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
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
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
39 #if !defined(IB_VERBS_H)
42 #include <linux/types.h>
43 #include <linux/device.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/netdevice.h>
60 #include <linux/if_link.h>
61 #include <linux/atomic.h>
62 #include <linux/mmu_notifier.h>
63 #include <linux/uaccess.h>
64 #include <linux/cgroup_rdma.h>
65 #include <uapi/rdma/ib_user_verbs.h>
67 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
69 extern struct workqueue_struct
*ib_wq
;
70 extern struct workqueue_struct
*ib_comp_wq
;
80 extern union ib_gid zgid
;
83 /* If link layer is Ethernet, this is RoCE V1 */
86 IB_GID_TYPE_ROCE_UDP_ENCAP
= 1,
90 #define ROCE_V2_UDP_DPORT 4791
92 enum ib_gid_type gid_type
;
93 struct net_device
*ndev
;
97 /* IB values map to NodeInfo:NodeType. */
107 /* set the local administered indication */
108 IB_SA_WELL_KNOWN_GUID
= BIT_ULL(57) | 2,
111 enum rdma_transport_type
{
113 RDMA_TRANSPORT_IWARP
,
114 RDMA_TRANSPORT_USNIC
,
115 RDMA_TRANSPORT_USNIC_UDP
118 enum rdma_protocol_type
{
122 RDMA_PROTOCOL_USNIC_UDP
125 __attribute_const__
enum rdma_transport_type
126 rdma_node_get_transport(enum rdma_node_type node_type
);
128 enum rdma_network_type
{
130 RDMA_NETWORK_ROCE_V1
= RDMA_NETWORK_IB
,
135 static inline enum ib_gid_type
ib_network_to_gid_type(enum rdma_network_type network_type
)
137 if (network_type
== RDMA_NETWORK_IPV4
||
138 network_type
== RDMA_NETWORK_IPV6
)
139 return IB_GID_TYPE_ROCE_UDP_ENCAP
;
141 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
142 return IB_GID_TYPE_IB
;
145 static inline enum rdma_network_type
ib_gid_to_network_type(enum ib_gid_type gid_type
,
148 if (gid_type
== IB_GID_TYPE_IB
)
149 return RDMA_NETWORK_IB
;
151 if (ipv6_addr_v4mapped((struct in6_addr
*)gid
))
152 return RDMA_NETWORK_IPV4
;
154 return RDMA_NETWORK_IPV6
;
157 enum rdma_link_layer
{
158 IB_LINK_LAYER_UNSPECIFIED
,
159 IB_LINK_LAYER_INFINIBAND
,
160 IB_LINK_LAYER_ETHERNET
,
163 enum ib_device_cap_flags
{
164 IB_DEVICE_RESIZE_MAX_WR
= (1 << 0),
165 IB_DEVICE_BAD_PKEY_CNTR
= (1 << 1),
166 IB_DEVICE_BAD_QKEY_CNTR
= (1 << 2),
167 IB_DEVICE_RAW_MULTI
= (1 << 3),
168 IB_DEVICE_AUTO_PATH_MIG
= (1 << 4),
169 IB_DEVICE_CHANGE_PHY_PORT
= (1 << 5),
170 IB_DEVICE_UD_AV_PORT_ENFORCE
= (1 << 6),
171 IB_DEVICE_CURR_QP_STATE_MOD
= (1 << 7),
172 IB_DEVICE_SHUTDOWN_PORT
= (1 << 8),
173 /* Not in use, former INIT_TYPE = (1 << 9),*/
174 IB_DEVICE_PORT_ACTIVE_EVENT
= (1 << 10),
175 IB_DEVICE_SYS_IMAGE_GUID
= (1 << 11),
176 IB_DEVICE_RC_RNR_NAK_GEN
= (1 << 12),
177 IB_DEVICE_SRQ_RESIZE
= (1 << 13),
178 IB_DEVICE_N_NOTIFY_CQ
= (1 << 14),
181 * This device supports a per-device lkey or stag that can be
182 * used without performing a memory registration for the local
183 * memory. Note that ULPs should never check this flag, but
184 * instead of use the local_dma_lkey flag in the ib_pd structure,
185 * which will always contain a usable lkey.
187 IB_DEVICE_LOCAL_DMA_LKEY
= (1 << 15),
188 /* Reserved, old SEND_W_INV = (1 << 16),*/
189 IB_DEVICE_MEM_WINDOW
= (1 << 17),
191 * Devices should set IB_DEVICE_UD_IP_SUM if they support
192 * insertion of UDP and TCP checksum on outgoing UD IPoIB
193 * messages and can verify the validity of checksum for
194 * incoming messages. Setting this flag implies that the
195 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
197 IB_DEVICE_UD_IP_CSUM
= (1 << 18),
198 IB_DEVICE_UD_TSO
= (1 << 19),
199 IB_DEVICE_XRC
= (1 << 20),
202 * This device supports the IB "base memory management extension",
203 * which includes support for fast registrations (IB_WR_REG_MR,
204 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
205 * also be set by any iWarp device which must support FRs to comply
206 * to the iWarp verbs spec. iWarp devices also support the
207 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
210 IB_DEVICE_MEM_MGT_EXTENSIONS
= (1 << 21),
211 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK
= (1 << 22),
212 IB_DEVICE_MEM_WINDOW_TYPE_2A
= (1 << 23),
213 IB_DEVICE_MEM_WINDOW_TYPE_2B
= (1 << 24),
214 IB_DEVICE_RC_IP_CSUM
= (1 << 25),
215 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
216 IB_DEVICE_RAW_IP_CSUM
= (1 << 26),
218 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
219 * support execution of WQEs that involve synchronization
220 * of I/O operations with single completion queue managed
223 IB_DEVICE_CROSS_CHANNEL
= (1 << 27),
224 IB_DEVICE_MANAGED_FLOW_STEERING
= (1 << 29),
225 IB_DEVICE_SIGNATURE_HANDOVER
= (1 << 30),
226 IB_DEVICE_ON_DEMAND_PAGING
= (1ULL << 31),
227 IB_DEVICE_SG_GAPS_REG
= (1ULL << 32),
228 IB_DEVICE_VIRTUAL_FUNCTION
= (1ULL << 33),
229 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
230 IB_DEVICE_RAW_SCATTER_FCS
= (1ULL << 34),
231 IB_DEVICE_RDMA_NETDEV_OPA_VNIC
= (1ULL << 35),
232 /* The device supports padding incoming writes to cacheline. */
233 IB_DEVICE_PCI_WRITE_END_PADDING
= (1ULL << 36),
236 enum ib_signature_prot_cap
{
237 IB_PROT_T10DIF_TYPE_1
= 1,
238 IB_PROT_T10DIF_TYPE_2
= 1 << 1,
239 IB_PROT_T10DIF_TYPE_3
= 1 << 2,
242 enum ib_signature_guard_cap
{
243 IB_GUARD_T10DIF_CRC
= 1,
244 IB_GUARD_T10DIF_CSUM
= 1 << 1,
253 enum ib_odp_general_cap_bits
{
254 IB_ODP_SUPPORT
= 1 << 0,
255 IB_ODP_SUPPORT_IMPLICIT
= 1 << 1,
258 enum ib_odp_transport_cap_bits
{
259 IB_ODP_SUPPORT_SEND
= 1 << 0,
260 IB_ODP_SUPPORT_RECV
= 1 << 1,
261 IB_ODP_SUPPORT_WRITE
= 1 << 2,
262 IB_ODP_SUPPORT_READ
= 1 << 3,
263 IB_ODP_SUPPORT_ATOMIC
= 1 << 4,
267 uint64_t general_caps
;
269 uint32_t rc_odp_caps
;
270 uint32_t uc_odp_caps
;
271 uint32_t ud_odp_caps
;
272 } per_transport_caps
;
276 /* Corresponding bit will be set if qp type from
277 * 'enum ib_qp_type' is supported, e.g.
278 * supported_qpts |= 1 << IB_QPT_UD
281 u32 max_rwq_indirection_tables
;
282 u32 max_rwq_indirection_table_size
;
285 enum ib_tm_cap_flags
{
286 /* Support tag matching on RC transport */
287 IB_TM_CAP_RC
= 1 << 0,
291 /* Max size of RNDV header */
292 u32 max_rndv_hdr_size
;
293 /* Max number of entries in tag matching list */
295 /* From enum ib_tm_cap_flags */
297 /* Max number of outstanding list operations */
299 /* Max number of SGE in tag matching entry */
303 enum ib_cq_creation_flags
{
304 IB_CQ_FLAGS_TIMESTAMP_COMPLETION
= 1 << 0,
305 IB_CQ_FLAGS_IGNORE_OVERRUN
= 1 << 1,
308 struct ib_cq_init_attr
{
314 enum ib_cq_attr_mask
{
315 IB_CQ_MODERATE
= 1 << 0,
319 u16 max_cq_moderation_count
;
320 u16 max_cq_moderation_period
;
323 struct ib_device_attr
{
325 __be64 sys_image_guid
;
333 u64 device_cap_flags
;
343 int max_qp_init_rd_atom
;
344 int max_ee_init_rd_atom
;
345 enum ib_atomic_cap atomic_cap
;
346 enum ib_atomic_cap masked_atomic_cap
;
353 int max_mcast_qp_attach
;
354 int max_total_mcast_qp_attach
;
361 unsigned int max_fast_reg_page_list_len
;
363 u8 local_ca_ack_delay
;
366 struct ib_odp_caps odp_caps
;
367 uint64_t timestamp_mask
;
368 uint64_t hca_core_clock
; /* in KHZ */
369 struct ib_rss_caps rss_caps
;
371 u32 raw_packet_caps
; /* Use ib_raw_packet_caps enum */
372 struct ib_tm_caps tm_caps
;
373 struct ib_cq_caps cq_caps
;
384 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu
)
387 case IB_MTU_256
: return 256;
388 case IB_MTU_512
: return 512;
389 case IB_MTU_1024
: return 1024;
390 case IB_MTU_2048
: return 2048;
391 case IB_MTU_4096
: return 4096;
396 static inline enum ib_mtu
ib_mtu_int_to_enum(int mtu
)
400 else if (mtu
>= 2048)
402 else if (mtu
>= 1024)
416 IB_PORT_ACTIVE_DEFER
= 5
419 enum ib_port_cap_flags
{
421 IB_PORT_NOTICE_SUP
= 1 << 2,
422 IB_PORT_TRAP_SUP
= 1 << 3,
423 IB_PORT_OPT_IPD_SUP
= 1 << 4,
424 IB_PORT_AUTO_MIGR_SUP
= 1 << 5,
425 IB_PORT_SL_MAP_SUP
= 1 << 6,
426 IB_PORT_MKEY_NVRAM
= 1 << 7,
427 IB_PORT_PKEY_NVRAM
= 1 << 8,
428 IB_PORT_LED_INFO_SUP
= 1 << 9,
429 IB_PORT_SM_DISABLED
= 1 << 10,
430 IB_PORT_SYS_IMAGE_GUID_SUP
= 1 << 11,
431 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP
= 1 << 12,
432 IB_PORT_EXTENDED_SPEEDS_SUP
= 1 << 14,
433 IB_PORT_CM_SUP
= 1 << 16,
434 IB_PORT_SNMP_TUNNEL_SUP
= 1 << 17,
435 IB_PORT_REINIT_SUP
= 1 << 18,
436 IB_PORT_DEVICE_MGMT_SUP
= 1 << 19,
437 IB_PORT_VENDOR_CLASS_SUP
= 1 << 20,
438 IB_PORT_DR_NOTICE_SUP
= 1 << 21,
439 IB_PORT_CAP_MASK_NOTICE_SUP
= 1 << 22,
440 IB_PORT_BOOT_MGMT_SUP
= 1 << 23,
441 IB_PORT_LINK_LATENCY_SUP
= 1 << 24,
442 IB_PORT_CLIENT_REG_SUP
= 1 << 25,
443 IB_PORT_IP_BASED_GIDS
= 1 << 26,
453 static inline int ib_width_enum_to_int(enum ib_port_width width
)
456 case IB_WIDTH_1X
: return 1;
457 case IB_WIDTH_4X
: return 4;
458 case IB_WIDTH_8X
: return 8;
459 case IB_WIDTH_12X
: return 12;
475 * struct rdma_hw_stats
476 * @timestamp - Used by the core code to track when the last update was
477 * @lifespan - Used by the core code to determine how old the counters
478 * should be before being updated again. Stored in jiffies, defaults
479 * to 10 milliseconds, drivers can override the default be specifying
480 * their own value during their allocation routine.
481 * @name - Array of pointers to static names used for the counters in
483 * @num_counters - How many hardware counters there are. If name is
484 * shorter than this number, a kernel oops will result. Driver authors
485 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
486 * in their code to prevent this.
