RDMA/rtrs: server: Fix some error return code
[linux/fpc-iii.git] / include / rdma / ib_verbs.h
blobdb58f11552f159dfdd75d7153ecbe05ff6c0c6b0
1 /*
2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
18 * conditions are met:
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer.
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 * SOFTWARE.
39 #if !defined(IB_VERBS_H)
40 #define IB_VERBS_H
42 #include <linux/types.h>
43 #include <linux/device.h>
44 #include <linux/dma-mapping.h>
45 #include <linux/kref.h>
46 #include <linux/list.h>
47 #include <linux/rwsem.h>
48 #include <linux/workqueue.h>
49 #include <linux/irq_poll.h>
50 #include <uapi/linux/if_ether.h>
51 #include <net/ipv6.h>
52 #include <net/ip.h>
53 #include <linux/string.h>
54 #include <linux/slab.h>
55 #include <linux/netdevice.h>
56 #include <linux/refcount.h>
57 #include <linux/if_link.h>
58 #include <linux/atomic.h>
59 #include <linux/mmu_notifier.h>
60 #include <linux/uaccess.h>
61 #include <linux/cgroup_rdma.h>
62 #include <linux/irqflags.h>
63 #include <linux/preempt.h>
64 #include <linux/dim.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 #include <rdma/rdma_counter.h>
67 #include <rdma/restrack.h>
68 #include <rdma/signature.h>
69 #include <uapi/rdma/rdma_user_ioctl.h>
70 #include <uapi/rdma/ib_user_ioctl_verbs.h>
72 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
74 struct ib_umem_odp;
75 struct ib_uqp_object;
76 struct ib_usrq_object;
77 struct ib_uwq_object;
79 extern struct workqueue_struct *ib_wq;
80 extern struct workqueue_struct *ib_comp_wq;
81 extern struct workqueue_struct *ib_comp_unbound_wq;
83 struct ib_ucq_object;
85 __printf(3, 4) __cold
86 void ibdev_printk(const char *level, const struct ib_device *ibdev,
87 const char *format, ...);
88 __printf(2, 3) __cold
89 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
90 __printf(2, 3) __cold
91 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
92 __printf(2, 3) __cold
93 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
94 __printf(2, 3) __cold
95 void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
96 __printf(2, 3) __cold
97 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
98 __printf(2, 3) __cold
99 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
100 __printf(2, 3) __cold
101 void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
103 #if defined(CONFIG_DYNAMIC_DEBUG)
104 #define ibdev_dbg(__dev, format, args...) \
105 dynamic_ibdev_dbg(__dev, format, ##args)
106 #else
107 __printf(2, 3) __cold
108 static inline
109 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
110 #endif
112 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
113 do { \
114 static DEFINE_RATELIMIT_STATE(_rs, \
115 DEFAULT_RATELIMIT_INTERVAL, \
116 DEFAULT_RATELIMIT_BURST); \
117 if (__ratelimit(&_rs)) \
118 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
119 } while (0)
121 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
122 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
123 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \
124 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
125 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \
126 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
127 #define ibdev_err_ratelimited(ibdev, fmt, ...) \
128 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
129 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \
130 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
131 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \
132 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
133 #define ibdev_info_ratelimited(ibdev, fmt, ...) \
134 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
136 #if defined(CONFIG_DYNAMIC_DEBUG)
137 /* descriptor check is first to prevent flooding with "callbacks suppressed" */
138 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
139 do { \
140 static DEFINE_RATELIMIT_STATE(_rs, \
141 DEFAULT_RATELIMIT_INTERVAL, \
142 DEFAULT_RATELIMIT_BURST); \
143 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
144 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
145 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
146 ##__VA_ARGS__); \
147 } while (0)
148 #else
149 __printf(2, 3) __cold
150 static inline
151 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
152 #endif
154 union ib_gid {
155 u8 raw[16];
156 struct {
157 __be64 subnet_prefix;
158 __be64 interface_id;
159 } global;
162 extern union ib_gid zgid;
164 enum ib_gid_type {
165 /* If link layer is Ethernet, this is RoCE V1 */
166 IB_GID_TYPE_IB = 0,
167 IB_GID_TYPE_ROCE = 0,
168 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
169 IB_GID_TYPE_SIZE
172 #define ROCE_V2_UDP_DPORT 4791
173 struct ib_gid_attr {
174 struct net_device __rcu *ndev;
175 struct ib_device *device;
176 union ib_gid gid;
177 enum ib_gid_type gid_type;
178 u16 index;
179 u8 port_num;
182 enum {
183 /* set the local administered indication */
184 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
187 enum rdma_transport_type {
188 RDMA_TRANSPORT_IB,
189 RDMA_TRANSPORT_IWARP,
190 RDMA_TRANSPORT_USNIC,
191 RDMA_TRANSPORT_USNIC_UDP,
192 RDMA_TRANSPORT_UNSPECIFIED,
195 enum rdma_protocol_type {
196 RDMA_PROTOCOL_IB,
197 RDMA_PROTOCOL_IBOE,
198 RDMA_PROTOCOL_IWARP,
199 RDMA_PROTOCOL_USNIC_UDP
202 __attribute_const__ enum rdma_transport_type
203 rdma_node_get_transport(unsigned int node_type);
205 enum rdma_network_type {
206 RDMA_NETWORK_IB,
207 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
208 RDMA_NETWORK_IPV4,
209 RDMA_NETWORK_IPV6
212 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
214 if (network_type == RDMA_NETWORK_IPV4 ||
215 network_type == RDMA_NETWORK_IPV6)
216 return IB_GID_TYPE_ROCE_UDP_ENCAP;
218 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
219 return IB_GID_TYPE_IB;
222 static inline enum rdma_network_type
223 rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
225 if (attr->gid_type == IB_GID_TYPE_IB)
226 return RDMA_NETWORK_IB;
228 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
229 return RDMA_NETWORK_IPV4;
230 else
231 return RDMA_NETWORK_IPV6;
234 enum rdma_link_layer {
235 IB_LINK_LAYER_UNSPECIFIED,
236 IB_LINK_LAYER_INFINIBAND,
237 IB_LINK_LAYER_ETHERNET,
240 enum ib_device_cap_flags {
241 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
242 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
243 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
244 IB_DEVICE_RAW_MULTI = (1 << 3),
245 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
246 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
247 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
248 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
249 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
250 /* Not in use, former INIT_TYPE = (1 << 9),*/
251 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
252 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
253 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
254 IB_DEVICE_SRQ_RESIZE = (1 << 13),
255 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
258 * This device supports a per-device lkey or stag that can be
259 * used without performing a memory registration for the local
260 * memory. Note that ULPs should never check this flag, but
261 * instead of use the local_dma_lkey flag in the ib_pd structure,
262 * which will always contain a usable lkey.
264 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
265 /* Reserved, old SEND_W_INV = (1 << 16),*/
266 IB_DEVICE_MEM_WINDOW = (1 << 17),
268 * Devices should set IB_DEVICE_UD_IP_SUM if they support
269 * insertion of UDP and TCP checksum on outgoing UD IPoIB
270 * messages and can verify the validity of checksum for
271 * incoming messages. Setting this flag implies that the
272 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
274 IB_DEVICE_UD_IP_CSUM = (1 << 18),
275 IB_DEVICE_UD_TSO = (1 << 19),
276 IB_DEVICE_XRC = (1 << 20),
279 * This device supports the IB "base memory management extension",
280 * which includes support for fast registrations (IB_WR_REG_MR,
281 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
282 * also be set by any iWarp device which must support FRs to comply
283 * to the iWarp verbs spec. iWarp devices also support the
284 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
285 * stag.
287 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
288 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
289 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
290 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
291 IB_DEVICE_RC_IP_CSUM = (1 << 25),
292 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
293 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
295 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
296 * support execution of WQEs that involve synchronization
297 * of I/O operations with single completion queue managed
298 * by hardware.
300 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
301 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
302 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
303 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
304 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
305 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
306 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
307 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
308 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
309 /* The device supports padding incoming writes to cacheline. */
310 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
311 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
314 enum ib_atomic_cap {
315 IB_ATOMIC_NONE,
316 IB_ATOMIC_HCA,
317 IB_ATOMIC_GLOB
320 enum ib_odp_general_cap_bits {
321 IB_ODP_SUPPORT = 1 << 0,
322 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
325 enum ib_odp_transport_cap_bits {
326 IB_ODP_SUPPORT_SEND = 1 << 0,
327 IB_ODP_SUPPORT_RECV = 1 << 1,
328 IB_ODP_SUPPORT_WRITE = 1 << 2,
329 IB_ODP_SUPPORT_READ = 1 << 3,
330 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
331 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
334 struct ib_odp_caps {
335 uint64_t general_caps;
336 struct {
337 uint32_t rc_odp_caps;
338 uint32_t uc_odp_caps;
339 uint32_t ud_odp_caps;
340 uint32_t xrc_odp_caps;
341 } per_transport_caps;
344 struct ib_rss_caps {
345 /* Corresponding bit will be set if qp type from
346 * 'enum ib_qp_type' is supported, e.g.
347 * supported_qpts |= 1 << IB_QPT_UD
349 u32 supported_qpts;
350 u32 max_rwq_indirection_tables;
351 u32 max_rwq_indirection_table_size;
354 enum ib_tm_cap_flags {
355 /* Support tag matching with rendezvous offload for RC transport */
356 IB_TM_CAP_RNDV_RC = 1 << 0,
359 struct ib_tm_caps {
360 /* Max size of RNDV header */
361 u32 max_rndv_hdr_size;
362 /* Max number of entries in tag matching list */
363 u32 max_num_tags;
364 /* From enum ib_tm_cap_flags */
365 u32 flags;
366 /* Max number of outstanding list operations */
367 u32 max_ops;
368 /* Max number of SGE in tag matching entry */
369 u32 max_sge;
372 struct ib_cq_init_attr {
373 unsigned int cqe;
374 u32 comp_vector;
375 u32 flags;
378 enum ib_cq_attr_mask {
379 IB_CQ_MODERATE = 1 << 0,
382 struct ib_cq_caps {
383 u16 max_cq_moderation_count;
384 u16 max_cq_moderation_period;
387 struct ib_dm_mr_attr {
388 u64 length;
389 u64 offset;
390 u32 access_flags;
393 struct ib_dm_alloc_attr {
394 u64 length;
395 u32 alignment;
396 u32 flags;
399 struct ib_device_attr {
400 u64 fw_ver;
401 __be64 sys_image_guid;
402 u64 max_mr_size;
403 u64 page_size_cap;
404 u32 vendor_id;
405 u32 vendor_part_id;
406 u32 hw_ver;
407 int max_qp;
408 int max_qp_wr;
409 u64 device_cap_flags;
410 int max_send_sge;
411 int max_recv_sge;
412 int max_sge_rd;
413 int max_cq;
414 int max_cqe;
415 int max_mr;
416 int max_pd;
417 int max_qp_rd_atom;
418 int max_ee_rd_atom;
419 int max_res_rd_atom;
420 int max_qp_init_rd_atom;
421 int max_ee_init_rd_atom;
422 enum ib_atomic_cap atomic_cap;
423 enum ib_atomic_cap masked_atomic_cap;
424 int max_ee;
425 int max_rdd;
426 int max_mw;
427 int max_raw_ipv6_qp;
428 int max_raw_ethy_qp;
429 int max_mcast_grp;
430 int max_mcast_qp_attach;
431 int max_total_mcast_qp_attach;
432 int max_ah;
433 int max_fmr;
434 int max_map_per_fmr;
435 int max_srq;
436 int max_srq_wr;
437 int max_srq_sge;
438 unsigned int max_fast_reg_page_list_len;
439 unsigned int max_pi_fast_reg_page_list_len;
440 u16 max_pkeys;
441 u8 local_ca_ack_delay;
442 int sig_prot_cap;
443 int sig_guard_cap;
444 struct ib_odp_caps odp_caps;
445 uint64_t timestamp_mask;
446 uint64_t hca_core_clock; /* in KHZ */
447 struct ib_rss_caps rss_caps;
448 u32 max_wq_type_rq;
449 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
450 struct ib_tm_caps tm_caps;
451 struct ib_cq_caps cq_caps;
452 u64 max_dm_size;
453 /* Max entries for sgl for optimized performance per READ */
454 u32 max_sgl_rd;
457 enum ib_mtu {
458 IB_MTU_256 = 1,
459 IB_MTU_512 = 2,
460 IB_MTU_1024 = 3,
461 IB_MTU_2048 = 4,
462 IB_MTU_4096 = 5
465 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
467 switch (mtu) {
468 case IB_MTU_256: return 256;
469 case IB_MTU_512: return 512;
470 case IB_MTU_1024: return 1024;
471 case IB_MTU_2048: return 2048;
472 case IB_MTU_4096: return 4096;
473 default: return -1;
477 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
479 if (mtu >= 4096)
480 return IB_MTU_4096;
481 else if (mtu >= 2048)
482 return IB_MTU_2048;
483 else if (mtu >= 1024)
484 return IB_MTU_1024;
485 else if (mtu >= 512)
486 return IB_MTU_512;
487 else
488 return IB_MTU_256;
491 enum ib_port_state {
492 IB_PORT_NOP = 0,
493 IB_PORT_DOWN = 1,
494 IB_PORT_INIT = 2,
495 IB_PORT_ARMED = 3,
496 IB_PORT_ACTIVE = 4,
497 IB_PORT_ACTIVE_DEFER = 5
500 enum ib_port_phys_state {
501 IB_PORT_PHYS_STATE_SLEEP = 1,
502 IB_PORT_PHYS_STATE_POLLING = 2,
503 IB_PORT_PHYS_STATE_DISABLED = 3,
504 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
505 IB_PORT_PHYS_STATE_LINK_UP = 5,
506 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
507 IB_PORT_PHYS_STATE_PHY_TEST = 7,
510 enum ib_port_width {
511 IB_WIDTH_1X = 1,
512 IB_WIDTH_2X = 16,
513 IB_WIDTH_4X = 2,
514 IB_WIDTH_8X = 4,
515 IB_WIDTH_12X = 8
518 static inline int ib_width_enum_to_int(enum ib_port_width width)
520 switch (width) {
521 case IB_WIDTH_1X: return 1;
522 case IB_WIDTH_2X: return 2;
523 case IB_WIDTH_4X: return 4;
524 case IB_WIDTH_8X: return 8;
525 case IB_WIDTH_12X: return 12;
526 default: return -1;
530 enum ib_port_speed {
531 IB_SPEED_SDR = 1,
532 IB_SPEED_DDR = 2,
533 IB_SPEED_QDR = 4,
534 IB_SPEED_FDR10 = 8,
535 IB_SPEED_FDR = 16,
536 IB_SPEED_EDR = 32,
537 IB_SPEED_HDR = 64
541 * struct rdma_hw_stats
542 * @lock - Mutex to protect parallel write access to lifespan and values
543 * of counters, which are 64bits and not guaranteeed to be written
544 * atomicaly on 32bits systems.
545 * @timestamp - Used by the core code to track when the last update was
546 * @lifespan - Used by the core code to determine how old the counters
547 * should be before being updated again. Stored in jiffies, defaults
548 * to 10 milliseconds, drivers can override the default be specifying
549 * their own value during their allocation routine.
550 * @name - Array of pointers to static names used for the counters in
551 * directory.
552 * @num_counters - How many hardware counters there are. If name is
553 * shorter than this number, a kernel oops will result. Driver authors
554 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
555 * in their code to prevent this.
556 * @value - Array of u64 counters that are accessed by the sysfs code and
557 * filled in by the drivers get_stats routine
559 struct rdma_hw_stats {
560 struct mutex lock; /* Protect lifespan and values[] */
561 unsigned long timestamp;
562 unsigned long lifespan;
563 const char * const *names;
564 int num_counters;
565 u64 value[];
568 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
570 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
571 * for drivers.
572 * @names - Array of static const char *
573 * @num_counters - How many elements in array
574 * @lifespan - How many milliseconds between updates
576 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
577 const char * const *names, int num_counters,
578 unsigned long lifespan)
580 struct rdma_hw_stats *stats;
582 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
583 GFP_KERNEL);
584 if (!stats)
585 return NULL;
586 stats->names = names;
587 stats->num_counters = num_counters;
588 stats->lifespan = msecs_to_jiffies(lifespan);
590 return stats;
594 /* Define bits for the various functionality this port needs to be supported by
595 * the core.
