offload for RC transport */
335 IB_TM_CAP_RNDV_RC = 1 << 0,
336 };
337
338 struct ib_tm_caps {
339 /* Max size of RNDV header */
340 u32 max_rndv_hdr_size;
341 /* Max number of entries in tag matching list */
342 u32 max_num_tags;
343 /* From enum ib_tm_cap_flags */
344 u32 flags;
345 /* Max number of outstanding list operations */
346 u32 max_ops;
347 /* Max number of SGE in tag matching entry */
348 u32 max_sge;
349 };
350
351 struct ib_cq_init_attr {
352 unsigned int cqe;
353 u32 comp_vector;
354 u32 flags;
355 };
356
357 enum ib_cq_attr_mask {
358 IB_CQ_MODERATE = 1 << 0,
359 };
360
361 struct ib_cq_caps {
362 u16 max_cq_moderation_count;
363 u16 max_cq_moderation_period;
364 };
365
366 struct ib_dm_mr_attr {
367 u64 length;
368 u64 offset;
369 u32 access_flags;
370 };
371
372 struct ib_dm_alloc_attr {
373 u64 length;
374 u32 alignment;
375 u32 flags;
376 };
377
378 struct ib_device_attr {
379 u64 fw_ver;
380 __be64 sys_image_guid;
381 u64 max_mr_size;
382 u64 page_size_cap;
383 u32 vendor_id;
384 u32 vendor_part_id;
385 u32 hw_ver;
386 int max_qp;
387 int max_qp_wr;
388 u64 device_cap_flags;
389 int max_send_sge;
390 int max_recv_sge;
391 int max_sge_rd;
392 int max_cq;
393 int max_cqe;
394 int max_mr;
395 int max_pd;
396 int max_qp_rd_atom;
397 int max_ee_rd_atom;
398 int max_res_rd_atom;
399 int max_qp_init_rd_atom;
400 int max_ee_init_rd_atom;
401 enum ib_atomic_cap atomic_cap;
402 enum ib_atomic_cap masked_atomic_cap;
403 int max_ee;
404 int max_rdd;
405 int max_mw;
406 int max_raw_ipv6_qp;
407 int max_raw_ethy_qp;
408 int max_mcast_grp;
409 int max_mcast_qp_attach;
410 int max_total_mcast_qp_attach;
411 int max_ah;
412 int max_srq;
413 int max_srq_wr;
414 int max_srq_sge;
415 unsigned int max_fast_reg_page_list_len;
416 unsigned int max_pi_fast_reg_page_list_len;
417 u16 max_pkeys;
418 u8 local_ca_ack_delay;
419 int sig_prot_cap;
420 int sig_guard_cap;
421 struct ib_odp_caps odp_caps;
422 uint64_t timestamp_mask;
423 uint64_t hca_core_clock; /* in KHZ */
424 struct ib_rss_caps rss_caps;
425 u32 max_wq_type_rq;
426 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
427 struct ib_tm_caps tm_caps;
428 struct ib_cq_caps cq_caps;
429 u64 max_dm_size;
430 /* Max entries for sgl for optimized performance per READ */
431 u32 max_sgl_rd;
432 };
433
434 enum ib_mtu {
435 IB_MTU_256 = 1,
436 IB_MTU_512 = 2,
437 IB_MTU_1024 = 3,
438 IB_MTU_2048 = 4,
439 IB_MTU_4096 = 5
440 };
441
442 enum opa_mtu {
443 OPA_MTU_8192 = 6,
444 OPA_MTU_10240 = 7
445 };
446
447 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
448 {
449 switch (mtu) {
450 case IB_MTU_256: return 256;
451 case IB_MTU_512: return 512;
452 case IB_MTU_1024: return 1024;
453 case IB_MTU_2048: return 2048;
454 case IB_MTU_4096: return 4096;
455 default: return -1;
456 }
457 }
458
459 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
460 {
461 if (mtu >= 4096)
462 return IB_MTU_4096;
463 else if (mtu >= 2048)
464 return IB_MTU_2048;
465 else if (mtu >= 1024)
466 return IB_MTU_1024;
467 else if (mtu >= 512)
468 return IB_MTU_512;
469 else
470 return IB_MTU_256;
471 }
472
473 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
474 {
475 switch (mtu) {
476 case OPA_MTU_8192:
477 return 8192;
478 case OPA_MTU_10240:
479 return 10240;
480 default:
481 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
482 }
483 }
484
485 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
486 {
487 if (mtu >= 10240)
488 return OPA_MTU_10240;
489 else if (mtu >= 8192)
490 return OPA_MTU_8192;
491 else
492 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
493 }
494
495 enum ib_port_state {
496 IB_PORT_NOP = 0,
497 IB_PORT_DOWN = 1,
498 IB_PORT_INIT = 2,
499 IB_PORT_ARMED = 3,
500 IB_PORT_ACTIVE = 4,
501 IB_PORT_ACTIVE_DEFER = 5
502 };
503
504 enum ib_port_phys_state {
505 IB_PORT_PHYS_STATE_SLEEP = 1,
506 IB_PORT_PHYS_STATE_POLLING = 2,
507 IB_PORT_PHYS_STATE_DISABLED = 3,
508 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
509 IB_PORT_PHYS_STATE_LINK_UP = 5,
510 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
511 IB_PORT_PHYS_STATE_PHY_TEST = 7,
512 };
513
514 enum ib_port_width {
515 IB_WIDTH_1X = 1,
516 IB_WIDTH_2X = 16,
517 IB_WIDTH_4X = 2,
518 IB_WIDTH_8X = 4,
519 IB_WIDTH_12X = 8
520 };
521
522 static inline int ib_width_enum_to_int(enum ib_port_width width)
523 {
524 switch (width) {
525 case IB_WIDTH_1X: return 1;
526 case IB_WIDTH_2X: return 2;
527 case IB_WIDTH_4X: return 4;
528 case IB_WIDTH_8X: return 8;
529 case IB_WIDTH_12X: return 12;
530 default: return -1;
531 }
532 }
533
534 enum ib_port_speed {
535 IB_SPEED_SDR = 1,
536 IB_SPEED_DDR = 2,
537 IB_SPEED_QDR = 4,
538 IB_SPEED_FDR10 = 8,
539 IB_SPEED_FDR = 16,
540 IB_SPEED_EDR = 32,
541 IB_SPEED_HDR = 64,
542 IB_SPEED_NDR = 128,
543 };
544
545 /**
546 * struct rdma_hw_stats
547 * @lock - Mutex to protect parallel write access to lifespan and values
548 * of counters, which are 64bits and not guaranteeed to be written
549 * atomicaly on 32bits systems.
550 * @timestamp - Used by the core code to track when the last update was
551 * @lifespan - Used by the core code to determine how old the counters
552 * should be before being updated again. Stored in jiffies, defaults
553 * to 10 milliseconds, drivers can override the default be specifying
554 * their own value during their allocation routine.
555 * @name - Array of pointers to static names used for the counters in
556 * directory.
557 * @num_counters - How many hardware counters there are. If name is
558 * shorter than this number, a kernel oops will result. Driver authors
559 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
560 * in their code to prevent this.
561 * @value - Array of u64 counters that are accessed by the sysfs code and
562 * filled in by the drivers get_stats routine
563 */
564 struct rdma_hw_stats {
565 struct mutex lock; /* Protect lifespan and values[] */
566 unsigned long timestamp;
567 unsigned long lifespan;
568 const char * const *names;
569 int num_counters;
570 u64 value[];
571 };
572
573 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
574 /**
575 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
576 * for drivers.
577 * @names - Array of static const char *
578 * @num_counters - How many elements in array
579 * @lifespan - How many milliseconds between updates
580 */
581 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
582 const char * const *names, int num_counters,
583 unsigned long lifespan)
584 {
585 struct rdma_hw_stats *stats;
586
587 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
588 GFP_KERNEL);
589 if (!stats)
590 return NULL;
591 stats->names = names;
592 stats->num_counters = num_counters;
593 stats->lifespan = msecs_to_jiffies(lifespan);
594
595 return stats;
596 }
597
598
599 /* Define bits for the various functionality this port needs to be supported by
600 * the core.
601 */
602 /* Management 0x00000FFF */
603 #define RDMA_CORE_CAP_IB_MAD 0x00000001
604 #define RDMA_CORE_CAP_IB_SMI 0x00000002
605 #define RDMA_CORE_CAP_IB_CM 0x00000004
606 #define RDMA_CORE_CAP_IW_CM 0x00000008
607 #define RDMA_CORE_CAP_IB_SA 0x00000010
608 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
609
610 /* Address format 0x000FF000 */
611 #define RDMA_CORE_CAP_AF_IB 0x00001000
612 #define RDMA_CORE_CAP_ETH_AH 0x00002000
613 #define RDMA_CORE_CAP_OPA_AH 0x00004000
614 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
615
616 /* Protocol 0xFFF00000 */
617 #define RDMA_CORE_CAP_PROT_IB 0x00100000
618 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
619 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
620 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
621 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
622 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
623
624 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
625 | RDMA_CORE_CAP_PROT_ROCE \
626 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
627
628 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
629 | RDMA_CORE_CAP_IB_MAD \
630 | RDMA_CORE_CAP_IB_SMI \
631 | RDMA_CORE_CAP_IB_CM \
632 | RDMA_CORE_CAP_IB_SA \
633 | RDMA_CORE_CAP_AF_IB)
634 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
635 | RDMA_CORE_CAP_IB_MAD \
636 | RDMA_CORE_CAP_IB_CM \
637 | RDMA_CORE_CAP_AF_IB \
638 | RDMA_CORE_CAP_ETH_AH)
639 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
640 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
641 | RDMA_CORE_CAP_IB_MAD \
642 | RDMA_CORE_CAP_IB_CM \
643 | RDMA_CORE_CAP_AF_IB \
644 | RDMA_CORE_CAP_ETH_AH)
645 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
646 | RDMA_CORE_CAP_IW_CM)
647 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
648 | RDMA_CORE_CAP_OPA_MAD)
649
650 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
651
652 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
653
654 struct ib_port_attr {
655 u64 subnet_prefix;
656 enum ib_port_state state;
657 enum ib_mtu max_mtu;
658 enum ib_mtu active_mtu;
659 u32 phys_mtu;
660 int gid_tbl_len;
661 unsigned int ip_gids:1;
662 /* This is the value from PortInfo CapabilityMask, defined by IBA */
663 u32 port_cap_flags;
664 u32 max_msg_sz;
665 u32 bad_pkey_cntr;
666 u32 qkey_viol_cntr;
667 u16 pkey_tbl_len;
668 u32 sm_lid;
669 u32 lid;
670 u8 lmc;
671 u8 max_vl_num;
672 u8 sm_sl;
673 u8 subnet_timeout;
674 u8 init_type_reply;
675 u8 active_width;
676 u16 active_speed;
677 u8 phys_state;
678 u16 port_cap_flags2;
679 };
680
681 enum ib_device_modify_flags {
682 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
683 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
684 };
685
686 #define IB_DEVICE_NODE_DESC_MAX 64
687
688 struct ib_device_modify {
689 u64 sys_image_guid;
690 char node_desc[IB_DEVICE_NODE_DESC_MAX];
691 };
692
693 enum ib_port_modify_flags {
694 IB_PORT_SHUTDOWN = 1,
695 IB_PORT_INIT_TYPE = (1<<2),
696 IB_PORT_RESET_QKEY_CNTR = (1<<3),
697 IB_PORT_OPA_MASK_CHG = (1<<4)
698 };
699
700 struct ib_port_modify {
701 u32 set_port_cap_mask;
702 u32 clr_port_cap_mask;
703 u8 init_type;
704 };
705
706 enum ib_event_type {
707 IB_EVENT_CQ_ERR,
708 IB_EVENT_QP_FATAL,
709 IB_EVENT_QP_REQ_ERR,
710 IB_EVENT_QP_ACCESS_ERR,
711 IB_EVENT_COMM_EST,
712 IB_EVENT_SQ_DRAINED,
713 IB_EVENT_PATH_MIG,
714 IB_EVENT_PATH_MIG_ERR,
715 IB_EVENT_DEVICE_FATAL,
716 IB_EVENT_PORT_ACTIVE,
717 IB_EVENT_PORT_ERR,
718 IB_EVENT_LID_CHANGE,
719 IB_EVENT_PKEY_CHANGE,
720 IB_EVENT_SM_CHANGE,
721 IB_EVENT_SRQ_ERR,
722 IB_EVENT_SRQ_LIMIT_REACHED,
723 IB_EVENT_QP_LAST_WQE_REACHED,
724 IB_EVENT_CLIENT_REREGISTER,
725 IB_EVENT_GID_CHANGE,
726 IB_EVENT_WQ_FATAL,
727 };
728
729 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
730
731 struct ib_event {
732 struct ib_device *device;
733 union {
734 struct ib_cq *cq;
735 struct ib_qp *qp;
736 struct ib_srq *srq;
737 struct ib_wq *wq;
738 u8 port_num;
739 } element;
740 enum ib_event_type event;
741 };
742
743 struct ib_event_handler {
744 struct ib_device *device;
745 void (*handler)(struct ib_event_handler *, struct ib_event *);
746 struct list_head list;
747 };
748
749 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
750 do { \
751 (_ptr)->device = _device; \
752 (_ptr)->handler = _handler; \
753 INIT_LIST_HEAD(&(_ptr)->list); \
754 } while (0)
755
756 struct ib_global_route {
757 const struct ib_gid_attr *sgid_attr;
758 union ib_gid dgid;
759 u32 flow_label;
760 u8 sgid_index;
761 u8 hop_limit;
762 u8 traffic_class;
763 };
764
765 struct ib_grh {
766 __be32 version_tclass_flow;
767 __be16 paylen;
768 u8 next_hdr;
769 u8 hop_limit;
770 union ib_gid sgid;
771 union ib_gid dgid;
772 };
773
774 union rdma_network_hdr {
775 struct ib_grh ibgrh;
776 struct {
777 /* The IB spec states that if it's IPv4, the header
778 * is located in the last 20 bytes of the header.
779 */
780 u8 reserved[20];
781 struct iphdr roce4grh;
782 };
783 };
784
785 #define IB_QPN_MASK 0xFFFFFF
786
787 enum {
788 IB_MULTICAST_QPN = 0xffffff
789 };
790
791 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
792 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
793
794 enum ib_ah_flags {
795 IB_AH_GRH = 1
796 };
797
798 enum ib_rate {
799 IB_RATE_PORT_CURRENT = 0,
800 IB_RATE_2_5_GBPS = 2,
801 IB_RATE_5_GBPS = 5,
802 IB_RATE_10_GBPS = 3,
803 IB_RATE_20_GBPS = 6,
804 IB_RATE_30_GBPS = 4,
805 IB_RATE_40_GBPS = 7,
806 IB_RATE_60_GBPS = 8,
807 IB_RATE_80_GBPS = 9,
808 IB_RATE_120_GBPS = 10,
809 IB_RATE_14_GBPS = 11,
810 IB_RATE_56_GBPS = 12,
811 IB_RATE_112_GBPS = 13,
812 IB_RATE_168_GBPS = 14,
813 IB_RATE_25_GBPS = 15,
814 IB_RATE_100_GBPS = 16,
815 IB_RATE_200_GBPS = 17,
816 IB_RATE_300_GBPS = 18,
817 IB_RATE_28_GBPS = 19,
818 IB_RATE_50_GBPS = 20,
819 IB_RATE_400_GBPS = 21,
820 IB_RATE_600_GBPS = 22,
821 };
822
823 /**
824 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
825 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
826 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
827 * @rate: rate to convert.
828 */
829 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
830
831 /**
832 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
833 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
834 * @rate: rate to convert.
835 */
836 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
837
838
839 /**
840 * enum ib_mr_type - memory region type
841 * @IB_MR_TYPE_MEM_REG: memory region that is used for
842 * normal registration
843 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
844 * register any arbitrary sg lists (without
845 * the normal mr constraints - see
846 * ib_map_mr_sg)
847 * @IB_MR_TYPE_DM: memory region that is used for device
848 * memory registration
849 * @IB_MR_TYPE_USER: memory region that is used for the user-space
850 * application
851 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
852 * without address translations (VA=PA)
853 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
854 * data integrity operations
855 */
856 enum ib_mr_type {
857 IB_MR_TYPE_MEM_REG,
858 IB_MR_TYPE_SG_GAPS,
859 IB_MR_TYPE_DM,
860 IB_MR_TYPE_USER,
861 IB_MR_TYPE_DMA,
862 IB_MR_TYPE_INTEGRITY,
863 };
864
865 enum ib_mr_status_check {
866 IB_MR_CHECK_SIG_STATUS = 1,
867 };
868
869 /**
870 * struct ib_mr_status - Memory region status container
871 *
872 * @fail_status: Bitmask of MR checks status. For each
873 * failed check a corresponding status bit is set.