487 * @value - Array of u64 counters that are accessed by the sysfs code and
488 * filled in by the drivers get_stats routine
490 struct rdma_hw_stats
{
491 unsigned long timestamp
;
492 unsigned long lifespan
;
493 const char * const *names
;
498 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
500 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
502 * @names - Array of static const char *
503 * @num_counters - How many elements in array
504 * @lifespan - How many milliseconds between updates
506 static inline struct rdma_hw_stats
*rdma_alloc_hw_stats_struct(
507 const char * const *names
, int num_counters
,
508 unsigned long lifespan
)
510 struct rdma_hw_stats
*stats
;
512 stats
= kzalloc(sizeof(*stats
) + num_counters
* sizeof(u64
),
516 stats
->names
= names
;
517 stats
->num_counters
= num_counters
;
518 stats
->lifespan
= msecs_to_jiffies(lifespan
);
524 /* Define bits for the various functionality this port needs to be supported by
527 /* Management 0x00000FFF */
528 #define RDMA_CORE_CAP_IB_MAD 0x00000001
529 #define RDMA_CORE_CAP_IB_SMI 0x00000002
530 #define RDMA_CORE_CAP_IB_CM 0x00000004
531 #define RDMA_CORE_CAP_IW_CM 0x00000008
532 #define RDMA_CORE_CAP_IB_SA 0x00000010
533 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
535 /* Address format 0x000FF000 */
536 #define RDMA_CORE_CAP_AF_IB 0x00001000
537 #define RDMA_CORE_CAP_ETH_AH 0x00002000
538 #define RDMA_CORE_CAP_OPA_AH 0x00004000
540 /* Protocol 0xFFF00000 */
541 #define RDMA_CORE_CAP_PROT_IB 0x00100000
542 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
543 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
544 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
545 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
546 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
548 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
549 | RDMA_CORE_CAP_IB_MAD \
550 | RDMA_CORE_CAP_IB_SMI \
551 | RDMA_CORE_CAP_IB_CM \
552 | RDMA_CORE_CAP_IB_SA \
553 | RDMA_CORE_CAP_AF_IB)
554 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
555 | RDMA_CORE_CAP_IB_MAD \
556 | RDMA_CORE_CAP_IB_CM \
557 | RDMA_CORE_CAP_AF_IB \
558 | RDMA_CORE_CAP_ETH_AH)
559 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
560 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
561 | RDMA_CORE_CAP_IB_MAD \
562 | RDMA_CORE_CAP_IB_CM \
563 | RDMA_CORE_CAP_AF_IB \
564 | RDMA_CORE_CAP_ETH_AH)
565 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
566 | RDMA_CORE_CAP_IW_CM)
567 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
568 | RDMA_CORE_CAP_OPA_MAD)
570 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
572 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
574 struct ib_port_attr
{
576 enum ib_port_state state
;
578 enum ib_mtu active_mtu
;
598 enum ib_device_modify_flags
{
599 IB_DEVICE_MODIFY_SYS_IMAGE_GUID
= 1 << 0,
600 IB_DEVICE_MODIFY_NODE_DESC
= 1 << 1
603 #define IB_DEVICE_NODE_DESC_MAX 64
605 struct ib_device_modify
{
607 char node_desc
[IB_DEVICE_NODE_DESC_MAX
];
610 enum ib_port_modify_flags
{
611 IB_PORT_SHUTDOWN
= 1,
612 IB_PORT_INIT_TYPE
= (1<<2),
613 IB_PORT_RESET_QKEY_CNTR
= (1<<3),
614 IB_PORT_OPA_MASK_CHG
= (1<<4)
617 struct ib_port_modify
{
618 u32 set_port_cap_mask
;
619 u32 clr_port_cap_mask
;
627 IB_EVENT_QP_ACCESS_ERR
,
631 IB_EVENT_PATH_MIG_ERR
,
632 IB_EVENT_DEVICE_FATAL
,
633 IB_EVENT_PORT_ACTIVE
,
636 IB_EVENT_PKEY_CHANGE
,
639 IB_EVENT_SRQ_LIMIT_REACHED
,
640 IB_EVENT_QP_LAST_WQE_REACHED
,
641 IB_EVENT_CLIENT_REREGISTER
,
646 const char *__attribute_const__
ib_event_msg(enum ib_event_type event
);
649 struct ib_device
*device
;
657 enum ib_event_type event
;
660 struct ib_event_handler
{
661 struct ib_device
*device
;
662 void (*handler
)(struct ib_event_handler
*, struct ib_event
*);
663 struct list_head list
;
666 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
668 (_ptr)->device = _device; \
669 (_ptr)->handler = _handler; \
670 INIT_LIST_HEAD(&(_ptr)->list); \
673 struct ib_global_route
{
682 __be32 version_tclass_flow
;
690 union rdma_network_hdr
{
693 /* The IB spec states that if it's IPv4, the header
694 * is located in the last 20 bytes of the header.
697 struct iphdr roce4grh
;
701 #define IB_QPN_MASK 0xFFFFFF
704 IB_MULTICAST_QPN
= 0xffffff
707 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
708 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
715 IB_RATE_PORT_CURRENT
= 0,
716 IB_RATE_2_5_GBPS
= 2,
724 IB_RATE_120_GBPS
= 10,
725 IB_RATE_14_GBPS
= 11,
726 IB_RATE_56_GBPS
= 12,
727 IB_RATE_112_GBPS
= 13,
728 IB_RATE_168_GBPS
= 14,
729 IB_RATE_25_GBPS
= 15,
730 IB_RATE_100_GBPS
= 16,
731 IB_RATE_200_GBPS
= 17,
732 IB_RATE_300_GBPS
= 18
736 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
737 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
738 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
739 * @rate: rate to convert.
741 __attribute_const__
int ib_rate_to_mult(enum ib_rate rate
);
744 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
745 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
746 * @rate: rate to convert.
748 __attribute_const__
int ib_rate_to_mbps(enum ib_rate rate
);
752 * enum ib_mr_type - memory region type
753 * @IB_MR_TYPE_MEM_REG: memory region that is used for
754 * normal registration
755 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
756 * signature operations (data-integrity
758 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
759 * register any arbitrary sg lists (without
760 * the normal mr constraints - see
765 IB_MR_TYPE_SIGNATURE
,
771 * IB_SIG_TYPE_NONE: Unprotected.
772 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
774 enum ib_signature_type
{
780 * Signature T10-DIF block-guard types
781 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
782 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
784 enum ib_t10_dif_bg_type
{
790 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
792 * @bg_type: T10-DIF block guard type (CRC|CSUM)
793 * @pi_interval: protection information interval.
794 * @bg: seed of guard computation.
795 * @app_tag: application tag of guard block
796 * @ref_tag: initial guard block reference tag.
797 * @ref_remap: Indicate wethear the reftag increments each block
798 * @app_escape: Indicate to skip block check if apptag=0xffff
799 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
800 * @apptag_check_mask: check bitmask of application tag.
802 struct ib_t10_dif_domain
{
803 enum ib_t10_dif_bg_type bg_type
;
811 u16 apptag_check_mask
;
815 * struct ib_sig_domain - Parameters for signature domain
816 * @sig_type: specific signauture type
817 * @sig: union of all signature domain attributes that may
818 * be used to set domain layout.
820 struct ib_sig_domain
{
821 enum ib_signature_type sig_type
;
823 struct ib_t10_dif_domain dif
;
828 * struct ib_sig_attrs - Parameters for signature handover operation
829 * @check_mask: bitmask for signature byte check (8 bytes)
830 * @mem: memory domain layout desciptor.
831 * @wire: wire domain layout desciptor.
833 struct ib_sig_attrs
{
835 struct ib_sig_domain mem
;
836 struct ib_sig_domain wire
;
839 enum ib_sig_err_type
{
846 * struct ib_sig_err - signature error descriptor
849 enum ib_sig_err_type err_type
;
856 enum ib_mr_status_check
{
857 IB_MR_CHECK_SIG_STATUS
= 1,
861 * struct ib_mr_status - Memory region status container
863 * @fail_status: Bitmask of MR checks status. For each
864 * failed check a corresponding status bit is set.
865 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
868 struct ib_mr_status
{
870 struct ib_sig_err sig_err
;
874 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
876 * @mult: multiple to convert.
878 __attribute_const__
enum ib_rate
mult_to_ib_rate(int mult
);
880 enum rdma_ah_attr_type
{
881 RDMA_AH_ATTR_TYPE_UNDEFINED
,
882 RDMA_AH_ATTR_TYPE_IB
,
883 RDMA_AH_ATTR_TYPE_ROCE
,
884 RDMA_AH_ATTR_TYPE_OPA
,
892 struct roce_ah_attr
{
902 struct rdma_ah_attr
{
903 struct ib_global_route grh
;
908 enum rdma_ah_attr_type type
;
910 struct ib_ah_attr ib
;
911 struct roce_ah_attr roce
;
912 struct opa_ah_attr opa
;
920 IB_WC_LOC_EEC_OP_ERR
,
925 IB_WC_LOC_ACCESS_ERR
,
926 IB_WC_REM_INV_REQ_ERR
,
927 IB_WC_REM_ACCESS_ERR
,
930 IB_WC_RNR_RETRY_EXC_ERR
,
931 IB_WC_LOC_RDD_VIOL_ERR
,
932 IB_WC_REM_INV_RD_REQ_ERR
,
935 IB_WC_INV_EEC_STATE_ERR
,
937 IB_WC_RESP_TIMEOUT_ERR
,
941 const char *__attribute_const__
ib_wc_status_msg(enum ib_wc_status status
);
952 IB_WC_MASKED_COMP_SWAP
,
953 IB_WC_MASKED_FETCH_ADD
,
955 * Set value of IB_WC_RECV so consumers can test if a completion is a
956 * receive by testing (opcode & IB_WC_RECV).
959 IB_WC_RECV_RDMA_WITH_IMM
964 IB_WC_WITH_IMM
= (1<<1),
965 IB_WC_WITH_INVALIDATE
= (1<<2),
966 IB_WC_IP_CSUM_OK
= (1<<3),
967 IB_WC_WITH_SMAC
= (1<<4),
968 IB_WC_WITH_VLAN
= (1<<5),
969 IB_WC_WITH_NETWORK_HDR_TYPE
= (1<<6),
975 struct ib_cqe
*wr_cqe
;
977 enum ib_wc_status status
;
978 enum ib_wc_opcode opcode
;
992 u8 port_num
; /* valid only for DR SMPs on switches */
998 enum ib_cq_notify_flags
{
999 IB_CQ_SOLICITED
= 1 << 0,
1000 IB_CQ_NEXT_COMP
= 1 << 1,
1001 IB_CQ_SOLICITED_MASK
= IB_CQ_SOLICITED
| IB_CQ_NEXT_COMP
,
1002 IB_CQ_REPORT_MISSED_EVENTS
= 1 << 2,
1011 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type
)
1013 return srq_type
== IB_SRQT_XRC
||
1014 srq_type
== IB_SRQT_TM
;
1017 enum ib_srq_attr_mask
{
1018 IB_SRQ_MAX_WR
= 1 << 0,
1019 IB_SRQ_LIMIT
= 1 << 1,
1022 struct ib_srq_attr
{
1028 struct ib_srq_init_attr
{
1029 void (*event_handler
)(struct ib_event
*, void *);
1031 struct ib_srq_attr attr
;
1032 enum ib_srq_type srq_type
;
1038 struct ib_xrcd
*xrcd
;
1053 u32 max_inline_data
;
1056 * Maximum number of rdma_rw_ctx structures in flight at a time.
1057 * ib_create_qp() will calculate the right amount of neededed WRs
1058 * and MRs based on this.
1070 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1071 * here (and in that order) since the MAD layer uses them as
1072 * indices into a 2-entry table.
1081 IB_QPT_RAW_ETHERTYPE
,
1082 IB_QPT_RAW_PACKET
= 8,
1086 /* Reserve a range for qp types internal to the low level driver.
1087 * These qp types will not be visible at the IB core layer, so the
1088 * IB_QPT_MAX usages should not be affected in the core layer
1090 IB_QPT_RESERVED1
= 0x1000,
1102 enum ib_qp_create_flags
{
1103 IB_QP_CREATE_IPOIB_UD_LSO
= 1 << 0,
1104 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK
= 1 << 1,
1105 IB_QP_CREATE_CROSS_CHANNEL
= 1 << 2,
1106 IB_QP_CREATE_MANAGED_SEND
= 1 << 3,
1107 IB_QP_CREATE_MANAGED_RECV
= 1 << 4,
1108 IB_QP_CREATE_NETIF_QP
= 1 << 5,
1109 IB_QP_CREATE_SIGNATURE_EN
= 1 << 6,
1110 /* FREE = 1 << 7, */
1111 IB_QP_CREATE_SCATTER_FCS
= 1 << 8,
1112 IB_QP_CREATE_CVLAN_STRIPPING
= 1 << 9,
1113 IB_QP_CREATE_SOURCE_QPN
= 1 << 10,
1114 IB_QP_CREATE_PCI_WRITE_END_PADDING
= 1 << 11,
1115 /* reserve bits 26-31 for low level drivers' internal use */
1116 IB_QP_CREATE_RESERVED_START
= 1 << 26,
1117 IB_QP_CREATE_RESERVED_END
= 1 << 31,
1121 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1122 * callback to destroy the passed in QP.
1125 struct ib_qp_init_attr
{
1126 void (*event_handler
)(struct ib_event
*, void *);
1128 struct ib_cq
*send_cq
;
1129 struct ib_cq
*recv_cq
;
1131 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
1132 struct ib_qp_cap cap
;
1133 enum ib_sig_type sq_sig_type
;
1134 enum ib_qp_type qp_type
;
1135 enum ib_qp_create_flags create_flags
;
1138 * Only needed for special QP types, or when using the RW API.