597 /* Management 0x00000FFF */
598 #define RDMA_CORE_CAP_IB_MAD 0x00000001
599 #define RDMA_CORE_CAP_IB_SMI 0x00000002
600 #define RDMA_CORE_CAP_IB_CM 0x00000004
601 #define RDMA_CORE_CAP_IW_CM 0x00000008
602 #define RDMA_CORE_CAP_IB_SA 0x00000010
603 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
605 /* Address format 0x000FF000 */
606 #define RDMA_CORE_CAP_AF_IB 0x00001000
607 #define RDMA_CORE_CAP_ETH_AH 0x00002000
608 #define RDMA_CORE_CAP_OPA_AH 0x00004000
609 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
611 /* Protocol 0xFFF00000 */
612 #define RDMA_CORE_CAP_PROT_IB 0x00100000
613 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
614 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
615 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
616 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
617 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
619 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
620 | RDMA_CORE_CAP_PROT_ROCE \
621 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
623 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
624 | RDMA_CORE_CAP_IB_MAD \
625 | RDMA_CORE_CAP_IB_SMI \
626 | RDMA_CORE_CAP_IB_CM \
627 | RDMA_CORE_CAP_IB_SA \
628 | RDMA_CORE_CAP_AF_IB)
629 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
630 | RDMA_CORE_CAP_IB_MAD \
631 | RDMA_CORE_CAP_IB_CM \
632 | RDMA_CORE_CAP_AF_IB \
633 | RDMA_CORE_CAP_ETH_AH)
634 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
635 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
636 | RDMA_CORE_CAP_IB_MAD \
637 | RDMA_CORE_CAP_IB_CM \
638 | RDMA_CORE_CAP_AF_IB \
639 | RDMA_CORE_CAP_ETH_AH)
640 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
641 | RDMA_CORE_CAP_IW_CM)
642 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
643 | RDMA_CORE_CAP_OPA_MAD)
645 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
647 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
649 struct ib_port_attr {
650 u64 subnet_prefix;
651 enum ib_port_state state;
652 enum ib_mtu max_mtu;
653 enum ib_mtu active_mtu;
654 int gid_tbl_len;
655 unsigned int ip_gids:1;
656 /* This is the value from PortInfo CapabilityMask, defined by IBA */
657 u32 port_cap_flags;
658 u32 max_msg_sz;
659 u32 bad_pkey_cntr;
660 u32 qkey_viol_cntr;
661 u16 pkey_tbl_len;
662 u32 sm_lid;
663 u32 lid;
664 u8 lmc;
665 u8 max_vl_num;
666 u8 sm_sl;
667 u8 subnet_timeout;
668 u8 init_type_reply;
669 u8 active_width;
670 u8 active_speed;
671 u8 phys_state;
672 u16 port_cap_flags2;
675 enum ib_device_modify_flags {
676 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
677 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
680 #define IB_DEVICE_NODE_DESC_MAX 64
682 struct ib_device_modify {
683 u64 sys_image_guid;
684 char node_desc[IB_DEVICE_NODE_DESC_MAX];
687 enum ib_port_modify_flags {
688 IB_PORT_SHUTDOWN = 1,
689 IB_PORT_INIT_TYPE = (1<<2),
690 IB_PORT_RESET_QKEY_CNTR = (1<<3),
691 IB_PORT_OPA_MASK_CHG = (1<<4)
694 struct ib_port_modify {
695 u32 set_port_cap_mask;
696 u32 clr_port_cap_mask;
697 u8 init_type;
700 enum ib_event_type {
701 IB_EVENT_CQ_ERR,
702 IB_EVENT_QP_FATAL,
703 IB_EVENT_QP_REQ_ERR,
704 IB_EVENT_QP_ACCESS_ERR,
705 IB_EVENT_COMM_EST,
706 IB_EVENT_SQ_DRAINED,
707 IB_EVENT_PATH_MIG,
708 IB_EVENT_PATH_MIG_ERR,
709 IB_EVENT_DEVICE_FATAL,
710 IB_EVENT_PORT_ACTIVE,
711 IB_EVENT_PORT_ERR,
712 IB_EVENT_LID_CHANGE,
713 IB_EVENT_PKEY_CHANGE,
714 IB_EVENT_SM_CHANGE,
715 IB_EVENT_SRQ_ERR,
716 IB_EVENT_SRQ_LIMIT_REACHED,
717 IB_EVENT_QP_LAST_WQE_REACHED,
718 IB_EVENT_CLIENT_REREGISTER,
719 IB_EVENT_GID_CHANGE,
720 IB_EVENT_WQ_FATAL,
723 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
725 struct ib_event {
726 struct ib_device *device;
727 union {
728 struct ib_cq *cq;
729 struct ib_qp *qp;
730 struct ib_srq *srq;
731 struct ib_wq *wq;
732 u8 port_num;
733 } element;
734 enum ib_event_type event;
737 struct ib_event_handler {
738 struct ib_device *device;
739 void (*handler)(struct ib_event_handler *, struct ib_event *);
740 struct list_head list;
743 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
744 do { \
745 (_ptr)->device = _device; \
746 (_ptr)->handler = _handler; \
747 INIT_LIST_HEAD(&(_ptr)->list); \
748 } while (0)
750 struct ib_global_route {
751 const struct ib_gid_attr *sgid_attr;
752 union ib_gid dgid;
753 u32 flow_label;
754 u8 sgid_index;
755 u8 hop_limit;
756 u8 traffic_class;
759 struct ib_grh {
760 __be32 version_tclass_flow;
761 __be16 paylen;
762 u8 next_hdr;
763 u8 hop_limit;
764 union ib_gid sgid;
765 union ib_gid dgid;
768 union rdma_network_hdr {
769 struct ib_grh ibgrh;
770 struct {
771 /* The IB spec states that if it's IPv4, the header
772 * is located in the last 20 bytes of the header.
774 u8 reserved[20];
775 struct iphdr roce4grh;
779 #define IB_QPN_MASK 0xFFFFFF
781 enum {
782 IB_MULTICAST_QPN = 0xffffff
785 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
786 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
788 enum ib_ah_flags {
789 IB_AH_GRH = 1
792 enum ib_rate {
793 IB_RATE_PORT_CURRENT = 0,
794 IB_RATE_2_5_GBPS = 2,
795 IB_RATE_5_GBPS = 5,
796 IB_RATE_10_GBPS = 3,
797 IB_RATE_20_GBPS = 6,
798 IB_RATE_30_GBPS = 4,
799 IB_RATE_40_GBPS = 7,
800 IB_RATE_60_GBPS = 8,
801 IB_RATE_80_GBPS = 9,
802 IB_RATE_120_GBPS = 10,
803 IB_RATE_14_GBPS = 11,
804 IB_RATE_56_GBPS = 12,
805 IB_RATE_112_GBPS = 13,
806 IB_RATE_168_GBPS = 14,
807 IB_RATE_25_GBPS = 15,
808 IB_RATE_100_GBPS = 16,
809 IB_RATE_200_GBPS = 17,
810 IB_RATE_300_GBPS = 18,
811 IB_RATE_28_GBPS = 19,
812 IB_RATE_50_GBPS = 20,
813 IB_RATE_400_GBPS = 21,
814 IB_RATE_600_GBPS = 22,
818 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
819 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
820 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
821 * @rate: rate to convert.
823 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
826 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
827 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
828 * @rate: rate to convert.
830 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
834 * enum ib_mr_type - memory region type
835 * @IB_MR_TYPE_MEM_REG: memory region that is used for
836 * normal registration
837 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
838 * register any arbitrary sg lists (without
839 * the normal mr constraints - see
840 * ib_map_mr_sg)
841 * @IB_MR_TYPE_DM: memory region that is used for device
842 * memory registration
843 * @IB_MR_TYPE_USER: memory region that is used for the user-space
844 * application
845 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
846 * without address translations (VA=PA)
847 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
848 * data integrity operations
850 enum ib_mr_type {
851 IB_MR_TYPE_MEM_REG,
852 IB_MR_TYPE_SG_GAPS,
853 IB_MR_TYPE_DM,
854 IB_MR_TYPE_USER,
855 IB_MR_TYPE_DMA,
856 IB_MR_TYPE_INTEGRITY,
859 enum ib_mr_status_check {
860 IB_MR_CHECK_SIG_STATUS = 1,
864 * struct ib_mr_status - Memory region status container
866 * @fail_status: Bitmask of MR checks status. For each
867 * failed check a corresponding status bit is set.
868 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
869 * failure.
871 struct ib_mr_status {
872 u32 fail_status;
873 struct ib_sig_err sig_err;
877 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
878 * enum.
879 * @mult: multiple to convert.
881 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
883 struct rdma_ah_init_attr {
884 struct rdma_ah_attr *ah_attr;
885 u32 flags;
886 struct net_device *xmit_slave;
889 enum rdma_ah_attr_type {
890 RDMA_AH_ATTR_TYPE_UNDEFINED,
891 RDMA_AH_ATTR_TYPE_IB,
892 RDMA_AH_ATTR_TYPE_ROCE,
893 RDMA_AH_ATTR_TYPE_OPA,
896 struct ib_ah_attr {
897 u16 dlid;
898 u8 src_path_bits;
901 struct roce_ah_attr {
902 u8 dmac[ETH_ALEN];
905 struct opa_ah_attr {
906 u32 dlid;
907 u8 src_path_bits;
908 bool make_grd;
911 struct rdma_ah_attr {
912 struct ib_global_route grh;
913 u8 sl;
914 u8 static_rate;
915 u8 port_num;
916 u8 ah_flags;
917 enum rdma_ah_attr_type type;
918 union {
919 struct ib_ah_attr ib;
920 struct roce_ah_attr roce;
921 struct opa_ah_attr opa;
925 enum ib_wc_status {
926 IB_WC_SUCCESS,
927 IB_WC_LOC_LEN_ERR,
928 IB_WC_LOC_QP_OP_ERR,
929 IB_WC_LOC_EEC_OP_ERR,
930 IB_WC_LOC_PROT_ERR,
931 IB_WC_WR_FLUSH_ERR,
932 IB_WC_MW_BIND_ERR,
933 IB_WC_BAD_RESP_ERR,
934 IB_WC_LOC_ACCESS_ERR,
935 IB_WC_REM_INV_REQ_ERR,
936 IB_WC_REM_ACCESS_ERR,
937 IB_WC_REM_OP_ERR,
938 IB_WC_RETRY_EXC_ERR,
939 IB_WC_RNR_RETRY_EXC_ERR,
940 IB_WC_LOC_RDD_VIOL_ERR,
941 IB_WC_REM_INV_RD_REQ_ERR,
942 IB_WC_REM_ABORT_ERR,
943 IB_WC_INV_EECN_ERR,
944 IB_WC_INV_EEC_STATE_ERR,
945 IB_WC_FATAL_ERR,
946 IB_WC_RESP_TIMEOUT_ERR,
947 IB_WC_GENERAL_ERR
950 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
952 enum ib_wc_opcode {
953 IB_WC_SEND,
954 IB_WC_RDMA_WRITE,
955 IB_WC_RDMA_READ,
956 IB_WC_COMP_SWAP,
957 IB_WC_FETCH_ADD,
958 IB_WC_LSO,
959 IB_WC_LOCAL_INV,
960 IB_WC_REG_MR,
961 IB_WC_MASKED_COMP_SWAP,
962 IB_WC_MASKED_FETCH_ADD,
964 * Set value of IB_WC_RECV so consumers can test if a completion is a
965 * receive by testing (opcode & IB_WC_RECV).
967 IB_WC_RECV = 1 << 7,
968 IB_WC_RECV_RDMA_WITH_IMM
971 enum ib_wc_flags {
972 IB_WC_GRH = 1,
973 IB_WC_WITH_IMM = (1<<1),
974 IB_WC_WITH_INVALIDATE = (1<<2),
975 IB_WC_IP_CSUM_OK = (1<<3),
976 IB_WC_WITH_SMAC = (1<<4),
977 IB_WC_WITH_VLAN = (1<<5),
978 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
981 struct ib_wc {
982 union {
983 u64 wr_id;
984 struct ib_cqe *wr_cqe;
986 enum ib_wc_status status;
987 enum ib_wc_opcode opcode;
988 u32 vendor_err;
989 u32 byte_len;
990 struct ib_qp *qp;
991 union {
992 __be32 imm_data;
993 u32 invalidate_rkey;
994 } ex;
995 u32 src_qp;
996 u32 slid;
997 int wc_flags;
998 u16 pkey_index;
999 u8 sl;
1000 u8 dlid_path_bits;
1001 u8 port_num; /* valid only for DR SMPs on switches */
1002 u8 smac[ETH_ALEN];
1003 u16 vlan_id;
1004 u8 network_hdr_type;
1007 enum ib_cq_notify_flags {
1008 IB_CQ_SOLICITED = 1 << 0,
1009 IB_CQ_NEXT_COMP = 1 << 1,
1010 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1011 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1014 enum ib_srq_type {
1015 IB_SRQT_BASIC,
1016 IB_SRQT_XRC,
1017 IB_SRQT_TM,
1020 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1022 return srq_type == IB_SRQT_XRC ||
1023 srq_type == IB_SRQT_TM;
1026 enum ib_srq_attr_mask {
1027 IB_SRQ_MAX_WR = 1 << 0,
1028 IB_SRQ_LIMIT = 1 << 1,
1031 struct ib_srq_attr {
1032 u32 max_wr;
1033 u32 max_sge;
1034 u32 srq_limit;
1037 struct ib_srq_init_attr {
1038 void (*event_handler)(struct ib_event *, void *);
1039 void *srq_context;
1040 struct ib_srq_attr attr;
1041 enum ib_srq_type srq_type;
1043 struct {
1044 struct ib_cq *cq;
1045 union {
1046 struct {
1047 struct ib_xrcd *xrcd;
1048 } xrc;
1050 struct {
1051 u32 max_num_tags;
1052 } tag_matching;
1054 } ext;
1057 struct ib_qp_cap {
1058 u32 max_send_wr;
1059 u32 max_recv_wr;
1060 u32 max_send_sge;
1061 u32 max_recv_sge;
1062 u32 max_inline_data;
1065 * Maximum number of rdma_rw_ctx structures in flight at a time.
1066 * ib_create_qp() will calculate the right amount of neededed WRs
1067 * and MRs based on this.
1069 u32 max_rdma_ctxs;
1072 enum ib_sig_type {
1073 IB_SIGNAL_ALL_WR,
1074 IB_SIGNAL_REQ_WR
1077 enum ib_qp_type {
1079 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1080 * here (and in that order) since the MAD layer uses them as
1081 * indices into a 2-entry table.
1083 IB_QPT_SMI,
1084 IB_QPT_GSI,
1086 IB_QPT_RC,
1087 IB_QPT_UC,
1088 IB_QPT_UD,
1089 IB_QPT_RAW_IPV6,
1090 IB_QPT_RAW_ETHERTYPE,
1091 IB_QPT_RAW_PACKET = 8,
1092 IB_QPT_XRC_INI = 9,
1093 IB_QPT_XRC_TGT,
1094 IB_QPT_MAX,
1095 IB_QPT_DRIVER = 0xFF,
1096 /* Reserve a range for qp types internal to the low level driver.
1097 * These qp types will not be visible at the IB core layer, so the
1098 * IB_QPT_MAX usages should not be affected in the core layer
1100 IB_QPT_RESERVED1 = 0x1000,
1101 IB_QPT_RESERVED2,
1102 IB_QPT_RESERVED3,
1103 IB_QPT_RESERVED4,
1104 IB_QPT_RESERVED5,
1105 IB_QPT_RESERVED6,
1106 IB_QPT_RESERVED7,
1107 IB_QPT_RESERVED8,
1108 IB_QPT_RESERVED9,
1109 IB_QPT_RESERVED10,
1112 enum ib_qp_create_flags {
1113 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1114 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1115 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1116 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1117 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1118 IB_QP_CREATE_NETIF_QP = 1 << 5,
1119 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1120 /* FREE = 1 << 7, */
1121 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1122 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
1123 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1124 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
1125 /* reserve bits 26-31 for low level drivers' internal use */
1126 IB_QP_CREATE_RESERVED_START = 1 << 26,
1127 IB_QP_CREATE_RESERVED_END = 1 << 31,
1131 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1132 * callback to destroy the passed in QP.
1135 struct ib_qp_init_attr {
1136 /* Consumer's event_handler callback must not block */
1137 void (*event_handler)(struct ib_event *, void *);
1139 void *qp_context;
1140 struct ib_cq *send_cq;
1141 struct ib_cq *recv_cq;
1142 struct ib_srq *srq;
1143 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1144 struct ib_qp_cap cap;
1145 enum ib_sig_type sq_sig_type;
1146 enum ib_qp_type qp_type;
1147 u32 create_flags;
1150 * Only needed for special QP types, or when using the RW API.