874 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
875 * failure.
876 */
877 struct ib_mr_status {
878 u32 fail_status;
879 struct ib_sig_err sig_err;
880 };
881
882 /**
883 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
884 * enum.
885 * @mult: multiple to convert.
886 */
887 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
888
889 struct rdma_ah_init_attr {
890 struct rdma_ah_attr *ah_attr;
891 u32 flags;
892 struct net_device *xmit_slave;
893 };
894
895 enum rdma_ah_attr_type {
896 RDMA_AH_ATTR_TYPE_UNDEFINED,
897 RDMA_AH_ATTR_TYPE_IB,
898 RDMA_AH_ATTR_TYPE_ROCE,
899 RDMA_AH_ATTR_TYPE_OPA,
900 };
901
902 struct ib_ah_attr {
903 u16 dlid;
904 u8 src_path_bits;
905 };
906
907 struct roce_ah_attr {
908 u8 dmac[ETH_ALEN];
909 };
910
911 struct opa_ah_attr {
912 u32 dlid;
913 u8 src_path_bits;
914 bool make_grd;
915 };
916
917 struct rdma_ah_attr {
918 struct ib_global_route grh;
919 u8 sl;
920 u8 static_rate;
921 u8 port_num;
922 u8 ah_flags;
923 enum rdma_ah_attr_type type;
924 union {
925 struct ib_ah_attr ib;
926 struct roce_ah_attr roce;
927 struct opa_ah_attr opa;
928 };
929 };
930
931 enum ib_wc_status {
932 IB_WC_SUCCESS,
933 IB_WC_LOC_LEN_ERR,
934 IB_WC_LOC_QP_OP_ERR,
935 IB_WC_LOC_EEC_OP_ERR,
936 IB_WC_LOC_PROT_ERR,
937 IB_WC_WR_FLUSH_ERR,
938 IB_WC_MW_BIND_ERR,
939 IB_WC_BAD_RESP_ERR,
940 IB_WC_LOC_ACCESS_ERR,
941 IB_WC_REM_INV_REQ_ERR,
942 IB_WC_REM_ACCESS_ERR,
943 IB_WC_REM_OP_ERR,
944 IB_WC_RETRY_EXC_ERR,
945 IB_WC_RNR_RETRY_EXC_ERR,
946 IB_WC_LOC_RDD_VIOL_ERR,
947 IB_WC_REM_INV_RD_REQ_ERR,
948 IB_WC_REM_ABORT_ERR,
949 IB_WC_INV_EECN_ERR,
950 IB_WC_INV_EEC_STATE_ERR,
951 IB_WC_FATAL_ERR,
952 IB_WC_RESP_TIMEOUT_ERR,
953 IB_WC_GENERAL_ERR
954 };
955
956 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
957
958 enum ib_wc_opcode {
959 IB_WC_SEND = IB_UVERBS_WC_SEND,
960 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
961 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
962 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
963 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
964 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
965 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
966 IB_WC_LSO = IB_UVERBS_WC_TSO,
967 IB_WC_REG_MR,
968 IB_WC_MASKED_COMP_SWAP,
969 IB_WC_MASKED_FETCH_ADD,
970 /*
971 * Set value of IB_WC_RECV so consumers can test if a completion is a
972 * receive by testing (opcode & IB_WC_RECV).
973 */
974 IB_WC_RECV = 1 << 7,
975 IB_WC_RECV_RDMA_WITH_IMM
976 };
977
978 enum ib_wc_flags {
979 IB_WC_GRH = 1,
980 IB_WC_WITH_IMM = (1<<1),
981 IB_WC_WITH_INVALIDATE = (1<<2),
982 IB_WC_IP_CSUM_OK = (1<<3),
983 IB_WC_WITH_SMAC = (1<<4),
984 IB_WC_WITH_VLAN = (1<<5),
985 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
986 };
987
988 struct ib_wc {
989 union {
990 u64 wr_id;
991 struct ib_cqe *wr_cqe;
992 };
993 enum ib_wc_status status;
994 enum ib_wc_opcode opcode;
995 u32 vendor_err;
996 u32 byte_len;
997 struct ib_qp *qp;
998 union {
999 __be32 imm_data;
1000 u32 invalidate_rkey;
1001 } ex;
1002 u32 src_qp;
1003 u32 slid;
1004 int wc_flags;
1005 u16 pkey_index;
1006 u8 sl;
1007 u8 dlid_path_bits;
1008 u8 port_num; /* valid only for DR SMPs on switches */
1009 u8 smac[ETH_ALEN];
1010 u16 vlan_id;
1011 u8 network_hdr_type;
1012 };
1013
1014 enum ib_cq_notify_flags {
1015 IB_CQ_SOLICITED = 1 << 0,
1016 IB_CQ_NEXT_COMP = 1 << 1,
1017 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1018 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1019 };
1020
1021 enum ib_srq_type {
1022 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1023 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1024 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1025 };
1026
1027 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1028 {
1029 return srq_type == IB_SRQT_XRC ||
1030 srq_type == IB_SRQT_TM;
1031 }
1032
1033 enum ib_srq_attr_mask {
1034 IB_SRQ_MAX_WR = 1 << 0,
1035 IB_SRQ_LIMIT = 1 << 1,
1036 };
1037
1038 struct ib_srq_attr {
1039 u32 max_wr;
1040 u32 max_sge;
1041 u32 srq_limit;
1042 };
1043
1044 struct ib_srq_init_attr {
1045 void (*event_handler)(struct ib_event *, void *);
1046 void *srq_context;
1047 struct ib_srq_attr attr;
1048 enum ib_srq_type srq_type;
1049
1050 struct {
1051 struct ib_cq *cq;
1052 union {
1053 struct {
1054 struct ib_xrcd *xrcd;
1055 } xrc;
1056
1057 struct {
1058 u32 max_num_tags;
1059 } tag_matching;
1060 };
1061 } ext;
1062 };
1063
1064 struct ib_qp_cap {
1065 u32 max_send_wr;
1066 u32 max_recv_wr;
1067 u32 max_send_sge;
1068 u32 max_recv_sge;
1069 u32 max_inline_data;
1070
1071 /*
1072 * Maximum number of rdma_rw_ctx structures in flight at a time.
1073 * ib_create_qp() will calculate the right amount of neededed WRs
1074 * and MRs based on this.
1075 */
1076 u32 max_rdma_ctxs;
1077 };
1078
1079 enum ib_sig_type {
1080 IB_SIGNAL_ALL_WR,
1081 IB_SIGNAL_REQ_WR
1082 };
1083
1084 enum ib_qp_type {
1085 /*
1086 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1087 * here (and in that order) since the MAD layer uses them as
1088 * indices into a 2-entry table.
1089 */
1090 IB_QPT_SMI,
1091 IB_QPT_GSI,
1092
1093 IB_QPT_RC = IB_UVERBS_QPT_RC,
1094 IB_QPT_UC = IB_UVERBS_QPT_UC,
1095 IB_QPT_UD = IB_UVERBS_QPT_UD,
1096 IB_QPT_RAW_IPV6,
1097 IB_QPT_RAW_ETHERTYPE,
1098 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1099 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1100 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1101 IB_QPT_MAX,
1102 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1103 /* Reserve a range for qp types internal to the low level driver.
1104 * These qp types will not be visible at the IB core layer, so the
1105 * IB_QPT_MAX usages should not be affected in the core layer
1106 */
1107 IB_QPT_RESERVED1 = 0x1000,
1108 IB_QPT_RESERVED2,
1109 IB_QPT_RESERVED3,
1110 IB_QPT_RESERVED4,
1111 IB_QPT_RESERVED5,
1112 IB_QPT_RESERVED6,
1113 IB_QPT_RESERVED7,
1114 IB_QPT_RESERVED8,
1115 IB_QPT_RESERVED9,
1116 IB_QPT_RESERVED10,
1117 };
1118
1119 enum ib_qp_create_flags {
1120 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1121 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1122 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1123 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1124 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1125 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1126 IB_QP_CREATE_NETIF_QP = 1 << 5,
1127 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1128 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1129 IB_QP_CREATE_SCATTER_FCS =
1130 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1131 IB_QP_CREATE_CVLAN_STRIPPING =
1132 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1133 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1134 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1135 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1136 /* reserve bits 26-31 for low level drivers' internal use */
1137 IB_QP_CREATE_RESERVED_START = 1 << 26,
1138 IB_QP_CREATE_RESERVED_END = 1 << 31,
1139 };
1140
1141 /*
1142 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1143 * callback to destroy the passed in QP.
1144 */
1145
1146 struct ib_qp_init_attr {
1147 /* Consumer's event_handler callback must not block */
1148 void (*event_handler)(struct ib_event *, void *);
1149
1150 void *qp_context;
1151 struct ib_cq *send_cq;
1152 struct ib_cq *recv_cq;
1153 struct ib_srq *srq;
1154 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1155 struct ib_qp_cap cap;
1156 enum ib_sig_type sq_sig_type;
1157 enum ib_qp_type qp_type;
1158 u32 create_flags;
1159
1160 /*
1161 * Only needed for special QP types, or when using the RW API.
1162 */
1163 u8 port_num;
1164 struct ib_rwq_ind_table *rwq_ind_tbl;
1165 u32 source_qpn;
1166 };
1167
1168 struct ib_qp_open_attr {
1169 void (*event_handler)(struct ib_event *, void *);
1170 void *qp_context;
1171 u32 qp_num;
1172 enum ib_qp_type qp_type;
1173 };
1174
1175 enum ib_rnr_timeout {
1176 IB_RNR_TIMER_655_36 = 0,
1177 IB_RNR_TIMER_000_01 = 1,
1178 IB_RNR_TIMER_000_02 = 2,
1179 IB_RNR_TIMER_000_03 = 3,
1180 IB_RNR_TIMER_000_04 = 4,
1181 IB_RNR_TIMER_000_06 = 5,
1182 IB_RNR_TIMER_000_08 = 6,
1183 IB_RNR_TIMER_000_12 = 7,
1184 IB_RNR_TIMER_000_16 = 8,
1185 IB_RNR_TIMER_000_24 = 9,
1186 IB_RNR_TIMER_000_32 = 10,
1187 IB_RNR_TIMER_000_48 = 11,
1188 IB_RNR_TIMER_000_64 = 12,
1189 IB_RNR_TIMER_000_96 = 13,
1190 IB_RNR_TIMER_001_28 = 14,
1191 IB_RNR_TIMER_001_92 = 15,
1192 IB_RNR_TIMER_002_56 = 16,
1193 IB_RNR_TIMER_003_84 = 17,
1194 IB_RNR_TIMER_005_12 = 18,
1195 IB_RNR_TIMER_007_68 = 19,
1196 IB_RNR_TIMER_010_24 = 20,
1197 IB_RNR_TIMER_015_36 = 21,
1198 IB_RNR_TIMER_020_48 = 22,
1199 IB_RNR_TIMER_030_72 = 23,
1200 IB_RNR_TIMER_040_96 = 24,
1201 IB_RNR_TIMER_061_44 = 25,
1202 IB_RNR_TIMER_081_92 = 26,
1203 IB_RNR_TIMER_122_88 = 27,
1204 IB_RNR_TIMER_163_84 = 28,
1205 IB_RNR_TIMER_245_76 = 29,
1206 IB_RNR_TIMER_327_68 = 30,
1207 IB_RNR_TIMER_491_52 = 31
1208 };
1209
1210 enum ib_qp_attr_mask {
1211 IB_QP_STATE = 1,
1212 IB_QP_CUR_STATE = (1<<1),
1213 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1214 IB_QP_ACCESS_FLAGS = (1<<3),
1215 IB_QP_PKEY_INDEX = (1<<4),
1216 IB_QP_PORT = (1<<5),
1217 IB_QP_QKEY = (1<<6),
1218 IB_QP_AV = (1<<7),
1219 IB_QP_PATH_MTU = (1<<8),
1220 IB_QP_TIMEOUT = (1<<9),
1221 IB_QP_RETRY_CNT = (1<<10),
1222 IB_QP_RNR_RETRY = (1<<11),
1223 IB_QP_RQ_PSN = (1<<12),
1224 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1225 IB_QP_ALT_PATH = (1<<14),
1226 IB_QP_MIN_RNR_TIMER = (1<<15),
1227 IB_QP_SQ_PSN = (1<<16),
1228 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1229 IB_QP_PATH_MIG_STATE = (1<<18),
1230 IB_QP_CAP = (1<<19),
1231 IB_QP_DEST_QPN = (1<<20),
1232 IB_QP_RESERVED1 = (1<<21),
1233 IB_QP_RESERVED2 = (1<<22),
1234 IB_QP_RESERVED3 = (1<<23),
1235 IB_QP_RESERVED4 = (1<<24),
1236 IB_QP_RATE_LIMIT = (1<<25),
1237 };
1238
1239 enum ib_qp_state {
1240 IB_QPS_RESET,
1241 IB_QPS_INIT,
1242 IB_QPS_RTR,
1243 IB_QPS_RTS,
1244 IB_QPS_SQD,
1245 IB_QPS_SQE,
1246 IB_QPS_ERR
1247 };
1248
1249 enum ib_mig_state {
1250 IB_MIG_MIGRATED,
1251 IB_MIG_REARM,
1252 IB_MIG_ARMED
1253 };
1254
1255 enum ib_mw_type {
1256 IB_MW_TYPE_1 = 1,
1257 IB_MW_TYPE_2 = 2
1258 };
1259
1260 struct ib_qp_attr {
1261 enum ib_qp_state qp_state;
1262 enum ib_qp_state cur_qp_state;
1263 enum ib_mtu path_mtu;
1264 enum ib_mig_state path_mig_state;
1265 u32 qkey;
1266 u32 rq_psn;
1267 u32 sq_psn;
1268 u32 dest_qp_num;
1269 int qp_access_flags;
1270 struct ib_qp_cap cap;
1271 struct rdma_ah_attr ah_attr;
1272 struct rdma_ah_attr alt_ah_attr;
1273 u16 pkey_index;
1274 u16 alt_pkey_index;
1275 u8 en_sqd_async_notify;
1276 u8 sq_draining;
1277 u8 max_rd_atomic;
1278 u8 max_dest_rd_atomic;
1279 u8 min_rnr_timer;
1280 u8 port_num;
1281 u8 timeout;
1282 u8 retry_cnt;
1283 u8 rnr_retry;
1284 u8 alt_port_num;
1285 u8 alt_timeout;
1286 u32 rate_limit;
1287 struct net_device *xmit_slave;
1288 };
1289
1290 enum ib_wr_opcode {
1291 /* These are shared with userspace */
1292 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1293 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1294 IB_WR_SEND = IB_UVERBS_WR_SEND,
1295 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1296 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1297 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1298 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1299 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1300 IB_WR_LSO = IB_UVERBS_WR_TSO,
1301 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1302 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1303 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1304 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1305 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1306 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1307 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1308
1309 /* These are kernel only and can not be issued by userspace */
1310 IB_WR_REG_MR = 0x20,
1311 IB_WR_REG_MR_INTEGRITY,
1312
1313 /* reserve values for low level drivers' internal use.
1314 * These values will not be used at all in the ib core layer.