1141 struct ib_rwq_ind_table
*rwq_ind_tbl
;
1145 struct ib_qp_open_attr
{
1146 void (*event_handler
)(struct ib_event
*, void *);
1149 enum ib_qp_type qp_type
;
1152 enum ib_rnr_timeout
{
1153 IB_RNR_TIMER_655_36
= 0,
1154 IB_RNR_TIMER_000_01
= 1,
1155 IB_RNR_TIMER_000_02
= 2,
1156 IB_RNR_TIMER_000_03
= 3,
1157 IB_RNR_TIMER_000_04
= 4,
1158 IB_RNR_TIMER_000_06
= 5,
1159 IB_RNR_TIMER_000_08
= 6,
1160 IB_RNR_TIMER_000_12
= 7,
1161 IB_RNR_TIMER_000_16
= 8,
1162 IB_RNR_TIMER_000_24
= 9,
1163 IB_RNR_TIMER_000_32
= 10,
1164 IB_RNR_TIMER_000_48
= 11,
1165 IB_RNR_TIMER_000_64
= 12,
1166 IB_RNR_TIMER_000_96
= 13,
1167 IB_RNR_TIMER_001_28
= 14,
1168 IB_RNR_TIMER_001_92
= 15,
1169 IB_RNR_TIMER_002_56
= 16,
1170 IB_RNR_TIMER_003_84
= 17,
1171 IB_RNR_TIMER_005_12
= 18,
1172 IB_RNR_TIMER_007_68
= 19,
1173 IB_RNR_TIMER_010_24
= 20,
1174 IB_RNR_TIMER_015_36
= 21,
1175 IB_RNR_TIMER_020_48
= 22,
1176 IB_RNR_TIMER_030_72
= 23,
1177 IB_RNR_TIMER_040_96
= 24,
1178 IB_RNR_TIMER_061_44
= 25,
1179 IB_RNR_TIMER_081_92
= 26,
1180 IB_RNR_TIMER_122_88
= 27,
1181 IB_RNR_TIMER_163_84
= 28,
1182 IB_RNR_TIMER_245_76
= 29,
1183 IB_RNR_TIMER_327_68
= 30,
1184 IB_RNR_TIMER_491_52
= 31
1187 enum ib_qp_attr_mask
{
1189 IB_QP_CUR_STATE
= (1<<1),
1190 IB_QP_EN_SQD_ASYNC_NOTIFY
= (1<<2),
1191 IB_QP_ACCESS_FLAGS
= (1<<3),
1192 IB_QP_PKEY_INDEX
= (1<<4),
1193 IB_QP_PORT
= (1<<5),
1194 IB_QP_QKEY
= (1<<6),
1196 IB_QP_PATH_MTU
= (1<<8),
1197 IB_QP_TIMEOUT
= (1<<9),
1198 IB_QP_RETRY_CNT
= (1<<10),
1199 IB_QP_RNR_RETRY
= (1<<11),
1200 IB_QP_RQ_PSN
= (1<<12),
1201 IB_QP_MAX_QP_RD_ATOMIC
= (1<<13),
1202 IB_QP_ALT_PATH
= (1<<14),
1203 IB_QP_MIN_RNR_TIMER
= (1<<15),
1204 IB_QP_SQ_PSN
= (1<<16),
1205 IB_QP_MAX_DEST_RD_ATOMIC
= (1<<17),
1206 IB_QP_PATH_MIG_STATE
= (1<<18),
1207 IB_QP_CAP
= (1<<19),
1208 IB_QP_DEST_QPN
= (1<<20),
1209 IB_QP_RESERVED1
= (1<<21),
1210 IB_QP_RESERVED2
= (1<<22),
1211 IB_QP_RESERVED3
= (1<<23),
1212 IB_QP_RESERVED4
= (1<<24),
1213 IB_QP_RATE_LIMIT
= (1<<25),
1238 enum ib_qp_state qp_state
;
1239 enum ib_qp_state cur_qp_state
;
1240 enum ib_mtu path_mtu
;
1241 enum ib_mig_state path_mig_state
;
1246 int qp_access_flags
;
1247 struct ib_qp_cap cap
;
1248 struct rdma_ah_attr ah_attr
;
1249 struct rdma_ah_attr alt_ah_attr
;
1252 u8 en_sqd_async_notify
;
1255 u8 max_dest_rd_atomic
;
1268 IB_WR_RDMA_WRITE_WITH_IMM
,
1270 IB_WR_SEND_WITH_IMM
,
1272 IB_WR_ATOMIC_CMP_AND_SWP
,
1273 IB_WR_ATOMIC_FETCH_AND_ADD
,
1275 IB_WR_SEND_WITH_INV
,
1276 IB_WR_RDMA_READ_WITH_INV
,
1279 IB_WR_MASKED_ATOMIC_CMP_AND_SWP
,
1280 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD
,
1282 /* reserve values for low level drivers' internal use.
1283 * These values will not be used at all in the ib core layer.
1285 IB_WR_RESERVED1
= 0xf0,
1297 enum ib_send_flags
{
1299 IB_SEND_SIGNALED
= (1<<1),
1300 IB_SEND_SOLICITED
= (1<<2),
1301 IB_SEND_INLINE
= (1<<3),
1302 IB_SEND_IP_CSUM
= (1<<4),
1304 /* reserve bits 26-31 for low level drivers' internal use */
1305 IB_SEND_RESERVED_START
= (1 << 26),
1306 IB_SEND_RESERVED_END
= (1 << 31),
1316 void (*done
)(struct ib_cq
*cq
, struct ib_wc
*wc
);
1320 struct ib_send_wr
*next
;
1323 struct ib_cqe
*wr_cqe
;
1325 struct ib_sge
*sg_list
;
1327 enum ib_wr_opcode opcode
;
1331 u32 invalidate_rkey
;
1336 struct ib_send_wr wr
;
1341 static inline struct ib_rdma_wr
*rdma_wr(struct ib_send_wr
*wr
)
1343 return container_of(wr
, struct ib_rdma_wr
, wr
);
1346 struct ib_atomic_wr
{
1347 struct ib_send_wr wr
;
1351 u64 compare_add_mask
;
1356 static inline struct ib_atomic_wr
*atomic_wr(struct ib_send_wr
*wr
)
1358 return container_of(wr
, struct ib_atomic_wr
, wr
);
1362 struct ib_send_wr wr
;
1369 u16 pkey_index
; /* valid for GSI only */
1370 u8 port_num
; /* valid for DR SMPs on switch only */
1373 static inline struct ib_ud_wr
*ud_wr(struct ib_send_wr
*wr
)
1375 return container_of(wr
, struct ib_ud_wr
, wr
);
1379 struct ib_send_wr wr
;
1385 static inline struct ib_reg_wr
*reg_wr(struct ib_send_wr
*wr
)
1387 return container_of(wr
, struct ib_reg_wr
, wr
);
1390 struct ib_sig_handover_wr
{
1391 struct ib_send_wr wr
;
1392 struct ib_sig_attrs
*sig_attrs
;
1393 struct ib_mr
*sig_mr
;
1395 struct ib_sge
*prot
;
1398 static inline struct ib_sig_handover_wr
*sig_handover_wr(struct ib_send_wr
*wr
)
1400 return container_of(wr
, struct ib_sig_handover_wr
, wr
);
1404 struct ib_recv_wr
*next
;
1407 struct ib_cqe
*wr_cqe
;
1409 struct ib_sge
*sg_list
;
1413 enum ib_access_flags
{
1414 IB_ACCESS_LOCAL_WRITE
= 1,
1415 IB_ACCESS_REMOTE_WRITE
= (1<<1),
1416 IB_ACCESS_REMOTE_READ
= (1<<2),
1417 IB_ACCESS_REMOTE_ATOMIC
= (1<<3),
1418 IB_ACCESS_MW_BIND
= (1<<4),
1419 IB_ZERO_BASED
= (1<<5),
1420 IB_ACCESS_ON_DEMAND
= (1<<6),
1421 IB_ACCESS_HUGETLB
= (1<<7),
1425 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1426 * are hidden here instead of a uapi header!
1428 enum ib_mr_rereg_flags
{
1429 IB_MR_REREG_TRANS
= 1,
1430 IB_MR_REREG_PD
= (1<<1),
1431 IB_MR_REREG_ACCESS
= (1<<2),
1432 IB_MR_REREG_SUPPORTED
= ((IB_MR_REREG_ACCESS
<< 1) - 1)
1435 struct ib_fmr_attr
{
1443 enum rdma_remove_reason
{
1444 /* Userspace requested uobject deletion. Call could fail */
1445 RDMA_REMOVE_DESTROY
,
1446 /* Context deletion. This call should delete the actual object itself */
1448 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1449 RDMA_REMOVE_DRIVER_REMOVE
,
1450 /* Context is being cleaned-up, but commit was just completed */
1451 RDMA_REMOVE_DURING_CLEANUP
,
1454 struct ib_rdmacg_object
{
1455 #ifdef CONFIG_CGROUP_RDMA
1456 struct rdma_cgroup
*cg
; /* owner rdma cgroup */
1460 struct ib_ucontext
{
1461 struct ib_device
*device
;
1462 struct ib_uverbs_file
*ufile
;
1465 /* locking the uobjects_list */
1466 struct mutex uobjects_lock
;
1467 struct list_head uobjects
;
1468 /* protects cleanup process from other actions */
1469 struct rw_semaphore cleanup_rwsem
;
1470 enum rdma_remove_reason cleanup_reason
;
1473 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1474 struct rb_root_cached umem_tree
;
1476 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1477 * mmu notifiers registration.
1479 struct rw_semaphore umem_rwsem
;
1480 void (*invalidate_range
)(struct ib_umem
*umem
,
1481 unsigned long start
, unsigned long end
);
1483 struct mmu_notifier mn
;
1484 atomic_t notifier_count
;
1485 /* A list of umems that don't have private mmu notifier counters yet. */
1486 struct list_head no_private_counters
;
1490 struct ib_rdmacg_object cg_obj
;
1494 u64 user_handle
; /* handle given to us by userspace */
1495 struct ib_ucontext
*context
; /* associated user context */
1496 void *object
; /* containing object */
1497 struct list_head list
; /* link to context's list */
1498 struct ib_rdmacg_object cg_obj
; /* rdmacg object */
1499 int id
; /* index into kernel idr */
1501 atomic_t usecnt
; /* protects exclusive access */
1502 struct rcu_head rcu
; /* kfree_rcu() overhead */
1504 const struct uverbs_obj_type
*type
;
1507 struct ib_uobject_file
{
1508 struct ib_uobject uobj
;
1509 /* ufile contains the lock between context release and file close */
1510 struct ib_uverbs_file
*ufile
;
1514 const void __user
*inbuf
;
1515 void __user
*outbuf
;
1523 struct ib_device
*device
;
1524 struct ib_uobject
*uobject
;
1525 atomic_t usecnt
; /* count all resources */
1527 u32 unsafe_global_rkey
;
1530 * Implementation details of the RDMA core, don't use in drivers:
1532 struct ib_mr
*__internal_mr
;
1536 struct ib_device
*device
;
1537 atomic_t usecnt
; /* count all exposed resources */
1538 struct inode
*inode
;
1540 struct mutex tgt_qp_mutex
;
1541 struct list_head tgt_qp_list
;
1545 struct ib_device
*device
;
1547 struct ib_uobject
*uobject
;
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 */
1560 struct ib_device
*device
;
1561 struct ib_uobject
*uobject
;
1562 ib_comp_handler comp_handler
;
1563 void (*event_handler
)(struct ib_event
*, void *);
1566 atomic_t usecnt
; /* count number of work queues */
1567 enum ib_poll_context poll_ctx
;
1570 struct irq_poll iop
;
1571 struct work_struct work
;
1576 struct ib_device
*device
;
1578 struct ib_uobject
*uobject
;
1579 void (*event_handler
)(struct ib_event
*, void *);
1581 enum ib_srq_type srq_type
;
1588 struct ib_xrcd
*xrcd
;
1595 enum ib_raw_packet_caps
{
1596 /* Strip cvlan from incoming packet and report it in the matching work
1597 * completion is supported.
1599 IB_RAW_PACKET_CAP_CVLAN_STRIPPING
= (1 << 0),
1600 /* Scatter FCS field of an incoming packet to host memory is supported.
1602 IB_RAW_PACKET_CAP_SCATTER_FCS
= (1 << 1),
1603 /* Checksum offloads are supported (for both send and receive). */
1604 IB_RAW_PACKET_CAP_IP_CSUM
= (1 << 2),
1605 /* When a packet is received for an RQ with no receive WQEs, the
1606 * packet processing is delayed.
1608 IB_RAW_PACKET_CAP_DELAY_DROP
= (1 << 3),
1622 struct ib_device
*device
;
1623 struct ib_uobject
*uobject
;
1625 void (*event_handler
)(struct ib_event
*, void *);
1629 enum ib_wq_state state
;
1630 enum ib_wq_type wq_type
;
1635 IB_WQ_FLAGS_CVLAN_STRIPPING
= 1 << 0,
1636 IB_WQ_FLAGS_SCATTER_FCS
= 1 << 1,
1637 IB_WQ_FLAGS_DELAY_DROP
= 1 << 2,
1638 IB_WQ_FLAGS_PCI_WRITE_END_PADDING
= 1 << 3,
1641 struct ib_wq_init_attr
{
1643 enum ib_wq_type wq_type
;
1647 void (*event_handler
)(struct ib_event
*, void *);
1648 u32 create_flags
; /* Use enum ib_wq_flags */
1651 enum ib_wq_attr_mask
{
1652 IB_WQ_STATE
= 1 << 0,
1653 IB_WQ_CUR_STATE
= 1 << 1,
1654 IB_WQ_FLAGS
= 1 << 2,
1658 enum ib_wq_state wq_state
;
1659 enum ib_wq_state curr_wq_state
;
1660 u32 flags
; /* Use enum ib_wq_flags */
1661 u32 flags_mask
; /* Use enum ib_wq_flags */
1664 struct ib_rwq_ind_table
{
1665 struct ib_device
*device
;
1666 struct ib_uobject
*uobject
;
1669 u32 log_ind_tbl_size
;
1670 struct ib_wq
**ind_tbl
;
1673 struct ib_rwq_ind_table_init_attr
{
1674 u32 log_ind_tbl_size
;
1675 /* Each entry is a pointer to Receive Work Queue */
1676 struct ib_wq
**ind_tbl
;
1679 enum port_pkey_state
{
1680 IB_PORT_PKEY_NOT_VALID
= 0,
1681 IB_PORT_PKEY_VALID
= 1,
1682 IB_PORT_PKEY_LISTED
= 2,
1685 struct ib_qp_security
;
1687 struct ib_port_pkey
{
1688 enum port_pkey_state state
;
1691 struct list_head qp_list
;
1692 struct list_head to_error_list
;
1693 struct ib_qp_security
*sec
;
1696 struct ib_ports_pkeys
{
1697 struct ib_port_pkey main
;
1698 struct ib_port_pkey alt
;
1701 struct ib_qp_security
{
1703 struct ib_device
*dev
;
1704 /* Hold this mutex when changing port and pkey settings. */
1706 struct ib_ports_pkeys
*ports_pkeys
;
1707 /* A list of all open shared QP handles. Required to enforce security
1708 * properly for all users of a shared QP.