1152 u8 port_num;
1153 struct ib_rwq_ind_table *rwq_ind_tbl;
1154 u32 source_qpn;
1157 struct ib_qp_open_attr {
1158 void (*event_handler)(struct ib_event *, void *);
1159 void *qp_context;
1160 u32 qp_num;
1161 enum ib_qp_type qp_type;
1164 enum ib_rnr_timeout {
1165 IB_RNR_TIMER_655_36 = 0,
1166 IB_RNR_TIMER_000_01 = 1,
1167 IB_RNR_TIMER_000_02 = 2,
1168 IB_RNR_TIMER_000_03 = 3,
1169 IB_RNR_TIMER_000_04 = 4,
1170 IB_RNR_TIMER_000_06 = 5,
1171 IB_RNR_TIMER_000_08 = 6,
1172 IB_RNR_TIMER_000_12 = 7,
1173 IB_RNR_TIMER_000_16 = 8,
1174 IB_RNR_TIMER_000_24 = 9,
1175 IB_RNR_TIMER_000_32 = 10,
1176 IB_RNR_TIMER_000_48 = 11,
1177 IB_RNR_TIMER_000_64 = 12,
1178 IB_RNR_TIMER_000_96 = 13,
1179 IB_RNR_TIMER_001_28 = 14,
1180 IB_RNR_TIMER_001_92 = 15,
1181 IB_RNR_TIMER_002_56 = 16,
1182 IB_RNR_TIMER_003_84 = 17,
1183 IB_RNR_TIMER_005_12 = 18,
1184 IB_RNR_TIMER_007_68 = 19,
1185 IB_RNR_TIMER_010_24 = 20,
1186 IB_RNR_TIMER_015_36 = 21,
1187 IB_RNR_TIMER_020_48 = 22,
1188 IB_RNR_TIMER_030_72 = 23,
1189 IB_RNR_TIMER_040_96 = 24,
1190 IB_RNR_TIMER_061_44 = 25,
1191 IB_RNR_TIMER_081_92 = 26,
1192 IB_RNR_TIMER_122_88 = 27,
1193 IB_RNR_TIMER_163_84 = 28,
1194 IB_RNR_TIMER_245_76 = 29,
1195 IB_RNR_TIMER_327_68 = 30,
1196 IB_RNR_TIMER_491_52 = 31
1199 enum ib_qp_attr_mask {
1200 IB_QP_STATE = 1,
1201 IB_QP_CUR_STATE = (1<<1),
1202 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1203 IB_QP_ACCESS_FLAGS = (1<<3),
1204 IB_QP_PKEY_INDEX = (1<<4),
1205 IB_QP_PORT = (1<<5),
1206 IB_QP_QKEY = (1<<6),
1207 IB_QP_AV = (1<<7),
1208 IB_QP_PATH_MTU = (1<<8),
1209 IB_QP_TIMEOUT = (1<<9),
1210 IB_QP_RETRY_CNT = (1<<10),
1211 IB_QP_RNR_RETRY = (1<<11),
1212 IB_QP_RQ_PSN = (1<<12),
1213 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1214 IB_QP_ALT_PATH = (1<<14),
1215 IB_QP_MIN_RNR_TIMER = (1<<15),
1216 IB_QP_SQ_PSN = (1<<16),
1217 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1218 IB_QP_PATH_MIG_STATE = (1<<18),
1219 IB_QP_CAP = (1<<19),
1220 IB_QP_DEST_QPN = (1<<20),
1221 IB_QP_RESERVED1 = (1<<21),
1222 IB_QP_RESERVED2 = (1<<22),
1223 IB_QP_RESERVED3 = (1<<23),
1224 IB_QP_RESERVED4 = (1<<24),
1225 IB_QP_RATE_LIMIT = (1<<25),
1228 enum ib_qp_state {
1229 IB_QPS_RESET,
1230 IB_QPS_INIT,
1231 IB_QPS_RTR,
1232 IB_QPS_RTS,
1233 IB_QPS_SQD,
1234 IB_QPS_SQE,
1235 IB_QPS_ERR
1238 enum ib_mig_state {
1239 IB_MIG_MIGRATED,
1240 IB_MIG_REARM,
1241 IB_MIG_ARMED
1244 enum ib_mw_type {
1245 IB_MW_TYPE_1 = 1,
1246 IB_MW_TYPE_2 = 2
1249 struct ib_qp_attr {
1250 enum ib_qp_state qp_state;
1251 enum ib_qp_state cur_qp_state;
1252 enum ib_mtu path_mtu;
1253 enum ib_mig_state path_mig_state;
1254 u32 qkey;
1255 u32 rq_psn;
1256 u32 sq_psn;
1257 u32 dest_qp_num;
1258 int qp_access_flags;
1259 struct ib_qp_cap cap;
1260 struct rdma_ah_attr ah_attr;
1261 struct rdma_ah_attr alt_ah_attr;
1262 u16 pkey_index;
1263 u16 alt_pkey_index;
1264 u8 en_sqd_async_notify;
1265 u8 sq_draining;
1266 u8 max_rd_atomic;
1267 u8 max_dest_rd_atomic;
1268 u8 min_rnr_timer;
1269 u8 port_num;
1270 u8 timeout;
1271 u8 retry_cnt;
1272 u8 rnr_retry;
1273 u8 alt_port_num;
1274 u8 alt_timeout;
1275 u32 rate_limit;
1276 struct net_device *xmit_slave;
1279 enum ib_wr_opcode {
1280 /* These are shared with userspace */
1281 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1282 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1283 IB_WR_SEND = IB_UVERBS_WR_SEND,
1284 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1285 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1286 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1287 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1288 IB_WR_LSO = IB_UVERBS_WR_TSO,
1289 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1290 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1291 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1292 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1293 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1294 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1295 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1297 /* These are kernel only and can not be issued by userspace */
1298 IB_WR_REG_MR = 0x20,
1299 IB_WR_REG_MR_INTEGRITY,
1301 /* reserve values for low level drivers' internal use.
1302 * These values will not be used at all in the ib core layer.
1304 IB_WR_RESERVED1 = 0xf0,
1305 IB_WR_RESERVED2,
1306 IB_WR_RESERVED3,
1307 IB_WR_RESERVED4,
1308 IB_WR_RESERVED5,
1309 IB_WR_RESERVED6,
1310 IB_WR_RESERVED7,
1311 IB_WR_RESERVED8,
1312 IB_WR_RESERVED9,
1313 IB_WR_RESERVED10,
1316 enum ib_send_flags {
1317 IB_SEND_FENCE = 1,
1318 IB_SEND_SIGNALED = (1<<1),
1319 IB_SEND_SOLICITED = (1<<2),
1320 IB_SEND_INLINE = (1<<3),
1321 IB_SEND_IP_CSUM = (1<<4),
1323 /* reserve bits 26-31 for low level drivers' internal use */
1324 IB_SEND_RESERVED_START = (1 << 26),
1325 IB_SEND_RESERVED_END = (1 << 31),
1328 struct ib_sge {
1329 u64 addr;
1330 u32 length;
1331 u32 lkey;
1334 struct ib_cqe {
1335 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1338 struct ib_send_wr {
1339 struct ib_send_wr *next;
1340 union {
1341 u64 wr_id;
1342 struct ib_cqe *wr_cqe;
1344 struct ib_sge *sg_list;
1345 int num_sge;
1346 enum ib_wr_opcode opcode;
1347 int send_flags;
1348 union {
1349 __be32 imm_data;
1350 u32 invalidate_rkey;
1351 } ex;
1354 struct ib_rdma_wr {
1355 struct ib_send_wr wr;
1356 u64 remote_addr;
1357 u32 rkey;
1360 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1362 return container_of(wr, struct ib_rdma_wr, wr);
1365 struct ib_atomic_wr {
1366 struct ib_send_wr wr;
1367 u64 remote_addr;
1368 u64 compare_add;
1369 u64 swap;
1370 u64 compare_add_mask;
1371 u64 swap_mask;
1372 u32 rkey;
1375 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1377 return container_of(wr, struct ib_atomic_wr, wr);
1380 struct ib_ud_wr {
1381 struct ib_send_wr wr;
1382 struct ib_ah *ah;
1383 void *header;
1384 int hlen;
1385 int mss;
1386 u32 remote_qpn;
1387 u32 remote_qkey;
1388 u16 pkey_index; /* valid for GSI only */
1389 u8 port_num; /* valid for DR SMPs on switch only */
1392 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1394 return container_of(wr, struct ib_ud_wr, wr);
1397 struct ib_reg_wr {
1398 struct ib_send_wr wr;
1399 struct ib_mr *mr;
1400 u32 key;
1401 int access;
1404 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1406 return container_of(wr, struct ib_reg_wr, wr);
1409 struct ib_recv_wr {
1410 struct ib_recv_wr *next;
1411 union {
1412 u64 wr_id;
1413 struct ib_cqe *wr_cqe;
1415 struct ib_sge *sg_list;
1416 int num_sge;
1419 enum ib_access_flags {
1420 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1421 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1422 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1423 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1424 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1425 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1426 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1427 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1428 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1430 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1431 IB_ACCESS_SUPPORTED =
1432 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1436 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1437 * are hidden here instead of a uapi header!
1439 enum ib_mr_rereg_flags {
1440 IB_MR_REREG_TRANS = 1,
1441 IB_MR_REREG_PD = (1<<1),
1442 IB_MR_REREG_ACCESS = (1<<2),
1443 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1446 struct ib_fmr_attr {
1447 int max_pages;
1448 int max_maps;
1449 u8 page_shift;
1452 struct ib_umem;
1454 enum rdma_remove_reason {
1456 * Userspace requested uobject deletion or initial try
1457 * to remove uobject via cleanup. Call could fail
1459 RDMA_REMOVE_DESTROY,
1460 /* Context deletion. This call should delete the actual object itself */
1461 RDMA_REMOVE_CLOSE,
1462 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1463 RDMA_REMOVE_DRIVER_REMOVE,
1464 /* uobj is being cleaned-up before being committed */
1465 RDMA_REMOVE_ABORT,
1468 struct ib_rdmacg_object {
1469 #ifdef CONFIG_CGROUP_RDMA
1470 struct rdma_cgroup *cg; /* owner rdma cgroup */
1471 #endif
1474 struct ib_ucontext {
1475 struct ib_device *device;
1476 struct ib_uverbs_file *ufile;
1478 * 'closing' can be read by the driver only during a destroy callback,
1479 * it is set when we are closing the file descriptor and indicates
1480 * that mm_sem may be locked.
1482 bool closing;
1484 bool cleanup_retryable;
1486 struct ib_rdmacg_object cg_obj;
1488 * Implementation details of the RDMA core, don't use in drivers:
1490 struct rdma_restrack_entry res;
1491 struct xarray mmap_xa;
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 */
1508 const struct uverbs_api_object *uapi_object;
1511 struct ib_udata {
1512 const void __user *inbuf;
1513 void __user *outbuf;
1514 size_t inlen;
1515 size_t outlen;
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 */
1525 u32 unsafe_global_rkey;
1528 * Implementation details of the RDMA core, don't use in drivers:
1530 struct ib_mr *__internal_mr;
1531 struct rdma_restrack_entry res;
1534 struct ib_xrcd {
1535 struct ib_device *device;
1536 atomic_t usecnt; /* count all exposed resources */
1537 struct inode *inode;
1539 struct mutex tgt_qp_mutex;
1540 struct list_head tgt_qp_list;
1543 struct ib_ah {
1544 struct ib_device *device;
1545 struct ib_pd *pd;
1546 struct ib_uobject *uobject;
1547 const struct ib_gid_attr *sgid_attr;
1548 enum rdma_ah_attr_type type;
1551 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1553 enum ib_poll_context {
1554 IB_POLL_DIRECT, /* caller context, no hw completions */
1555 IB_POLL_SOFTIRQ, /* poll from softirq context */
1556 IB_POLL_WORKQUEUE, /* poll from workqueue */
1557 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1560 struct ib_cq {
1561 struct ib_device *device;
1562 struct ib_ucq_object *uobject;
1563 ib_comp_handler comp_handler;
1564 void (*event_handler)(struct ib_event *, void *);
1565 void *cq_context;
1566 int cqe;
1567 atomic_t usecnt; /* count number of work queues */
1568 enum ib_poll_context poll_ctx;
1569 struct ib_wc *wc;
1570 union {
1571 struct irq_poll iop;
1572 struct work_struct work;
1574 struct workqueue_struct *comp_wq;
1575 struct dim *dim;
1577 /* updated only by trace points */
1578 ktime_t timestamp;
1579 bool interrupt;
1582 * Implementation details of the RDMA core, don't use in drivers:
1584 struct rdma_restrack_entry res;
1587 struct ib_srq {
1588 struct ib_device *device;
1589 struct ib_pd *pd;
1590 struct ib_usrq_object *uobject;
1591 void (*event_handler)(struct ib_event *, void *);
1592 void *srq_context;
1593 enum ib_srq_type srq_type;
1594 atomic_t usecnt;
1596 struct {
1597 struct ib_cq *cq;
1598 union {
1599 struct {
1600 struct ib_xrcd *xrcd;
1601 u32 srq_num;
1602 } xrc;
1604 } ext;
1607 enum ib_raw_packet_caps {
1608 /* Strip cvlan from incoming packet and report it in the matching work
1609 * completion is supported.
1611 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1612 /* Scatter FCS field of an incoming packet to host memory is supported.
1614 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1615 /* Checksum offloads are supported (for both send and receive). */
1616 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1617 /* When a packet is received for an RQ with no receive WQEs, the
1618 * packet processing is delayed.
1620 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1623 enum ib_wq_type {
1624 IB_WQT_RQ
1627 enum ib_wq_state {
1628 IB_WQS_RESET,
1629 IB_WQS_RDY,
1630 IB_WQS_ERR
1633 struct ib_wq {
1634 struct ib_device *device;
1635 struct ib_uwq_object *uobject;
1636 void *wq_context;
1637 void (*event_handler)(struct ib_event *, void *);
1638 struct ib_pd *pd;
1639 struct ib_cq *cq;
1640 u32 wq_num;
1641 enum ib_wq_state state;
1642 enum ib_wq_type wq_type;
1643 atomic_t usecnt;
1646 enum ib_wq_flags {
1647 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
1648 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
1649 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
1650 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
1653 struct ib_wq_init_attr {
1654 void *wq_context;
1655 enum ib_wq_type wq_type;
1656 u32 max_wr;
1657 u32 max_sge;
1658 struct ib_cq *cq;
1659 void (*event_handler)(struct ib_event *, void *);
1660 u32 create_flags; /* Use enum ib_wq_flags */
1663 enum ib_wq_attr_mask {
1664 IB_WQ_STATE = 1 << 0,
1665 IB_WQ_CUR_STATE = 1 << 1,
1666 IB_WQ_FLAGS = 1 << 2,
1669 struct ib_wq_attr {
1670 enum ib_wq_state wq_state;
1671 enum ib_wq_state curr_wq_state;
1672 u32 flags; /* Use enum ib_wq_flags */
1673 u32 flags_mask; /* Use enum ib_wq_flags */
1676 struct ib_rwq_ind_table {
1677 struct ib_device *device;
1678 struct ib_uobject *uobject;
1679 atomic_t usecnt;
1680 u32 ind_tbl_num;
1681 u32 log_ind_tbl_size;
1682 struct ib_wq **ind_tbl;
1685 struct ib_rwq_ind_table_init_attr {
1686 u32 log_ind_tbl_size;
1687 /* Each entry is a pointer to Receive Work Queue */
1688 struct ib_wq **ind_tbl;
1691 enum port_pkey_state {
1692 IB_PORT_PKEY_NOT_VALID = 0,
1693 IB_PORT_PKEY_VALID = 1,
1694 IB_PORT_PKEY_LISTED = 2,
1697 struct ib_qp_security;
1699 struct ib_port_pkey {
1700 enum port_pkey_state state;
1701 u16 pkey_index;
1702 u8 port_num;
1703 struct list_head qp_list;
1704 struct list_head to_error_list;
1705 struct ib_qp_security *sec;
1708 struct ib_ports_pkeys {
1709 struct ib_port_pkey main;
1710 struct ib_port_pkey alt;
1713 struct ib_qp_security {
1714 struct ib_qp *qp;
1715 struct ib_device *dev;
1716 /* Hold this mutex when changing port and pkey settings. */
1717 struct mutex mutex;
1718 struct ib_ports_pkeys *ports_pkeys;
1719 /* A list of all open shared QP handles. Required to enforce security
1720 * properly for all users of a shared QP.