1315 */
1316 IB_WR_RESERVED1 = 0xf0,
1317 IB_WR_RESERVED2,
1318 IB_WR_RESERVED3,
1319 IB_WR_RESERVED4,
1320 IB_WR_RESERVED5,
1321 IB_WR_RESERVED6,
1322 IB_WR_RESERVED7,
1323 IB_WR_RESERVED8,
1324 IB_WR_RESERVED9,
1325 IB_WR_RESERVED10,
1326 };
1327
1328 enum ib_send_flags {
1329 IB_SEND_FENCE = 1,
1330 IB_SEND_SIGNALED = (1<<1),
1331 IB_SEND_SOLICITED = (1<<2),
1332 IB_SEND_INLINE = (1<<3),
1333 IB_SEND_IP_CSUM = (1<<4),
1334
1335 /* reserve bits 26-31 for low level drivers' internal use */
1336 IB_SEND_RESERVED_START = (1 << 26),
1337 IB_SEND_RESERVED_END = (1 << 31),
1338 };
1339
1340 struct ib_sge {
1341 u64 addr;
1342 u32 length;
1343 u32 lkey;
1344 };
1345
1346 struct ib_cqe {
1347 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1348 };
1349
1350 struct ib_send_wr {
1351 struct ib_send_wr *next;
1352 union {
1353 u64 wr_id;
1354 struct ib_cqe *wr_cqe;
1355 };
1356 struct ib_sge *sg_list;
1357 int num_sge;
1358 enum ib_wr_opcode opcode;
1359 int send_flags;
1360 union {
1361 __be32 imm_data;
1362 u32 invalidate_rkey;
1363 } ex;
1364 };
1365
1366 struct ib_rdma_wr {
1367 struct ib_send_wr wr;
1368 u64 remote_addr;
1369 u32 rkey;
1370 };
1371
1372 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1373 {
1374 return container_of(wr, struct ib_rdma_wr, wr);
1375 }
1376
1377 struct ib_atomic_wr {
1378 struct ib_send_wr wr;
1379 u64 remote_addr;
1380 u64 compare_add;
1381 u64 swap;
1382 u64 compare_add_mask;
1383 u64 swap_mask;
1384 u32 rkey;
1385 };
1386
1387 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1388 {
1389 return container_of(wr, struct ib_atomic_wr, wr);
1390 }
1391
1392 struct ib_ud_wr {
1393 struct ib_send_wr wr;
1394 struct ib_ah *ah;
1395 void *header;
1396 int hlen;
1397 int mss;
1398 u32 remote_qpn;
1399 u32 remote_qkey;
1400 u16 pkey_index; /* valid for GSI only */
1401 u8 port_num; /* valid for DR SMPs on switch only */
1402 };
1403
1404 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1405 {
1406 return container_of(wr, struct ib_ud_wr, wr);
1407 }
1408
1409 struct ib_reg_wr {
1410 struct ib_send_wr wr;
1411 struct ib_mr *mr;
1412 u32 key;
1413 int access;
1414 };
1415
1416 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1417 {
1418 return container_of(wr, struct ib_reg_wr, wr);
1419 }
1420
1421 struct ib_recv_wr {
1422 struct ib_recv_wr *next;
1423 union {
1424 u64 wr_id;
1425 struct ib_cqe *wr_cqe;
1426 };
1427 struct ib_sge *sg_list;
1428 int num_sge;
1429 };
1430
1431 enum ib_access_flags {
1432 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1433 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1434 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1435 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1436 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1437 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1438 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1439 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1440 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1441
1442 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1443 IB_ACCESS_SUPPORTED =
1444 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1445 };
1446
1447 /*
1448 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1449 * are hidden here instead of a uapi header!
1450 */
1451 enum ib_mr_rereg_flags {
1452 IB_MR_REREG_TRANS = 1,
1453 IB_MR_REREG_PD = (1<<1),
1454 IB_MR_REREG_ACCESS = (1<<2),
1455 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1456 };
1457
1458 struct ib_umem;
1459
1460 enum rdma_remove_reason {
1461 /*
1462 * Userspace requested uobject deletion or initial try
1463 * to remove uobject via cleanup. Call could fail
1464 */
1465 RDMA_REMOVE_DESTROY,
1466 /* Context deletion. This call should delete the actual object itself */
1467 RDMA_REMOVE_CLOSE,
1468 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1469 RDMA_REMOVE_DRIVER_REMOVE,
1470 /* uobj is being cleaned-up before being committed */
1471 RDMA_REMOVE_ABORT,
1472 };
1473
1474 struct ib_rdmacg_object {
1475 #ifdef CONFIG_CGROUP_RDMA
1476 struct rdma_cgroup *cg; /* owner rdma cgroup */
1477 #endif
1478 };
1479
1480 struct ib_ucontext {
1481 struct ib_device *device;
1482 struct ib_uverbs_file *ufile;
1483
1484 bool cleanup_retryable;
1485
1486 struct ib_rdmacg_object cg_obj;
1487 /*
1488 * Implementation details of the RDMA core, don't use in drivers:
1489 */
1490 struct rdma_restrack_entry res;
1491 struct xarray mmap_xa;
1492 };
1493
1494 struct ib_uobject {
1495 u64 user_handle; /* handle given to us by userspace */
1496 /* ufile & ucontext owning this object */
1497 struct ib_uverbs_file *ufile;
1498 /* FIXME, save memory: ufile->context == context */
1499 struct ib_ucontext *context; /* associated user context */
1500 void *object; /* containing object */
1501 struct list_head list; /* link to context's list */
1502 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1503 int id; /* index into kernel idr */
1504 struct kref ref;
1505 atomic_t usecnt; /* protects exclusive access */
1506 struct rcu_head rcu; /* kfree_rcu() overhead */
1507
1508 const struct uverbs_api_object *uapi_object;
1509 };
1510
1511 struct ib_udata {
1512 const void __user *inbuf;
1513 void __user *outbuf;
1514 size_t inlen;
1515 size_t outlen;
1516 };
1517
1518 struct ib_pd {
1519 u32 local_dma_lkey;
1520 u32 flags;
1521 struct ib_device *device;
1522 struct ib_uobject *uobject;
1523 atomic_t usecnt; /* count all resources */
1524
1525 u32 unsafe_global_rkey;
1526
1527 /*
1528 * Implementation details of the RDMA core, don't use in drivers:
1529 */
1530 struct ib_mr *__internal_mr;
1531 struct rdma_restrack_entry res;
1532 };
1533
1534 struct ib_xrcd {
1535 struct ib_device *device;
1536 atomic_t usecnt; /* count all exposed resources */
1537 struct inode *inode;
1538 struct rw_semaphore tgt_qps_rwsem;
1539 struct xarray tgt_qps;
1540 };
1541
1542 struct ib_ah {
1543 struct ib_device *device;
1544 struct ib_pd *pd;
1545 struct ib_uobject *uobject;
1546 const struct ib_gid_attr *sgid_attr;
1547 enum rdma_ah_attr_type type;
1548 };
1549
1550 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1551
1552 enum ib_poll_context {
1553 IB_POLL_SOFTIRQ, /* poll from softirq context */
1554 IB_POLL_WORKQUEUE, /* poll from workqueue */
1555 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1556 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1557
1558 IB_POLL_DIRECT, /* caller context, no hw completions */
1559 };
1560
1561 struct ib_cq {
1562 struct ib_device *device;
1563 struct ib_ucq_object *uobject;
1564 ib_comp_handler comp_handler;
1565 void (*event_handler)(struct ib_event *, void *);
1566 void *cq_context;
1567 int cqe;
1568 unsigned int cqe_used;
1569 atomic_t usecnt; /* count number of work queues */
1570 enum ib_poll_context poll_ctx;
1571 struct ib_wc *wc;
1572 struct list_head pool_entry;
1573 union {
1574 struct irq_poll iop;
1575 struct work_struct work;
1576 };
1577 struct workqueue_struct *comp_wq;
1578 struct dim *dim;
1579
1580 /* updated only by trace points */
1581 ktime_t timestamp;
1582 u8 interrupt:1;
1583 u8 shared:1;
1584 unsigned int comp_vector;
1585
1586 /*
1587 * Implementation details of the RDMA core, don't use in drivers:
1588 */
1589 struct rdma_restrack_entry res;
1590 };
1591
1592 struct ib_srq {
1593 struct ib_device *device;
1594 struct ib_pd *pd;
1595 struct ib_usrq_object *uobject;
1596 void (*event_handler)(struct ib_event *, void *);
1597 void *srq_context;
1598 enum ib_srq_type srq_type;
1599 atomic_t usecnt;
1600
1601 struct {
1602 struct ib_cq *cq;
1603 union {
1604 struct {
1605 struct ib_xrcd *xrcd;
1606 u32 srq_num;
1607 } xrc;
1608 };
1609 } ext;
1610 };
1611
1612 enum ib_raw_packet_caps {
1613 /* Strip cvlan from incoming packet and report it in the matching work
1614 * completion is supported.
1615 */
1616 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1617 /* Scatter FCS field of an incoming packet to host memory is supported.
1618 */
1619 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1620 /* Checksum offloads are supported (for both send and receive). */
1621 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1622 /* When a packet is received for an RQ with no receive WQEs, the
1623 * packet processing is delayed.
1624 */
1625 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1626 };
1627
1628 enum ib_wq_type {
1629 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1630 };
1631
1632 enum ib_wq_state {
1633 IB_WQS_RESET,
1634 IB_WQS_RDY,
1635 IB_WQS_ERR
1636 };
1637
1638 struct ib_wq {
1639 struct ib_device *device;
1640 struct ib_uwq_object *uobject;
1641 void *wq_context;
1642 void (*event_handler)(struct ib_event *, void *);
1643 struct ib_pd *pd;
1644 struct ib_cq *cq;
1645 u32 wq_num;
1646 enum ib_wq_state state;
1647 enum ib_wq_type wq_type;
1648 atomic_t usecnt;
1649 };
1650
1651 enum ib_wq_flags {
1652 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1653 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1654 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1655 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1656 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1657 };
1658
1659 struct ib_wq_init_attr {
1660 void *wq_context;
1661 enum ib_wq_type wq_type;
1662 u32 max_wr;
1663 u32 max_sge;
1664 struct ib_cq *cq;
1665 void (*event_handler)(struct ib_event *, void *);
1666 u32 create_flags; /* Use enum ib_wq_flags */
1667 };
1668
1669 enum ib_wq_attr_mask {
1670 IB_WQ_STATE = 1 << 0,
1671 IB_WQ_CUR_STATE = 1 << 1,
1672 IB_WQ_FLAGS = 1 << 2,
1673 };
1674
1675 struct ib_wq_attr {
1676 enum ib_wq_state wq_state;
1677 enum ib_wq_state curr_wq_state;
1678 u32 flags; /* Use enum ib_wq_flags */
1679 u32 flags_mask; /* Use enum ib_wq_flags */
1680 };
1681
1682 struct ib_rwq_ind_table {
1683 struct ib_device *device;
1684 struct ib_uobject *uobject;
1685 atomic_t usecnt;
1686 u32 ind_tbl_num;
1687 u32 log_ind_tbl_size;
1688 struct ib_wq **ind_tbl;
1689 };
1690
1691 struct ib_rwq_ind_table_init_attr {
1692 u32 log_ind_tbl_size;
1693 /* Each entry is a pointer to Receive Work Queue */
1694 struct ib_wq **ind_tbl;
1695 };
1696
1697 enum port_pkey_state {
1698 IB_PORT_PKEY_NOT_VALID = 0,
1699 IB_PORT_PKEY_VALID = 1,
1700 IB_PORT_PKEY_LISTED = 2,
1701 };
1702
1703 struct ib_qp_security;
1704
1705 struct ib_port_pkey {
1706 enum port_pkey_state state;
1707 u16 pkey_index;
1708 u8 port_num;
1709 struct list_head qp_list;
1710 struct list_head to_error_list;
1711 struct ib_qp_security *sec;
1712 };
1713
1714 struct ib_ports_pkeys {
1715 struct ib_port_pkey main;
1716 struct ib_port_pkey alt;
1717 };
1718
1719 struct ib_qp_security {
1720 struct ib_qp *qp;
1721 struct ib_device *dev;
1722 /* Hold this mutex when changing port and pkey settings. */
1723 struct mutex mutex;
1724 struct ib_ports_pkeys *ports_pkeys;
1725 /* A list of all open shared QP handles. Required to enforce security
1726 * properly for all users of a shared QP.
1727 */
1728 struct list_head shared_qp_list;
1729 void *security;
1730 bool destroying;
1731 atomic_t error_list_count;
1732 struct completion error_complete;
1733 int error_comps_pending;
1734 };
1735
1736 /*
1737 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1738 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1739 */
1740 struct ib_qp {
1741 struct ib_device *device;
1742 struct ib_pd *pd;
1743 struct ib_cq *send_cq;
1744 struct ib_cq *recv_cq;
1745 spinlock_t mr_lock;
1746 int mrs_used;
1747 struct list_head rdma_mrs;
1748 struct list_head sig_mrs;
1749 struct ib_srq *srq;
1750 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1751 struct list_head xrcd_list;
1752
1753 /* count times opened, mcast attaches, flow attaches */
1754 atomic_t usecnt;
1755 struct list_head open_list;
1756 struct ib_qp *real_qp;
1757 struct ib_uqp_object *uobject;
1758 void (*event_handler)(struct ib_event *, void *);
1759 void *qp_context;
1760 /* sgid_attrs associated with the AV's */
1761 const struct ib_gid_attr *av_sgid_attr;
1762 const struct ib_gid_attr *alt_path_sgid_attr;
1763 u32 qp_num;
1764 u32 max_write_sge;
1765 u32 max_read_sge;
1766 enum ib_qp_type qp_type;
1767 struct ib_rwq_ind_table *rwq_ind_tbl;
1768 struct ib_qp_security *qp_sec;
1769 u8 port;
1770
1771 bool integrity_en;
1772 /*
1773 * Implementation details of the RDMA core, don't use in drivers:
1774 */
1775 struct rdma_restrack_entry res;
1776
1777 /* The counter the qp is bind to */
1778 struct rdma_counter *counter;
1779 };
1780
1781 struct ib_dm {
1782 struct ib_device *device;
1783 u32 length;
1784 u32 flags;
1785 struct ib_uobject *uobject;
1786 atomic_t usecnt;
1787 };
1788
1789 struct ib_mr {
1790 struct ib_device *device;
1791 struct ib_pd *pd;
1792 u32 lkey;
1793 u32 rkey;
1794 u64 iova;
1795 u64 length;
1796 unsigned int page_size;
1797 enum ib_mr_type type;
1798 bool need_inval;
1799 union {
1800 struct ib_uobject *uobject; /* user */
1801 struct list_head qp_entry; /* FR */
1802 };
1803
1804 struct ib_dm *dm;
1805 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1806 /*
1807 * Implementation details of the RDMA core, don't use in drivers:
1808 */
1809 struct rdma_restrack_entry res;
1810 };
1811
1812 struct ib_mw {
1813 struct ib_device *device;
1814 struct ib_pd *pd;
1815 struct ib_uobject *uobject;
1816 u32 rkey;
1817 enum ib_mw_type type;
1818 };
1819
1820 /* Supported steering options */
1821 enum ib_flow_attr_type {
1822 /* steering according to rule specifications */
1823 IB_FLOW_ATTR_NORMAL = 0x0,
1824 /* default unicast and multicast rule -
1825 * receive all Eth traffic which isn't steered to any QP
1826 */
1827 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1828 /* default multicast rule -
1829 * receive all Eth multicast traffic which isn't steered to any QP
1830 */
1831 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1832 /* sniffer rule - receive all port traffic */
1833 IB_FLOW_ATTR_SNIFFER = 0x3
1834 };
1835
1836 /* Supported steering header types */
1837 enum ib_flow_spec_type {
1838 /* L2 headers*/
1839 IB_FLOW_SPEC_ETH = 0x20,
1840 IB_FLOW_SPEC_IB = 0x22,
1841 /* L3 header*/
1842 IB_FLOW_SPEC_IPV4 = 0x30,
1843 IB_FLOW_SPEC_IPV6 = 0x31,
1844 IB_FLOW_SPEC_ESP = 0x34,
1845 /* L4 headers*/
1846 IB_FLOW_SPEC_TCP = 0x40,
1847 IB_FLOW_SPEC_UDP = 0x41,
1848 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1849 IB_FLOW_SPEC_GRE = 0x51,
1850 IB_FLOW_SPEC_MPLS = 0x60,
1851 IB_FLOW_SPEC_INNER = 0x100,
1852 /* Actions */
1853 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1854 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1855 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1856 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1857 };
1858 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1859 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1860
1861 enum ib_flow_flags {
1862 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1863 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1864 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1865 };
1866
1867 struct ib_flow_eth_filter {
1868 u8 dst_mac[6];
1869 u8 src_mac[6];
1870 __be16 ether_type;
1871 __be16 vlan_tag;
1872 /* Must be last */
1873 u8 real_sz[];
1874 };
1875
1876 struct ib_flow_spec_eth {
1877 u32 type;
1878 u16 size;
1879 struct ib_flow_eth_filter val;