1710 struct list_head shared_qp_list
;
1713 atomic_t error_list_count
;
1714 struct completion error_complete
;
1715 int error_comps_pending
;
1719 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1720 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1723 struct ib_device
*device
;
1725 struct ib_cq
*send_cq
;
1726 struct ib_cq
*recv_cq
;
1729 struct list_head rdma_mrs
;
1730 struct list_head sig_mrs
;
1732 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
1733 struct list_head xrcd_list
;
1735 /* count times opened, mcast attaches, flow attaches */
1737 struct list_head open_list
;
1738 struct ib_qp
*real_qp
;
1739 struct ib_uobject
*uobject
;
1740 void (*event_handler
)(struct ib_event
*, void *);
1745 enum ib_qp_type qp_type
;
1746 struct ib_rwq_ind_table
*rwq_ind_tbl
;
1747 struct ib_qp_security
*qp_sec
;
1752 struct ib_device
*device
;
1758 unsigned int page_size
;
1761 struct ib_uobject
*uobject
; /* user */
1762 struct list_head qp_entry
; /* FR */
1767 struct ib_device
*device
;
1769 struct ib_uobject
*uobject
;
1771 enum ib_mw_type type
;
1775 struct ib_device
*device
;
1777 struct list_head list
;
1782 /* Supported steering options */
1783 enum ib_flow_attr_type
{
1784 /* steering according to rule specifications */
1785 IB_FLOW_ATTR_NORMAL
= 0x0,
1786 /* default unicast and multicast rule -
1787 * receive all Eth traffic which isn't steered to any QP
1789 IB_FLOW_ATTR_ALL_DEFAULT
= 0x1,
1790 /* default multicast rule -
1791 * receive all Eth multicast traffic which isn't steered to any QP
1793 IB_FLOW_ATTR_MC_DEFAULT
= 0x2,
1794 /* sniffer rule - receive all port traffic */
1795 IB_FLOW_ATTR_SNIFFER
= 0x3
1798 /* Supported steering header types */
1799 enum ib_flow_spec_type
{
1801 IB_FLOW_SPEC_ETH
= 0x20,
1802 IB_FLOW_SPEC_IB
= 0x22,
1804 IB_FLOW_SPEC_IPV4
= 0x30,
1805 IB_FLOW_SPEC_IPV6
= 0x31,
1807 IB_FLOW_SPEC_TCP
= 0x40,
1808 IB_FLOW_SPEC_UDP
= 0x41,
1809 IB_FLOW_SPEC_VXLAN_TUNNEL
= 0x50,
1810 IB_FLOW_SPEC_INNER
= 0x100,
1812 IB_FLOW_SPEC_ACTION_TAG
= 0x1000,
1813 IB_FLOW_SPEC_ACTION_DROP
= 0x1001,
1815 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1816 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1818 /* Flow steering rule priority is set according to it's domain.
1819 * Lower domain value means higher priority.
1821 enum ib_flow_domain
{
1822 IB_FLOW_DOMAIN_USER
,
1823 IB_FLOW_DOMAIN_ETHTOOL
,
1826 IB_FLOW_DOMAIN_NUM
/* Must be last */
1829 enum ib_flow_flags
{
1830 IB_FLOW_ATTR_FLAGS_DONT_TRAP
= 1UL << 1, /* Continue match, no steal */
1831 IB_FLOW_ATTR_FLAGS_RESERVED
= 1UL << 2 /* Must be last */
1834 struct ib_flow_eth_filter
{
1843 struct ib_flow_spec_eth
{
1846 struct ib_flow_eth_filter val
;
1847 struct ib_flow_eth_filter mask
;
1850 struct ib_flow_ib_filter
{
1857 struct ib_flow_spec_ib
{
1860 struct ib_flow_ib_filter val
;
1861 struct ib_flow_ib_filter mask
;
1864 /* IPv4 header flags */
1865 enum ib_ipv4_flags
{
1866 IB_IPV4_DONT_FRAG
= 0x2, /* Don't enable packet fragmentation */
1867 IB_IPV4_MORE_FRAG
= 0X4 /* For All fragmented packets except the
1868 last have this flag set */
1871 struct ib_flow_ipv4_filter
{
1882 struct ib_flow_spec_ipv4
{
1885 struct ib_flow_ipv4_filter val
;
1886 struct ib_flow_ipv4_filter mask
;
1889 struct ib_flow_ipv6_filter
{
1900 struct ib_flow_spec_ipv6
{
1903 struct ib_flow_ipv6_filter val
;
1904 struct ib_flow_ipv6_filter mask
;
1907 struct ib_flow_tcp_udp_filter
{
1914 struct ib_flow_spec_tcp_udp
{
1917 struct ib_flow_tcp_udp_filter val
;
1918 struct ib_flow_tcp_udp_filter mask
;
1921 struct ib_flow_tunnel_filter
{
1926 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1927 * the tunnel_id from val has the vni value
1929 struct ib_flow_spec_tunnel
{
1932 struct ib_flow_tunnel_filter val
;
1933 struct ib_flow_tunnel_filter mask
;
1936 struct ib_flow_spec_action_tag
{
1937 enum ib_flow_spec_type type
;
1942 struct ib_flow_spec_action_drop
{
1943 enum ib_flow_spec_type type
;
1947 union ib_flow_spec
{
1952 struct ib_flow_spec_eth eth
;
1953 struct ib_flow_spec_ib ib
;
1954 struct ib_flow_spec_ipv4 ipv4
;
1955 struct ib_flow_spec_tcp_udp tcp_udp
;
1956 struct ib_flow_spec_ipv6 ipv6
;
1957 struct ib_flow_spec_tunnel tunnel
;
1958 struct ib_flow_spec_action_tag flow_tag
;
1959 struct ib_flow_spec_action_drop drop
;
1962 struct ib_flow_attr
{
1963 enum ib_flow_attr_type type
;
1969 /* Following are the optional layers according to user request
1970 * struct ib_flow_spec_xxx
1971 * struct ib_flow_spec_yyy
1977 struct ib_uobject
*uobject
;
1983 enum ib_process_mad_flags
{
1984 IB_MAD_IGNORE_MKEY
= 1,
1985 IB_MAD_IGNORE_BKEY
= 2,
1986 IB_MAD_IGNORE_ALL
= IB_MAD_IGNORE_MKEY
| IB_MAD_IGNORE_BKEY
1989 enum ib_mad_result
{
1990 IB_MAD_RESULT_FAILURE
= 0, /* (!SUCCESS is the important flag) */
1991 IB_MAD_RESULT_SUCCESS
= 1 << 0, /* MAD was successfully processed */
1992 IB_MAD_RESULT_REPLY
= 1 << 1, /* Reply packet needs to be sent */
1993 IB_MAD_RESULT_CONSUMED
= 1 << 2 /* Packet consumed: stop processing */
1996 struct ib_port_cache
{
1998 struct ib_pkey_cache
*pkey
;
1999 struct ib_gid_table
*gid
;
2001 enum ib_port_state port_state
;
2006 struct ib_event_handler event_handler
;
2007 struct ib_port_cache
*ports
;
2012 struct ib_port_immutable
{
2019 /* rdma netdev type - specifies protocol type */
2020 enum rdma_netdev_t
{
2021 RDMA_NETDEV_OPA_VNIC
,
2026 * struct rdma_netdev - rdma netdev
2027 * For cases where netstack interfacing is required.
2029 struct rdma_netdev
{
2031 struct ib_device
*hca
;
2034 /* cleanup function must be specified */
2035 void (*free_rdma_netdev
)(struct net_device
*netdev
);
2037 /* control functions */
2038 void (*set_id
)(struct net_device
*netdev
, int id
);
2040 int (*send
)(struct net_device
*dev
, struct sk_buff
*skb
,
2041 struct ib_ah
*address
, u32 dqpn
);
2043 int (*attach_mcast
)(struct net_device
*dev
, struct ib_device
*hca
,
2044 union ib_gid
*gid
, u16 mlid
,
2045 int set_qkey
, u32 qkey
);
2046 int (*detach_mcast
)(struct net_device
*dev
, struct ib_device
*hca
,
2047 union ib_gid
*gid
, u16 mlid
);
2050 struct ib_port_pkey_list
{
2051 /* Lock to hold while modifying the list. */
2052 spinlock_t list_lock
;
2053 struct list_head pkey_list
;
2057 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2058 struct device
*dma_device
;
2060 char name
[IB_DEVICE_NAME_MAX
];
2062 struct list_head event_handler_list
;
2063 spinlock_t event_handler_lock
;
2065 spinlock_t client_data_lock
;
2066 struct list_head core_list
;
2067 /* Access to the client_data_list is protected by the client_data_lock
2068 * spinlock and the lists_rwsem read-write semaphore */
2069 struct list_head client_data_list
;
2071 struct ib_cache cache
;
2073 * port_immutable is indexed by port number
2075 struct ib_port_immutable
*port_immutable
;
2077 int num_comp_vectors
;
2079 struct ib_port_pkey_list
*port_pkey_list
;
2081 struct iw_cm_verbs
*iwcm
;
2084 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2085 * driver initialized data. The struct is kfree()'ed by the sysfs
2086 * core when the device is removed. A lifespan of -1 in the return
2087 * struct tells the core to set a default lifespan.
2089 struct rdma_hw_stats
*(*alloc_hw_stats
)(struct ib_device
*device
,
2092 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2093 * @index - The index in the value array we wish to have updated, or
2094 * num_counters if we want all stats updated
2096 * < 0 - Error, no counters updated
2097 * index - Updated the single counter pointed to by index
2098 * num_counters - Updated all counters (will reset the timestamp
2099 * and prevent further calls for lifespan milliseconds)
2100 * Drivers are allowed to update all counters in leiu of just the
2101 * one given in index at their option
2103 int (*get_hw_stats
)(struct ib_device
*device
,
2104 struct rdma_hw_stats
*stats
,
2105 u8 port
, int index
);
2106 int (*query_device
)(struct ib_device
*device
,
2107 struct ib_device_attr
*device_attr
,
2108 struct ib_udata
*udata
);
2109 int (*query_port
)(struct ib_device
*device
,
2111 struct ib_port_attr
*port_attr
);
2112 enum rdma_link_layer (*get_link_layer
)(struct ib_device
*device
,
2114 /* When calling get_netdev, the HW vendor's driver should return the
2115 * net device of device @device at port @port_num or NULL if such
2116 * a net device doesn't exist. The vendor driver should call dev_hold
2117 * on this net device. The HW vendor's device driver must guarantee
2118 * that this function returns NULL before the net device reaches
2119 * NETDEV_UNREGISTER_FINAL state.
2121 struct net_device
*(*get_netdev
)(struct ib_device
*device
,
2123 int (*query_gid
)(struct ib_device
*device
,
2124 u8 port_num
, int index
,
2126 /* When calling add_gid, the HW vendor's driver should
2127 * add the gid of device @device at gid index @index of
2128 * port @port_num to be @gid. Meta-info of that gid (for example,
2129 * the network device related to this gid is available
2130 * at @attr. @context allows the HW vendor driver to store extra
2131 * information together with a GID entry. The HW vendor may allocate
2132 * memory to contain this information and store it in @context when a
2133 * new GID entry is written to. Params are consistent until the next
2134 * call of add_gid or delete_gid. The function should return 0 on
2135 * success or error otherwise. The function could be called
2136 * concurrently for different ports. This function is only called
2137 * when roce_gid_table is used.
2139 int (*add_gid
)(struct ib_device
*device
,
2142 const union ib_gid
*gid
,
2143 const struct ib_gid_attr
*attr
,
2145 /* When calling del_gid, the HW vendor's driver should delete the
2146 * gid of device @device at gid index @index of port @port_num.
2147 * Upon the deletion of a GID entry, the HW vendor must free any
2148 * allocated memory. The caller will clear @context afterwards.
2149 * This function is only called when roce_gid_table is used.