1722 struct list_head shared_qp_list;
1723 void *security;
1724 bool destroying;
1725 atomic_t error_list_count;
1726 struct completion error_complete;
1727 int error_comps_pending;
1731 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1732 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1734 struct ib_qp {
1735 struct ib_device *device;
1736 struct ib_pd *pd;
1737 struct ib_cq *send_cq;
1738 struct ib_cq *recv_cq;
1739 spinlock_t mr_lock;
1740 int mrs_used;
1741 struct list_head rdma_mrs;
1742 struct list_head sig_mrs;
1743 struct ib_srq *srq;
1744 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1745 struct list_head xrcd_list;
1747 /* count times opened, mcast attaches, flow attaches */
1748 atomic_t usecnt;
1749 struct list_head open_list;
1750 struct ib_qp *real_qp;
1751 struct ib_uqp_object *uobject;
1752 void (*event_handler)(struct ib_event *, void *);
1753 void *qp_context;
1754 /* sgid_attrs associated with the AV's */
1755 const struct ib_gid_attr *av_sgid_attr;
1756 const struct ib_gid_attr *alt_path_sgid_attr;
1757 u32 qp_num;
1758 u32 max_write_sge;
1759 u32 max_read_sge;
1760 enum ib_qp_type qp_type;
1761 struct ib_rwq_ind_table *rwq_ind_tbl;
1762 struct ib_qp_security *qp_sec;
1763 u8 port;
1765 bool integrity_en;
1767 * Implementation details of the RDMA core, don't use in drivers:
1769 struct rdma_restrack_entry res;
1771 /* The counter the qp is bind to */
1772 struct rdma_counter *counter;
1775 struct ib_dm {
1776 struct ib_device *device;
1777 u32 length;
1778 u32 flags;
1779 struct ib_uobject *uobject;
1780 atomic_t usecnt;
1783 struct ib_mr {
1784 struct ib_device *device;
1785 struct ib_pd *pd;
1786 u32 lkey;
1787 u32 rkey;
1788 u64 iova;
1789 u64 length;
1790 unsigned int page_size;
1791 enum ib_mr_type type;
1792 bool need_inval;
1793 union {
1794 struct ib_uobject *uobject; /* user */
1795 struct list_head qp_entry; /* FR */
1798 struct ib_dm *dm;
1799 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1801 * Implementation details of the RDMA core, don't use in drivers:
1803 struct rdma_restrack_entry res;
1806 struct ib_mw {
1807 struct ib_device *device;
1808 struct ib_pd *pd;
1809 struct ib_uobject *uobject;
1810 u32 rkey;
1811 enum ib_mw_type type;
1814 struct ib_fmr {
1815 struct ib_device *device;
1816 struct ib_pd *pd;
1817 struct list_head list;
1818 u32 lkey;
1819 u32 rkey;
1822 /* Supported steering options */
1823 enum ib_flow_attr_type {
1824 /* steering according to rule specifications */
1825 IB_FLOW_ATTR_NORMAL = 0x0,
1826 /* default unicast and multicast rule -
1827 * receive all Eth traffic which isn't steered to any QP
1829 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1830 /* default multicast rule -
1831 * receive all Eth multicast traffic which isn't steered to any QP
1833 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1834 /* sniffer rule - receive all port traffic */
1835 IB_FLOW_ATTR_SNIFFER = 0x3
1838 /* Supported steering header types */
1839 enum ib_flow_spec_type {
1840 /* L2 headers*/
1841 IB_FLOW_SPEC_ETH = 0x20,
1842 IB_FLOW_SPEC_IB = 0x22,
1843 /* L3 header*/
1844 IB_FLOW_SPEC_IPV4 = 0x30,
1845 IB_FLOW_SPEC_IPV6 = 0x31,
1846 IB_FLOW_SPEC_ESP = 0x34,
1847 /* L4 headers*/
1848 IB_FLOW_SPEC_TCP = 0x40,
1849 IB_FLOW_SPEC_UDP = 0x41,
1850 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1851 IB_FLOW_SPEC_GRE = 0x51,
1852 IB_FLOW_SPEC_MPLS = 0x60,
1853 IB_FLOW_SPEC_INNER = 0x100,
1854 /* Actions */
1855 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1856 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1857 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1858 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1860 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1861 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1863 /* Flow steering rule priority is set according to it's domain.
1864 * Lower domain value means higher priority.
1866 enum ib_flow_domain {
1867 IB_FLOW_DOMAIN_USER,
1868 IB_FLOW_DOMAIN_ETHTOOL,
1869 IB_FLOW_DOMAIN_RFS,
1870 IB_FLOW_DOMAIN_NIC,
1871 IB_FLOW_DOMAIN_NUM /* Must be last */
1874 enum ib_flow_flags {
1875 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1876 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1877 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1880 struct ib_flow_eth_filter {
1881 u8 dst_mac[6];
1882 u8 src_mac[6];
1883 __be16 ether_type;
1884 __be16 vlan_tag;
1885 /* Must be last */
1886 u8 real_sz[];
1889 struct ib_flow_spec_eth {
1890 u32 type;
1891 u16 size;
1892 struct ib_flow_eth_filter val;
1893 struct ib_flow_eth_filter mask;
1896 struct ib_flow_ib_filter {
1897 __be16 dlid;
1898 __u8 sl;
1899 /* Must be last */
1900 u8 real_sz[];
1903 struct ib_flow_spec_ib {
1904 u32 type;
1905 u16 size;
1906 struct ib_flow_ib_filter val;
1907 struct ib_flow_ib_filter mask;
1910 /* IPv4 header flags */
1911 enum ib_ipv4_flags {
1912 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1913 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1914 last have this flag set */
1917 struct ib_flow_ipv4_filter {
1918 __be32 src_ip;
1919 __be32 dst_ip;
1920 u8 proto;
1921 u8 tos;
1922 u8 ttl;
1923 u8 flags;
1924 /* Must be last */
1925 u8 real_sz[];
1928 struct ib_flow_spec_ipv4 {
1929 u32 type;
1930 u16 size;
1931 struct ib_flow_ipv4_filter val;
1932 struct ib_flow_ipv4_filter mask;
1935 struct ib_flow_ipv6_filter {
1936 u8 src_ip[16];
1937 u8 dst_ip[16];
1938 __be32 flow_label;
1939 u8 next_hdr;
1940 u8 traffic_class;
1941 u8 hop_limit;
1942 /* Must be last */
1943 u8 real_sz[];
1946 struct ib_flow_spec_ipv6 {
1947 u32 type;
1948 u16 size;
1949 struct ib_flow_ipv6_filter val;
1950 struct ib_flow_ipv6_filter mask;
1953 struct ib_flow_tcp_udp_filter {
1954 __be16 dst_port;
1955 __be16 src_port;
1956 /* Must be last */
1957 u8 real_sz[];
1960 struct ib_flow_spec_tcp_udp {
1961 u32 type;
1962 u16 size;
1963 struct ib_flow_tcp_udp_filter val;
1964 struct ib_flow_tcp_udp_filter mask;
1967 struct ib_flow_tunnel_filter {
1968 __be32 tunnel_id;
1969 u8 real_sz[];
1972 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1973 * the tunnel_id from val has the vni value
1975 struct ib_flow_spec_tunnel {
1976 u32 type;
1977 u16 size;
1978 struct ib_flow_tunnel_filter val;
1979 struct ib_flow_tunnel_filter mask;
1982 struct ib_flow_esp_filter {
1983 __be32 spi;
1984 __be32 seq;
1985 /* Must be last */
1986 u8 real_sz[];
1989 struct ib_flow_spec_esp {
1990 u32 type;
1991 u16 size;
1992 struct ib_flow_esp_filter val;
1993 struct ib_flow_esp_filter mask;
1996 struct ib_flow_gre_filter {
1997 __be16 c_ks_res0_ver;
1998 __be16 protocol;
1999 __be32 key;
2000 /* Must be last */
2001 u8 real_sz[];
2004 struct ib_flow_spec_gre {
2005 u32 type;
2006 u16 size;
2007 struct ib_flow_gre_filter val;
2008 struct ib_flow_gre_filter mask;
2011 struct ib_flow_mpls_filter {
2012 __be32 tag;
2013 /* Must be last */
2014 u8 real_sz[];
2017 struct ib_flow_spec_mpls {
2018 u32 type;
2019 u16 size;
2020 struct ib_flow_mpls_filter val;
2021 struct ib_flow_mpls_filter mask;
2024 struct ib_flow_spec_action_tag {
2025 enum ib_flow_spec_type type;
2026 u16 size;
2027 u32 tag_id;
2030 struct ib_flow_spec_action_drop {
2031 enum ib_flow_spec_type type;
2032 u16 size;
2035 struct ib_flow_spec_action_handle {
2036 enum ib_flow_spec_type type;
2037 u16 size;
2038 struct ib_flow_action *act;
2041 enum ib_counters_description {
2042 IB_COUNTER_PACKETS,
2043 IB_COUNTER_BYTES,
2046 struct ib_flow_spec_action_count {
2047 enum ib_flow_spec_type type;
2048 u16 size;
2049 struct ib_counters *counters;
2052 union ib_flow_spec {
2053 struct {
2054 u32 type;
2055 u16 size;
2057 struct ib_flow_spec_eth eth;
2058 struct ib_flow_spec_ib ib;
2059 struct ib_flow_spec_ipv4 ipv4;
2060 struct ib_flow_spec_tcp_udp tcp_udp;
2061 struct ib_flow_spec_ipv6 ipv6;
2062 struct ib_flow_spec_tunnel tunnel;
2063 struct ib_flow_spec_esp esp;
2064 struct ib_flow_spec_gre gre;
2065 struct ib_flow_spec_mpls mpls;
2066 struct ib_flow_spec_action_tag flow_tag;
2067 struct ib_flow_spec_action_drop drop;
2068 struct ib_flow_spec_action_handle action;
2069 struct ib_flow_spec_action_count flow_count;
2072 struct ib_flow_attr {
2073 enum ib_flow_attr_type type;
2074 u16 size;
2075 u16 priority;
2076 u32 flags;
2077 u8 num_of_specs;
2078 u8 port;
2079 union ib_flow_spec flows[];
2082 struct ib_flow {
2083 struct ib_qp *qp;
2084 struct ib_device *device;
2085 struct ib_uobject *uobject;
2088 enum ib_flow_action_type {
2089 IB_FLOW_ACTION_UNSPECIFIED,
2090 IB_FLOW_ACTION_ESP = 1,
2093 struct ib_flow_action_attrs_esp_keymats {
2094 enum ib_uverbs_flow_action_esp_keymat protocol;
2095 union {
2096 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2097 } keymat;
2100 struct ib_flow_action_attrs_esp_replays {
2101 enum ib_uverbs_flow_action_esp_replay protocol;
2102 union {
2103 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2104 } replay;
2107 enum ib_flow_action_attrs_esp_flags {
2108 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2109 * This is done in order to share the same flags between user-space and
2110 * kernel and spare an unnecessary translation.
2113 /* Kernel flags */
2114 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2115 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2118 struct ib_flow_spec_list {
2119 struct ib_flow_spec_list *next;
2120 union ib_flow_spec spec;
2123 struct ib_flow_action_attrs_esp {
2124 struct ib_flow_action_attrs_esp_keymats *keymat;
2125 struct ib_flow_action_attrs_esp_replays *replay;
2126 struct ib_flow_spec_list *encap;
2127 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2128 * Value of 0 is a valid value.
2130 u32 esn;
2131 u32 spi;
2132 u32 seq;
2133 u32 tfc_pad;
2134 /* Use enum ib_flow_action_attrs_esp_flags */
2135 u64 flags;
2136 u64 hard_limit_pkts;
2139 struct ib_flow_action {
2140 struct ib_device *device;
2141 struct ib_uobject *uobject;
2142 enum ib_flow_action_type type;
2143 atomic_t usecnt;
2146 struct ib_mad;
2147 struct ib_grh;
2149 enum ib_process_mad_flags {
2150 IB_MAD_IGNORE_MKEY = 1,
2151 IB_MAD_IGNORE_BKEY = 2,
2152 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2155 enum ib_mad_result {
2156 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2157 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2158 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2159 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2162 struct ib_port_cache {
2163 u64 subnet_prefix;
2164 struct ib_pkey_cache *pkey;
2165 struct ib_gid_table *gid;
2166 u8 lmc;
2167 enum ib_port_state port_state;
2170 struct ib_port_immutable {
2171 int pkey_tbl_len;
2172 int gid_tbl_len;
2173 u32 core_cap_flags;
2174 u32 max_mad_size;
2177 struct ib_port_data {
2178 struct ib_device *ib_dev;
2180 struct ib_port_immutable immutable;
2182 spinlock_t pkey_list_lock;
2183 struct list_head pkey_list;
2185 struct ib_port_cache cache;
2187 spinlock_t netdev_lock;
2188 struct net_device __rcu *netdev;
2189 struct hlist_node ndev_hash_link;
2190 struct rdma_port_counter port_counter;
2191 struct rdma_hw_stats *hw_stats;
2194 /* rdma netdev type - specifies protocol type */
2195 enum rdma_netdev_t {
2196 RDMA_NETDEV_OPA_VNIC,
2197 RDMA_NETDEV_IPOIB,
2201 * struct rdma_netdev - rdma netdev
2202 * For cases where netstack interfacing is required.
2204 struct rdma_netdev {
2205 void *clnt_priv;
2206 struct ib_device *hca;
2207 u8 port_num;
2210 * cleanup function must be specified.
2211 * FIXME: This is only used for OPA_VNIC and that usage should be
2212 * removed too.
2214 void (*free_rdma_netdev)(struct net_device *netdev);
2216 /* control functions */
2217 void (*set_id)(struct net_device *netdev, int id);
2218 /* send packet */
2219 int (*send)(struct net_device *dev, struct sk_buff *skb,
2220 struct ib_ah *address, u32 dqpn);
2221 /* multicast */
2222 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2223 union ib_gid *gid, u16 mlid,
2224 int set_qkey, u32 qkey);
2225 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2226 union ib_gid *gid, u16 mlid);
2229 struct rdma_netdev_alloc_params {
2230 size_t sizeof_priv;
2231 unsigned int txqs;
2232 unsigned int rxqs;
2233 void *param;
2235 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2236 struct net_device *netdev, void *param);
2239 struct ib_odp_counters {
2240 atomic64_t faults;
2241 atomic64_t invalidations;
2244 struct ib_counters {
2245 struct ib_device *device;
2246 struct ib_uobject *uobject;
2247 /* num of objects attached */
2248 atomic_t usecnt;
2251 struct ib_counters_read_attr {
2252 u64 *counters_buff;
2253 u32 ncounters;
2254 u32 flags; /* use enum ib_read_counters_flags */
2257 struct uverbs_attr_bundle;
2258 struct iw_cm_id;
2259 struct iw_cm_conn_param;
2261 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2262 .size_##ib_struct = \
2263 (sizeof(struct drv_struct) + \
2264 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2265 BUILD_BUG_ON_ZERO( \
2266 !__same_type(((struct drv_struct *)NULL)->member, \
2267 struct ib_struct)))
2269 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2270 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2272 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2273 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2275 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2277 struct rdma_user_mmap_entry {
2278 struct kref ref;
2279 struct ib_ucontext *ucontext;
2280 unsigned long start_pgoff;
2281 size_t npages;
2282 bool driver_removed;
2285 /* Return the offset (in bytes) the user should pass to libc's mmap() */
2286 static inline u64
2287 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2289 return (u64)entry->start_pgoff << PAGE_SHIFT;
2293 * struct ib_device_ops - InfiniBand device operations
2294 * This structure defines all the InfiniBand device operations, providers will
2295 * need to define the supported operations, otherwise they will be set to null.