1880 struct ib_flow_eth_filter mask;
1881 };
1882
1883 struct ib_flow_ib_filter {
1884 __be16 dlid;
1885 __u8 sl;
1886 /* Must be last */
1887 u8 real_sz[];
1888 };
1889
1890 struct ib_flow_spec_ib {
1891 u32 type;
1892 u16 size;
1893 struct ib_flow_ib_filter val;
1894 struct ib_flow_ib_filter mask;
1895 };
1896
1897 /* IPv4 header flags */
1898 enum ib_ipv4_flags {
1899 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1900 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1901 last have this flag set */
1902 };
1903
1904 struct ib_flow_ipv4_filter {
1905 __be32 src_ip;
1906 __be32 dst_ip;
1907 u8 proto;
1908 u8 tos;
1909 u8 ttl;
1910 u8 flags;
1911 /* Must be last */
1912 u8 real_sz[];
1913 };
1914
1915 struct ib_flow_spec_ipv4 {
1916 u32 type;
1917 u16 size;
1918 struct ib_flow_ipv4_filter val;
1919 struct ib_flow_ipv4_filter mask;
1920 };
1921
1922 struct ib_flow_ipv6_filter {
1923 u8 src_ip[16];
1924 u8 dst_ip[16];
1925 __be32 flow_label;
1926 u8 next_hdr;
1927 u8 traffic_class;
1928 u8 hop_limit;
1929 /* Must be last */
1930 u8 real_sz[];
1931 };
1932
1933 struct ib_flow_spec_ipv6 {
1934 u32 type;
1935 u16 size;
1936 struct ib_flow_ipv6_filter val;
1937 struct ib_flow_ipv6_filter mask;
1938 };
1939
1940 struct ib_flow_tcp_udp_filter {
1941 __be16 dst_port;
1942 __be16 src_port;
1943 /* Must be last */
1944 u8 real_sz[];
1945 };
1946
1947 struct ib_flow_spec_tcp_udp {
1948 u32 type;
1949 u16 size;
1950 struct ib_flow_tcp_udp_filter val;
1951 struct ib_flow_tcp_udp_filter mask;
1952 };
1953
1954 struct ib_flow_tunnel_filter {
1955 __be32 tunnel_id;
1956 u8 real_sz[];
1957 };
1958
1959 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1960 * the tunnel_id from val has the vni value
1961 */
1962 struct ib_flow_spec_tunnel {
1963 u32 type;
1964 u16 size;
1965 struct ib_flow_tunnel_filter val;
1966 struct ib_flow_tunnel_filter mask;
1967 };
1968
1969 struct ib_flow_esp_filter {
1970 __be32 spi;
1971 __be32 seq;
1972 /* Must be last */
1973 u8 real_sz[];
1974 };
1975
1976 struct ib_flow_spec_esp {
1977 u32 type;
1978 u16 size;
1979 struct ib_flow_esp_filter val;
1980 struct ib_flow_esp_filter mask;
1981 };
1982
1983 struct ib_flow_gre_filter {
1984 __be16 c_ks_res0_ver;
1985 __be16 protocol;
1986 __be32 key;
1987 /* Must be last */
1988 u8 real_sz[];
1989 };
1990
1991 struct ib_flow_spec_gre {
1992 u32 type;
1993 u16 size;
1994 struct ib_flow_gre_filter val;
1995 struct ib_flow_gre_filter mask;
1996 };
1997
1998 struct ib_flow_mpls_filter {
1999 __be32 tag;
2000 /* Must be last */
2001 u8 real_sz[];
2002 };
2003
2004 struct ib_flow_spec_mpls {
2005 u32 type;
2006 u16 size;
2007 struct ib_flow_mpls_filter val;
2008 struct ib_flow_mpls_filter mask;
2009 };
2010
2011 struct ib_flow_spec_action_tag {
2012 enum ib_flow_spec_type type;
2013 u16 size;
2014 u32 tag_id;
2015 };
2016
2017 struct ib_flow_spec_action_drop {
2018 enum ib_flow_spec_type type;
2019 u16 size;
2020 };
2021
2022 struct ib_flow_spec_action_handle {
2023 enum ib_flow_spec_type type;
2024 u16 size;
2025 struct ib_flow_action *act;
2026 };
2027
2028 enum ib_counters_description {
2029 IB_COUNTER_PACKETS,
2030 IB_COUNTER_BYTES,
2031 };
2032
2033 struct ib_flow_spec_action_count {
2034 enum ib_flow_spec_type type;
2035 u16 size;
2036 struct ib_counters *counters;
2037 };
2038
2039 union ib_flow_spec {
2040 struct {
2041 u32 type;
2042 u16 size;
2043 };
2044 struct ib_flow_spec_eth eth;
2045 struct ib_flow_spec_ib ib;
2046 struct ib_flow_spec_ipv4 ipv4;
2047 struct ib_flow_spec_tcp_udp tcp_udp;
2048 struct ib_flow_spec_ipv6 ipv6;
2049 struct ib_flow_spec_tunnel tunnel;
2050 struct ib_flow_spec_esp esp;
2051 struct ib_flow_spec_gre gre;
2052 struct ib_flow_spec_mpls mpls;
2053 struct ib_flow_spec_action_tag flow_tag;
2054 struct ib_flow_spec_action_drop drop;
2055 struct ib_flow_spec_action_handle action;
2056 struct ib_flow_spec_action_count flow_count;
2057 };
2058
2059 struct ib_flow_attr {
2060 enum ib_flow_attr_type type;
2061 u16 size;
2062 u16 priority;
2063 u32 flags;
2064 u8 num_of_specs;
2065 u8 port;
2066 union ib_flow_spec flows[];
2067 };
2068
2069 struct ib_flow {
2070 struct ib_qp *qp;
2071 struct ib_device *device;
2072 struct ib_uobject *uobject;
2073 };
2074
2075 enum ib_flow_action_type {
2076 IB_FLOW_ACTION_UNSPECIFIED,
2077 IB_FLOW_ACTION_ESP = 1,
2078 };
2079
2080 struct ib_flow_action_attrs_esp_keymats {
2081 enum ib_uverbs_flow_action_esp_keymat protocol;
2082 union {
2083 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2084 } keymat;
2085 };
2086
2087 struct ib_flow_action_attrs_esp_replays {
2088 enum ib_uverbs_flow_action_esp_replay protocol;
2089 union {
2090 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2091 } replay;
2092 };
2093
2094 enum ib_flow_action_attrs_esp_flags {
2095 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2096 * This is done in order to share the same flags between user-space and
2097 * kernel and spare an unnecessary translation.
2098 */
2099
2100 /* Kernel flags */
2101 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2102 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2103 };
2104
2105 struct ib_flow_spec_list {
2106 struct ib_flow_spec_list *next;
2107 union ib_flow_spec spec;
2108 };
2109
2110 struct ib_flow_action_attrs_esp {
2111 struct ib_flow_action_attrs_esp_keymats *keymat;
2112 struct ib_flow_action_attrs_esp_replays *replay;
2113 struct ib_flow_spec_list *encap;
2114 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2115 * Value of 0 is a valid value.
2116 */
2117 u32 esn;
2118 u32 spi;
2119 u32 seq;
2120 u32 tfc_pad;
2121 /* Use enum ib_flow_action_attrs_esp_flags */
2122 u64 flags;
2123 u64 hard_limit_pkts;
2124 };
2125
2126 struct ib_flow_action {
2127 struct ib_device *device;
2128 struct ib_uobject *uobject;
2129 enum ib_flow_action_type type;
2130 atomic_t usecnt;
2131 };
2132
2133 struct ib_mad;
2134 struct ib_grh;
2135
2136 enum ib_process_mad_flags {
2137 IB_MAD_IGNORE_MKEY = 1,
2138 IB_MAD_IGNORE_BKEY = 2,
2139 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2140 };
2141
2142 enum ib_mad_result {
2143 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2144 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2145 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2146 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2147 };
2148
2149 struct ib_port_cache {
2150 u64 subnet_prefix;
2151 struct ib_pkey_cache *pkey;
2152 struct ib_gid_table *gid;
2153 u8 lmc;
2154 enum ib_port_state port_state;
2155 };
2156
2157 struct ib_port_immutable {
2158 int pkey_tbl_len;
2159 int gid_tbl_len;
2160 u32 core_cap_flags;
2161 u32 max_mad_size;
2162 };
2163
2164 struct ib_port_data {
2165 struct ib_device *ib_dev;
2166
2167 struct ib_port_immutable immutable;
2168
2169 spinlock_t pkey_list_lock;
2170 struct list_head pkey_list;
2171
2172 struct ib_port_cache cache;
2173
2174 spinlock_t netdev_lock;
2175 struct net_device __rcu *netdev;
2176 struct hlist_node ndev_hash_link;
2177 struct rdma_port_counter port_counter;
2178 struct rdma_hw_stats *hw_stats;
2179 };
2180
2181 /* rdma netdev type - specifies protocol type */
2182 enum rdma_netdev_t {
2183 RDMA_NETDEV_OPA_VNIC,
2184 RDMA_NETDEV_IPOIB,
2185 };
2186
2187 /**
2188 * struct rdma_netdev - rdma netdev
2189 * For cases where netstack interfacing is required.
2190 */
2191 struct rdma_netdev {
2192 void *clnt_priv;
2193 struct ib_device *hca;
2194 u8 port_num;
2195 int mtu;
2196
2197 /*
2198 * cleanup function must be specified.
2199 * FIXME: This is only used for OPA_VNIC and that usage should be
2200 * removed too.
2201 */
2202 void (*free_rdma_netdev)(struct net_device *netdev);
2203
2204 /* control functions */
2205 void (*set_id)(struct net_device *netdev, int id);
2206 /* send packet */
2207 int (*send)(struct net_device *dev, struct sk_buff *skb,
2208 struct ib_ah *address, u32 dqpn);
2209 /* multicast */
2210 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2211 union ib_gid *gid, u16 mlid,
2212 int set_qkey, u32 qkey);
2213 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2214 union ib_gid *gid, u16 mlid);
2215 };
2216
2217 struct rdma_netdev_alloc_params {
2218 size_t sizeof_priv;
2219 unsigned int txqs;
2220 unsigned int rxqs;
2221 void *param;
2222
2223 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2224 struct net_device *netdev, void *param);
2225 };
2226
2227 struct ib_odp_counters {
2228 atomic64_t faults;
2229 atomic64_t invalidations;
2230 atomic64_t prefetch;
2231 };
2232
2233 struct ib_counters {
2234 struct ib_device *device;
2235 struct ib_uobject *uobject;
2236 /* num of objects attached */
2237 atomic_t usecnt;
2238 };
2239
2240 struct ib_counters_read_attr {
2241 u64 *counters_buff;
2242 u32 ncounters;
2243 u32 flags; /* use enum ib_read_counters_flags */
2244 };
2245
2246 struct uverbs_attr_bundle;
2247 struct iw_cm_id;
2248 struct iw_cm_conn_param;
2249
2250 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2251 .size_##ib_struct = \
2252 (sizeof(struct drv_struct) + \
2253 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2254 BUILD_BUG_ON_ZERO( \
2255 !__same_type(((struct drv_struct *)NULL)->member, \
2256 struct ib_struct)))
2257
2258 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2259 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2260
2261 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2262 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2263
2264 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2265
2266 struct rdma_user_mmap_entry {
2267 struct kref ref;
2268 struct ib_ucontext *ucontext;
2269 unsigned long start_pgoff;
2270 size_t npages;
2271 bool driver_removed;
2272 };
2273
2274 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2275 static inline u64
2276 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2277 {
2278 return (u64)entry->start_pgoff << PAGE_SHIFT;
2279 }
2280
2281 /**
2282 * struct ib_device_ops - InfiniBand device operations
2283 * This structure defines all the InfiniBand device operations, providers will
2284 * need to define the supported operations, otherwise they will be set to null.
2285 */
2286 struct ib_device_ops {
2287 struct module *owner;
2288 enum rdma_driver_id driver_id;
2289 u32 uverbs_abi_ver;
2290 unsigned int uverbs_no_driver_id_binding:1;
2291
2292 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2293 const struct ib_send_wr **bad_send_wr);
2294 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2295 const struct ib_recv_wr **bad_recv_wr);
2296 void (*drain_rq)(struct ib_qp *qp);
2297 void (*drain_sq)(struct ib_qp *qp);
2298 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2299 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2300 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2301 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2302 int (*post_srq_recv)(struct ib_srq *srq,
2303 const struct ib_recv_wr *recv_wr,
2304 const struct ib_recv_wr **bad_recv_wr);
2305 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2306 u8 port_num, const struct ib_wc *in_wc,
2307 const struct ib_grh *in_grh,
2308 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2309 size_t *out_mad_size, u16 *out_mad_pkey_index);
2310 int (*query_device)(struct ib_device *device,
2311 struct ib_device_attr *device_attr,
2312 struct ib_udata *udata);
2313 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2314 struct ib_device_modify *device_modify);
2315 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2316 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2317 int comp_vector);
2318 int (*query_port)(struct ib_device *device, u8 port_num,
2319 struct ib_port_attr *port_attr);
2320 int (*modify_port)(struct ib_device *device, u8 port_num,
2321 int port_modify_mask,
2322 struct ib_port_modify *port_modify);
2323 /**
2324 * The following mandatory functions are used only at device
2325 * registration. Keep functions such as these at the end of this
2326 * structure to avoid cache line misses when accessing struct ib_device
2327 * in fast paths.
2328 */
2329 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2330 struct ib_port_immutable *immutable);
2331 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2332 u8 port_num);
2333 /**
2334 * When calling get_netdev, the HW vendor's driver should return the
2335 * net device of device @device at port @port_num or NULL if such
2336 * a net device doesn't exist. The vendor driver should call dev_hold
2337 * on this net device. The HW vendor's device driver must guarantee
2338 * that this function returns NULL before the net device has finished
2339 * NETDEV_UNREGISTER state.
2340 */
2341 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2342 /**
2343 * rdma netdev operation
2344 *
2345 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2346 * must return -EOPNOTSUPP if it doesn't support the specified type.
2347 */
2348 struct net_device *(*alloc_rdma_netdev)(
2349 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2350 const char *name, unsigned char name_assign_type,
2351 void (*setup)(struct net_device *));
2352
2353 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2354 enum rdma_netdev_t type,
2355 struct rdma_netdev_alloc_params *params);
2356 /**
2357 * query_gid should be return GID value for @device, when @port_num
2358 * link layer is either IB or iWarp. It is no-op if @port_num port
2359 * is RoCE link layer.
2360 */
2361 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2362 union ib_gid *gid);
2363 /**
2364 * When calling add_gid, the HW vendor's driver should add the gid
2365 * of device of port at gid index available at @attr. Meta-info of
2366 * that gid (for example, the network device related to this gid) is
2367 * available at @attr. @context allows the HW vendor driver to store
2368 * extra information together with a GID entry. The HW vendor driver may
2369 * allocate memory to contain this information and store it in @context
2370 * when a new GID entry is written to. Params are consistent until the
2371 * next call of add_gid or delete_gid. The function should return 0 on
2372 * success or error otherwise. The function could be called
2373 * concurrently for different ports. This function is only called when
2374 * roce_gid_table is used.
2375 */
2376 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2377 /**
2378 * When calling del_gid, the HW vendor's driver should delete the
2379 * gid of device @device at gid index gid_index of port port_num
2380 * available in @attr.
2381 * Upon the deletion of a GID entry, the HW vendor must free any
2382 * allocated memory. The caller will clear @context afterwards.
2383 * This function is only called when roce_gid_table is used.
2384 */
2385 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2386 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2387 u16 *pkey);
2388 int (*alloc_ucontext)(struct ib_ucontext *context,
2389 struct ib_udata *udata);
2390 void (*dealloc_ucontext)(struct ib_ucontext *context);
2391 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2392 /**
2393 * This will be called once refcount of an entry in mmap_xa reaches
2394 * zero. The type of the memory that was mapped may differ between
2395 * entries and is opaque to the rdma_user_mmap interface.
2396 * Therefore needs to be implemented by the driver in mmap_free.