2151 int (*del_gid
)(struct ib_device
*device
,
2155 int (*query_pkey
)(struct ib_device
*device
,
2156 u8 port_num
, u16 index
, u16
*pkey
);
2157 int (*modify_device
)(struct ib_device
*device
,
2158 int device_modify_mask
,
2159 struct ib_device_modify
*device_modify
);
2160 int (*modify_port
)(struct ib_device
*device
,
2161 u8 port_num
, int port_modify_mask
,
2162 struct ib_port_modify
*port_modify
);
2163 struct ib_ucontext
* (*alloc_ucontext
)(struct ib_device
*device
,
2164 struct ib_udata
*udata
);
2165 int (*dealloc_ucontext
)(struct ib_ucontext
*context
);
2166 int (*mmap
)(struct ib_ucontext
*context
,
2167 struct vm_area_struct
*vma
);
2168 struct ib_pd
* (*alloc_pd
)(struct ib_device
*device
,
2169 struct ib_ucontext
*context
,
2170 struct ib_udata
*udata
);
2171 int (*dealloc_pd
)(struct ib_pd
*pd
);
2172 struct ib_ah
* (*create_ah
)(struct ib_pd
*pd
,
2173 struct rdma_ah_attr
*ah_attr
,
2174 struct ib_udata
*udata
);
2175 int (*modify_ah
)(struct ib_ah
*ah
,
2176 struct rdma_ah_attr
*ah_attr
);
2177 int (*query_ah
)(struct ib_ah
*ah
,
2178 struct rdma_ah_attr
*ah_attr
);
2179 int (*destroy_ah
)(struct ib_ah
*ah
);
2180 struct ib_srq
* (*create_srq
)(struct ib_pd
*pd
,
2181 struct ib_srq_init_attr
*srq_init_attr
,
2182 struct ib_udata
*udata
);
2183 int (*modify_srq
)(struct ib_srq
*srq
,
2184 struct ib_srq_attr
*srq_attr
,
2185 enum ib_srq_attr_mask srq_attr_mask
,
2186 struct ib_udata
*udata
);
2187 int (*query_srq
)(struct ib_srq
*srq
,
2188 struct ib_srq_attr
*srq_attr
);
2189 int (*destroy_srq
)(struct ib_srq
*srq
);
2190 int (*post_srq_recv
)(struct ib_srq
*srq
,
2191 struct ib_recv_wr
*recv_wr
,
2192 struct ib_recv_wr
**bad_recv_wr
);
2193 struct ib_qp
* (*create_qp
)(struct ib_pd
*pd
,
2194 struct ib_qp_init_attr
*qp_init_attr
,
2195 struct ib_udata
*udata
);
2196 int (*modify_qp
)(struct ib_qp
*qp
,
2197 struct ib_qp_attr
*qp_attr
,
2199 struct ib_udata
*udata
);
2200 int (*query_qp
)(struct ib_qp
*qp
,
2201 struct ib_qp_attr
*qp_attr
,
2203 struct ib_qp_init_attr
*qp_init_attr
);
2204 int (*destroy_qp
)(struct ib_qp
*qp
);
2205 int (*post_send
)(struct ib_qp
*qp
,
2206 struct ib_send_wr
*send_wr
,
2207 struct ib_send_wr
**bad_send_wr
);
2208 int (*post_recv
)(struct ib_qp
*qp
,
2209 struct ib_recv_wr
*recv_wr
,
2210 struct ib_recv_wr
**bad_recv_wr
);
2211 struct ib_cq
* (*create_cq
)(struct ib_device
*device
,
2212 const struct ib_cq_init_attr
*attr
,
2213 struct ib_ucontext
*context
,
2214 struct ib_udata
*udata
);
2215 int (*modify_cq
)(struct ib_cq
*cq
, u16 cq_count
,
2217 int (*destroy_cq
)(struct ib_cq
*cq
);
2218 int (*resize_cq
)(struct ib_cq
*cq
, int cqe
,
2219 struct ib_udata
*udata
);
2220 int (*poll_cq
)(struct ib_cq
*cq
, int num_entries
,
2222 int (*peek_cq
)(struct ib_cq
*cq
, int wc_cnt
);
2223 int (*req_notify_cq
)(struct ib_cq
*cq
,
2224 enum ib_cq_notify_flags flags
);
2225 int (*req_ncomp_notif
)(struct ib_cq
*cq
,
2227 struct ib_mr
* (*get_dma_mr
)(struct ib_pd
*pd
,
2228 int mr_access_flags
);
2229 struct ib_mr
* (*reg_user_mr
)(struct ib_pd
*pd
,
2230 u64 start
, u64 length
,
2232 int mr_access_flags
,
2233 struct ib_udata
*udata
);
2234 int (*rereg_user_mr
)(struct ib_mr
*mr
,
2236 u64 start
, u64 length
,
2238 int mr_access_flags
,
2240 struct ib_udata
*udata
);
2241 int (*dereg_mr
)(struct ib_mr
*mr
);
2242 struct ib_mr
* (*alloc_mr
)(struct ib_pd
*pd
,
2243 enum ib_mr_type mr_type
,
2245 int (*map_mr_sg
)(struct ib_mr
*mr
,
2246 struct scatterlist
*sg
,
2248 unsigned int *sg_offset
);
2249 struct ib_mw
* (*alloc_mw
)(struct ib_pd
*pd
,
2250 enum ib_mw_type type
,
2251 struct ib_udata
*udata
);
2252 int (*dealloc_mw
)(struct ib_mw
*mw
);
2253 struct ib_fmr
* (*alloc_fmr
)(struct ib_pd
*pd
,
2254 int mr_access_flags
,
2255 struct ib_fmr_attr
*fmr_attr
);
2256 int (*map_phys_fmr
)(struct ib_fmr
*fmr
,
2257 u64
*page_list
, int list_len
,
2259 int (*unmap_fmr
)(struct list_head
*fmr_list
);
2260 int (*dealloc_fmr
)(struct ib_fmr
*fmr
);
2261 int (*attach_mcast
)(struct ib_qp
*qp
,
2264 int (*detach_mcast
)(struct ib_qp
*qp
,
2267 int (*process_mad
)(struct ib_device
*device
,
2268 int process_mad_flags
,
2270 const struct ib_wc
*in_wc
,
2271 const struct ib_grh
*in_grh
,
2272 const struct ib_mad_hdr
*in_mad
,
2274 struct ib_mad_hdr
*out_mad
,
2275 size_t *out_mad_size
,
2276 u16
*out_mad_pkey_index
);
2277 struct ib_xrcd
* (*alloc_xrcd
)(struct ib_device
*device
,
2278 struct ib_ucontext
*ucontext
,
2279 struct ib_udata
*udata
);
2280 int (*dealloc_xrcd
)(struct ib_xrcd
*xrcd
);
2281 struct ib_flow
* (*create_flow
)(struct ib_qp
*qp
,
2285 int (*destroy_flow
)(struct ib_flow
*flow_id
);
2286 int (*check_mr_status
)(struct ib_mr
*mr
, u32 check_mask
,
2287 struct ib_mr_status
*mr_status
);
2288 void (*disassociate_ucontext
)(struct ib_ucontext
*ibcontext
);
2289 void (*drain_rq
)(struct ib_qp
*qp
);
2290 void (*drain_sq
)(struct ib_qp
*qp
);
2291 int (*set_vf_link_state
)(struct ib_device
*device
, int vf
, u8 port
,
2293 int (*get_vf_config
)(struct ib_device
*device
, int vf
, u8 port
,
2294 struct ifla_vf_info
*ivf
);
2295 int (*get_vf_stats
)(struct ib_device
*device
, int vf
, u8 port
,
2296 struct ifla_vf_stats
*stats
);
2297 int (*set_vf_guid
)(struct ib_device
*device
, int vf
, u8 port
, u64 guid
,
2299 struct ib_wq
* (*create_wq
)(struct ib_pd
*pd
,
2300 struct ib_wq_init_attr
*init_attr
,
2301 struct ib_udata
*udata
);
2302 int (*destroy_wq
)(struct ib_wq
*wq
);
2303 int (*modify_wq
)(struct ib_wq
*wq
,
2304 struct ib_wq_attr
*attr
,
2306 struct ib_udata
*udata
);
2307 struct ib_rwq_ind_table
* (*create_rwq_ind_table
)(struct ib_device
*device
,
2308 struct ib_rwq_ind_table_init_attr
*init_attr
,
2309 struct ib_udata
*udata
);
2310 int (*destroy_rwq_ind_table
)(struct ib_rwq_ind_table
*wq_ind_table
);
2312 * rdma netdev operation
2314 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2315 * doesn't support the specified rdma netdev type.
2317 struct net_device
*(*alloc_rdma_netdev
)(
2318 struct ib_device
*device
,
2320 enum rdma_netdev_t type
,
2322 unsigned char name_assign_type
,
2323 void (*setup
)(struct net_device
*));
2325 struct module
*owner
;
2327 struct kobject
*ports_parent
;
2328 struct list_head port_list
;
2331 IB_DEV_UNINITIALIZED
,
2337 u64 uverbs_cmd_mask
;
2338 u64 uverbs_ex_cmd_mask
;
2340 char node_desc
[IB_DEVICE_NODE_DESC_MAX
];
2346 struct ib_device_attr attrs
;
2347 struct attribute_group
*hw_stats_ag
;
2348 struct rdma_hw_stats
*hw_stats
;
2350 #ifdef CONFIG_CGROUP_RDMA
2351 struct rdmacg_device cg_device
;
2357 * The following mandatory functions are used only at device
2358 * registration. Keep functions such as these at the end of this
2359 * structure to avoid cache line misses when accessing struct ib_device
2362 int (*get_port_immutable
)(struct ib_device
*, u8
, struct ib_port_immutable
*);
2363 void (*get_dev_fw_str
)(struct ib_device
*, char *str
);
2364 const struct cpumask
*(*get_vector_affinity
)(struct ib_device
*ibdev
,
2367 struct uverbs_root_spec
*specs_root
;
2372 void (*add
) (struct ib_device
*);
2373 void (*remove
)(struct ib_device
*, void *client_data
);
2375 /* Returns the net_dev belonging to this ib_client and matching the
2377 * @dev: An RDMA device that the net_dev use for communication.
2378 * @port: A physical port number on the RDMA device.
2379 * @pkey: P_Key that the net_dev uses if applicable.
2380 * @gid: A GID that the net_dev uses to communicate.
2381 * @addr: An IP address the net_dev is configured with.
2382 * @client_data: The device's client data set by ib_set_client_data().
2384 * An ib_client that implements a net_dev on top of RDMA devices
2385 * (such as IP over IB) should implement this callback, allowing the
2386 * rdma_cm module to find the right net_dev for a given request.
2388 * The caller is responsible for calling dev_put on the returned
2390 struct net_device
*(*get_net_dev_by_params
)(
2391 struct ib_device
*dev
,
2394 const union ib_gid
*gid
,
2395 const struct sockaddr
*addr
,
2397 struct list_head list
;
2400 struct ib_device
*ib_alloc_device(size_t size
);
2401 void ib_dealloc_device(struct ib_device
*device
);
2403 void ib_get_device_fw_str(struct ib_device
*device
, char *str
);
2405 int ib_register_device(struct ib_device
*device
,
2406 int (*port_callback
)(struct ib_device
*,
2407 u8
, struct kobject
*));
2408 void ib_unregister_device(struct ib_device
*device
);
2410 int ib_register_client (struct ib_client
*client
);
2411 void ib_unregister_client(struct ib_client
*client
);
2413 void *ib_get_client_data(struct ib_device
*device
, struct ib_client
*client
);
2414 void ib_set_client_data(struct ib_device
*device
, struct ib_client
*client
,
2417 static inline int ib_copy_from_udata(void *dest
, struct ib_udata
*udata
, size_t len
)
2419 return copy_from_user(dest
, udata
->inbuf
, len
) ? -EFAULT
: 0;
2422 static inline int ib_copy_to_udata(struct ib_udata
*udata
, void *src
, size_t len
)
2424 return copy_to_user(udata
->outbuf
, src
, len
) ? -EFAULT
: 0;
2427 static inline bool ib_is_udata_cleared(struct ib_udata
*udata
,
2431 const void __user
*p
= udata
->inbuf
+ offset
;
2435 if (len
> USHRT_MAX
)
2438 buf
= memdup_user(p
, len
);
2442 ret
= !memchr_inv(buf
, 0, len
);
2448 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2449 * contains all required attributes and no attributes not allowed for
2450 * the given QP state transition.
2451 * @cur_state: Current QP state
2452 * @next_state: Next QP state
2454 * @mask: Mask of supplied QP attributes
2455 * @ll : link layer of port
2457 * This function is a helper function that a low-level driver's
2458 * modify_qp method can use to validate the consumer's input. It
2459 * checks that cur_state and next_state are valid QP states, that a
2460 * transition from cur_state to next_state is allowed by the IB spec,
2461 * and that the attribute mask supplied is allowed for the transition.
2463 int ib_modify_qp_is_ok(enum ib_qp_state cur_state
, enum ib_qp_state next_state
,
2464 enum ib_qp_type type
, enum ib_qp_attr_mask mask
,
2465 enum rdma_link_layer ll
);
2467 void ib_register_event_handler(struct ib_event_handler
*event_handler
);
2468 void ib_unregister_event_handler(struct ib_event_handler
*event_handler
);
2469 void ib_dispatch_event(struct ib_event
*event
);
2471 int ib_query_port(struct ib_device
*device
,
2472 u8 port_num
, struct ib_port_attr
*port_attr
);
2474 enum rdma_link_layer
rdma_port_get_link_layer(struct ib_device
*device
,
2478 * rdma_cap_ib_switch - Check if the device is IB switch
2479 * @device: Device to check
2481 * Device driver is responsible for setting is_switch bit on
2482 * in ib_device structure at init time.
2484 * Return: true if the device is IB switch.
2486 static inline bool rdma_cap_ib_switch(const struct ib_device
*device
)
2488 return device
->is_switch
;
2492 * rdma_start_port - Return the first valid port number for the device
2495 * @device: Device to be checked
2497 * Return start port number
2499 static inline u8
rdma_start_port(const struct ib_device
*device
)
2501 return rdma_cap_ib_switch(device
) ? 0 : 1;
2505 * rdma_end_port - Return the last valid port number for the device
2508 * @device: Device to be checked
2510 * Return last port number
2512 static inline u8
rdma_end_port(const struct ib_device
*device
)
2514 return rdma_cap_ib_switch(device
) ? 0 : device
->phys_port_cnt
;
2517 static inline int rdma_is_port_valid(const struct ib_device
*device
,
2520 return (port
>= rdma_start_port(device
) &&
2521 port
<= rdma_end_port(device
));
2524 static inline bool rdma_protocol_ib(const struct ib_device
*device
, u8 port_num
)
2526 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IB
;
2529 static inline bool rdma_protocol_roce(const struct ib_device
*device
, u8 port_num
)
2531 return device
->port_immutable
[port_num
].core_cap_flags
&
2532 (RDMA_CORE_CAP_PROT_ROCE
| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP
);
2535 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device
*device
, u8 port_num
)
2537 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP
;
2540 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device
*device
, u8 port_num
)
2542 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_ROCE
;
2545 static inline bool rdma_protocol_iwarp(const struct ib_device
*device
, u8 port_num
)
2547 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IWARP
;
2550 static inline bool rdma_ib_or_roce(const struct ib_device
*device
, u8 port_num
)
2552 return rdma_protocol_ib(device
, port_num
) ||
2553 rdma_protocol_roce(device
, port_num
);
2556 static inline bool rdma_protocol_raw_packet(const struct ib_device
*device
, u8 port_num
)
2558 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_RAW_PACKET
;
2561 static inline bool rdma_protocol_usnic(const struct ib_device
*device
, u8 port_num
)
2563 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_USNIC
;
2567 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2568 * Management Datagrams.
2569 * @device: Device to check
2570 * @port_num: Port number to check
2572 * Management Datagrams (MAD) are a required part of the InfiniBand
2573 * specification and are supported on all InfiniBand devices. A slightly
2574 * extended version are also supported on OPA interfaces.
2576 * Return: true if the port supports sending/receiving of MAD packets.
2578 static inline bool rdma_cap_ib_mad(const struct ib_device
*device
, u8 port_num
)
2580 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_MAD
;
2584 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2585 * Management Datagrams.
2586 * @device: Device to check
2587 * @port_num: Port number to check
2589 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2590 * datagrams with their own versions. These OPA MADs share many but not all of
2591 * the characteristics of InfiniBand MADs.
2593 * OPA MADs differ in the following ways:
2595 * 1) MADs are variable size up to 2K
2596 * IBTA defined MADs remain fixed at 256 bytes
2597 * 2) OPA SMPs must carry valid PKeys
2598 * 3) OPA SMP packets are a different format
2600 * Return: true if the port supports OPA MAD packet formats.