2297 struct ib_device_ops {
2298 struct module *owner;
2299 enum rdma_driver_id driver_id;
2300 u32 uverbs_abi_ver;
2301 unsigned int uverbs_no_driver_id_binding:1;
2303 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2304 const struct ib_send_wr **bad_send_wr);
2305 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2306 const struct ib_recv_wr **bad_recv_wr);
2307 void (*drain_rq)(struct ib_qp *qp);
2308 void (*drain_sq)(struct ib_qp *qp);
2309 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2310 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2311 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2312 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2313 int (*post_srq_recv)(struct ib_srq *srq,
2314 const struct ib_recv_wr *recv_wr,
2315 const struct ib_recv_wr **bad_recv_wr);
2316 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2317 u8 port_num, const struct ib_wc *in_wc,
2318 const struct ib_grh *in_grh,
2319 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2320 size_t *out_mad_size, u16 *out_mad_pkey_index);
2321 int (*query_device)(struct ib_device *device,
2322 struct ib_device_attr *device_attr,
2323 struct ib_udata *udata);
2324 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2325 struct ib_device_modify *device_modify);
2326 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2327 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2328 int comp_vector);
2329 int (*query_port)(struct ib_device *device, u8 port_num,
2330 struct ib_port_attr *port_attr);
2331 int (*modify_port)(struct ib_device *device, u8 port_num,
2332 int port_modify_mask,
2333 struct ib_port_modify *port_modify);
2335 * The following mandatory functions are used only at device
2336 * registration. Keep functions such as these at the end of this
2337 * structure to avoid cache line misses when accessing struct ib_device
2338 * in fast paths.
2340 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2341 struct ib_port_immutable *immutable);
2342 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2343 u8 port_num);
2345 * When calling get_netdev, the HW vendor's driver should return the
2346 * net device of device @device at port @port_num or NULL if such
2347 * a net device doesn't exist. The vendor driver should call dev_hold
2348 * on this net device. The HW vendor's device driver must guarantee
2349 * that this function returns NULL before the net device has finished
2350 * NETDEV_UNREGISTER state.
2352 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2354 * rdma netdev operation
2356 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2357 * must return -EOPNOTSUPP if it doesn't support the specified type.
2359 struct net_device *(*alloc_rdma_netdev)(
2360 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2361 const char *name, unsigned char name_assign_type,
2362 void (*setup)(struct net_device *));
2364 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2365 enum rdma_netdev_t type,
2366 struct rdma_netdev_alloc_params *params);
2368 * query_gid should be return GID value for @device, when @port_num
2369 * link layer is either IB or iWarp. It is no-op if @port_num port
2370 * is RoCE link layer.
2372 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2373 union ib_gid *gid);
2375 * When calling add_gid, the HW vendor's driver should add the gid
2376 * of device of port at gid index available at @attr. Meta-info of
2377 * that gid (for example, the network device related to this gid) is
2378 * available at @attr. @context allows the HW vendor driver to store
2379 * extra information together with a GID entry. The HW vendor driver may
2380 * allocate memory to contain this information and store it in @context
2381 * when a new GID entry is written to. Params are consistent until the
2382 * next call of add_gid or delete_gid. The function should return 0 on
2383 * success or error otherwise. The function could be called
2384 * concurrently for different ports. This function is only called when
2385 * roce_gid_table is used.
2387 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2389 * When calling del_gid, the HW vendor's driver should delete the
2390 * gid of device @device at gid index gid_index of port port_num
2391 * available in @attr.
2392 * Upon the deletion of a GID entry, the HW vendor must free any
2393 * allocated memory. The caller will clear @context afterwards.
2394 * This function is only called when roce_gid_table is used.
2396 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2397 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2398 u16 *pkey);
2399 int (*alloc_ucontext)(struct ib_ucontext *context,
2400 struct ib_udata *udata);
2401 void (*dealloc_ucontext)(struct ib_ucontext *context);
2402 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2404 * This will be called once refcount of an entry in mmap_xa reaches
2405 * zero. The type of the memory that was mapped may differ between
2406 * entries and is opaque to the rdma_user_mmap interface.
2407 * Therefore needs to be implemented by the driver in mmap_free.
2409 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2410 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2411 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2412 void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2413 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2414 struct ib_udata *udata);
2415 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2416 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2417 void (*destroy_ah)(struct ib_ah *ah, u32 flags);
2418 int (*create_srq)(struct ib_srq *srq,
2419 struct ib_srq_init_attr *srq_init_attr,
2420 struct ib_udata *udata);
2421 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2422 enum ib_srq_attr_mask srq_attr_mask,
2423 struct ib_udata *udata);
2424 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2425 void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2426 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2427 struct ib_qp_init_attr *qp_init_attr,
2428 struct ib_udata *udata);
2429 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2430 int qp_attr_mask, struct ib_udata *udata);
2431 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2432 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2433 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2434 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2435 struct ib_udata *udata);
2436 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2437 void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2438 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2439 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2440 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2441 u64 virt_addr, int mr_access_flags,
2442 struct ib_udata *udata);
2443 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2444 u64 virt_addr, int mr_access_flags,
2445 struct ib_pd *pd, struct ib_udata *udata);
2446 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2447 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2448 u32 max_num_sg, struct ib_udata *udata);
2449 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2450 u32 max_num_data_sg,
2451 u32 max_num_meta_sg);
2452 int (*advise_mr)(struct ib_pd *pd,
2453 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2454 struct ib_sge *sg_list, u32 num_sge,
2455 struct uverbs_attr_bundle *attrs);
2456 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2457 unsigned int *sg_offset);
2458 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2459 struct ib_mr_status *mr_status);
2460 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2461 struct ib_udata *udata);
2462 int (*dealloc_mw)(struct ib_mw *mw);
2463 struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2464 struct ib_fmr_attr *fmr_attr);
2465 int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2466 u64 iova);
2467 int (*unmap_fmr)(struct list_head *fmr_list);
2468 int (*dealloc_fmr)(struct ib_fmr *fmr);
2469 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2470 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2471 struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
2472 struct ib_udata *udata);
2473 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2474 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2475 struct ib_flow_attr *flow_attr,
2476 int domain, struct ib_udata *udata);
2477 int (*destroy_flow)(struct ib_flow *flow_id);
2478 struct ib_flow_action *(*create_flow_action_esp)(
2479 struct ib_device *device,
2480 const struct ib_flow_action_attrs_esp *attr,
2481 struct uverbs_attr_bundle *attrs);
2482 int (*destroy_flow_action)(struct ib_flow_action *action);
2483 int (*modify_flow_action_esp)(
2484 struct ib_flow_action *action,
2485 const struct ib_flow_action_attrs_esp *attr,
2486 struct uverbs_attr_bundle *attrs);
2487 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2488 int state);
2489 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2490 struct ifla_vf_info *ivf);
2491 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2492 struct ifla_vf_stats *stats);
2493 int (*get_vf_guid)(struct ib_device *device, int vf, u8 port,
2494 struct ifla_vf_guid *node_guid,
2495 struct ifla_vf_guid *port_guid);
2496 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2497 int type);
2498 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2499 struct ib_wq_init_attr *init_attr,
2500 struct ib_udata *udata);
2501 void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2502 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2503 u32 wq_attr_mask, struct ib_udata *udata);
2504 struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2505 struct ib_device *device,
2506 struct ib_rwq_ind_table_init_attr *init_attr,
2507 struct ib_udata *udata);
2508 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2509 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2510 struct ib_ucontext *context,
2511 struct ib_dm_alloc_attr *attr,
2512 struct uverbs_attr_bundle *attrs);
2513 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2514 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2515 struct ib_dm_mr_attr *attr,
2516 struct uverbs_attr_bundle *attrs);
2517 struct ib_counters *(*create_counters)(
2518 struct ib_device *device, struct uverbs_attr_bundle *attrs);
2519 int (*destroy_counters)(struct ib_counters *counters);
2520 int (*read_counters)(struct ib_counters *counters,
2521 struct ib_counters_read_attr *counters_read_attr,
2522 struct uverbs_attr_bundle *attrs);
2523 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2524 int data_sg_nents, unsigned int *data_sg_offset,
2525 struct scatterlist *meta_sg, int meta_sg_nents,
2526 unsigned int *meta_sg_offset);
2529 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2530 * driver initialized data. The struct is kfree()'ed by the sysfs
2531 * core when the device is removed. A lifespan of -1 in the return
2532 * struct tells the core to set a default lifespan.
2534 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2535 u8 port_num);
2537 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2538 * @index - The index in the value array we wish to have updated, or
2539 * num_counters if we want all stats updated
2540 * Return codes -
2541 * < 0 - Error, no counters updated
2542 * index - Updated the single counter pointed to by index
2543 * num_counters - Updated all counters (will reset the timestamp
2544 * and prevent further calls for lifespan milliseconds)
2545 * Drivers are allowed to update all counters in leiu of just the
2546 * one given in index at their option
2548 int (*get_hw_stats)(struct ib_device *device,
2549 struct rdma_hw_stats *stats, u8 port, int index);
2551 * This function is called once for each port when a ib device is
2552 * registered.
2554 int (*init_port)(struct ib_device *device, u8 port_num,
2555 struct kobject *port_sysfs);
2557 * Allows rdma drivers to add their own restrack attributes.
2559 int (*fill_res_entry)(struct sk_buff *msg,
2560 struct rdma_restrack_entry *entry);
2562 /* Device lifecycle callbacks */
2564 * Called after the device becomes registered, before clients are
2565 * attached
2567 int (*enable_driver)(struct ib_device *dev);
2569 * This is called as part of ib_dealloc_device().
2571 void (*dealloc_driver)(struct ib_device *dev);
2573 /* iWarp CM callbacks */
2574 void (*iw_add_ref)(struct ib_qp *qp);
2575 void (*iw_rem_ref)(struct ib_qp *qp);
2576 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2577 int (*iw_connect)(struct iw_cm_id *cm_id,
2578 struct iw_cm_conn_param *conn_param);
2579 int (*iw_accept)(struct iw_cm_id *cm_id,
2580 struct iw_cm_conn_param *conn_param);
2581 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2582 u8 pdata_len);
2583 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2584 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2586 * counter_bind_qp - Bind a QP to a counter.
2587 * @counter - The counter to be bound. If counter->id is zero then
2588 * the driver needs to allocate a new counter and set counter->id
2590 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2592 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2593 * counter and bind it onto the default one
2595 int (*counter_unbind_qp)(struct ib_qp *qp);
2597 * counter_dealloc -De-allocate the hw counter
2599 int (*counter_dealloc)(struct rdma_counter *counter);
2601 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2602 * the driver initialized data.
2604 struct rdma_hw_stats *(*counter_alloc_stats)(
2605 struct rdma_counter *counter);
2607 * counter_update_stats - Query the stats value of this counter
2609 int (*counter_update_stats)(struct rdma_counter *counter);
2612 * Allows rdma drivers to add their own restrack attributes
2613 * dumped via 'rdma stat' iproute2 command.
2615 int (*fill_stat_entry)(struct sk_buff *msg,
2616 struct rdma_restrack_entry *entry);
2618 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2619 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2620 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2621 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2622 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2625 struct ib_core_device {
2626 /* device must be the first element in structure until,
2627 * union of ib_core_device and device exists in ib_device.
2629 struct device dev;
2630 possible_net_t rdma_net;
2631 struct kobject *ports_kobj;
2632 struct list_head port_list;
2633 struct ib_device *owner; /* reach back to owner ib_device */
2636 struct rdma_restrack_root;
2637 struct ib_device {
2638 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2639 struct device *dma_device;
2640 struct ib_device_ops ops;
2641 char name[IB_DEVICE_NAME_MAX];
2642 struct rcu_head rcu_head;
2644 struct list_head event_handler_list;
2645 /* Protects event_handler_list */
2646 struct rw_semaphore event_handler_rwsem;
2648 /* Protects QP's event_handler calls and open_qp list */
2649 spinlock_t qp_open_list_lock;
2651 struct rw_semaphore client_data_rwsem;
2652 struct xarray client_data;
2653 struct mutex unregistration_lock;
2655 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2656 rwlock_t cache_lock;
2658 * port_data is indexed by port number
2660 struct ib_port_data *port_data;
2662 int num_comp_vectors;
2664 union {
2665 struct device dev;
2666 struct ib_core_device coredev;
2669 /* First group for device attributes,
2670 * Second group for driver provided attributes (optional).
2671 * It is NULL terminated array.
2673 const struct attribute_group *groups[3];
2675 u64 uverbs_cmd_mask;
2676 u64 uverbs_ex_cmd_mask;
2678 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2679 __be64 node_guid;
2680 u32 local_dma_lkey;
2681 u16 is_switch:1;
2682 /* Indicates kernel verbs support, should not be used in drivers */
2683 u16 kverbs_provider:1;
2684 /* CQ adaptive moderation (RDMA DIM) */
2685 u16 use_cq_dim:1;
2686 u8 node_type;
2687 u8 phys_port_cnt;
2688 struct ib_device_attr attrs;
2689 struct attribute_group *hw_stats_ag;
2690 struct rdma_hw_stats *hw_stats;
2692 #ifdef CONFIG_CGROUP_RDMA
2693 struct rdmacg_device cg_device;
2694 #endif
2696 u32 index;
2697 struct rdma_restrack_root *res;
2699 const struct uapi_definition *driver_def;
2702 * Positive refcount indicates that the device is currently
2703 * registered and cannot be unregistered.
2705 refcount_t refcount;
2706 struct completion unreg_completion;
2707 struct work_struct unregistration_work;
2709 const struct rdma_link_ops *link_ops;
2711 /* Protects compat_devs xarray modifications */
2712 struct mutex compat_devs_mutex;
2713 /* Maintains compat devices for each net namespace */
2714 struct xarray compat_devs;
2716 /* Used by iWarp CM */
2717 char iw_ifname[IFNAMSIZ];
2718 u32 iw_driver_flags;
2719 u32 lag_flags;
2722 struct ib_client_nl_info;
2723 struct ib_client {
2724 const char *name;
2725 int (*add)(struct ib_device *ibdev);
2726 void (*remove)(struct ib_device *, void *client_data);
2727 void (*rename)(struct ib_device *dev, void *client_data);
2728 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2729 struct ib_client_nl_info *res);
2730 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2732 /* Returns the net_dev belonging to this ib_client and matching the
2733 * given parameters.
2734 * @dev: An RDMA device that the net_dev use for communication.
2735 * @port: A physical port number on the RDMA device.
2736 * @pkey: P_Key that the net_dev uses if applicable.
2737 * @gid: A GID that the net_dev uses to communicate.
2738 * @addr: An IP address the net_dev is configured with.
2739 * @client_data: The device's client data set by ib_set_client_data().
2741 * An ib_client that implements a net_dev on top of RDMA devices
2742 * (such as IP over IB) should implement this callback, allowing the
2743 * rdma_cm module to find the right net_dev for a given request.
2745 * The caller is responsible for calling dev_put on the returned
2746 * netdev. */
2747 struct net_device *(*get_net_dev_by_params)(
2748 struct ib_device *dev,
2749 u8 port,
2750 u16 pkey,
2751 const union ib_gid *gid,
2752 const struct sockaddr *addr,
2753 void *client_data);
2755 refcount_t uses;
2756 struct completion uses_zero;
2757 u32 client_id;
2759 /* kverbs are not required by the client */
2760 u8 no_kverbs_req:1;
2764 * IB block DMA iterator
2766 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2767 * to a HW supported page size.
2769 struct ib_block_iter {
2770 /* internal states */
2771 struct scatterlist *__sg; /* sg holding the current aligned block */
2772 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2773 unsigned int __sg_nents; /* number of SG entries */
2774 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2775 unsigned int __pg_bit; /* alignment of current block */
2778 struct ib_device *_ib_alloc_device(size_t size);
2779 #define ib_alloc_device(drv_struct, member) \
2780 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2781 BUILD_BUG_ON_ZERO(offsetof( \
2782 struct drv_struct, member))), \
2783 struct drv_struct, member)
2785 void ib_dealloc_device(struct ib_device *device);
2787 void ib_get_device_fw_str(struct ib_device *device, char *str);
2789 int ib_register_device(struct ib_device *device, const char *name);
2790 void ib_unregister_device(struct ib_device *device);
2791 void ib_unregister_driver(enum rdma_driver_id driver_id);
2792 void ib_unregister_device_and_put(struct ib_device *device);
2793 void ib_unregister_device_queued(struct ib_device *ib_dev);
2795 int ib_register_client (struct ib_client *client);
2796 void ib_unregister_client(struct ib_client *client);
2798 void __rdma_block_iter_start(struct ib_block_iter *biter,
2799 struct scatterlist *sglist,
2800 unsigned int nents,
2801 unsigned long pgsz);
2802 bool __rdma_block_iter_next(struct ib_block_iter *biter);
2805 * rdma_block_iter_dma_address - get the aligned dma address of the current
2806 * block held by the block iterator.
2807 * @biter: block iterator holding the memory block
2809 static inline dma_addr_t
2810 rdma_block_iter_dma_address(struct ib_block_iter *biter)
2812 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2816 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2817 * @sglist: sglist to iterate over
2818 * @biter: block iterator holding the memory block
2819 * @nents: maximum number of sg entries to iterate over
2820 * @pgsz: best HW supported page size to use
2822 * Callers may use rdma_block_iter_dma_address() to get each
2823 * blocks aligned DMA address.