2397 */
2398 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2399 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2400 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2401 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2402 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2403 struct ib_udata *udata);
2404 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2405 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2406 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2407 int (*create_srq)(struct ib_srq *srq,
2408 struct ib_srq_init_attr *srq_init_attr,
2409 struct ib_udata *udata);
2410 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2411 enum ib_srq_attr_mask srq_attr_mask,
2412 struct ib_udata *udata);
2413 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2414 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2415 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2416 struct ib_qp_init_attr *qp_init_attr,
2417 struct ib_udata *udata);
2418 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2419 int qp_attr_mask, struct ib_udata *udata);
2420 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2421 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2422 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2423 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2424 struct ib_udata *udata);
2425 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2426 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2427 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2428 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2429 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2430 u64 virt_addr, int mr_access_flags,
2431 struct ib_udata *udata);
2432 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2433 u64 virt_addr, int mr_access_flags,
2434 struct ib_pd *pd, struct ib_udata *udata);
2435 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2436 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2437 u32 max_num_sg);
2438 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2439 u32 max_num_data_sg,
2440 u32 max_num_meta_sg);
2441 int (*advise_mr)(struct ib_pd *pd,
2442 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2443 struct ib_sge *sg_list, u32 num_sge,
2444 struct uverbs_attr_bundle *attrs);
2445 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2446 unsigned int *sg_offset);
2447 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2448 struct ib_mr_status *mr_status);
2449 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2450 int (*dealloc_mw)(struct ib_mw *mw);
2451 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2452 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2453 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2454 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2455 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2456 struct ib_flow_attr *flow_attr,
2457 struct ib_udata *udata);
2458 int (*destroy_flow)(struct ib_flow *flow_id);
2459 struct ib_flow_action *(*create_flow_action_esp)(
2460 struct ib_device *device,
2461 const struct ib_flow_action_attrs_esp *attr,
2462 struct uverbs_attr_bundle *attrs);
2463 int (*destroy_flow_action)(struct ib_flow_action *action);
2464 int (*modify_flow_action_esp)(
2465 struct ib_flow_action *action,
2466 const struct ib_flow_action_attrs_esp *attr,
2467 struct uverbs_attr_bundle *attrs);
2468 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2469 int state);
2470 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2471 struct ifla_vf_info *ivf);
2472 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2473 struct ifla_vf_stats *stats);
2474 int (*get_vf_guid)(struct ib_device *device, int vf, u8 port,
2475 struct ifla_vf_guid *node_guid,
2476 struct ifla_vf_guid *port_guid);
2477 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2478 int type);
2479 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2480 struct ib_wq_init_attr *init_attr,
2481 struct ib_udata *udata);
2482 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2483 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2484 u32 wq_attr_mask, struct ib_udata *udata);
2485 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2486 struct ib_rwq_ind_table_init_attr *init_attr,
2487 struct ib_udata *udata);
2488 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2489 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2490 struct ib_ucontext *context,
2491 struct ib_dm_alloc_attr *attr,
2492 struct uverbs_attr_bundle *attrs);
2493 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2494 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2495 struct ib_dm_mr_attr *attr,
2496 struct uverbs_attr_bundle *attrs);
2497 int (*create_counters)(struct ib_counters *counters,
2498 struct uverbs_attr_bundle *attrs);
2499 int (*destroy_counters)(struct ib_counters *counters);
2500 int (*read_counters)(struct ib_counters *counters,
2501 struct ib_counters_read_attr *counters_read_attr,
2502 struct uverbs_attr_bundle *attrs);
2503 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2504 int data_sg_nents, unsigned int *data_sg_offset,
2505 struct scatterlist *meta_sg, int meta_sg_nents,
2506 unsigned int *meta_sg_offset);
2507
2508 /**
2509 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2510 * driver initialized data. The struct is kfree()'ed by the sysfs
2511 * core when the device is removed. A lifespan of -1 in the return
2512 * struct tells the core to set a default lifespan.
2513 */
2514 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2515 u8 port_num);
2516 /**
2517 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2518 * @index - The index in the value array we wish to have updated, or
2519 * num_counters if we want all stats updated
2520 * Return codes -
2521 * < 0 - Error, no counters updated
2522 * index - Updated the single counter pointed to by index
2523 * num_counters - Updated all counters (will reset the timestamp
2524 * and prevent further calls for lifespan milliseconds)
2525 * Drivers are allowed to update all counters in leiu of just the
2526 * one given in index at their option
2527 */
2528 int (*get_hw_stats)(struct ib_device *device,
2529 struct rdma_hw_stats *stats, u8 port, int index);
2530 /*
2531 * This function is called once for each port when a ib device is
2532 * registered.
2533 */
2534 int (*init_port)(struct ib_device *device, u8 port_num,
2535 struct kobject *port_sysfs);
2536 /**
2537 * Allows rdma drivers to add their own restrack attributes.
2538 */
2539 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2540 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2541 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2542 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2543 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2544 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2545 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2546
2547 /* Device lifecycle callbacks */
2548 /*
2549 * Called after the device becomes registered, before clients are
2550 * attached
2551 */
2552 int (*enable_driver)(struct ib_device *dev);
2553 /*
2554 * This is called as part of ib_dealloc_device().
2555 */
2556 void (*dealloc_driver)(struct ib_device *dev);
2557
2558 /* iWarp CM callbacks */
2559 void (*iw_add_ref)(struct ib_qp *qp);
2560 void (*iw_rem_ref)(struct ib_qp *qp);
2561 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2562 int (*iw_connect)(struct iw_cm_id *cm_id,
2563 struct iw_cm_conn_param *conn_param);
2564 int (*iw_accept)(struct iw_cm_id *cm_id,
2565 struct iw_cm_conn_param *conn_param);
2566 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2567 u8 pdata_len);
2568 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2569 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2570 /**
2571 * counter_bind_qp - Bind a QP to a counter.
2572 * @counter - The counter to be bound. If counter->id is zero then
2573 * the driver needs to allocate a new counter and set counter->id
2574 */
2575 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2576 /**
2577 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2578 * counter and bind it onto the default one
2579 */
2580 int (*counter_unbind_qp)(struct ib_qp *qp);
2581 /**
2582 * counter_dealloc -De-allocate the hw counter
2583 */
2584 int (*counter_dealloc)(struct rdma_counter *counter);
2585 /**
2586 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2587 * the driver initialized data.
2588 */
2589 struct rdma_hw_stats *(*counter_alloc_stats)(
2590 struct rdma_counter *counter);
2591 /**
2592 * counter_update_stats - Query the stats value of this counter
2593 */
2594 int (*counter_update_stats)(struct rdma_counter *counter);
2595
2596 /**
2597 * Allows rdma drivers to add their own restrack attributes
2598 * dumped via 'rdma stat' iproute2 command.
2599 */
2600 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2601
2602 /* query driver for its ucontext properties */
2603 int (*query_ucontext)(struct ib_ucontext *context,
2604 struct uverbs_attr_bundle *attrs);
2605
2606 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2607 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2608 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2609 DECLARE_RDMA_OBJ_SIZE(ib_mw);
2610 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2611 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2612 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2613 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2614 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2615 };
2616
2617 struct ib_core_device {
2618 /* device must be the first element in structure until,
2619 * union of ib_core_device and device exists in ib_device.
2620 */
2621 struct device dev;
2622 possible_net_t rdma_net;
2623 struct kobject *ports_kobj;
2624 struct list_head port_list;
2625 struct ib_device *owner; /* reach back to owner ib_device */
2626 };
2627
2628 struct rdma_restrack_root;
2629 struct ib_device {
2630 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2631 struct device *dma_device;
2632 struct ib_device_ops ops;
2633 char name[IB_DEVICE_NAME_MAX];
2634 struct rcu_head rcu_head;
2635
2636 struct list_head event_handler_list;
2637 /* Protects event_handler_list */
2638 struct rw_semaphore event_handler_rwsem;
2639
2640 /* Protects QP's event_handler calls and open_qp list */
2641 spinlock_t qp_open_list_lock;
2642
2643 struct rw_semaphore client_data_rwsem;
2644 struct xarray client_data;
2645 struct mutex unregistration_lock;
2646
2647 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2648 rwlock_t cache_lock;
2649 /**
2650 * port_data is indexed by port number
2651 */
2652 struct ib_port_data *port_data;
2653
2654 int num_comp_vectors;
2655
2656 union {
2657 struct device dev;
2658 struct ib_core_device coredev;
2659 };
2660
2661 /* First group for device attributes,
2662 * Second group for driver provided attributes (optional).
2663 * It is NULL terminated array.
2664 */
2665 const struct attribute_group *groups[3];
2666
2667 u64 uverbs_cmd_mask;
2668 u64 uverbs_ex_cmd_mask;
2669
2670 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2671 __be64 node_guid;
2672 u32 local_dma_lkey;
2673 u16 is_switch:1;
2674 /* Indicates kernel verbs support, should not be used in drivers */
2675 u16 kverbs_provider:1;
2676 /* CQ adaptive moderation (RDMA DIM) */
2677 u16 use_cq_dim:1;
2678 u8 node_type;
2679 u8 phys_port_cnt;
2680 struct ib_device_attr attrs;
2681 struct attribute_group *hw_stats_ag;
2682 struct rdma_hw_stats *hw_stats;
2683
2684 #ifdef CONFIG_CGROUP_RDMA
2685 struct rdmacg_device cg_device;
2686 #endif
2687
2688 u32 index;
2689
2690 spinlock_t cq_pools_lock;
2691 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2692
2693 struct rdma_restrack_root *res;
2694
2695 const struct uapi_definition *driver_def;
2696
2697 /*
2698 * Positive refcount indicates that the device is currently
2699 * registered and cannot be unregistered.
2700 */
2701 refcount_t refcount;
2702 struct completion unreg_completion;
2703 struct work_struct unregistration_work;
2704
2705 const struct rdma_link_ops *link_ops;
2706
2707 /* Protects compat_devs xarray modifications */
2708 struct mutex compat_devs_mutex;
2709 /* Maintains compat devices for each net namespace */
2710 struct xarray compat_devs;
2711
2712 /* Used by iWarp CM */
2713 char iw_ifname[IFNAMSIZ];
2714 u32 iw_driver_flags;
2715 u32 lag_flags;
2716 };
2717
2718 struct ib_client_nl_info;
2719 struct ib_client {
2720 const char *name;
2721 int (*add)(struct ib_device *ibdev);
2722 void (*remove)(struct ib_device *, void *client_data);
2723 void (*rename)(struct ib_device *dev, void *client_data);
2724 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2725 struct ib_client_nl_info *res);
2726 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2727
2728 /* Returns the net_dev belonging to this ib_client and matching the
2729 * given parameters.
2730 * @dev: An RDMA device that the net_dev use for communication.
2731 * @port: A physical port number on the RDMA device.
2732 * @pkey: P_Key that the net_dev uses if applicable.
2733 * @gid: A GID that the net_dev uses to communicate.
2734 * @addr: An IP address the net_dev is configured with.
2735 * @client_data: The device's client data set by ib_set_client_data().
2736 *
2737 * An ib_client that implements a net_dev on top of RDMA devices
2738 * (such as IP over IB) should implement this callback, allowing the
2739 * rdma_cm module to find the right net_dev for a given request.
2740 *
2741 * The caller is responsible for calling dev_put on the returned
2742 * netdev. */
2743 struct net_device *(*get_net_dev_by_params)(
2744 struct ib_device *dev,
2745 u8 port,
2746 u16 pkey,
2747 const union ib_gid *gid,
2748 const struct sockaddr *addr,
2749 void *client_data);
2750
2751 refcount_t uses;
2752 struct completion uses_zero;
2753 u32 client_id;
2754
2755 /* kverbs are not required by the client */
2756 u8 no_kverbs_req:1;
2757 };
2758
2759 /*
2760 * IB block DMA iterator
2761 *
2762 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2763 * to a HW supported page size.
2764 */
2765 struct ib_block_iter {
2766 /* internal states */
2767 struct scatterlist *__sg; /* sg holding the current aligned block */
2768 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2769 unsigned int __sg_nents; /* number of SG entries */
2770 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2771 unsigned int __pg_bit; /* alignment of current block */
2772 };
2773
2774 struct ib_device *_ib_alloc_device(size_t size);
2775 #define ib_alloc_device(drv_struct, member) \
2776 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2777 BUILD_BUG_ON_ZERO(offsetof( \
2778 struct drv_struct, member))), \
2779 struct drv_struct, member)
2780
2781 void ib_dealloc_device(struct ib_device *device);
2782
2783 void ib_get_device_fw_str(struct ib_device *device, char *str);
2784
2785 int ib_register_device(struct ib_device *device, const char *name,
2786 struct device *dma_device);
2787 void ib_unregister_device(struct ib_device *device);
2788 void ib_unregister_driver(enum rdma_driver_id driver_id);
2789 void ib_unregister_device_and_put(struct ib_device *device);
2790 void ib_unregister_device_queued(struct ib_device *ib_dev);
2791
2792 int ib_register_client (struct ib_client *client);
2793 void ib_unregister_client(struct ib_client *client);
2794
2795 void __rdma_block_iter_start(struct ib_block_iter *biter,
2796 struct scatterlist *sglist,
2797 unsigned int nents,
2798 unsigned long pgsz);
2799 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2800
2801 /**
2802 * rdma_block_iter_dma_address - get the aligned dma address of the current
2803 * block held by the block iterator.
2804 * @biter: block iterator holding the memory block
2805 */
2806 static inline dma_addr_t
2807 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2808 {
2809 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2810 }
2811
2812 /**
2813 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2814 * @sglist: sglist to iterate over
2815 * @biter: block iterator holding the memory block
2816 * @nents: maximum number of sg entries to iterate over
2817 * @pgsz: best HW supported page size to use
2818 *
2819 * Callers may use rdma_block_iter_dma_address() to get each
2820 * blocks aligned DMA address.
2821 */
2822 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2823 for (__rdma_block_iter_start(biter, sglist, nents, \
2824 pgsz); \
2825 __rdma_block_iter_next(biter);)
2826
2827 /**
2828 * ib_get_client_data - Get IB client context
2829 * @device:Device to get context for
2830 * @client:Client to get context for
2831 *
2832 * ib_get_client_data() returns the client context data set with
2833 * ib_set_client_data(). This can only be called while the client is
2834 * registered to the device, once the ib_client remove() callback returns this
2835 * cannot be called.
2836 */
2837 static inline void *ib_get_client_data(struct ib_device *device,
2838 struct ib_client *client)
2839 {
2840 return xa_load(&device->client_data, client->client_id);
2841 }
2842 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2843 void *data);
2844 void ib_set_device_ops(struct ib_device *device,
2845 const struct ib_device_ops *ops);
2846
2847 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2848 unsigned long pfn, unsigned long size, pgprot_t prot,
2849 struct rdma_user_mmap_entry *entry);
2850 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2851 struct rdma_user_mmap_entry *entry,
2852 size_t length);
2853 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2854 struct rdma_user_mmap_entry *entry,
2855 size_t length, u32 min_pgoff,
2856 u32 max_pgoff);
2857
2858 struct rdma_user_mmap_entry *
2859 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2860 unsigned long pgoff);
2861 struct rdma_user_mmap_entry *
2862 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2863 struct vm_area_struct *vma);
2864 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2865
2866 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2867
2868 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2869 {
2870 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2871 }
2872
2873 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2874 {
2875 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2876 }
2877
2878 static inline bool ib_is_buffer_cleared(const void __user *p,
2879 size_t len)
2880 {
2881 bool ret;
2882 u8 *buf;
2883
2884 if (len > USHRT_MAX)
2885 return false;
2886
2887 buf = memdup_user(p, len);
2888 if (IS_ERR(buf))
2889 return false;
2890
2891 ret = !memchr_inv(buf, 0, len);
2892 kfree(buf);
2893 return ret;
2894 }
2895
2896 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2897 size_t offset,
2898 size_t len)
2899 {
2900 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2901 }
2902
2903 /**
2904 * ib_is_destroy_retryable - Check whether the uobject destruction
2905 * is retryable.
2906 * @ret: The initial destruction return code
2907 * @why: remove reason
2908 * @uobj: The uobject that is destroyed
2909 *
2910 * This function is a helper function that IB layer and low-level drivers
2911 * can use to consider whether the destruction of the given uobject is
2912 * retry-able.
2913 * It checks the original return code, if it wasn't success the destruction
2914 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2915 * the remove reason. (i.e. why).
2916 * Must be called with the object locked for destroy.
2917 */
2918 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2919 struct ib_uobject *uobj)
2920 {
2921 return ret && (why == RDMA_REMOVE_DESTROY ||
2922 uobj->context->cleanup_retryable);
2923 }
2924
2925 /**
2926 * ib_destroy_usecnt - Called during destruction to check the usecnt
2927 * @usecnt: The usecnt atomic
2928 * @why: remove reason
2929 * @uobj: The uobject that is destroyed
2930 *
2931 * Non-zero usecnts will block destruction unless destruction was triggered by
2932 * a ucontext cleanup.
2933 */
2934 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2935 enum rdma_remove_reason why,
2936 struct ib_uobject *uobj)
2937 {
2938 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2939 return -EBUSY;
2940 return 0;
2941 }
2942
2943 /**
2944 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2945 * contains all required attributes and no attributes not allowed for
2946 * the given QP state transition.
2947 * @cur_state: Current QP state
2948 * @next_state: Next QP state
2949 * @type: QP type
2950 * @mask: Mask of supplied QP attributes
2951 *
2952 * This function is a helper function that a low-level driver's
2953 * modify_qp method can use to validate the consumer's input. It
2954 * checks that cur_state and next_state are valid QP states, that a
2955 * transition from cur_state to next_state is allowed by the IB spec,
2956 * and that the attribute mask supplied is allowed for the transition.
2957 */
2958 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2959 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2960
2961 void ib_register_event_handler(struct ib_event_handler *event_handler);
2962 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2963 void ib_dispatch_event(const struct ib_event *event);
2964
2965 int ib_query_port(struct ib_device *device,
2966 u8 port_num, struct ib_port_attr *port_attr);
2967
2968 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2969 u8 port_num);
2970
2971 /**
2972 * rdma_cap_ib_switch - Check if the device is IB switch
2973 * @device: Device to check
2974 *
2975 * Device driver is responsible for setting is_switch bit on
2976 * in ib_device structure at init time.
2977 *
2978 * Return: true if the device is IB switch.