2602 static inline bool rdma_cap_opa_mad(struct ib_device
*device
, u8 port_num
)
2604 return (device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_OPA_MAD
)
2605 == RDMA_CORE_CAP_OPA_MAD
;
2609 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2610 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2611 * @device: Device to check
2612 * @port_num: Port number to check
2614 * Each InfiniBand node is required to provide a Subnet Management Agent
2615 * that the subnet manager can access. Prior to the fabric being fully
2616 * configured by the subnet manager, the SMA is accessed via a well known
2617 * interface called the Subnet Management Interface (SMI). This interface
2618 * uses directed route packets to communicate with the SM to get around the
2619 * chicken and egg problem of the SM needing to know what's on the fabric
2620 * in order to configure the fabric, and needing to configure the fabric in
2621 * order to send packets to the devices on the fabric. These directed
2622 * route packets do not need the fabric fully configured in order to reach
2623 * their destination. The SMI is the only method allowed to send
2624 * directed route packets on an InfiniBand fabric.
2626 * Return: true if the port provides an SMI.
2628 static inline bool rdma_cap_ib_smi(const struct ib_device
*device
, u8 port_num
)
2630 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SMI
;
2634 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2635 * Communication Manager.
2636 * @device: Device to check
2637 * @port_num: Port number to check
2639 * The InfiniBand Communication Manager is one of many pre-defined General
2640 * Service Agents (GSA) that are accessed via the General Service
2641 * Interface (GSI). It's role is to facilitate establishment of connections
2642 * between nodes as well as other management related tasks for established
2645 * Return: true if the port supports an IB CM (this does not guarantee that
2646 * a CM is actually running however).
2648 static inline bool rdma_cap_ib_cm(const struct ib_device
*device
, u8 port_num
)
2650 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_CM
;
2654 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2655 * Communication Manager.
2656 * @device: Device to check
2657 * @port_num: Port number to check
2659 * Similar to above, but specific to iWARP connections which have a different
2660 * managment protocol than InfiniBand.
2662 * Return: true if the port supports an iWARP CM (this does not guarantee that
2663 * a CM is actually running however).
2665 static inline bool rdma_cap_iw_cm(const struct ib_device
*device
, u8 port_num
)
2667 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IW_CM
;
2671 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2672 * Subnet Administration.
2673 * @device: Device to check
2674 * @port_num: Port number to check
2676 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2677 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2678 * fabrics, devices should resolve routes to other hosts by contacting the
2679 * SA to query the proper route.
2681 * Return: true if the port should act as a client to the fabric Subnet
2682 * Administration interface. This does not imply that the SA service is
2685 static inline bool rdma_cap_ib_sa(const struct ib_device
*device
, u8 port_num
)
2687 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SA
;
2691 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2693 * @device: Device to check
2694 * @port_num: Port number to check
2696 * InfiniBand multicast registration is more complex than normal IPv4 or
2697 * IPv6 multicast registration. Each Host Channel Adapter must register
2698 * with the Subnet Manager when it wishes to join a multicast group. It
2699 * should do so only once regardless of how many queue pairs it subscribes
2700 * to this group. And it should leave the group only after all queue pairs
2701 * attached to the group have been detached.
2703 * Return: true if the port must undertake the additional adminstrative
2704 * overhead of registering/unregistering with the SM and tracking of the
2705 * total number of queue pairs attached to the multicast group.
2707 static inline bool rdma_cap_ib_mcast(const struct ib_device
*device
, u8 port_num
)
2709 return rdma_cap_ib_sa(device
, port_num
);
2713 * rdma_cap_af_ib - Check if the port of device has the capability
2714 * Native Infiniband Address.
2715 * @device: Device to check
2716 * @port_num: Port number to check
2718 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2719 * GID. RoCE uses a different mechanism, but still generates a GID via
2720 * a prescribed mechanism and port specific data.
2722 * Return: true if the port uses a GID address to identify devices on the
2725 static inline bool rdma_cap_af_ib(const struct ib_device
*device
, u8 port_num
)
2727 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_AF_IB
;
2731 * rdma_cap_eth_ah - Check if the port of device has the capability
2732 * Ethernet Address Handle.
2733 * @device: Device to check
2734 * @port_num: Port number to check
2736 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2737 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2738 * port. Normally, packet headers are generated by the sending host
2739 * adapter, but when sending connectionless datagrams, we must manually
2740 * inject the proper headers for the fabric we are communicating over.
2742 * Return: true if we are running as a RoCE port and must force the
2743 * addition of a Global Route Header built from our Ethernet Address
2744 * Handle into our header list for connectionless packets.
2746 static inline bool rdma_cap_eth_ah(const struct ib_device
*device
, u8 port_num
)
2748 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_ETH_AH
;
2752 * rdma_cap_opa_ah - Check if the port of device supports
2753 * OPA Address handles
2754 * @device: Device to check
2755 * @port_num: Port number to check
2757 * Return: true if we are running on an OPA device which supports
2758 * the extended OPA addressing.
2760 static inline bool rdma_cap_opa_ah(struct ib_device
*device
, u8 port_num
)
2762 return (device
->port_immutable
[port_num
].core_cap_flags
&
2763 RDMA_CORE_CAP_OPA_AH
) == RDMA_CORE_CAP_OPA_AH
;
2767 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2770 * @port_num: Port number
2772 * This MAD size includes the MAD headers and MAD payload. No other headers
2775 * Return the max MAD size required by the Port. Will return 0 if the port
2776 * does not support MADs
2778 static inline size_t rdma_max_mad_size(const struct ib_device
*device
, u8 port_num
)
2780 return device
->port_immutable
[port_num
].max_mad_size
;
2784 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2785 * @device: Device to check
2786 * @port_num: Port number to check
2788 * RoCE GID table mechanism manages the various GIDs for a device.
2790 * NOTE: if allocating the port's GID table has failed, this call will still
2791 * return true, but any RoCE GID table API will fail.
2793 * Return: true if the port uses RoCE GID table mechanism in order to manage
2796 static inline bool rdma_cap_roce_gid_table(const struct ib_device
*device
,
2799 return rdma_protocol_roce(device
, port_num
) &&
2800 device
->add_gid
&& device
->del_gid
;
2804 * Check if the device supports READ W/ INVALIDATE.
2806 static inline bool rdma_cap_read_inv(struct ib_device
*dev
, u32 port_num
)
2809 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2810 * has support for it yet.
2812 return rdma_protocol_iwarp(dev
, port_num
);
2815 int ib_query_gid(struct ib_device
*device
,
2816 u8 port_num
, int index
, union ib_gid
*gid
,
2817 struct ib_gid_attr
*attr
);
2819 int ib_set_vf_link_state(struct ib_device
*device
, int vf
, u8 port
,
2821 int ib_get_vf_config(struct ib_device
*device
, int vf
, u8 port
,
2822 struct ifla_vf_info
*info
);
2823 int ib_get_vf_stats(struct ib_device
*device
, int vf
, u8 port
,
2824 struct ifla_vf_stats
*stats
);
2825 int ib_set_vf_guid(struct ib_device
*device
, int vf
, u8 port
, u64 guid
,
2828 int ib_query_pkey(struct ib_device
*device
,
2829 u8 port_num
, u16 index
, u16
*pkey
);
2831 int ib_modify_device(struct ib_device
*device
,
2832 int device_modify_mask
,
2833 struct ib_device_modify
*device_modify
);
2835 int ib_modify_port(struct ib_device
*device
,
2836 u8 port_num
, int port_modify_mask
,
2837 struct ib_port_modify
*port_modify
);
2839 int ib_find_gid(struct ib_device
*device
, union ib_gid
*gid
,
2840 enum ib_gid_type gid_type
, struct net_device
*ndev
,
2841 u8
*port_num
, u16
*index
);
2843 int ib_find_pkey(struct ib_device
*device
,
2844 u8 port_num
, u16 pkey
, u16
*index
);
2848 * Create a memory registration for all memory in the system and place
2849 * the rkey for it into pd->unsafe_global_rkey. This can be used by
2850 * ULPs to avoid the overhead of dynamic MRs.
2852 * This flag is generally considered unsafe and must only be used in
2853 * extremly trusted environments. Every use of it will log a warning
2854 * in the kernel log.
2856 IB_PD_UNSAFE_GLOBAL_RKEY
= 0x01,
2859 struct ib_pd
*__ib_alloc_pd(struct ib_device
*device
, unsigned int flags
,
2860 const char *caller
);
2861 #define ib_alloc_pd(device, flags) \
2862 __ib_alloc_pd((device), (flags), __func__)
2863 void ib_dealloc_pd(struct ib_pd
*pd
);
2866 * rdma_create_ah - Creates an address handle for the given address vector.
2867 * @pd: The protection domain associated with the address handle.
2868 * @ah_attr: The attributes of the address vector.
2870 * The address handle is used to reference a local or global destination
2871 * in all UD QP post sends.
2873 struct ib_ah
*rdma_create_ah(struct ib_pd
*pd
, struct rdma_ah_attr
*ah_attr
);
2876 * rdma_create_user_ah - Creates an address handle for the given address vector.
2877 * It resolves destination mac address for ah attribute of RoCE type.
2878 * @pd: The protection domain associated with the address handle.
2879 * @ah_attr: The attributes of the address vector.
2880 * @udata: pointer to user's input output buffer information need by
2883 * It returns 0 on success and returns appropriate error code on error.
2884 * The address handle is used to reference a local or global destination
2885 * in all UD QP post sends.
2887 struct ib_ah
*rdma_create_user_ah(struct ib_pd
*pd
,
2888 struct rdma_ah_attr
*ah_attr
,
2889 struct ib_udata
*udata
);
2891 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2893 * @hdr: the L3 header to parse
2894 * @net_type: type of header to parse
2895 * @sgid: place to store source gid
2896 * @dgid: place to store destination gid
2898 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr
*hdr
,
2899 enum rdma_network_type net_type
,
2900 union ib_gid
*sgid
, union ib_gid
*dgid
);
2903 * ib_get_rdma_header_version - Get the header version
2904 * @hdr: the L3 header to parse
2906 int ib_get_rdma_header_version(const union rdma_network_hdr
*hdr
);
2909 * ib_init_ah_from_wc - Initializes address handle attributes from a
2911 * @device: Device on which the received message arrived.
2912 * @port_num: Port on which the received message arrived.
2913 * @wc: Work completion associated with the received message.
2914 * @grh: References the received global route header. This parameter is
2915 * ignored unless the work completion indicates that the GRH is valid.
2916 * @ah_attr: Returned attributes that can be used when creating an address
2917 * handle for replying to the message.
2919 int ib_init_ah_from_wc(struct ib_device
*device
, u8 port_num
,
2920 const struct ib_wc
*wc
, const struct ib_grh
*grh
,
2921 struct rdma_ah_attr
*ah_attr
);
2924 * ib_create_ah_from_wc - Creates an address handle associated with the
2925 * sender of the specified work completion.
2926 * @pd: The protection domain associated with the address handle.
2927 * @wc: Work completion information associated with a received message.
2928 * @grh: References the received global route header. This parameter is
2929 * ignored unless the work completion indicates that the GRH is valid.
2930 * @port_num: The outbound port number to associate with the address.
2932 * The address handle is used to reference a local or global destination
2933 * in all UD QP post sends.
2935 struct ib_ah
*ib_create_ah_from_wc(struct ib_pd
*pd
, const struct ib_wc
*wc
,
2936 const struct ib_grh
*grh
, u8 port_num
);
2939 * rdma_modify_ah - Modifies the address vector associated with an address
2941 * @ah: The address handle to modify.
2942 * @ah_attr: The new address vector attributes to associate with the
2945 int rdma_modify_ah(struct ib_ah
*ah
, struct rdma_ah_attr
*ah_attr
);
2948 * rdma_query_ah - Queries the address vector associated with an address
2950 * @ah: The address handle to query.
2951 * @ah_attr: The address vector attributes associated with the address
2954 int rdma_query_ah(struct ib_ah
*ah
, struct rdma_ah_attr
*ah_attr
);
2957 * rdma_destroy_ah - Destroys an address handle.
2958 * @ah: The address handle to destroy.
2960 int rdma_destroy_ah(struct ib_ah
*ah
);
2963 * ib_create_srq - Creates a SRQ associated with the specified protection
2965 * @pd: The protection domain associated with the SRQ.
2966 * @srq_init_attr: A list of initial attributes required to create the
2967 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2968 * the actual capabilities of the created SRQ.
2970 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2971 * requested size of the SRQ, and set to the actual values allocated
2972 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2973 * will always be at least as large as the requested values.
2975 struct ib_srq
*ib_create_srq(struct ib_pd
*pd
,
2976 struct ib_srq_init_attr
*srq_init_attr
);
2979 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2980 * @srq: The SRQ to modify.
2981 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2982 * the current values of selected SRQ attributes are returned.
2983 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2984 * are being modified.
2986 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2987 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2988 * the number of receives queued drops below the limit.
2990 int ib_modify_srq(struct ib_srq
*srq
,
2991 struct ib_srq_attr
*srq_attr
,
2992 enum ib_srq_attr_mask srq_attr_mask
);
2995 * ib_query_srq - Returns the attribute list and current values for the
2997 * @srq: The SRQ to query.
2998 * @srq_attr: The attributes of the specified SRQ.
3000 int ib_query_srq(struct ib_srq
*srq
,
3001 struct ib_srq_attr
*srq_attr
);
3004 * ib_destroy_srq - Destroys the specified SRQ.
3005 * @srq: The SRQ to destroy.
3007 int ib_destroy_srq(struct ib_srq
*srq
);
3010 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3011 * @srq: The SRQ to post the work request on.
3012 * @recv_wr: A list of work requests to post on the receive queue.
3013 * @bad_recv_wr: On an immediate failure, this parameter will reference
3014 * the work request that failed to be posted on the QP.