2825 #define rdma_for_each_block(sglist, biter, nents, pgsz) \
2826 for (__rdma_block_iter_start(biter, sglist, nents, \
2827 pgsz); \
2828 __rdma_block_iter_next(biter);)
2831 * ib_get_client_data - Get IB client context
2832 * @device:Device to get context for
2833 * @client:Client to get context for
2835 * ib_get_client_data() returns the client context data set with
2836 * ib_set_client_data(). This can only be called while the client is
2837 * registered to the device, once the ib_client remove() callback returns this
2838 * cannot be called.
2840 static inline void *ib_get_client_data(struct ib_device *device,
2841 struct ib_client *client)
2843 return xa_load(&device->client_data, client->client_id);
2845 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2846 void *data);
2847 void ib_set_device_ops(struct ib_device *device,
2848 const struct ib_device_ops *ops);
2850 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2851 unsigned long pfn, unsigned long size, pgprot_t prot,
2852 struct rdma_user_mmap_entry *entry);
2853 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2854 struct rdma_user_mmap_entry *entry,
2855 size_t length);
2856 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2857 struct rdma_user_mmap_entry *entry,
2858 size_t length, u32 min_pgoff,
2859 u32 max_pgoff);
2861 struct rdma_user_mmap_entry *
2862 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2863 unsigned long pgoff);
2864 struct rdma_user_mmap_entry *
2865 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2866 struct vm_area_struct *vma);
2867 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2869 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2871 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2873 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2876 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2878 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2881 static inline bool ib_is_buffer_cleared(const void __user *p,
2882 size_t len)
2884 bool ret;
2885 u8 *buf;
2887 if (len > USHRT_MAX)
2888 return false;
2890 buf = memdup_user(p, len);
2891 if (IS_ERR(buf))
2892 return false;
2894 ret = !memchr_inv(buf, 0, len);
2895 kfree(buf);
2896 return ret;
2899 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2900 size_t offset,
2901 size_t len)
2903 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2907 * ib_is_destroy_retryable - Check whether the uobject destruction
2908 * is retryable.
2909 * @ret: The initial destruction return code
2910 * @why: remove reason
2911 * @uobj: The uobject that is destroyed
2913 * This function is a helper function that IB layer and low-level drivers
2914 * can use to consider whether the destruction of the given uobject is
2915 * retry-able.
2916 * It checks the original return code, if it wasn't success the destruction
2917 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2918 * the remove reason. (i.e. why).
2919 * Must be called with the object locked for destroy.
2921 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2922 struct ib_uobject *uobj)
2924 return ret && (why == RDMA_REMOVE_DESTROY ||
2925 uobj->context->cleanup_retryable);
2929 * ib_destroy_usecnt - Called during destruction to check the usecnt
2930 * @usecnt: The usecnt atomic
2931 * @why: remove reason
2932 * @uobj: The uobject that is destroyed
2934 * Non-zero usecnts will block destruction unless destruction was triggered by
2935 * a ucontext cleanup.
2937 static inline int ib_destroy_usecnt(atomic_t *usecnt,
2938 enum rdma_remove_reason why,
2939 struct ib_uobject *uobj)
2941 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2942 return -EBUSY;
2943 return 0;
2947 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2948 * contains all required attributes and no attributes not allowed for
2949 * the given QP state transition.
2950 * @cur_state: Current QP state
2951 * @next_state: Next QP state
2952 * @type: QP type
2953 * @mask: Mask of supplied QP attributes
2955 * This function is a helper function that a low-level driver's
2956 * modify_qp method can use to validate the consumer's input. It
2957 * checks that cur_state and next_state are valid QP states, that a
2958 * transition from cur_state to next_state is allowed by the IB spec,
2959 * and that the attribute mask supplied is allowed for the transition.
2961 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2962 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2964 void ib_register_event_handler(struct ib_event_handler *event_handler);
2965 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2966 void ib_dispatch_event(const struct ib_event *event);
2968 int ib_query_port(struct ib_device *device,
2969 u8 port_num, struct ib_port_attr *port_attr);
2971 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2972 u8 port_num);
2975 * rdma_cap_ib_switch - Check if the device is IB switch
2976 * @device: Device to check
2978 * Device driver is responsible for setting is_switch bit on
2979 * in ib_device structure at init time.
2981 * Return: true if the device is IB switch.
2983 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2985 return device->is_switch;
2989 * rdma_start_port - Return the first valid port number for the device
2990 * specified
2992 * @device: Device to be checked
2994 * Return start port number
2996 static inline u8 rdma_start_port(const struct ib_device *device)
2998 return rdma_cap_ib_switch(device) ? 0 : 1;
3002 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3003 * @device - The struct ib_device * to iterate over
3004 * @iter - The unsigned int to store the port number
3006 #define rdma_for_each_port(device, iter) \
3007 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
3008 unsigned int, iter))); \
3009 iter <= rdma_end_port(device); (iter)++)
3012 * rdma_end_port - Return the last valid port number for the device
3013 * specified
3015 * @device: Device to be checked
3017 * Return last port number
3019 static inline u8 rdma_end_port(const struct ib_device *device)
3021 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3024 static inline int rdma_is_port_valid(const struct ib_device *device,
3025 unsigned int port)
3027 return (port >= rdma_start_port(device) &&
3028 port <= rdma_end_port(device));
3031 static inline bool rdma_is_grh_required(const struct ib_device *device,
3032 u8 port_num)
3034 return device->port_data[port_num].immutable.core_cap_flags &
3035 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3038 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
3040 return device->port_data[port_num].immutable.core_cap_flags &
3041 RDMA_CORE_CAP_PROT_IB;
3044 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
3046 return device->port_data[port_num].immutable.core_cap_flags &
3047 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3050 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
3052 return device->port_data[port_num].immutable.core_cap_flags &
3053 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3056 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
3058 return device->port_data[port_num].immutable.core_cap_flags &
3059 RDMA_CORE_CAP_PROT_ROCE;
3062 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
3064 return device->port_data[port_num].immutable.core_cap_flags &
3065 RDMA_CORE_CAP_PROT_IWARP;
3068 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
3070 return rdma_protocol_ib(device, port_num) ||
3071 rdma_protocol_roce(device, port_num);
3074 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
3076 return device->port_data[port_num].immutable.core_cap_flags &
3077 RDMA_CORE_CAP_PROT_RAW_PACKET;
3080 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
3082 return device->port_data[port_num].immutable.core_cap_flags &
3083 RDMA_CORE_CAP_PROT_USNIC;
3087 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3088 * Management Datagrams.
3089 * @device: Device to check
3090 * @port_num: Port number to check
3092 * Management Datagrams (MAD) are a required part of the InfiniBand
3093 * specification and are supported on all InfiniBand devices. A slightly
3094 * extended version are also supported on OPA interfaces.
3096 * Return: true if the port supports sending/receiving of MAD packets.
3098 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
3100 return device->port_data[port_num].immutable.core_cap_flags &
3101 RDMA_CORE_CAP_IB_MAD;
3105 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3106 * Management Datagrams.
3107 * @device: Device to check
3108 * @port_num: Port number to check
3110 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3111 * datagrams with their own versions. These OPA MADs share many but not all of
3112 * the characteristics of InfiniBand MADs.
3114 * OPA MADs differ in the following ways:
3116 * 1) MADs are variable size up to 2K
3117 * IBTA defined MADs remain fixed at 256 bytes
3118 * 2) OPA SMPs must carry valid PKeys
3119 * 3) OPA SMP packets are a different format
3121 * Return: true if the port supports OPA MAD packet formats.
3123 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3125 return device->port_data[port_num].immutable.core_cap_flags &
3126 RDMA_CORE_CAP_OPA_MAD;
3130 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3131 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3132 * @device: Device to check
3133 * @port_num: Port number to check
3135 * Each InfiniBand node is required to provide a Subnet Management Agent
3136 * that the subnet manager can access. Prior to the fabric being fully
3137 * configured by the subnet manager, the SMA is accessed via a well known
3138 * interface called the Subnet Management Interface (SMI). This interface
3139 * uses directed route packets to communicate with the SM to get around the
3140 * chicken and egg problem of the SM needing to know what's on the fabric
3141 * in order to configure the fabric, and needing to configure the fabric in
3142 * order to send packets to the devices on the fabric. These directed
3143 * route packets do not need the fabric fully configured in order to reach
3144 * their destination. The SMI is the only method allowed to send
3145 * directed route packets on an InfiniBand fabric.
3147 * Return: true if the port provides an SMI.
3149 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3151 return device->port_data[port_num].immutable.core_cap_flags &
3152 RDMA_CORE_CAP_IB_SMI;
3156 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3157 * Communication Manager.
3158 * @device: Device to check
3159 * @port_num: Port number to check
3161 * The InfiniBand Communication Manager is one of many pre-defined General
3162 * Service Agents (GSA) that are accessed via the General Service
3163 * Interface (GSI). It's role is to facilitate establishment of connections
3164 * between nodes as well as other management related tasks for established
3165 * connections.
3167 * Return: true if the port supports an IB CM (this does not guarantee that
3168 * a CM is actually running however).
3170 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3172 return device->port_data[port_num].immutable.core_cap_flags &
3173 RDMA_CORE_CAP_IB_CM;
3177 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3178 * Communication Manager.
3179 * @device: Device to check
3180 * @port_num: Port number to check
3182 * Similar to above, but specific to iWARP connections which have a different
3183 * managment protocol than InfiniBand.
3185 * Return: true if the port supports an iWARP CM (this does not guarantee that
3186 * a CM is actually running however).
3188 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3190 return device->port_data[port_num].immutable.core_cap_flags &
3191 RDMA_CORE_CAP_IW_CM;
3195 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3196 * Subnet Administration.
3197 * @device: Device to check
3198 * @port_num: Port number to check
3200 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3201 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3202 * fabrics, devices should resolve routes to other hosts by contacting the
3203 * SA to query the proper route.
3205 * Return: true if the port should act as a client to the fabric Subnet
3206 * Administration interface. This does not imply that the SA service is
3207 * running locally.
3209 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3211 return device->port_data[port_num].immutable.core_cap_flags &
3212 RDMA_CORE_CAP_IB_SA;
3216 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3217 * Multicast.
3218 * @device: Device to check
3219 * @port_num: Port number to check
3221 * InfiniBand multicast registration is more complex than normal IPv4 or
3222 * IPv6 multicast registration. Each Host Channel Adapter must register
3223 * with the Subnet Manager when it wishes to join a multicast group. It
3224 * should do so only once regardless of how many queue pairs it subscribes
3225 * to this group. And it should leave the group only after all queue pairs
3226 * attached to the group have been detached.
3228 * Return: true if the port must undertake the additional adminstrative
3229 * overhead of registering/unregistering with the SM and tracking of the
3230 * total number of queue pairs attached to the multicast group.
3232 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3234 return rdma_cap_ib_sa(device, port_num);
3238 * rdma_cap_af_ib - Check if the port of device has the capability
3239 * Native Infiniband Address.
3240 * @device: Device to check
3241 * @port_num: Port number to check
3243 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3244 * GID. RoCE uses a different mechanism, but still generates a GID via
3245 * a prescribed mechanism and port specific data.
3247 * Return: true if the port uses a GID address to identify devices on the
3248 * network.
3250 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3252 return device->port_data[port_num].immutable.core_cap_flags &
3253 RDMA_CORE_CAP_AF_IB;
3257 * rdma_cap_eth_ah - Check if the port of device has the capability
3258 * Ethernet Address Handle.
3259 * @device: Device to check
3260 * @port_num: Port number to check
3262 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3263 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3264 * port. Normally, packet headers are generated by the sending host
3265 * adapter, but when sending connectionless datagrams, we must manually
3266 * inject the proper headers for the fabric we are communicating over.
3268 * Return: true if we are running as a RoCE port and must force the
3269 * addition of a Global Route Header built from our Ethernet Address
3270 * Handle into our header list for connectionless packets.
3272 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3274 return device->port_data[port_num].immutable.core_cap_flags &
3275 RDMA_CORE_CAP_ETH_AH;
3279 * rdma_cap_opa_ah - Check if the port of device supports
3280 * OPA Address handles
3281 * @device: Device to check
3282 * @port_num: Port number to check
3284 * Return: true if we are running on an OPA device which supports
3285 * the extended OPA addressing.
3287 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3289 return (device->port_data[port_num].immutable.core_cap_flags &
3290 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3294 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3296 * @device: Device
3297 * @port_num: Port number
3299 * This MAD size includes the MAD headers and MAD payload. No other headers
3300 * are included.
3302 * Return the max MAD size required by the Port. Will return 0 if the port
3303 * does not support MADs
3305 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3307 return device->port_data[port_num].immutable.max_mad_size;
3311 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3312 * @device: Device to check
3313 * @port_num: Port number to check
3315 * RoCE GID table mechanism manages the various GIDs for a device.
3317 * NOTE: if allocating the port's GID table has failed, this call will still
3318 * return true, but any RoCE GID table API will fail.
3320 * Return: true if the port uses RoCE GID table mechanism in order to manage
3321 * its GIDs.
3323 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3324 u8 port_num)
3326 return rdma_protocol_roce(device, port_num) &&
3327 device->ops.add_gid && device->ops.del_gid;
3331 * Check if the device supports READ W/ INVALIDATE.
3333 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3336 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3337 * has support for it yet.
3339 return rdma_protocol_iwarp(dev, port_num);
3343 * rdma_find_pg_bit - Find page bit given address and HW supported page sizes
3345 * @addr: address
3346 * @pgsz_bitmap: bitmap of HW supported page sizes
3348 static inline unsigned int rdma_find_pg_bit(unsigned long addr,
3349 unsigned long pgsz_bitmap)
3351 unsigned long align;
3352 unsigned long pgsz;
3354 align = addr & -addr;
3356 /* Find page bit such that addr is aligned to the highest supported
3357 * HW page size
3359 pgsz = pgsz_bitmap & ~(-align << 1);
3360 if (!pgsz)
3361 return __ffs(pgsz_bitmap);
3363 return __fls(pgsz);
3366 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3367 int state);
3368 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3369 struct ifla_vf_info *info);
3370 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3371 struct ifla_vf_stats *stats);
3372 int ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
3373 struct ifla_vf_guid *node_guid,
3374 struct ifla_vf_guid *port_guid);
3375 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3376 int type);
3378 int ib_query_pkey(struct ib_device *device,
3379 u8 port_num, u16 index, u16 *pkey);
3381 int ib_modify_device(struct ib_device *device,
3382 int device_modify_mask,
3383 struct ib_device_modify *device_modify);
3385 int ib_modify_port(struct ib_device *device,
3386 u8 port_num, int port_modify_mask,
3387 struct ib_port_modify *port_modify);
3389 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3390 u8 *port_num, u16 *index);
3392 int ib_find_pkey(struct ib_device *device,
3393 u8 port_num, u16 pkey, u16 *index);
3395 enum ib_pd_flags {
3397 * Create a memory registration for all memory in the system and place
3398 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3399 * ULPs to avoid the overhead of dynamic MRs.
3401 * This flag is generally considered unsafe and must only be used in
3402 * extremly trusted environments. Every use of it will log a warning
3403 * in the kernel log.
3405 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3408 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3409 const char *caller);
3411 #define ib_alloc_pd(device, flags) \
3412 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3415 * ib_dealloc_pd_user - Deallocate kernel/user PD
3416 * @pd: The protection domain
3417 * @udata: Valid user data or NULL for kernel objects
3419 void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3422 * ib_dealloc_pd - Deallocate kernel PD
3423 * @pd: The protection domain
3425 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3427 static inline void ib_dealloc_pd(struct ib_pd *pd)
3429 ib_dealloc_pd_user(pd, NULL);
3432 enum rdma_create_ah_flags {
3433 /* In a sleepable context */
3434 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3438 * rdma_create_ah - Creates an address handle for the given address vector.
3439 * @pd: The protection domain associated with the address handle.
3440 * @ah_attr: The attributes of the address vector.
3441 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3443 * The address handle is used to reference a local or global destination
3444 * in all UD QP post sends.
3446 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3447 u32 flags);
3450 * rdma_create_user_ah - Creates an address handle for the given address vector.
3451 * It resolves destination mac address for ah attribute of RoCE type.
3452 * @pd: The protection domain associated with the address handle.
3453 * @ah_attr: The attributes of the address vector.
3454 * @udata: pointer to user's input output buffer information need by
3455 * provider driver.
3457 * It returns 0 on success and returns appropriate error code on error.