2979 */
2980 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2981 {
2982 return device->is_switch;
2983 }
2984
2985 /**
2986 * rdma_start_port - Return the first valid port number for the device
2987 * specified
2988 *
2989 * @device: Device to be checked
2990 *
2991 * Return start port number
2992 */
2993 static inline u8 rdma_start_port(const struct ib_device *device)
2994 {
2995 return rdma_cap_ib_switch(device) ? 0 : 1;
2996 }
2997
2998 /**
2999 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3000 * @device - The struct ib_device * to iterate over
3001 * @iter - The unsigned int to store the port number
3002 */
3003 #define rdma_for_each_port(device, iter) \
3004 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
3005 unsigned int, iter))); \
3006 iter <= rdma_end_port(device); (iter)++)
3007
3008 /**
3009 * rdma_end_port - Return the last valid port number for the device
3010 * specified
3011 *
3012 * @device: Device to be checked
3013 *
3014 * Return last port number
3015 */
3016 static inline u8 rdma_end_port(const struct ib_device *device)
3017 {
3018 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3019 }
3020
3021 static inline int rdma_is_port_valid(const struct ib_device *device,
3022 unsigned int port)
3023 {
3024 return (port >= rdma_start_port(device) &&
3025 port <= rdma_end_port(device));
3026 }
3027
3028 static inline bool rdma_is_grh_required(const struct ib_device *device,
3029 u8 port_num)
3030 {
3031 return device->port_data[port_num].immutable.core_cap_flags &
3032 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3033 }
3034
3035 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
3036 {
3037 return device->port_data[port_num].immutable.core_cap_flags &
3038 RDMA_CORE_CAP_PROT_IB;
3039 }
3040
3041 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
3042 {
3043 return device->port_data[port_num].immutable.core_cap_flags &
3044 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3045 }
3046
3047 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
3048 {
3049 return device->port_data[port_num].immutable.core_cap_flags &
3050 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3051 }
3052
3053 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
3054 {
3055 return device->port_data[port_num].immutable.core_cap_flags &
3056 RDMA_CORE_CAP_PROT_ROCE;
3057 }
3058
3059 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
3060 {
3061 return device->port_data[port_num].immutable.core_cap_flags &
3062 RDMA_CORE_CAP_PROT_IWARP;
3063 }
3064
3065 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
3066 {
3067 return rdma_protocol_ib(device, port_num) ||
3068 rdma_protocol_roce(device, port_num);
3069 }
3070
3071 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
3072 {
3073 return device->port_data[port_num].immutable.core_cap_flags &
3074 RDMA_CORE_CAP_PROT_RAW_PACKET;
3075 }
3076
3077 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
3078 {
3079 return device->port_data[port_num].immutable.core_cap_flags &
3080 RDMA_CORE_CAP_PROT_USNIC;
3081 }
3082
3083 /**
3084 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3085 * Management Datagrams.
3086 * @device: Device to check
3087 * @port_num: Port number to check
3088 *
3089 * Management Datagrams (MAD) are a required part of the InfiniBand
3090 * specification and are supported on all InfiniBand devices. A slightly
3091 * extended version are also supported on OPA interfaces.
3092 *
3093 * Return: true if the port supports sending/receiving of MAD packets.
3094 */
3095 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
3096 {
3097 return device->port_data[port_num].immutable.core_cap_flags &
3098 RDMA_CORE_CAP_IB_MAD;
3099 }
3100
3101 /**
3102 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3103 * Management Datagrams.
3104 * @device: Device to check
3105 * @port_num: Port number to check
3106 *
3107 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3108 * datagrams with their own versions. These OPA MADs share many but not all of
3109 * the characteristics of InfiniBand MADs.
3110 *
3111 * OPA MADs differ in the following ways:
3112 *
3113 * 1) MADs are variable size up to 2K
3114 * IBTA defined MADs remain fixed at 256 bytes
3115 * 2) OPA SMPs must carry valid PKeys
3116 * 3) OPA SMP packets are a different format
3117 *
3118 * Return: true if the port supports OPA MAD packet formats.
3119 */
3120 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3121 {
3122 return device->port_data[port_num].immutable.core_cap_flags &
3123 RDMA_CORE_CAP_OPA_MAD;
3124 }
3125
3126 /**
3127 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3128 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3129 * @device: Device to check
3130 * @port_num: Port number to check
3131 *
3132 * Each InfiniBand node is required to provide a Subnet Management Agent
3133 * that the subnet manager can access. Prior to the fabric being fully
3134 * configured by the subnet manager, the SMA is accessed via a well known
3135 * interface called the Subnet Management Interface (SMI). This interface
3136 * uses directed route packets to communicate with the SM to get around the
3137 * chicken and egg problem of the SM needing to know what's on the fabric
3138 * in order to configure the fabric, and needing to configure the fabric in
3139 * order to send packets to the devices on the fabric. These directed
3140 * route packets do not need the fabric fully configured in order to reach
3141 * their destination. The SMI is the only method allowed to send
3142 * directed route packets on an InfiniBand fabric.
3143 *
3144 * Return: true if the port provides an SMI.
3145 */
3146 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3147 {
3148 return device->port_data[port_num].immutable.core_cap_flags &
3149 RDMA_CORE_CAP_IB_SMI;
3150 }
3151
3152 /**
3153 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3154 * Communication Manager.
3155 * @device: Device to check
3156 * @port_num: Port number to check
3157 *
3158 * The InfiniBand Communication Manager is one of many pre-defined General
3159 * Service Agents (GSA) that are accessed via the General Service
3160 * Interface (GSI). It's role is to facilitate establishment of connections
3161 * between nodes as well as other management related tasks for established
3162 * connections.
3163 *
3164 * Return: true if the port supports an IB CM (this does not guarantee that
3165 * a CM is actually running however).
3166 */
3167 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3168 {
3169 return device->port_data[port_num].immutable.core_cap_flags &
3170 RDMA_CORE_CAP_IB_CM;
3171 }
3172
3173 /**
3174 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3175 * Communication Manager.
3176 * @device: Device to check
3177 * @port_num: Port number to check
3178 *
3179 * Similar to above, but specific to iWARP connections which have a different
3180 * managment protocol than InfiniBand.
3181 *
3182 * Return: true if the port supports an iWARP CM (this does not guarantee that
3183 * a CM is actually running however).
3184 */
3185 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3186 {
3187 return device->port_data[port_num].immutable.core_cap_flags &
3188 RDMA_CORE_CAP_IW_CM;
3189 }
3190
3191 /**
3192 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3193 * Subnet Administration.
3194 * @device: Device to check
3195 * @port_num: Port number to check
3196 *
3197 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3198 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3199 * fabrics, devices should resolve routes to other hosts by contacting the
3200 * SA to query the proper route.
3201 *
3202 * Return: true if the port should act as a client to the fabric Subnet
3203 * Administration interface. This does not imply that the SA service is
3204 * running locally.
3205 */
3206 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3207 {
3208 return device->port_data[port_num].immutable.core_cap_flags &
3209 RDMA_CORE_CAP_IB_SA;
3210 }
3211
3212 /**
3213 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3214 * Multicast.
3215 * @device: Device to check
3216 * @port_num: Port number to check
3217 *
3218 * InfiniBand multicast registration is more complex than normal IPv4 or
3219 * IPv6 multicast registration. Each Host Channel Adapter must register
3220 * with the Subnet Manager when it wishes to join a multicast group. It
3221 * should do so only once regardless of how many queue pairs it subscribes
3222 * to this group. And it should leave the group only after all queue pairs
3223 * attached to the group have been detached.
3224 *
3225 * Return: true if the port must undertake the additional adminstrative
3226 * overhead of registering/unregistering with the SM and tracking of the
3227 * total number of queue pairs attached to the multicast group.
3228 */
3229 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3230 {
3231 return rdma_cap_ib_sa(device, port_num);
3232 }
3233
3234 /**
3235 * rdma_cap_af_ib - Check if the port of device has the capability
3236 * Native Infiniband Address.
3237 * @device: Device to check
3238 * @port_num: Port number to check
3239 *
3240 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3241 * GID. RoCE uses a different mechanism, but still generates a GID via
3242 * a prescribed mechanism and port specific data.
3243 *
3244 * Return: true if the port uses a GID address to identify devices on the
3245 * network.
3246 */
3247 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3248 {
3249 return device->port_data[port_num].immutable.core_cap_flags &
3250 RDMA_CORE_CAP_AF_IB;
3251 }
3252
3253 /**
3254 * rdma_cap_eth_ah - Check if the port of device has the capability
3255 * Ethernet Address Handle.
3256 * @device: Device to check
3257 * @port_num: Port number to check
3258 *
3259 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3260 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3261 * port. Normally, packet headers are generated by the sending host
3262 * adapter, but when sending connectionless datagrams, we must manually
3263 * inject the proper headers for the fabric we are communicating over.
3264 *
3265 * Return: true if we are running as a RoCE port and must force the
3266 * addition of a Global Route Header built from our Ethernet Address
3267 * Handle into our header list for connectionless packets.
3268 */
3269 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3270 {
3271 return device->port_data[port_num].immutable.core_cap_flags &
3272 RDMA_CORE_CAP_ETH_AH;
3273 }
3274
3275 /**
3276 * rdma_cap_opa_ah - Check if the port of device supports
3277 * OPA Address handles
3278 * @device: Device to check
3279 * @port_num: Port number to check
3280 *
3281 * Return: true if we are running on an OPA device which supports
3282 * the extended OPA addressing.
3283 */
3284 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3285 {
3286 return (device->port_data[port_num].immutable.core_cap_flags &
3287 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3288 }
3289
3290 /**
3291 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3292 *
3293 * @device: Device
3294 * @port_num: Port number
3295 *
3296 * This MAD size includes the MAD headers and MAD payload. No other headers
3297 * are included.
3298 *
3299 * Return the max MAD size required by the Port. Will return 0 if the port
3300 * does not support MADs
3301 */
3302 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3303 {
3304 return device->port_data[port_num].immutable.max_mad_size;
3305 }
3306
3307 /**
3308 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3309 * @device: Device to check
3310 * @port_num: Port number to check
3311 *
3312 * RoCE GID table mechanism manages the various GIDs for a device.
3313 *
3314 * NOTE: if allocating the port's GID table has failed, this call will still
3315 * return true, but any RoCE GID table API will fail.
3316 *
3317 * Return: true if the port uses RoCE GID table mechanism in order to manage
3318 * its GIDs.
3319 */
3320 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3321 u8 port_num)
3322 {
3323 return rdma_protocol_roce(device, port_num) &&
3324 device->ops.add_gid && device->ops.del_gid;
3325 }
3326
3327 /*
3328 * Check if the device supports READ W/ INVALIDATE.
3329 */
3330 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3331 {
3332 /*
3333 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3334 * has support for it yet.
3335 */
3336 return rdma_protocol_iwarp(dev, port_num);
3337 }
3338
3339 /**
3340 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3341 * @device: Device
3342 * @port_num: 1 based Port number
3343 *
3344 * Return true if port is an Intel OPA port , false if not
3345 */
3346 static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3347 u32 port_num)
3348 {
3349 return (device->port_data[port_num].immutable.core_cap_flags &
3350 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3351 }
3352
3353 /**
3354 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3355 * @device: Device
3356 * @port_num: Port number
3357 * @mtu: enum value of MTU
3358 *
3359 * Return the MTU size supported by the port as an integer value. Will return
3360 * -1 if enum value of mtu is not supported.
3361 */
3362 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u8 port,
3363 int mtu)
3364 {
3365 if (rdma_core_cap_opa_port(device, port))
3366 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3367 else
3368 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3369 }
3370
3371 /**
3372 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3373 * @device: Device
3374 * @port_num: Port number
3375 * @attr: port attribute
3376 *
3377 * Return the MTU size supported by the port as an integer value.
3378 */
3379 static inline int rdma_mtu_from_attr(struct ib_device *device, u8 port,
3380 struct ib_port_attr *attr)
3381 {
3382 if (rdma_core_cap_opa_port(device, port))
3383 return attr->phys_mtu;
3384 else
3385 return ib_mtu_enum_to_int(attr->max_mtu);
3386 }
3387
3388 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3389 int state);
3390 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3391 struct ifla_vf_info *info);
3392 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3393 struct ifla_vf_stats *stats);
3394 int ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
3395 struct ifla_vf_guid *node_guid,
3396 struct ifla_vf_guid *port_guid);
3397 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3398 int type);
3399
3400 int ib_query_pkey(struct ib_device *device,
3401 u8 port_num, u16 index, u16 *pkey);
3402
3403 int ib_modify_device(struct ib_device *device,
3404 int device_modify_mask,
3405 struct ib_device_modify *device_modify);
3406
3407 int ib_modify_port(struct ib_device *device,
3408 u8 port_num, int port_modify_mask,
3409 struct ib_port_modify *port_modify);
3410
3411 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3412 u8 *port_num, u16 *index);
3413
3414 int ib_find_pkey(struct ib_device *device,
3415 u8 port_num, u16 pkey, u16 *index);
3416
3417 enum ib_pd_flags {
3418 /*
3419 * Create a memory registration for all memory in the system and place
3420 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3421 * ULPs to avoid the overhead of dynamic MRs.
3422 *
3423 * This flag is generally considered unsafe and must only be used in
3424 * extremly trusted environments. Every use of it will log a warning
3425 * in the kernel log.
3426 */
3427 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3428 };
3429
3430 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3431 const char *caller);
3432
3433 #define ib_alloc_pd(device, flags) \
3434 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3435
3436 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3437
3438 /**
3439 * ib_dealloc_pd - Deallocate kernel PD
3440 * @pd: The protection domain
3441 *
3442 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3443 */
3444 static inline void ib_dealloc_pd(struct ib_pd *pd)
3445 {
3446 int ret = ib_dealloc_pd_user(pd, NULL);
3447
3448 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3449 }
3450
3451 enum rdma_create_ah_flags {
3452 /* In a sleepable context */
3453 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3454 };
3455
3456 /**
3457 * rdma_create_ah - Creates an address handle for the given address vector.
3458 * @pd: The protection domain associated with the address handle.
3459 * @ah_attr: The attributes of the address vector.
3460 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3461 *
3462 * The address handle is used to reference a local or global destination
3463 * in all UD QP post sends.
3464 */
3465 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3466 u32 flags);
3467
3468 /**
3469 * rdma_create_user_ah - Creates an address handle for the given address vector.
3470 * It resolves destination mac address for ah attribute of RoCE type.
3471 * @pd: The protection domain associated with the address handle.
3472 * @ah_attr: The attributes of the address vector.
3473 * @udata: pointer to user's input output buffer information need by
3474 * provider driver.
3475 *
3476 * It returns 0 on success and returns appropriate error code on error.
3477 * The address handle is used to reference a local or global destination
3478 * in all UD QP post sends.
3479 */
3480 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3481 struct rdma_ah_attr *ah_attr,
3482 struct ib_udata *udata);
3483 /**
3484 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3485 * work completion.
3486 * @hdr: the L3 header to parse
3487 * @net_type: type of header to parse
3488 * @sgid: place to store source gid
3489 * @dgid: place to store destination gid
3490 */
3491 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3492 enum rdma_network_type net_type,
3493 union ib_gid *sgid, union ib_gid *dgid);
3494
3495 /**
3496 * ib_get_rdma_header_version - Get the header version
3497 * @hdr: the L3 header to parse
3498 */
3499 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3500
3501 /**
3502 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3503 * work completion.
3504 * @device: Device on which the received message arrived.
3505 * @port_num: Port on which the received message arrived.
3506 * @wc: Work completion associated with the received message.
3507 * @grh: References the received global route header. This parameter is
3508 * ignored unless the work completion indicates that the GRH is valid.
3509 * @ah_attr: Returned attributes that can be used when creating an address
3510 * handle for replying to the message.
3511 * When ib_init_ah_attr_from_wc() returns success,
3512 * (a) for IB link layer it optionally contains a reference to SGID attribute
3513 * when GRH is present for IB link layer.
3514 * (b) for RoCE link layer it contains a reference to SGID attribute.
3515 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3516 * attributes which are initialized using ib_init_ah_attr_from_wc().
3517 *
3518 */
3519 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3520 const struct ib_wc *wc, const struct ib_grh *grh,
3521 struct rdma_ah_attr *ah_attr);
3522
3523 /**
3524 * ib_create_ah_from_wc - Creates an address handle associated with the
3525 * sender of the specified work completion.
3526 * @pd: The protection domain associated with the address handle.
3527 * @wc: Work completion information associated with a received message.
3528 * @grh: References the received global route header. This parameter is
3529 * ignored unless the work completion indicates that the GRH is valid.
3530 * @port_num: The outbound port number to associate with the address.
3531 *
3532 * The address handle is used to reference a local or global destination
3533 * in all UD QP post sends.