3016 static inline int ib_post_srq_recv(struct ib_srq
*srq
,
3017 struct ib_recv_wr
*recv_wr
,
3018 struct ib_recv_wr
**bad_recv_wr
)
3020 return srq
->device
->post_srq_recv(srq
, recv_wr
, bad_recv_wr
);
3024 * ib_create_qp - Creates a QP associated with the specified protection
3026 * @pd: The protection domain associated with the QP.
3027 * @qp_init_attr: A list of initial attributes required to create the
3028 * QP. If QP creation succeeds, then the attributes are updated to
3029 * the actual capabilities of the created QP.
3031 struct ib_qp
*ib_create_qp(struct ib_pd
*pd
,
3032 struct ib_qp_init_attr
*qp_init_attr
);
3035 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3036 * @qp: The QP to modify.
3037 * @attr: On input, specifies the QP attributes to modify. On output,
3038 * the current values of selected QP attributes are returned.
3039 * @attr_mask: A bit-mask used to specify which attributes of the QP
3040 * are being modified.
3041 * @udata: pointer to user's input output buffer information
3042 * are being modified.
3043 * It returns 0 on success and returns appropriate error code on error.
3045 int ib_modify_qp_with_udata(struct ib_qp
*qp
,
3046 struct ib_qp_attr
*attr
,
3048 struct ib_udata
*udata
);
3051 * ib_modify_qp - Modifies the attributes for the specified QP and then
3052 * transitions the QP to the given state.
3053 * @qp: The QP to modify.
3054 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3055 * the current values of selected QP attributes are returned.
3056 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3057 * are being modified.
3059 int ib_modify_qp(struct ib_qp
*qp
,
3060 struct ib_qp_attr
*qp_attr
,
3064 * ib_query_qp - Returns the attribute list and current values for the
3066 * @qp: The QP to query.
3067 * @qp_attr: The attributes of the specified QP.
3068 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3069 * @qp_init_attr: Additional attributes of the selected QP.
3071 * The qp_attr_mask may be used to limit the query to gathering only the
3072 * selected attributes.
3074 int ib_query_qp(struct ib_qp
*qp
,
3075 struct ib_qp_attr
*qp_attr
,
3077 struct ib_qp_init_attr
*qp_init_attr
);
3080 * ib_destroy_qp - Destroys the specified QP.
3081 * @qp: The QP to destroy.
3083 int ib_destroy_qp(struct ib_qp
*qp
);
3086 * ib_open_qp - Obtain a reference to an existing sharable QP.
3087 * @xrcd - XRC domain
3088 * @qp_open_attr: Attributes identifying the QP to open.
3090 * Returns a reference to a sharable QP.
3092 struct ib_qp
*ib_open_qp(struct ib_xrcd
*xrcd
,
3093 struct ib_qp_open_attr
*qp_open_attr
);
3096 * ib_close_qp - Release an external reference to a QP.
3097 * @qp: The QP handle to release
3099 * The opened QP handle is released by the caller. The underlying
3100 * shared QP is not destroyed until all internal references are released.
3102 int ib_close_qp(struct ib_qp
*qp
);
3105 * ib_post_send - Posts a list of work requests to the send queue of
3107 * @qp: The QP to post the work request on.
3108 * @send_wr: A list of work requests to post on the send queue.
3109 * @bad_send_wr: On an immediate failure, this parameter will reference
3110 * the work request that failed to be posted on the QP.
3112 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3113 * error is returned, the QP state shall not be affected,
3114 * ib_post_send() will return an immediate error after queueing any
3115 * earlier work requests in the list.
3117 static inline int ib_post_send(struct ib_qp
*qp
,
3118 struct ib_send_wr
*send_wr
,
3119 struct ib_send_wr
**bad_send_wr
)
3121 return qp
->device
->post_send(qp
, send_wr
, bad_send_wr
);
3125 * ib_post_recv - Posts a list of work requests to the receive queue of
3127 * @qp: The QP to post the work request on.
3128 * @recv_wr: A list of work requests to post on the receive queue.
3129 * @bad_recv_wr: On an immediate failure, this parameter will reference
3130 * the work request that failed to be posted on the QP.
3132 static inline int ib_post_recv(struct ib_qp
*qp
,
3133 struct ib_recv_wr
*recv_wr
,
3134 struct ib_recv_wr
**bad_recv_wr
)
3136 return qp
->device
->post_recv(qp
, recv_wr
, bad_recv_wr
);
3139 struct ib_cq
*ib_alloc_cq(struct ib_device
*dev
, void *private,
3140 int nr_cqe
, int comp_vector
, enum ib_poll_context poll_ctx
);
3141 void ib_free_cq(struct ib_cq
*cq
);
3142 int ib_process_cq_direct(struct ib_cq
*cq
, int budget
);
3145 * ib_create_cq - Creates a CQ on the specified device.
3146 * @device: The device on which to create the CQ.
3147 * @comp_handler: A user-specified callback that is invoked when a
3148 * completion event occurs on the CQ.
3149 * @event_handler: A user-specified callback that is invoked when an
3150 * asynchronous event not associated with a completion occurs on the CQ.
3151 * @cq_context: Context associated with the CQ returned to the user via
3152 * the associated completion and event handlers.
3153 * @cq_attr: The attributes the CQ should be created upon.
3155 * Users can examine the cq structure to determine the actual CQ size.
3157 struct ib_cq
*ib_create_cq(struct ib_device
*device
,
3158 ib_comp_handler comp_handler
,
3159 void (*event_handler
)(struct ib_event
*, void *),
3161 const struct ib_cq_init_attr
*cq_attr
);
3164 * ib_resize_cq - Modifies the capacity of the CQ.
3165 * @cq: The CQ to resize.
3166 * @cqe: The minimum size of the CQ.
3168 * Users can examine the cq structure to determine the actual CQ size.
3170 int ib_resize_cq(struct ib_cq
*cq
, int cqe
);
3173 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3174 * @cq: The CQ to modify.
3175 * @cq_count: number of CQEs that will trigger an event
3176 * @cq_period: max period of time in usec before triggering an event
3179 int rdma_set_cq_moderation(struct ib_cq
*cq
, u16 cq_count
, u16 cq_period
);
3182 * ib_destroy_cq - Destroys the specified CQ.
3183 * @cq: The CQ to destroy.
3185 int ib_destroy_cq(struct ib_cq
*cq
);
3188 * ib_poll_cq - poll a CQ for completion(s)
3189 * @cq:the CQ being polled
3190 * @num_entries:maximum number of completions to return
3191 * @wc:array of at least @num_entries &struct ib_wc where completions
3194 * Poll a CQ for (possibly multiple) completions. If the return value
3195 * is < 0, an error occurred. If the return value is >= 0, it is the
3196 * number of completions returned. If the return value is
3197 * non-negative and < num_entries, then the CQ was emptied.
3199 static inline int ib_poll_cq(struct ib_cq
*cq
, int num_entries
,
3202 return cq
->device
->poll_cq(cq
, num_entries
, wc
);
3206 * ib_peek_cq - Returns the number of unreaped completions currently
3207 * on the specified CQ.
3208 * @cq: The CQ to peek.
3209 * @wc_cnt: A minimum number of unreaped completions to check for.
3211 * If the number of unreaped completions is greater than or equal to wc_cnt,
3212 * this function returns wc_cnt, otherwise, it returns the actual number of
3213 * unreaped completions.
3215 int ib_peek_cq(struct ib_cq
*cq
, int wc_cnt
);
3218 * ib_req_notify_cq - Request completion notification on a CQ.
3219 * @cq: The CQ to generate an event for.
3221 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3222 * to request an event on the next solicited event or next work
3223 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3224 * may also be |ed in to request a hint about missed events, as
3228 * < 0 means an error occurred while requesting notification
3229 * == 0 means notification was requested successfully, and if
3230 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3231 * were missed and it is safe to wait for another event. In
3232 * this case is it guaranteed that any work completions added
3233 * to the CQ since the last CQ poll will trigger a completion
3234 * notification event.
3235 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3236 * in. It means that the consumer must poll the CQ again to
3237 * make sure it is empty to avoid missing an event because of a
3238 * race between requesting notification and an entry being
3239 * added to the CQ. This return value means it is possible
3240 * (but not guaranteed) that a work completion has been added
3241 * to the CQ since the last poll without triggering a
3242 * completion notification event.
3244 static inline int ib_req_notify_cq(struct ib_cq
*cq
,
3245 enum ib_cq_notify_flags flags
)
3247 return cq
->device
->req_notify_cq(cq
, flags
);
3251 * ib_req_ncomp_notif - Request completion notification when there are
3252 * at least the specified number of unreaped completions on the CQ.
3253 * @cq: The CQ to generate an event for.
3254 * @wc_cnt: The number of unreaped completions that should be on the
3255 * CQ before an event is generated.
3257 static inline int ib_req_ncomp_notif(struct ib_cq
*cq
, int wc_cnt
)
3259 return cq
->device
->req_ncomp_notif
?
3260 cq
->device
->req_ncomp_notif(cq
, wc_cnt
) :
3265 * ib_dma_mapping_error - check a DMA addr for error
3266 * @dev: The device for which the dma_addr was created
3267 * @dma_addr: The DMA address to check
3269 static inline int ib_dma_mapping_error(struct ib_device
*dev
, u64 dma_addr
)
3271 return dma_mapping_error(dev
->dma_device
, dma_addr
);
3275 * ib_dma_map_single - Map a kernel virtual address to DMA address
3276 * @dev: The device for which the dma_addr is to be created
3277 * @cpu_addr: The kernel virtual address
3278 * @size: The size of the region in bytes
3279 * @direction: The direction of the DMA
3281 static inline u64
ib_dma_map_single(struct ib_device
*dev
,
3282 void *cpu_addr
, size_t size
,
3283 enum dma_data_direction direction
)
3285 return dma_map_single(dev
->dma_device
, cpu_addr
, size
, direction
);
3289 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3290 * @dev: The device for which the DMA address was created
3291 * @addr: The DMA address
3292 * @size: The size of the region in bytes
3293 * @direction: The direction of the DMA
3295 static inline void ib_dma_unmap_single(struct ib_device
*dev
,
3296 u64 addr
, size_t size
,
3297 enum dma_data_direction direction
)
3299 dma_unmap_single(dev
->dma_device
, addr
, size
, direction
);
3303 * ib_dma_map_page - Map a physical page to DMA address
3304 * @dev: The device for which the dma_addr is to be created
3305 * @page: The page to be mapped
3306 * @offset: The offset within the page
3307 * @size: The size of the region in bytes
3308 * @direction: The direction of the DMA
3310 static inline u64
ib_dma_map_page(struct ib_device
*dev
,
3312 unsigned long offset
,
3314 enum dma_data_direction direction
)
3316 return dma_map_page(dev
->dma_device
, page
, offset
, size
, direction
);
3320 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3321 * @dev: The device for which the DMA address was created
3322 * @addr: The DMA address
3323 * @size: The size of the region in bytes
3324 * @direction: The direction of the DMA
3326 static inline void ib_dma_unmap_page(struct ib_device
*dev
,
3327 u64 addr
, size_t size
,
3328 enum dma_data_direction direction
)
3330 dma_unmap_page(dev
->dma_device
, addr
, size
, direction
);
3334 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3335 * @dev: The device for which the DMA addresses are to be created
3336 * @sg: The array of scatter/gather entries
3337 * @nents: The number of scatter/gather entries
3338 * @direction: The direction of the DMA
3340 static inline int ib_dma_map_sg(struct ib_device
*dev
,
3341 struct scatterlist
*sg
, int nents
,
3342 enum dma_data_direction direction
)
3344 return dma_map_sg(dev
->dma_device
, sg
, nents
, direction
);
3348 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3349 * @dev: The device for which the DMA addresses were created
3350 * @sg: The array of scatter/gather entries
3351 * @nents: The number of scatter/gather entries
3352 * @direction: The direction of the DMA
3354 static inline void ib_dma_unmap_sg(struct ib_device
*dev
,
3355 struct scatterlist
*sg
, int nents
,
3356 enum dma_data_direction direction
)
3358 dma_unmap_sg(dev
->dma_device
, sg
, nents
, direction
);
3361 static inline int ib_dma_map_sg_attrs(struct ib_device
*dev
,
3362 struct scatterlist
*sg
, int nents
,
3363 enum dma_data_direction direction
,
3364 unsigned long dma_attrs
)
3366 return dma_map_sg_attrs(dev
->dma_device
, sg
, nents
, direction
,
3370 static inline void ib_dma_unmap_sg_attrs(struct ib_device
*dev
,
3371 struct scatterlist
*sg
, int nents
,
3372 enum dma_data_direction direction
,
3373 unsigned long dma_attrs
)
3375 dma_unmap_sg_attrs(dev
->dma_device
, sg
, nents
, direction
, dma_attrs
);
3378 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3379 * @dev: The device for which the DMA addresses were created
3380 * @sg: The scatter/gather entry
3382 * Note: this function is obsolete. To do: change all occurrences of
3383 * ib_sg_dma_address() into sg_dma_address().
3385 static inline u64
ib_sg_dma_address(struct ib_device
*dev
,
3386 struct scatterlist
*sg
)
3388 return sg_dma_address(sg
);
3392 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3393 * @dev: The device for which the DMA addresses were created
3394 * @sg: The scatter/gather entry
3396 * Note: this function is obsolete. To do: change all occurrences of
3397 * ib_sg_dma_len() into sg_dma_len().