3458 * The address handle is used to reference a local or global destination
3459 * in all UD QP post sends.
3461 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3462 struct rdma_ah_attr *ah_attr,
3463 struct ib_udata *udata);
3465 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3466 * work completion.
3467 * @hdr: the L3 header to parse
3468 * @net_type: type of header to parse
3469 * @sgid: place to store source gid
3470 * @dgid: place to store destination gid
3472 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3473 enum rdma_network_type net_type,
3474 union ib_gid *sgid, union ib_gid *dgid);
3477 * ib_get_rdma_header_version - Get the header version
3478 * @hdr: the L3 header to parse
3480 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3483 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3484 * work completion.
3485 * @device: Device on which the received message arrived.
3486 * @port_num: Port on which the received message arrived.
3487 * @wc: Work completion associated with the received message.
3488 * @grh: References the received global route header. This parameter is
3489 * ignored unless the work completion indicates that the GRH is valid.
3490 * @ah_attr: Returned attributes that can be used when creating an address
3491 * handle for replying to the message.
3492 * When ib_init_ah_attr_from_wc() returns success,
3493 * (a) for IB link layer it optionally contains a reference to SGID attribute
3494 * when GRH is present for IB link layer.
3495 * (b) for RoCE link layer it contains a reference to SGID attribute.
3496 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3497 * attributes which are initialized using ib_init_ah_attr_from_wc().
3500 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3501 const struct ib_wc *wc, const struct ib_grh *grh,
3502 struct rdma_ah_attr *ah_attr);
3505 * ib_create_ah_from_wc - Creates an address handle associated with the
3506 * sender of the specified work completion.
3507 * @pd: The protection domain associated with the address handle.
3508 * @wc: Work completion information associated with a received message.
3509 * @grh: References the received global route header. This parameter is
3510 * ignored unless the work completion indicates that the GRH is valid.
3511 * @port_num: The outbound port number to associate with the address.
3513 * The address handle is used to reference a local or global destination
3514 * in all UD QP post sends.
3516 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3517 const struct ib_grh *grh, u8 port_num);
3520 * rdma_modify_ah - Modifies the address vector associated with an address
3521 * handle.
3522 * @ah: The address handle to modify.
3523 * @ah_attr: The new address vector attributes to associate with the
3524 * address handle.
3526 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3529 * rdma_query_ah - Queries the address vector associated with an address
3530 * handle.
3531 * @ah: The address handle to query.
3532 * @ah_attr: The address vector attributes associated with the address
3533 * handle.
3535 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3537 enum rdma_destroy_ah_flags {
3538 /* In a sleepable context */
3539 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3543 * rdma_destroy_ah_user - Destroys an address handle.
3544 * @ah: The address handle to destroy.
3545 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3546 * @udata: Valid user data or NULL for kernel objects
3548 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3551 * rdma_destroy_ah - Destroys an kernel address handle.
3552 * @ah: The address handle to destroy.
3553 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3555 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3557 static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3559 return rdma_destroy_ah_user(ah, flags, NULL);
3562 struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3563 struct ib_srq_init_attr *srq_init_attr,
3564 struct ib_usrq_object *uobject,
3565 struct ib_udata *udata);
3566 static inline struct ib_srq *
3567 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3569 if (!pd->device->ops.create_srq)
3570 return ERR_PTR(-EOPNOTSUPP);
3572 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3576 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3577 * @srq: The SRQ to modify.
3578 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3579 * the current values of selected SRQ attributes are returned.
3580 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3581 * are being modified.
3583 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3584 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3585 * the number of receives queued drops below the limit.
3587 int ib_modify_srq(struct ib_srq *srq,
3588 struct ib_srq_attr *srq_attr,
3589 enum ib_srq_attr_mask srq_attr_mask);
3592 * ib_query_srq - Returns the attribute list and current values for the
3593 * specified SRQ.
3594 * @srq: The SRQ to query.
3595 * @srq_attr: The attributes of the specified SRQ.
3597 int ib_query_srq(struct ib_srq *srq,
3598 struct ib_srq_attr *srq_attr);
3601 * ib_destroy_srq_user - Destroys the specified SRQ.
3602 * @srq: The SRQ to destroy.
3603 * @udata: Valid user data or NULL for kernel objects
3605 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3608 * ib_destroy_srq - Destroys the specified kernel SRQ.
3609 * @srq: The SRQ to destroy.
3611 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3613 static inline int ib_destroy_srq(struct ib_srq *srq)
3615 return ib_destroy_srq_user(srq, NULL);
3619 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3620 * @srq: The SRQ to post the work request on.
3621 * @recv_wr: A list of work requests to post on the receive queue.
3622 * @bad_recv_wr: On an immediate failure, this parameter will reference
3623 * the work request that failed to be posted on the QP.
3625 static inline int ib_post_srq_recv(struct ib_srq *srq,
3626 const struct ib_recv_wr *recv_wr,
3627 const struct ib_recv_wr **bad_recv_wr)
3629 const struct ib_recv_wr *dummy;
3631 return srq->device->ops.post_srq_recv(srq, recv_wr,
3632 bad_recv_wr ? : &dummy);
3635 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3636 struct ib_qp_init_attr *qp_init_attr);
3639 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3640 * @qp: The QP to modify.
3641 * @attr: On input, specifies the QP attributes to modify. On output,
3642 * the current values of selected QP attributes are returned.
3643 * @attr_mask: A bit-mask used to specify which attributes of the QP
3644 * are being modified.
3645 * @udata: pointer to user's input output buffer information
3646 * are being modified.
3647 * It returns 0 on success and returns appropriate error code on error.
3649 int ib_modify_qp_with_udata(struct ib_qp *qp,
3650 struct ib_qp_attr *attr,
3651 int attr_mask,
3652 struct ib_udata *udata);
3655 * ib_modify_qp - Modifies the attributes for the specified QP and then
3656 * transitions the QP to the given state.
3657 * @qp: The QP to modify.
3658 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3659 * the current values of selected QP attributes are returned.
3660 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3661 * are being modified.
3663 int ib_modify_qp(struct ib_qp *qp,
3664 struct ib_qp_attr *qp_attr,
3665 int qp_attr_mask);
3668 * ib_query_qp - Returns the attribute list and current values for the
3669 * specified QP.
3670 * @qp: The QP to query.
3671 * @qp_attr: The attributes of the specified QP.
3672 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3673 * @qp_init_attr: Additional attributes of the selected QP.
3675 * The qp_attr_mask may be used to limit the query to gathering only the
3676 * selected attributes.
3678 int ib_query_qp(struct ib_qp *qp,
3679 struct ib_qp_attr *qp_attr,
3680 int qp_attr_mask,
3681 struct ib_qp_init_attr *qp_init_attr);
3684 * ib_destroy_qp - Destroys the specified QP.
3685 * @qp: The QP to destroy.
3686 * @udata: Valid udata or NULL for kernel objects
3688 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3691 * ib_destroy_qp - Destroys the specified kernel QP.
3692 * @qp: The QP to destroy.
3694 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3696 static inline int ib_destroy_qp(struct ib_qp *qp)
3698 return ib_destroy_qp_user(qp, NULL);
3702 * ib_open_qp - Obtain a reference to an existing sharable QP.
3703 * @xrcd - XRC domain
3704 * @qp_open_attr: Attributes identifying the QP to open.
3706 * Returns a reference to a sharable QP.
3708 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3709 struct ib_qp_open_attr *qp_open_attr);
3712 * ib_close_qp - Release an external reference to a QP.
3713 * @qp: The QP handle to release
3715 * The opened QP handle is released by the caller. The underlying
3716 * shared QP is not destroyed until all internal references are released.
3718 int ib_close_qp(struct ib_qp *qp);
3721 * ib_post_send - Posts a list of work requests to the send queue of
3722 * the specified QP.
3723 * @qp: The QP to post the work request on.
3724 * @send_wr: A list of work requests to post on the send queue.
3725 * @bad_send_wr: On an immediate failure, this parameter will reference
3726 * the work request that failed to be posted on the QP.
3728 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3729 * error is returned, the QP state shall not be affected,
3730 * ib_post_send() will return an immediate error after queueing any
3731 * earlier work requests in the list.
3733 static inline int ib_post_send(struct ib_qp *qp,
3734 const struct ib_send_wr *send_wr,
3735 const struct ib_send_wr **bad_send_wr)
3737 const struct ib_send_wr *dummy;
3739 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3743 * ib_post_recv - Posts a list of work requests to the receive queue of
3744 * the specified QP.
3745 * @qp: The QP to post the work request on.
3746 * @recv_wr: A list of work requests to post on the receive queue.
3747 * @bad_recv_wr: On an immediate failure, this parameter will reference
3748 * the work request that failed to be posted on the QP.
3750 static inline int ib_post_recv(struct ib_qp *qp,
3751 const struct ib_recv_wr *recv_wr,
3752 const struct ib_recv_wr **bad_recv_wr)
3754 const struct ib_recv_wr *dummy;
3756 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3759 struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
3760 int nr_cqe, int comp_vector,
3761 enum ib_poll_context poll_ctx,
3762 const char *caller, struct ib_udata *udata);
3765 * ib_alloc_cq_user: Allocate kernel/user CQ
3766 * @dev: The IB device
3767 * @private: Private data attached to the CQE
3768 * @nr_cqe: Number of CQEs in the CQ
3769 * @comp_vector: Completion vector used for the IRQs
3770 * @poll_ctx: Context used for polling the CQ
3771 * @udata: Valid user data or NULL for kernel objects
3773 static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
3774 void *private, int nr_cqe,
3775 int comp_vector,
3776 enum ib_poll_context poll_ctx,
3777 struct ib_udata *udata)
3779 return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3780 KBUILD_MODNAME, udata);
3784 * ib_alloc_cq: Allocate kernel CQ
3785 * @dev: The IB device
3786 * @private: Private data attached to the CQE
3787 * @nr_cqe: Number of CQEs in the CQ
3788 * @comp_vector: Completion vector used for the IRQs
3789 * @poll_ctx: Context used for polling the CQ
3791 * NOTE: for user cq use ib_alloc_cq_user with valid udata!
3793 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3794 int nr_cqe, int comp_vector,
3795 enum ib_poll_context poll_ctx)
3797 return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3798 NULL);
3801 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3802 int nr_cqe, enum ib_poll_context poll_ctx,
3803 const char *caller);
3806 * ib_alloc_cq_any: Allocate kernel CQ
3807 * @dev: The IB device
3808 * @private: Private data attached to the CQE
3809 * @nr_cqe: Number of CQEs in the CQ
3810 * @poll_ctx: Context used for polling the CQ
3812 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3813 void *private, int nr_cqe,
3814 enum ib_poll_context poll_ctx)
3816 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3817 KBUILD_MODNAME);
3821 * ib_free_cq_user - Free kernel/user CQ
3822 * @cq: The CQ to free
3823 * @udata: Valid user data or NULL for kernel objects
3825 void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3828 * ib_free_cq - Free kernel CQ
3829 * @cq: The CQ to free
3831 * NOTE: for user cq use ib_free_cq_user with valid udata!
3833 static inline void ib_free_cq(struct ib_cq *cq)
3835 ib_free_cq_user(cq, NULL);
3838 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3841 * ib_create_cq - Creates a CQ on the specified device.
3842 * @device: The device on which to create the CQ.
3843 * @comp_handler: A user-specified callback that is invoked when a
3844 * completion event occurs on the CQ.
3845 * @event_handler: A user-specified callback that is invoked when an
3846 * asynchronous event not associated with a completion occurs on the CQ.
3847 * @cq_context: Context associated with the CQ returned to the user via
3848 * the associated completion and event handlers.
3849 * @cq_attr: The attributes the CQ should be created upon.
3851 * Users can examine the cq structure to determine the actual CQ size.
3853 struct ib_cq *__ib_create_cq(struct ib_device *device,
3854 ib_comp_handler comp_handler,
3855 void (*event_handler)(struct ib_event *, void *),
3856 void *cq_context,
3857 const struct ib_cq_init_attr *cq_attr,
3858 const char *caller);
3859 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3860 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3863 * ib_resize_cq - Modifies the capacity of the CQ.
3864 * @cq: The CQ to resize.
3865 * @cqe: The minimum size of the CQ.
3867 * Users can examine the cq structure to determine the actual CQ size.
3869 int ib_resize_cq(struct ib_cq *cq, int cqe);
3872 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3873 * @cq: The CQ to modify.
3874 * @cq_count: number of CQEs that will trigger an event
3875 * @cq_period: max period of time in usec before triggering an event
3878 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3881 * ib_destroy_cq_user - Destroys the specified CQ.
3882 * @cq: The CQ to destroy.
3883 * @udata: Valid user data or NULL for kernel objects
3885 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3888 * ib_destroy_cq - Destroys the specified kernel CQ.
3889 * @cq: The CQ to destroy.
3891 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3893 static inline void ib_destroy_cq(struct ib_cq *cq)
3895 ib_destroy_cq_user(cq, NULL);
3899 * ib_poll_cq - poll a CQ for completion(s)
3900 * @cq:the CQ being polled
3901 * @num_entries:maximum number of completions to return
3902 * @wc:array of at least @num_entries &struct ib_wc where completions
3903 * will be returned
3905 * Poll a CQ for (possibly multiple) completions. If the return value
3906 * is < 0, an error occurred. If the return value is >= 0, it is the
3907 * number of completions returned. If the return value is
3908 * non-negative and < num_entries, then the CQ was emptied.
3910 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3911 struct ib_wc *wc)
3913 return cq->device->ops.poll_cq(cq, num_entries, wc);
3917 * ib_req_notify_cq - Request completion notification on a CQ.
3918 * @cq: The CQ to generate an event for.
3919 * @flags:
3920 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3921 * to request an event on the next solicited event or next work
3922 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3923 * may also be |ed in to request a hint about missed events, as
3924 * described below.
3926 * Return Value:
3927 * < 0 means an error occurred while requesting notification
3928 * == 0 means notification was requested successfully, and if
3929 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3930 * were missed and it is safe to wait for another event. In
3931 * this case is it guaranteed that any work completions added
3932 * to the CQ since the last CQ poll will trigger a completion
3933 * notification event.
3934 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3935 * in. It means that the consumer must poll the CQ again to
3936 * make sure it is empty to avoid missing an event because of a
3937 * race between requesting notification and an entry being
3938 * added to the CQ. This return value means it is possible
3939 * (but not guaranteed) that a work completion has been added
3940 * to the CQ since the last poll without triggering a
3941 * completion notification event.
3943 static inline int ib_req_notify_cq(struct ib_cq *cq,
3944 enum ib_cq_notify_flags flags)
3946 return cq->device->ops.req_notify_cq(cq, flags);
3950 * ib_req_ncomp_notif - Request completion notification when there are
3951 * at least the specified number of unreaped completions on the CQ.
3952 * @cq: The CQ to generate an event for.
3953 * @wc_cnt: The number of unreaped completions that should be on the
3954 * CQ before an event is generated.
3956 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3958 return cq->device->ops.req_ncomp_notif ?