3534 */
3535 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3536 const struct ib_grh *grh, u8 port_num);
3537
3538 /**
3539 * rdma_modify_ah - Modifies the address vector associated with an address
3540 * handle.
3541 * @ah: The address handle to modify.
3542 * @ah_attr: The new address vector attributes to associate with the
3543 * address handle.
3544 */
3545 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3546
3547 /**
3548 * rdma_query_ah - Queries the address vector associated with an address
3549 * handle.
3550 * @ah: The address handle to query.
3551 * @ah_attr: The address vector attributes associated with the address
3552 * handle.
3553 */
3554 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3555
3556 enum rdma_destroy_ah_flags {
3557 /* In a sleepable context */
3558 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3559 };
3560
3561 /**
3562 * rdma_destroy_ah_user - Destroys an address handle.
3563 * @ah: The address handle to destroy.
3564 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3565 * @udata: Valid user data or NULL for kernel objects
3566 */
3567 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3568
3569 /**
3570 * rdma_destroy_ah - Destroys an kernel address handle.
3571 * @ah: The address handle to destroy.
3572 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3573 *
3574 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3575 */
3576 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3577 {
3578 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3579
3580 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3581 }
3582
3583 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3584 struct ib_srq_init_attr *srq_init_attr,
3585 struct ib_usrq_object *uobject,
3586 struct ib_udata *udata);
3587 static inline struct ib_srq *
3588 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3589 {
3590 if (!pd->device->ops.create_srq)
3591 return ERR_PTR(-EOPNOTSUPP);
3592
3593 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3594 }
3595
3596 /**
3597 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3598 * @srq: The SRQ to modify.
3599 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3600 * the current values of selected SRQ attributes are returned.
3601 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3602 * are being modified.
3603 *
3604 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3605 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3606 * the number of receives queued drops below the limit.
3607 */
3608 int ib_modify_srq(struct ib_srq *srq,
3609 struct ib_srq_attr *srq_attr,
3610 enum ib_srq_attr_mask srq_attr_mask);
3611
3612 /**
3613 * ib_query_srq - Returns the attribute list and current values for the
3614 * specified SRQ.
3615 * @srq: The SRQ to query.
3616 * @srq_attr: The attributes of the specified SRQ.
3617 */
3618 int ib_query_srq(struct ib_srq *srq,
3619 struct ib_srq_attr *srq_attr);
3620
3621 /**
3622 * ib_destroy_srq_user - Destroys the specified SRQ.
3623 * @srq: The SRQ to destroy.
3624 * @udata: Valid user data or NULL for kernel objects
3625 */
3626 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3627
3628 /**
3629 * ib_destroy_srq - Destroys the specified kernel SRQ.
3630 * @srq: The SRQ to destroy.
3631 *
3632 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3633 */
3634 static inline void ib_destroy_srq(struct ib_srq *srq)
3635 {
3636 int ret = ib_destroy_srq_user(srq, NULL);
3637
3638 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3639 }
3640
3641 /**
3642 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3643 * @srq: The SRQ to post the work request on.
3644 * @recv_wr: A list of work requests to post on the receive queue.
3645 * @bad_recv_wr: On an immediate failure, this parameter will reference
3646 * the work request that failed to be posted on the QP.
3647 */
3648 static inline int ib_post_srq_recv(struct ib_srq *srq,
3649 const struct ib_recv_wr *recv_wr,
3650 const struct ib_recv_wr **bad_recv_wr)
3651 {
3652 const struct ib_recv_wr *dummy;
3653
3654 return srq->device->ops.post_srq_recv(srq, recv_wr,
3655 bad_recv_wr ? : &dummy);
3656 }
3657
3658 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3659 struct ib_qp_init_attr *qp_init_attr);
3660
3661 /**
3662 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3663 * @qp: The QP to modify.
3664 * @attr: On input, specifies the QP attributes to modify. On output,
3665 * the current values of selected QP attributes are returned.
3666 * @attr_mask: A bit-mask used to specify which attributes of the QP
3667 * are being modified.
3668 * @udata: pointer to user's input output buffer information
3669 * are being modified.
3670 * It returns 0 on success and returns appropriate error code on error.
3671 */
3672 int ib_modify_qp_with_udata(struct ib_qp *qp,
3673 struct ib_qp_attr *attr,
3674 int attr_mask,
3675 struct ib_udata *udata);
3676
3677 /**
3678 * ib_modify_qp - Modifies the attributes for the specified QP and then
3679 * transitions the QP to the given state.
3680 * @qp: The QP to modify.
3681 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3682 * the current values of selected QP attributes are returned.
3683 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3684 * are being modified.
3685 */
3686 int ib_modify_qp(struct ib_qp *qp,
3687 struct ib_qp_attr *qp_attr,
3688 int qp_attr_mask);
3689
3690 /**
3691 * ib_query_qp - Returns the attribute list and current values for the
3692 * specified QP.
3693 * @qp: The QP to query.
3694 * @qp_attr: The attributes of the specified QP.
3695 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3696 * @qp_init_attr: Additional attributes of the selected QP.
3697 *
3698 * The qp_attr_mask may be used to limit the query to gathering only the
3699 * selected attributes.
3700 */
3701 int ib_query_qp(struct ib_qp *qp,
3702 struct ib_qp_attr *qp_attr,
3703 int qp_attr_mask,
3704 struct ib_qp_init_attr *qp_init_attr);
3705
3706 /**
3707 * ib_destroy_qp - Destroys the specified QP.
3708 * @qp: The QP to destroy.
3709 * @udata: Valid udata or NULL for kernel objects
3710 */
3711 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3712
3713 /**
3714 * ib_destroy_qp - Destroys the specified kernel QP.
3715 * @qp: The QP to destroy.
3716 *
3717 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3718 */
3719 static inline int ib_destroy_qp(struct ib_qp *qp)
3720 {
3721 return ib_destroy_qp_user(qp, NULL);
3722 }
3723
3724 /**
3725 * ib_open_qp - Obtain a reference to an existing sharable QP.
3726 * @xrcd - XRC domain
3727 * @qp_open_attr: Attributes identifying the QP to open.
3728 *
3729 * Returns a reference to a sharable QP.
3730 */
3731 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3732 struct ib_qp_open_attr *qp_open_attr);
3733
3734 /**
3735 * ib_close_qp - Release an external reference to a QP.
3736 * @qp: The QP handle to release
3737 *
3738 * The opened QP handle is released by the caller. The underlying
3739 * shared QP is not destroyed until all internal references are released.
3740 */
3741 int ib_close_qp(struct ib_qp *qp);
3742
3743 /**
3744 * ib_post_send - Posts a list of work requests to the send queue of
3745 * the specified QP.
3746 * @qp: The QP to post the work request on.
3747 * @send_wr: A list of work requests to post on the send queue.
3748 * @bad_send_wr: On an immediate failure, this parameter will reference
3749 * the work request that failed to be posted on the QP.
3750 *
3751 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3752 * error is returned, the QP state shall not be affected,
3753 * ib_post_send() will return an immediate error after queueing any
3754 * earlier work requests in the list.
3755 */
3756 static inline int ib_post_send(struct ib_qp *qp,
3757 const struct ib_send_wr *send_wr,
3758 const struct ib_send_wr **bad_send_wr)
3759 {
3760 const struct ib_send_wr *dummy;
3761
3762 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3763 }
3764
3765 /**
3766 * ib_post_recv - Posts a list of work requests to the receive queue of
3767 * the specified QP.
3768 * @qp: The QP to post the work request on.
3769 * @recv_wr: A list of work requests to post on the receive queue.
3770 * @bad_recv_wr: On an immediate failure, this parameter will reference
3771 * the work request that failed to be posted on the QP.
3772 */
3773 static inline int ib_post_recv(struct ib_qp *qp,
3774 const struct ib_recv_wr *recv_wr,
3775 const struct ib_recv_wr **bad_recv_wr)
3776 {
3777 const struct ib_recv_wr *dummy;
3778
3779 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3780 }
3781
3782 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3783 int comp_vector, enum ib_poll_context poll_ctx,
3784 const char *caller);
3785 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3786 int nr_cqe, int comp_vector,
3787 enum ib_poll_context poll_ctx)
3788 {
3789 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3790 KBUILD_MODNAME);
3791 }
3792
3793 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3794 int nr_cqe, enum ib_poll_context poll_ctx,
3795 const char *caller);
3796
3797 /**
3798 * ib_alloc_cq_any: Allocate kernel CQ
3799 * @dev: The IB device
3800 * @private: Private data attached to the CQE
3801 * @nr_cqe: Number of CQEs in the CQ
3802 * @poll_ctx: Context used for polling the CQ
3803 */
3804 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3805 void *private, int nr_cqe,
3806 enum ib_poll_context poll_ctx)
3807 {
3808 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3809 KBUILD_MODNAME);
3810 }
3811
3812 void ib_free_cq(struct ib_cq *cq);
3813 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3814
3815 /**
3816 * ib_create_cq - Creates a CQ on the specified device.
3817 * @device: The device on which to create the CQ.
3818 * @comp_handler: A user-specified callback that is invoked when a
3819 * completion event occurs on the CQ.
3820 * @event_handler: A user-specified callback that is invoked when an
3821 * asynchronous event not associated with a completion occurs on the CQ.
3822 * @cq_context: Context associated with the CQ returned to the user via
3823 * the associated completion and event handlers.
3824 * @cq_attr: The attributes the CQ should be created upon.
3825 *
3826 * Users can examine the cq structure to determine the actual CQ size.
3827 */
3828 struct ib_cq *__ib_create_cq(struct ib_device *device,
3829 ib_comp_handler comp_handler,
3830 void (*event_handler)(struct ib_event *, void *),
3831 void *cq_context,
3832 const struct ib_cq_init_attr *cq_attr,
3833 const char *caller);
3834 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3835 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3836
3837 /**
3838 * ib_resize_cq - Modifies the capacity of the CQ.
3839 * @cq: The CQ to resize.
3840 * @cqe: The minimum size of the CQ.
3841 *
3842 * Users can examine the cq structure to determine the actual CQ size.
3843 */
3844 int ib_resize_cq(struct ib_cq *cq, int cqe);
3845
3846 /**
3847 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3848 * @cq: The CQ to modify.
3849 * @cq_count: number of CQEs that will trigger an event
3850 * @cq_period: max period of time in usec before triggering an event
3851 *
3852 */
3853 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3854
3855 /**
3856 * ib_destroy_cq_user - Destroys the specified CQ.
3857 * @cq: The CQ to destroy.
3858 * @udata: Valid user data or NULL for kernel objects
3859 */
3860 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3861
3862 /**
3863 * ib_destroy_cq - Destroys the specified kernel CQ.
3864 * @cq: The CQ to destroy.
3865 *
3866 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3867 */
3868 static inline void ib_destroy_cq(struct ib_cq *cq)
3869 {
3870 int ret = ib_destroy_cq_user(cq, NULL);
3871
3872 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3873 }
3874
3875 /**
3876 * ib_poll_cq - poll a CQ for completion(s)
3877 * @cq:the CQ being polled
3878 * @num_entries:maximum number of completions to return
3879 * @wc:array of at least @num_entries &struct ib_wc where completions
3880 * will be returned
3881 *
3882 * Poll a CQ for (possibly multiple) completions. If the return value
3883 * is < 0, an error occurred. If the return value is >= 0, it is the
3884 * number of completions returned. If the return value is
3885 * non-negative and < num_entries, then the CQ was emptied.
3886 */
3887 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3888 struct ib_wc *wc)
3889 {
3890 return cq->device->ops.poll_cq(cq, num_entries, wc);
3891 }
3892
3893 /**
3894 * ib_req_notify_cq - Request completion notification on a CQ.
3895 * @cq: The CQ to generate an event for.
3896 * @flags:
3897 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3898 * to request an event on the next solicited event or next work
3899 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3900 * may also be |ed in to request a hint about missed events, as
3901 * described below.
3902 *
3903 * Return Value:
3904 * < 0 means an error occurred while requesting notification
3905 * == 0 means notification was requested successfully, and if
3906 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3907 * were missed and it is safe to wait for another event. In
3908 * this case is it guaranteed that any work completions added
3909 * to the CQ since the last CQ poll will trigger a completion
3910 * notification event.
3911 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3912 * in. It means that the consumer must poll the CQ again to
3913 * make sure it is empty to avoid missing an event because of a
3914 * race between requesting notification and an entry being
3915 * added to the CQ. This return value means it is possible
3916 * (but not guaranteed) that a work completion has been added
3917 * to the CQ since the last poll without triggering a
3918 * completion notification event.
3919 */
3920 static inline int ib_req_notify_cq(struct ib_cq *cq,
3921 enum ib_cq_notify_flags flags)
3922 {
3923 return cq->device->ops.req_notify_cq(cq, flags);
3924 }
3925
3926 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
3927 int comp_vector_hint,
3928 enum ib_poll_context poll_ctx);
3929
3930 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
3931
3932 /**
3933 * ib_req_ncomp_notif - Request completion notification when there are
3934 * at least the specified number of unreaped completions on the CQ.
3935 * @cq: The CQ to generate an event for.
3936 * @wc_cnt: The number of unreaped completions that should be on the
3937 * CQ before an event is generated.
3938 */
3939 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3940 {
3941 return cq->device->ops.req_ncomp_notif ?
3942 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3943 -ENOSYS;
3944 }
3945
3946 /*
3947 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
3948 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
3949 * address into the dma address.
3950 */
3951 static inline bool ib_uses_virt_dma(struct ib_device *dev)
3952 {
3953 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
3954 }
3955
3956 /**
3957 * ib_dma_mapping_error - check a DMA addr for error
3958 * @dev: The device for which the dma_addr was created
3959 * @dma_addr: The DMA address to check
3960 */
3961 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3962 {
3963 if (ib_uses_virt_dma(dev))
3964 return 0;
3965 return dma_mapping_error(dev->dma_device, dma_addr);
3966 }
3967
3968 /**
3969 * ib_dma_map_single - Map a kernel virtual address to DMA address
3970 * @dev: The device for which the dma_addr is to be created
3971 * @cpu_addr: The kernel virtual address
3972 * @size: The size of the region in bytes
3973 * @direction: The direction of the DMA
3974 */
3975 static inline u64 ib_dma_map_single(struct ib_device *dev,
3976 void *cpu_addr, size_t size,
3977 enum dma_data_direction direction)
3978 {
3979 if (ib_uses_virt_dma(dev))
3980 return (uintptr_t)cpu_addr;
3981 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3982 }
3983
3984 /**
3985 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3986 * @dev: The device for which the DMA address was created
3987 * @addr: The DMA address
3988 * @size: The size of the region in bytes
3989 * @direction: The direction of the DMA
3990 */
3991 static inline void ib_dma_unmap_single(struct ib_device *dev,
3992 u64 addr, size_t size,
3993 enum dma_data_direction direction)
3994 {
3995 if (!ib_uses_virt_dma(dev))
3996 dma_unmap_single(dev->dma_device, addr, size, direction);
3997 }
3998
3999 /**
4000 * ib_dma_map_page - Map a physical page to DMA address
4001 * @dev: The device for which the dma_addr is to be created
4002 * @page: The page to be mapped
4003 * @offset: The offset within the page
4004 * @size: The size of the region in bytes
4005 * @direction: The direction of the DMA
4006 */
4007 static inline u64 ib_dma_map_page(struct ib_device *dev,
4008 struct page *page,
4009 unsigned long offset,
4010 size_t size,
4011 enum dma_data_direction direction)
4012 {
4013 if (ib_uses_virt_dma(dev))
4014 return (uintptr_t)(page_address(page) + offset);
4015 return dma_map_page(dev->dma_device, page, offset, size, direction);
4016 }
4017
4018 /**
4019 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4020 * @dev: The device for which the DMA address was created
4021 * @addr: The DMA address
4022 * @size: The size of the region in bytes
4023 * @direction: The direction of the DMA
4024 */
4025 static inline void ib_dma_unmap_page(struct ib_device *dev,
4026 u64 addr, size_t size,
4027 enum dma_data_direction direction)
4028 {
4029 if (!ib_uses_virt_dma(dev))
4030 dma_unmap_page(dev->dma_device, addr, size, direction);
4031 }
4032
4033 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4034 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4035 struct scatterlist *sg, int nents,
4036 enum dma_data_direction direction,
4037 unsigned long dma_attrs)
4038 {
4039 if (ib_uses_virt_dma(dev))
4040 return ib_dma_virt_map_sg(dev, sg, nents);
4041 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4042 dma_attrs);
4043 }
4044
4045 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4046 struct scatterlist *sg, int nents,
4047 enum dma_data_direction direction,
4048 unsigned long dma_attrs)
4049 {
4050 if (!ib_uses_virt_dma(dev))
4051 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4052 dma_attrs);
4053 }
4054
4055 /**
4056 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4057 * @dev: The device for which the DMA addresses are to be created
4058 * @sg: The array of scatter/gather entries
4059 * @nents: The number of scatter/gather entries
4060 * @direction: The direction of the DMA
4061 */
4062 static inline int ib_dma_map_sg(struct ib_device *dev,
4063 struct scatterlist *sg, int nents,
4064 enum dma_data_direction direction)
4065 {
4066 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4067 }
4068
4069 /**
4070 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4071 * @dev: The device for which the DMA addresses were created
4072 * @sg: The array of scatter/gather entries
4073 * @nents: The number of scatter/gather entries
4074 * @direction: The direction of the DMA
4075 */
4076 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4077 struct scatterlist *sg, int nents,
4078 enum dma_data_direction direction)
4079 {
4080 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4081 }
4082
4083 /**
4084 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4085 * @dev: The device to query
4086 *
4087 * The returned value represents a size in bytes.