3399 static inline unsigned int ib_sg_dma_len(struct ib_device
*dev
,
3400 struct scatterlist
*sg
)
3402 return sg_dma_len(sg
);
3406 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3407 * @dev: The device for which the DMA address was created
3408 * @addr: The DMA address
3409 * @size: The size of the region in bytes
3410 * @dir: The direction of the DMA
3412 static inline void ib_dma_sync_single_for_cpu(struct ib_device
*dev
,
3415 enum dma_data_direction dir
)
3417 dma_sync_single_for_cpu(dev
->dma_device
, addr
, size
, dir
);
3421 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3422 * @dev: The device for which the DMA address was created
3423 * @addr: The DMA address
3424 * @size: The size of the region in bytes
3425 * @dir: The direction of the DMA
3427 static inline void ib_dma_sync_single_for_device(struct ib_device
*dev
,
3430 enum dma_data_direction dir
)
3432 dma_sync_single_for_device(dev
->dma_device
, addr
, size
, dir
);
3436 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3437 * @dev: The device for which the DMA address is requested
3438 * @size: The size of the region to allocate in bytes
3439 * @dma_handle: A pointer for returning the DMA address of the region
3440 * @flag: memory allocator flags
3442 static inline void *ib_dma_alloc_coherent(struct ib_device
*dev
,
3444 dma_addr_t
*dma_handle
,
3447 return dma_alloc_coherent(dev
->dma_device
, size
, dma_handle
, flag
);
3451 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3452 * @dev: The device for which the DMA addresses were allocated
3453 * @size: The size of the region
3454 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3455 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3457 static inline void ib_dma_free_coherent(struct ib_device
*dev
,
3458 size_t size
, void *cpu_addr
,
3459 dma_addr_t dma_handle
)
3461 dma_free_coherent(dev
->dma_device
, size
, cpu_addr
, dma_handle
);
3465 * ib_dereg_mr - Deregisters a memory region and removes it from the
3466 * HCA translation table.
3467 * @mr: The memory region to deregister.
3469 * This function can fail, if the memory region has memory windows bound to it.
3471 int ib_dereg_mr(struct ib_mr
*mr
);
3473 struct ib_mr
*ib_alloc_mr(struct ib_pd
*pd
,
3474 enum ib_mr_type mr_type
,
3478 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3480 * @mr - struct ib_mr pointer to be updated.
3481 * @newkey - new key to be used.
3483 static inline void ib_update_fast_reg_key(struct ib_mr
*mr
, u8 newkey
)
3485 mr
->lkey
= (mr
->lkey
& 0xffffff00) | newkey
;
3486 mr
->rkey
= (mr
->rkey
& 0xffffff00) | newkey
;
3490 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3491 * for calculating a new rkey for type 2 memory windows.
3492 * @rkey - the rkey to increment.
3494 static inline u32
ib_inc_rkey(u32 rkey
)
3496 const u32 mask
= 0x000000ff;
3497 return ((rkey
+ 1) & mask
) | (rkey
& ~mask
);
3501 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3502 * @pd: The protection domain associated with the unmapped region.
3503 * @mr_access_flags: Specifies the memory access rights.
3504 * @fmr_attr: Attributes of the unmapped region.
3506 * A fast memory region must be mapped before it can be used as part of
3509 struct ib_fmr
*ib_alloc_fmr(struct ib_pd
*pd
,
3510 int mr_access_flags
,
3511 struct ib_fmr_attr
*fmr_attr
);
3514 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3515 * @fmr: The fast memory region to associate with the pages.
3516 * @page_list: An array of physical pages to map to the fast memory region.
3517 * @list_len: The number of pages in page_list.
3518 * @iova: The I/O virtual address to use with the mapped region.
3520 static inline int ib_map_phys_fmr(struct ib_fmr
*fmr
,
3521 u64
*page_list
, int list_len
,
3524 return fmr
->device
->map_phys_fmr(fmr
, page_list
, list_len
, iova
);
3528 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3529 * @fmr_list: A linked list of fast memory regions to unmap.
3531 int ib_unmap_fmr(struct list_head
*fmr_list
);
3534 * ib_dealloc_fmr - Deallocates a fast memory region.
3535 * @fmr: The fast memory region to deallocate.
3537 int ib_dealloc_fmr(struct ib_fmr
*fmr
);
3540 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3541 * @qp: QP to attach to the multicast group. The QP must be type
3543 * @gid: Multicast group GID.
3544 * @lid: Multicast group LID in host byte order.
3546 * In order to send and receive multicast packets, subnet
3547 * administration must have created the multicast group and configured
3548 * the fabric appropriately. The port associated with the specified
3549 * QP must also be a member of the multicast group.
3551 int ib_attach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
3554 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3555 * @qp: QP to detach from the multicast group.
3556 * @gid: Multicast group GID.
3557 * @lid: Multicast group LID in host byte order.
3559 int ib_detach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
3562 * ib_alloc_xrcd - Allocates an XRC domain.
3563 * @device: The device on which to allocate the XRC domain.
3565 struct ib_xrcd
*ib_alloc_xrcd(struct ib_device
*device
);
3568 * ib_dealloc_xrcd - Deallocates an XRC domain.
3569 * @xrcd: The XRC domain to deallocate.
3571 int ib_dealloc_xrcd(struct ib_xrcd
*xrcd
);
3573 struct ib_flow
*ib_create_flow(struct ib_qp
*qp
,
3574 struct ib_flow_attr
*flow_attr
, int domain
);
3575 int ib_destroy_flow(struct ib_flow
*flow_id
);
3577 static inline int ib_check_mr_access(int flags
)
3580 * Local write permission is required if remote write or
3581 * remote atomic permission is also requested.
3583 if (flags
& (IB_ACCESS_REMOTE_ATOMIC
| IB_ACCESS_REMOTE_WRITE
) &&
3584 !(flags
& IB_ACCESS_LOCAL_WRITE
))
3590 static inline bool ib_access_writable(int access_flags
)
3593 * We have writable memory backing the MR if any of the following
3594 * access flags are set. "Local write" and "remote write" obviously
3595 * require write access. "Remote atomic" can do things like fetch and
3596 * add, which will modify memory, and "MW bind" can change permissions
3597 * by binding a window.
3599 return access_flags
&
3600 (IB_ACCESS_LOCAL_WRITE
| IB_ACCESS_REMOTE_WRITE
|
3601 IB_ACCESS_REMOTE_ATOMIC
| IB_ACCESS_MW_BIND
);
3605 * ib_check_mr_status: lightweight check of MR status.
3606 * This routine may provide status checks on a selected
3607 * ib_mr. first use is for signature status check.
3609 * @mr: A memory region.
3610 * @check_mask: Bitmask of which checks to perform from
3611 * ib_mr_status_check enumeration.
3612 * @mr_status: The container of relevant status checks.
3613 * failed checks will be indicated in the status bitmask
3614 * and the relevant info shall be in the error item.
3616 int ib_check_mr_status(struct ib_mr
*mr
, u32 check_mask
,
3617 struct ib_mr_status
*mr_status
);
3619 struct net_device
*ib_get_net_dev_by_params(struct ib_device
*dev
, u8 port
,
3620 u16 pkey
, const union ib_gid
*gid
,
3621 const struct sockaddr
*addr
);
3622 struct ib_wq
*ib_create_wq(struct ib_pd
*pd
,
3623 struct ib_wq_init_attr
*init_attr
);
3624 int ib_destroy_wq(struct ib_wq
*wq
);
3625 int ib_modify_wq(struct ib_wq
*wq
, struct ib_wq_attr
*attr
,
3627 struct ib_rwq_ind_table
*ib_create_rwq_ind_table(struct ib_device
*device
,
3628 struct ib_rwq_ind_table_init_attr
*
3629 wq_ind_table_init_attr
);
3630 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table
*wq_ind_table
);
3632 int ib_map_mr_sg(struct ib_mr
*mr
, struct scatterlist
*sg
, int sg_nents
,
3633 unsigned int *sg_offset
, unsigned int page_size
);
3636 ib_map_mr_sg_zbva(struct ib_mr
*mr
, struct scatterlist
*sg
, int sg_nents
,
3637 unsigned int *sg_offset
, unsigned int page_size
)
3641 n
= ib_map_mr_sg(mr
, sg
, sg_nents
, sg_offset
, page_size
);
3647 int ib_sg_to_pages(struct ib_mr
*mr
, struct scatterlist
*sgl
, int sg_nents
,
3648 unsigned int *sg_offset
, int (*set_page
)(struct ib_mr
*, u64
));
3650 void ib_drain_rq(struct ib_qp
*qp
);
3651 void ib_drain_sq(struct ib_qp
*qp
);
3652 void ib_drain_qp(struct ib_qp
*qp
);
3654 int ib_get_eth_speed(struct ib_device
*dev
, u8 port_num
, u8
*speed
, u8
*width
);
3656 static inline u8
*rdma_ah_retrieve_dmac(struct rdma_ah_attr
*attr
)
3658 if (attr
->type
== RDMA_AH_ATTR_TYPE_ROCE
)
3659 return attr
->roce
.dmac
;
3663 static inline void rdma_ah_set_dlid(struct rdma_ah_attr
*attr
, u32 dlid
)
3665 if (attr
->type
== RDMA_AH_ATTR_TYPE_IB
)
3666 attr
->ib
.dlid
= (u16
)dlid
;
3667 else if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3668 attr
->opa
.dlid
= dlid
;
3671 static inline u32
rdma_ah_get_dlid(const struct rdma_ah_attr
*attr
)
3673 if (attr
->type
== RDMA_AH_ATTR_TYPE_IB
)
3674 return attr
->ib
.dlid
;
3675 else if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3676 return attr
->opa
.dlid
;
3680 static inline void rdma_ah_set_sl(struct rdma_ah_attr
*attr
, u8 sl
)
3685 static inline u8
rdma_ah_get_sl(const struct rdma_ah_attr
*attr
)
3690 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr
*attr
,
3693 if (attr
->type
== RDMA_AH_ATTR_TYPE_IB
)
3694 attr
->ib
.src_path_bits
= src_path_bits
;
3695 else if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3696 attr
->opa
.src_path_bits
= src_path_bits
;
3699 static inline u8
rdma_ah_get_path_bits(const struct rdma_ah_attr
*attr
)
3701 if (attr
->type
== RDMA_AH_ATTR_TYPE_IB
)
3702 return attr
->ib
.src_path_bits
;
3703 else if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3704 return attr
->opa
.src_path_bits
;
3708 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr
*attr
,
3711 if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3712 attr
->opa
.make_grd
= make_grd
;
3715 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr
*attr
)
3717 if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3718 return attr
->opa
.make_grd
;
3722 static inline void rdma_ah_set_port_num(struct rdma_ah_attr
*attr
, u8 port_num
)
3724 attr
->port_num
= port_num
;
3727 static inline u8
rdma_ah_get_port_num(const struct rdma_ah_attr
*attr
)
3729 return attr
->port_num
;
3732 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr
*attr
,
3735 attr
->static_rate
= static_rate
;
3738 static inline u8
rdma_ah_get_static_rate(const struct rdma_ah_attr
*attr
)
3740 return attr
->static_rate
;
3743 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr
*attr
,
3744 enum ib_ah_flags flag
)
3746 attr
->ah_flags
= flag
;
3749 static inline enum ib_ah_flags
3750 rdma_ah_get_ah_flags(const struct rdma_ah_attr
*attr
)
3752 return attr
->ah_flags
;
3755 static inline const struct ib_global_route
3756 *rdma_ah_read_grh(const struct rdma_ah_attr
*attr
)
3761 /*To retrieve and modify the grh */
3762 static inline struct ib_global_route
3763 *rdma_ah_retrieve_grh(struct rdma_ah_attr
*attr
)
3768 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr
*attr
, void *dgid
)
3770 struct ib_global_route
*grh
= rdma_ah_retrieve_grh(attr
);
3772 memcpy(grh
->dgid
.raw
, dgid
, sizeof(grh
->dgid
));
3775 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr
*attr
,
3778 struct ib_global_route
*grh
= rdma_ah_retrieve_grh(attr
);
3780 grh
->dgid
.global
.subnet_prefix
= prefix
;
3783 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr
*attr
,
3786 struct ib_global_route
*grh
= rdma_ah_retrieve_grh(attr
);
3788 grh
->dgid
.global
.interface_id
= if_id
;
3791 static inline void rdma_ah_set_grh(struct rdma_ah_attr
*attr
,
3792 union ib_gid
*dgid
, u32 flow_label
,
3793 u8 sgid_index
, u8 hop_limit
,
3796 struct ib_global_route
*grh
= rdma_ah_retrieve_grh(attr
);
3798 attr
->ah_flags
= IB_AH_GRH
;
3801 grh
->flow_label
= flow_label
;
3802 grh
->sgid_index
= sgid_index
;
3803 grh
->hop_limit
= hop_limit
;
3804 grh
->traffic_class
= traffic_class
;
3808 * rdma_ah_find_type - Return address handle type.
3810 * @dev: Device to be checked
3811 * @port_num: Port number
3813 static inline enum rdma_ah_attr_type
rdma_ah_find_type(struct ib_device
*dev
,
3816 if (rdma_protocol_roce(dev
, port_num
))
3817 return RDMA_AH_ATTR_TYPE_ROCE
;
3818 if (rdma_protocol_ib(dev
, port_num
)) {
3819 if (rdma_cap_opa_ah(dev
, port_num
))
3820 return RDMA_AH_ATTR_TYPE_OPA
;
3821 return RDMA_AH_ATTR_TYPE_IB
;
3824 return RDMA_AH_ATTR_TYPE_UNDEFINED
;
3828 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
3829 * In the current implementation the only way to get
3830 * get the 32bit lid is from other sources for OPA.
3831 * For IB, lids will always be 16bits so cast the
3832 * value accordingly.
3836 static inline u16
ib_lid_cpu16(u32 lid
)
3838 WARN_ON_ONCE(lid
& 0xFFFF0000);
3843 * ib_lid_be16 - Return lid in 16bit BE encoding.
3847 static inline __be16
ib_lid_be16(u32 lid
)
3849 WARN_ON_ONCE(lid
& 0xFFFF0000);
3850 return cpu_to_be16((u16
)lid
);
3854 * ib_get_vector_affinity - Get the affinity mappings of a given completion
3856 * @device: the rdma device
3857 * @comp_vector: index of completion vector
3859 * Returns NULL on failure, otherwise a corresponding cpu map of the
3860 * completion vector (returns all-cpus map if the device driver doesn't
3861 * implement get_vector_affinity).
3863 static inline const struct cpumask
*
3864 ib_get_vector_affinity(struct ib_device
*device
, int comp_vector
)
3866 if (comp_vector
< 0 || comp_vector
>= device
->num_comp_vectors
||
3867 !device
->get_vector_affinity
)
3870 return device
->get_vector_affinity(device
, comp_vector
);
3874 #endif /* IB_VERBS_H */