3959 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
3960 -ENOSYS;
3964 * ib_dma_mapping_error - check a DMA addr for error
3965 * @dev: The device for which the dma_addr was created
3966 * @dma_addr: The DMA address to check
3968 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3970 return dma_mapping_error(dev->dma_device, dma_addr);
3974 * ib_dma_map_single - Map a kernel virtual address to DMA address
3975 * @dev: The device for which the dma_addr is to be created
3976 * @cpu_addr: The kernel virtual address
3977 * @size: The size of the region in bytes
3978 * @direction: The direction of the DMA
3980 static inline u64 ib_dma_map_single(struct ib_device *dev,
3981 void *cpu_addr, size_t size,
3982 enum dma_data_direction direction)
3984 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3988 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3989 * @dev: The device for which the DMA address was created
3990 * @addr: The DMA address
3991 * @size: The size of the region in bytes
3992 * @direction: The direction of the DMA
3994 static inline void ib_dma_unmap_single(struct ib_device *dev,
3995 u64 addr, size_t size,
3996 enum dma_data_direction direction)
3998 dma_unmap_single(dev->dma_device, addr, size, direction);
4002 * ib_dma_map_page - Map a physical page to DMA address
4003 * @dev: The device for which the dma_addr is to be created
4004 * @page: The page to be mapped
4005 * @offset: The offset within the page
4006 * @size: The size of the region in bytes
4007 * @direction: The direction of the DMA
4009 static inline u64 ib_dma_map_page(struct ib_device *dev,
4010 struct page *page,
4011 unsigned long offset,
4012 size_t size,
4013 enum dma_data_direction direction)
4015 return dma_map_page(dev->dma_device, page, offset, size, direction);
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
4025 static inline void ib_dma_unmap_page(struct ib_device *dev,
4026 u64 addr, size_t size,
4027 enum dma_data_direction direction)
4029 dma_unmap_page(dev->dma_device, addr, size, direction);
4033 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4034 * @dev: The device for which the DMA addresses are to be created
4035 * @sg: The array of scatter/gather entries
4036 * @nents: The number of scatter/gather entries
4037 * @direction: The direction of the DMA
4039 static inline int ib_dma_map_sg(struct ib_device *dev,
4040 struct scatterlist *sg, int nents,
4041 enum dma_data_direction direction)
4043 return dma_map_sg(dev->dma_device, sg, nents, direction);
4047 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4048 * @dev: The device for which the DMA addresses were created
4049 * @sg: The array of scatter/gather entries
4050 * @nents: The number of scatter/gather entries
4051 * @direction: The direction of the DMA
4053 static inline void ib_dma_unmap_sg(struct ib_device *dev,
4054 struct scatterlist *sg, int nents,
4055 enum dma_data_direction direction)
4057 dma_unmap_sg(dev->dma_device, sg, nents, direction);
4060 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4061 struct scatterlist *sg, int nents,
4062 enum dma_data_direction direction,
4063 unsigned long dma_attrs)
4065 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4066 dma_attrs);
4069 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4070 struct scatterlist *sg, int nents,
4071 enum dma_data_direction direction,
4072 unsigned long dma_attrs)
4074 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
4078 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4079 * @dev: The device to query
4081 * The returned value represents a size in bytes.
4083 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4085 return dma_get_max_seg_size(dev->dma_device);
4089 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4090 * @dev: The device for which the DMA address was created
4091 * @addr: The DMA address
4092 * @size: The size of the region in bytes
4093 * @dir: The direction of the DMA
4095 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4096 u64 addr,
4097 size_t size,
4098 enum dma_data_direction dir)
4100 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4104 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4105 * @dev: The device for which the DMA address was created
4106 * @addr: The DMA address
4107 * @size: The size of the region in bytes
4108 * @dir: The direction of the DMA
4110 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4111 u64 addr,
4112 size_t size,
4113 enum dma_data_direction dir)
4115 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4119 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4120 * @dev: The device for which the DMA address is requested
4121 * @size: The size of the region to allocate in bytes
4122 * @dma_handle: A pointer for returning the DMA address of the region
4123 * @flag: memory allocator flags
4125 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4126 size_t size,
4127 dma_addr_t *dma_handle,
4128 gfp_t flag)
4130 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4134 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4135 * @dev: The device for which the DMA addresses were allocated
4136 * @size: The size of the region
4137 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4138 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4140 static inline void ib_dma_free_coherent(struct ib_device *dev,
4141 size_t size, void *cpu_addr,
4142 dma_addr_t dma_handle)
4144 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4147 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4148 * space. This function should be called when 'current' is the owning MM.
4150 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4151 u64 virt_addr, int mr_access_flags);
4153 /* ib_advise_mr - give an advice about an address range in a memory region */
4154 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4155 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4157 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4158 * HCA translation table.
4159 * @mr: The memory region to deregister.
4160 * @udata: Valid user data or NULL for kernel object
4162 * This function can fail, if the memory region has memory windows bound to it.
4164 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4167 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4168 * HCA translation table.
4169 * @mr: The memory region to deregister.
4171 * This function can fail, if the memory region has memory windows bound to it.
4173 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4175 static inline int ib_dereg_mr(struct ib_mr *mr)
4177 return ib_dereg_mr_user(mr, NULL);
4180 struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type,
4181 u32 max_num_sg, struct ib_udata *udata);
4183 static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
4184 enum ib_mr_type mr_type, u32 max_num_sg)
4186 return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL);
4189 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4190 u32 max_num_data_sg,
4191 u32 max_num_meta_sg);
4194 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4195 * R_Key and L_Key.
4196 * @mr - struct ib_mr pointer to be updated.
4197 * @newkey - new key to be used.
4199 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4201 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4202 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4206 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4207 * for calculating a new rkey for type 2 memory windows.
4208 * @rkey - the rkey to increment.
4210 static inline u32 ib_inc_rkey(u32 rkey)
4212 const u32 mask = 0x000000ff;
4213 return ((rkey + 1) & mask) | (rkey & ~mask);
4217 * ib_alloc_fmr - Allocates a unmapped fast memory region.
4218 * @pd: The protection domain associated with the unmapped region.
4219 * @mr_access_flags: Specifies the memory access rights.
4220 * @fmr_attr: Attributes of the unmapped region.
4222 * A fast memory region must be mapped before it can be used as part of
4223 * a work request.
4225 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
4226 int mr_access_flags,
4227 struct ib_fmr_attr *fmr_attr);
4230 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
4231 * @fmr: The fast memory region to associate with the pages.
4232 * @page_list: An array of physical pages to map to the fast memory region.
4233 * @list_len: The number of pages in page_list.
4234 * @iova: The I/O virtual address to use with the mapped region.
4236 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
4237 u64 *page_list, int list_len,
4238 u64 iova)
4240 return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
4244 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
4245 * @fmr_list: A linked list of fast memory regions to unmap.
4247 int ib_unmap_fmr(struct list_head *fmr_list);
4250 * ib_dealloc_fmr - Deallocates a fast memory region.
4251 * @fmr: The fast memory region to deallocate.
4253 int ib_dealloc_fmr(struct ib_fmr *fmr);
4256 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4257 * @qp: QP to attach to the multicast group. The QP must be type
4258 * IB_QPT_UD.
4259 * @gid: Multicast group GID.
4260 * @lid: Multicast group LID in host byte order.
4262 * In order to send and receive multicast packets, subnet
4263 * administration must have created the multicast group and configured
4264 * the fabric appropriately. The port associated with the specified
4265 * QP must also be a member of the multicast group.
4267 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4270 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4271 * @qp: QP to detach from the multicast group.
4272 * @gid: Multicast group GID.
4273 * @lid: Multicast group LID in host byte order.
4275 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4278 * ib_alloc_xrcd - Allocates an XRC domain.
4279 * @device: The device on which to allocate the XRC domain.
4280 * @caller: Module name for kernel consumers
4282 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
4283 #define ib_alloc_xrcd(device) \
4284 __ib_alloc_xrcd((device), KBUILD_MODNAME)
4287 * ib_dealloc_xrcd - Deallocates an XRC domain.
4288 * @xrcd: The XRC domain to deallocate.
4289 * @udata: Valid user data or NULL for kernel object
4291 int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
4293 static inline int ib_check_mr_access(int flags)
4296 * Local write permission is required if remote write or
4297 * remote atomic permission is also requested.
4299 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4300 !(flags & IB_ACCESS_LOCAL_WRITE))
4301 return -EINVAL;
4303 if (flags & ~IB_ACCESS_SUPPORTED)
4304 return -EINVAL;
4306 return 0;
4309 static inline bool ib_access_writable(int access_flags)
4312 * We have writable memory backing the MR if any of the following
4313 * access flags are set. "Local write" and "remote write" obviously
4314 * require write access. "Remote atomic" can do things like fetch and
4315 * add, which will modify memory, and "MW bind" can change permissions
4316 * by binding a window.
4318 return access_flags &
4319 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4320 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4324 * ib_check_mr_status: lightweight check of MR status.
4325 * This routine may provide status checks on a selected
4326 * ib_mr. first use is for signature status check.
4328 * @mr: A memory region.
4329 * @check_mask: Bitmask of which checks to perform from
4330 * ib_mr_status_check enumeration.
4331 * @mr_status: The container of relevant status checks.
4332 * failed checks will be indicated in the status bitmask
4333 * and the relevant info shall be in the error item.
4335 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4336 struct ib_mr_status *mr_status);
4339 * ib_device_try_get: Hold a registration lock
4340 * device: The device to lock
4342 * A device under an active registration lock cannot become unregistered. It
4343 * is only possible to obtain a registration lock on a device that is fully
4344 * registered, otherwise this function returns false.
4346 * The registration lock is only necessary for actions which require the
4347 * device to still be registered. Uses that only require the device pointer to
4348 * be valid should use get_device(&ibdev->dev) to hold the memory.
4351 static inline bool ib_device_try_get(struct ib_device *dev)
4353 return refcount_inc_not_zero(&dev->refcount);
4356 void ib_device_put(struct ib_device *device);
4357 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4358 enum rdma_driver_id driver_id);
4359 struct ib_device *ib_device_get_by_name(const char *name,
4360 enum rdma_driver_id driver_id);
4361 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4362 u16 pkey, const union ib_gid *gid,
4363 const struct sockaddr *addr);
4364 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4365 unsigned int port);
4366 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4368 struct ib_wq *ib_create_wq(struct ib_pd *pd,
4369 struct ib_wq_init_attr *init_attr);
4370 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
4371 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4372 u32 wq_attr_mask);
4373 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4374 struct ib_rwq_ind_table_init_attr*
4375 wq_ind_table_init_attr);
4376 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4378 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4379 unsigned int *sg_offset, unsigned int page_size);
4380 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4381 int data_sg_nents, unsigned int *data_sg_offset,
4382 struct scatterlist *meta_sg, int meta_sg_nents,
4383 unsigned int *meta_sg_offset, unsigned int page_size);
4385 static inline int
4386 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4387 unsigned int *sg_offset, unsigned int page_size)
4389 int n;
4391 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4392 mr->iova = 0;
4394 return n;
4397 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4398 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4400 void ib_drain_rq(struct ib_qp *qp);
4401 void ib_drain_sq(struct ib_qp *qp);
4402 void ib_drain_qp(struct ib_qp *qp);
4404 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4406 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4408 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4409 return attr->roce.dmac;
4410 return NULL;
4413 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4415 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4416 attr->ib.dlid = (u16)dlid;
4417 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4418 attr->opa.dlid = dlid;
4421 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4423 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4424 return attr->ib.dlid;
4425 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4426 return attr->opa.dlid;
4427 return 0;
4430 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4432 attr->sl = sl;
4435 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4437 return attr->sl;
4440 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4441 u8 src_path_bits)
4443 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4444 attr->ib.src_path_bits = src_path_bits;
4445 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4446 attr->opa.src_path_bits = src_path_bits;
4449 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4451 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4452 return attr->ib.src_path_bits;
4453 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4454 return attr->opa.src_path_bits;
4455 return 0;
4458 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4459 bool make_grd)
4461 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4462 attr->opa.make_grd = make_grd;
4465 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4467 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4468 return attr->opa.make_grd;
4469 return false;
4472 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4474 attr->port_num = port_num;
4477 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4479 return attr->port_num;
4482 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4483 u8 static_rate)
4485 attr->static_rate = static_rate;
4488 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4490 return attr->static_rate;
4493 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4494 enum ib_ah_flags flag)
4496 attr->ah_flags = flag;
4499 static inline enum ib_ah_flags
4500 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4502 return attr->ah_flags;
4505 static inline const struct ib_global_route
4506 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4508 return &attr->grh;
4511 /*To retrieve and modify the grh */
4512 static inline struct ib_global_route
4513 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4515 return &attr->grh;
4518 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4520 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4522 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4525 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4526 __be64 prefix)
4528 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4530 grh->dgid.global.subnet_prefix = prefix;
4533 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4534 __be64 if_id)
4536 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4538 grh->dgid.global.interface_id = if_id;
4541 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4542 union ib_gid *dgid, u32 flow_label,
4543 u8 sgid_index, u8 hop_limit,
4544 u8 traffic_class)
4546 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4548 attr->ah_flags = IB_AH_GRH;
4549 if (dgid)
4550 grh->dgid = *dgid;
4551 grh->flow_label = flow_label;
4552 grh->sgid_index = sgid_index;
4553 grh->hop_limit = hop_limit;
4554 grh->traffic_class = traffic_class;
4555 grh->sgid_attr = NULL;
4558 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4559 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4560 u32 flow_label, u8 hop_limit, u8 traffic_class,
4561 const struct ib_gid_attr *sgid_attr);
4562 void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4563 const struct rdma_ah_attr *src);
4564 void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4565 const struct rdma_ah_attr *new);
4566 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4569 * rdma_ah_find_type - Return address handle type.
4571 * @dev: Device to be checked
4572 * @port_num: Port number
4574 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4575 u8 port_num)
4577 if (rdma_protocol_roce(dev, port_num))
4578 return RDMA_AH_ATTR_TYPE_ROCE;
4579 if (rdma_protocol_ib(dev, port_num)) {
4580 if (rdma_cap_opa_ah(dev, port_num))
4581 return RDMA_AH_ATTR_TYPE_OPA;
4582 return RDMA_AH_ATTR_TYPE_IB;
4585 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4589 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4590 * In the current implementation the only way to get
4591 * get the 32bit lid is from other sources for OPA.
4592 * For IB, lids will always be 16bits so cast the
4593 * value accordingly.
4595 * @lid: A 32bit LID
4597 static inline u16 ib_lid_cpu16(u32 lid)
4599 WARN_ON_ONCE(lid & 0xFFFF0000);
4600 return (u16)lid;
4604 * ib_lid_be16 - Return lid in 16bit BE encoding.
4606 * @lid: A 32bit LID
4608 static inline __be16 ib_lid_be16(u32 lid)
4610 WARN_ON_ONCE(lid & 0xFFFF0000);
4611 return cpu_to_be16((u16)lid);
4615 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4616 * vector
4617 * @device: the rdma device
4618 * @comp_vector: index of completion vector
4620 * Returns NULL on failure, otherwise a corresponding cpu map of the
4621 * completion vector (returns all-cpus map if the device driver doesn't
4622 * implement get_vector_affinity).
4624 static inline const struct cpumask *
4625 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4627 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4628 !device->ops.get_vector_affinity)
4629 return NULL;
4631 return device->ops.get_vector_affinity(device, comp_vector);
4636 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4637 * and add their gids, as needed, to the relevant RoCE devices.
4639 * @device: the rdma device
4641 void rdma_roce_rescan_device(struct ib_device *ibdev);
4643 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4645 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4647 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4648 enum rdma_netdev_t type, const char *name,
4649 unsigned char name_assign_type,
4650 void (*setup)(struct net_device *));
4652 int rdma_init_netdev(struct ib_device *device, u8 port_num,
4653 enum rdma_netdev_t type, const char *name,
4654 unsigned char name_assign_type,
4655 void (*setup)(struct net_device *),
4656 struct net_device *netdev);
4659 * rdma_set_device_sysfs_group - Set device attributes group to have
4660 * driver specific sysfs entries at
4661 * for infiniband class.
4663 * @device: device pointer for which attributes to be created
4664 * @group: Pointer to group which should be added when device
4665 * is registered with sysfs.
4666 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4667 * group per device to have sysfs attributes.
4669 * NOTE: New drivers should not make use of this API; instead new device
4670 * parameter should be exposed via netlink command. This API and mechanism
4671 * exist only for existing drivers.
4673 static inline void
4674 rdma_set_device_sysfs_group(struct ib_device *dev,
4675 const struct attribute_group *group)
4677 dev->groups[1] = group;
4681 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4683 * @device: device pointer for which ib_device pointer to retrieve
4685 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4688 static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4690 struct ib_core_device *coredev =
4691 container_of(device, struct ib_core_device, dev);
4693 return coredev->owner;
4697 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4698 * ib_device holder structure from device pointer.
4700 * NOTE: New drivers should not make use of this API; This API is only for
4701 * existing drivers who have exposed sysfs entries using
4702 * rdma_set_device_sysfs_group().
4704 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4705 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4707 bool rdma_dev_access_netns(const struct ib_device *device,
4708 const struct net *net);
4710 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4711 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4714 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4715 * on the flow_label
4717 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4718 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4719 * convention.
4721 static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4723 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4725 fl_low ^= fl_high >> 14;
4726 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4730 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4731 * local and remote qpn values
4733 * This function folded the multiplication results of two qpns, 24 bit each,
4734 * fields, and converts it to a 20 bit results.
4736 * This function will create symmetric flow_label value based on the local
4737 * and remote qpn values. this will allow both the requester and responder
4738 * to calculate the same flow_label for a given connection.
4740 * This helper function should be used by driver in case the upper layer
4741 * provide a zero flow_label value. This is to improve entropy of RDMA
4742 * traffic in the network.
4744 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4746 u64 v = (u64)lqpn * rqpn;
4748 v ^= v >> 20;
4749 v ^= v >> 40;
4751 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4753 #endif /* IB_VERBS_H */