4088 */
4089 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4090 {
4091 if (ib_uses_virt_dma(dev))
4092 return UINT_MAX;
4093 return dma_get_max_seg_size(dev->dma_device);
4094 }
4095
4096 /**
4097 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4098 * @dev: The device for which the DMA address was created
4099 * @addr: The DMA address
4100 * @size: The size of the region in bytes
4101 * @dir: The direction of the DMA
4102 */
4103 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4104 u64 addr,
4105 size_t size,
4106 enum dma_data_direction dir)
4107 {
4108 if (!ib_uses_virt_dma(dev))
4109 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4110 }
4111
4112 /**
4113 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4114 * @dev: The device for which the DMA address was created
4115 * @addr: The DMA address
4116 * @size: The size of the region in bytes
4117 * @dir: The direction of the DMA
4118 */
4119 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4120 u64 addr,
4121 size_t size,
4122 enum dma_data_direction dir)
4123 {
4124 if (!ib_uses_virt_dma(dev))
4125 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4126 }
4127
4128 /**
4129 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4130 * @dev: The device for which the DMA address is requested
4131 * @size: The size of the region to allocate in bytes
4132 * @dma_handle: A pointer for returning the DMA address of the region
4133 * @flag: memory allocator flags
4134 */
4135 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4136 size_t size,
4137 dma_addr_t *dma_handle,
4138 gfp_t flag)
4139 {
4140 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4141 }
4142
4143 /**
4144 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4145 * @dev: The device for which the DMA addresses were allocated
4146 * @size: The size of the region
4147 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4148 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4149 */
4150 static inline void ib_dma_free_coherent(struct ib_device *dev,
4151 size_t size, void *cpu_addr,
4152 dma_addr_t dma_handle)
4153 {
4154 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4155 }
4156
4157 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4158 * space. This function should be called when 'current' is the owning MM.
4159 */
4160 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4161 u64 virt_addr, int mr_access_flags);
4162
4163 /* ib_advise_mr - give an advice about an address range in a memory region */
4164 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4165 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4166 /**
4167 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4168 * HCA translation table.
4169 * @mr: The memory region to deregister.
4170 * @udata: Valid user data or NULL for kernel object
4171 *
4172 * This function can fail, if the memory region has memory windows bound to it.
4173 */
4174 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4175
4176 /**
4177 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4178 * HCA translation table.
4179 * @mr: The memory region to deregister.
4180 *
4181 * This function can fail, if the memory region has memory windows bound to it.
4182 *
4183 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4184 */
4185 static inline int ib_dereg_mr(struct ib_mr *mr)
4186 {
4187 return ib_dereg_mr_user(mr, NULL);
4188 }
4189
4190 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4191 u32 max_num_sg);
4192
4193 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4194 u32 max_num_data_sg,
4195 u32 max_num_meta_sg);
4196
4197 /**
4198 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4199 * R_Key and L_Key.
4200 * @mr - struct ib_mr pointer to be updated.
4201 * @newkey - new key to be used.
4202 */
4203 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4204 {
4205 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4206 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4207 }
4208
4209 /**
4210 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4211 * for calculating a new rkey for type 2 memory windows.
4212 * @rkey - the rkey to increment.
4213 */
4214 static inline u32 ib_inc_rkey(u32 rkey)
4215 {
4216 const u32 mask = 0x000000ff;
4217 return ((rkey + 1) & mask) | (rkey & ~mask);
4218 }
4219
4220 /**
4221 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4222 * @qp: QP to attach to the multicast group. The QP must be type
4223 * IB_QPT_UD.
4224 * @gid: Multicast group GID.
4225 * @lid: Multicast group LID in host byte order.
4226 *
4227 * In order to send and receive multicast packets, subnet
4228 * administration must have created the multicast group and configured
4229 * the fabric appropriately. The port associated with the specified
4230 * QP must also be a member of the multicast group.
4231 */
4232 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4233
4234 /**
4235 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4236 * @qp: QP to detach from the multicast group.
4237 * @gid: Multicast group GID.
4238 * @lid: Multicast group LID in host byte order.
4239 */
4240 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4241
4242 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4243 struct inode *inode, struct ib_udata *udata);
4244 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4245
4246 static inline int ib_check_mr_access(int flags)
4247 {
4248 /*
4249 * Local write permission is required if remote write or
4250 * remote atomic permission is also requested.
4251 */
4252 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4253 !(flags & IB_ACCESS_LOCAL_WRITE))
4254 return -EINVAL;
4255
4256 if (flags & ~IB_ACCESS_SUPPORTED)
4257 return -EINVAL;
4258
4259 return 0;
4260 }
4261
4262 static inline bool ib_access_writable(int access_flags)
4263 {
4264 /*
4265 * We have writable memory backing the MR if any of the following
4266 * access flags are set. "Local write" and "remote write" obviously
4267 * require write access. "Remote atomic" can do things like fetch and
4268 * add, which will modify memory, and "MW bind" can change permissions
4269 * by binding a window.
4270 */
4271 return access_flags &
4272 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4273 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4274 }
4275
4276 /**
4277 * ib_check_mr_status: lightweight check of MR status.
4278 * This routine may provide status checks on a selected
4279 * ib_mr. first use is for signature status check.
4280 *
4281 * @mr: A memory region.
4282 * @check_mask: Bitmask of which checks to perform from
4283 * ib_mr_status_check enumeration.
4284 * @mr_status: The container of relevant status checks.
4285 * failed checks will be indicated in the status bitmask
4286 * and the relevant info shall be in the error item.
4287 */
4288 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4289 struct ib_mr_status *mr_status);
4290
4291 /**
4292 * ib_device_try_get: Hold a registration lock
4293 * device: The device to lock
4294 *
4295 * A device under an active registration lock cannot become unregistered. It
4296 * is only possible to obtain a registration lock on a device that is fully
4297 * registered, otherwise this function returns false.
4298 *
4299 * The registration lock is only necessary for actions which require the
4300 * device to still be registered. Uses that only require the device pointer to
4301 * be valid should use get_device(&ibdev->dev) to hold the memory.
4302 *
4303 */
4304 static inline bool ib_device_try_get(struct ib_device *dev)
4305 {
4306 return refcount_inc_not_zero(&dev->refcount);
4307 }
4308
4309 void ib_device_put(struct ib_device *device);
4310 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4311 enum rdma_driver_id driver_id);
4312 struct ib_device *ib_device_get_by_name(const char *name,
4313 enum rdma_driver_id driver_id);
4314 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4315 u16 pkey, const union ib_gid *gid,
4316 const struct sockaddr *addr);
4317 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4318 unsigned int port);
4319 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4320
4321 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4322 struct ib_wq_init_attr *init_attr);
4323 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4324 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4325 u32 wq_attr_mask);
4326
4327 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4328 unsigned int *sg_offset, unsigned int page_size);
4329 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4330 int data_sg_nents, unsigned int *data_sg_offset,
4331 struct scatterlist *meta_sg, int meta_sg_nents,
4332 unsigned int *meta_sg_offset, unsigned int page_size);
4333
4334 static inline int
4335 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4336 unsigned int *sg_offset, unsigned int page_size)
4337 {
4338 int n;
4339
4340 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4341 mr->iova = 0;
4342
4343 return n;
4344 }
4345
4346 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4347 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4348
4349 void ib_drain_rq(struct ib_qp *qp);
4350 void ib_drain_sq(struct ib_qp *qp);
4351 void ib_drain_qp(struct ib_qp *qp);
4352
4353 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u16 *speed, u8 *width);
4354
4355 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4356 {
4357 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4358 return attr->roce.dmac;
4359 return NULL;
4360 }
4361
4362 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4363 {
4364 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4365 attr->ib.dlid = (u16)dlid;
4366 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4367 attr->opa.dlid = dlid;
4368 }
4369
4370 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4371 {
4372 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4373 return attr->ib.dlid;
4374 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4375 return attr->opa.dlid;
4376 return 0;
4377 }
4378
4379 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4380 {
4381 attr->sl = sl;
4382 }
4383
4384 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4385 {
4386 return attr->sl;
4387 }
4388
4389 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4390 u8 src_path_bits)
4391 {
4392 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4393 attr->ib.src_path_bits = src_path_bits;
4394 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4395 attr->opa.src_path_bits = src_path_bits;
4396 }
4397
4398 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4399 {
4400 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4401 return attr->ib.src_path_bits;
4402 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4403 return attr->opa.src_path_bits;
4404 return 0;
4405 }
4406
4407 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4408 bool make_grd)
4409 {
4410 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4411 attr->opa.make_grd = make_grd;
4412 }
4413
4414 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4415 {
4416 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4417 return attr->opa.make_grd;
4418 return false;
4419 }
4420
4421 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4422 {
4423 attr->port_num = port_num;
4424 }
4425
4426 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4427 {
4428 return attr->port_num;
4429 }
4430
4431 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4432 u8 static_rate)
4433 {
4434 attr->static_rate = static_rate;
4435 }
4436
4437 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4438 {
4439 return attr->static_rate;
4440 }
4441
4442 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4443 enum ib_ah_flags flag)
4444 {
4445 attr->ah_flags = flag;
4446 }
4447
4448 static inline enum ib_ah_flags
4449 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4450 {
4451 return attr->ah_flags;
4452 }
4453
4454 static inline const struct ib_global_route
4455 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4456 {
4457 return &attr->grh;
4458 }
4459
4460 /*To retrieve and modify the grh */
4461 static inline struct ib_global_route
4462 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4463 {
4464 return &attr->grh;
4465 }
4466
4467 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4468 {
4469 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4470
4471 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4472 }
4473
4474 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4475 __be64 prefix)
4476 {
4477 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4478
4479 grh->dgid.global.subnet_prefix = prefix;
4480 }
4481
4482 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4483 __be64 if_id)
4484 {
4485 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4486
4487 grh->dgid.global.interface_id = if_id;
4488 }
4489
4490 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4491 union ib_gid *dgid, u32 flow_label,
4492 u8 sgid_index, u8 hop_limit,
4493 u8 traffic_class)
4494 {
4495 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4496
4497 attr->ah_flags = IB_AH_GRH;
4498 if (dgid)
4499 grh->dgid = *dgid;
4500 grh->flow_label = flow_label;
4501 grh->sgid_index = sgid_index;
4502 grh->hop_limit = hop_limit;
4503 grh->traffic_class = traffic_class;
4504 grh->sgid_attr = NULL;
4505 }
4506
4507 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4508 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4509 u32 flow_label, u8 hop_limit, u8 traffic_class,
4510 const struct ib_gid_attr *sgid_attr);
4511 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4512 const struct rdma_ah_attr *src);
4513 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4514 const struct rdma_ah_attr *new);
4515 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4516
4517 /**
4518 * rdma_ah_find_type - Return address handle type.
4519 *
4520 * @dev: Device to be checked
4521 * @port_num: Port number
4522 */
4523 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4524 u8 port_num)
4525 {
4526 if (rdma_protocol_roce(dev, port_num))
4527 return RDMA_AH_ATTR_TYPE_ROCE;
4528 if (rdma_protocol_ib(dev, port_num)) {
4529 if (rdma_cap_opa_ah(dev, port_num))
4530 return RDMA_AH_ATTR_TYPE_OPA;
4531 return RDMA_AH_ATTR_TYPE_IB;
4532 }
4533
4534 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4535 }
4536
4537 /**
4538 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4539 * In the current implementation the only way to get
4540 * get the 32bit lid is from other sources for OPA.
4541 * For IB, lids will always be 16bits so cast the
4542 * value accordingly.
4543 *
4544 * @lid: A 32bit LID
4545 */
4546 static inline u16 ib_lid_cpu16(u32 lid)
4547 {
4548 WARN_ON_ONCE(lid & 0xFFFF0000);
4549 return (u16)lid;
4550 }
4551
4552 /**
4553 * ib_lid_be16 - Return lid in 16bit BE encoding.
4554 *
4555 * @lid: A 32bit LID
4556 */
4557 static inline __be16 ib_lid_be16(u32 lid)
4558 {
4559 WARN_ON_ONCE(lid & 0xFFFF0000);
4560 return cpu_to_be16((u16)lid);
4561 }
4562
4563 /**
4564 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4565 * vector
4566 * @device: the rdma device
4567 * @comp_vector: index of completion vector
4568 *
4569 * Returns NULL on failure, otherwise a corresponding cpu map of the
4570 * completion vector (returns all-cpus map if the device driver doesn't
4571 * implement get_vector_affinity).
4572 */
4573 static inline const struct cpumask *
4574 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4575 {
4576 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4577 !device->ops.get_vector_affinity)
4578 return NULL;
4579
4580 return device->ops.get_vector_affinity(device, comp_vector);
4581
4582 }
4583
4584 /**
4585 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4586 * and add their gids, as needed, to the relevant RoCE devices.
4587 *
4588 * @device: the rdma device
4589 */
4590 void rdma_roce_rescan_device(struct ib_device *ibdev);
4591
4592 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4593
4594 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4595
4596 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4597 enum rdma_netdev_t type, const char *name,
4598 unsigned char name_assign_type,
4599 void (*setup)(struct net_device *));
4600
4601 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4602 enum rdma_netdev_t type, const char *name,
4603 unsigned char name_assign_type,
4604 void (*setup)(struct net_device *),
4605 struct net_device *netdev);
4606
4607 /**
4608 * rdma_set_device_sysfs_group - Set device attributes group to have
4609 * driver specific sysfs entries at
4610 * for infiniband class.
4611 *
4612 * @device: device pointer for which attributes to be created
4613 * @group: Pointer to group which should be added when device
4614 * is registered with sysfs.
4615 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4616 * group per device to have sysfs attributes.
4617 *
4618 * NOTE: New drivers should not make use of this API; instead new device
4619 * parameter should be exposed via netlink command. This API and mechanism
4620 * exist only for existing drivers.
4621 */
4622 static inline void
4623 rdma_set_device_sysfs_group(struct ib_device *dev,
4624 const struct attribute_group *group)
4625 {
4626 dev->groups[1] = group;
4627 }
4628
4629 /**
4630 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4631 *
4632 * @device: device pointer for which ib_device pointer to retrieve
4633 *
4634 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4635 *
4636 */
4637 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4638 {
4639 struct ib_core_device *coredev =
4640 container_of(device, struct ib_core_device, dev);
4641
4642 return coredev->owner;
4643 }
4644
4645 /**
4646 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4647 * ib_device holder structure from device pointer.
4648 *
4649 * NOTE: New drivers should not make use of this API; This API is only for
4650 * existing drivers who have exposed sysfs entries using
4651 * rdma_set_device_sysfs_group().
4652 */
4653 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4654 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4655
4656 bool rdma_dev_access_netns(const struct ib_device *device,
4657 const struct net *net);
4658
4659 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4660 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4661 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4662
4663 /**
4664 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4665 * on the flow_label
4666 *
4667 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4668 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4669 * convention.
4670 */
4671 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4672 {
4673 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4674
4675 fl_low ^= fl_high >> 14;
4676 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4677 }
4678
4679 /**
4680 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4681 * local and remote qpn values
4682 *
4683 * This function folded the multiplication results of two qpns, 24 bit each,
4684 * fields, and converts it to a 20 bit results.
4685 *
4686 * This function will create symmetric flow_label value based on the local
4687 * and remote qpn values. this will allow both the requester and responder
4688 * to calculate the same flow_label for a given connection.
4689 *
4690 * This helper function should be used by driver in case the upper layer
4691 * provide a zero flow_label value. This is to improve entropy of RDMA
4692 * traffic in the network.
4693 */
4694 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4695 {
4696 u64 v = (u64)lqpn * rqpn;
4697
4698 v ^= v >> 20;
4699 v ^= v >> 40;
4700
4701 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4702 }
4703 #endif /* IB_VERBS_H */
Linux with added FPC-III support