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 <uapi/linux/if_ether.h>
54 #include <linux/atomic.h>
55 #include <linux/mmu_notifier.h>
56 #include <asm/uaccess.h>
58 extern struct workqueue_struct
*ib_wq
;
69 /* IB values map to NodeInfo:NodeType. */
78 enum rdma_transport_type
{
82 RDMA_TRANSPORT_USNIC_UDP
85 enum rdma_protocol_type
{
89 RDMA_PROTOCOL_USNIC_UDP
92 __attribute_const__
enum rdma_transport_type
93 rdma_node_get_transport(enum rdma_node_type node_type
);
95 enum rdma_link_layer
{
96 IB_LINK_LAYER_UNSPECIFIED
,
97 IB_LINK_LAYER_INFINIBAND
,
98 IB_LINK_LAYER_ETHERNET
,
101 enum ib_device_cap_flags
{
102 IB_DEVICE_RESIZE_MAX_WR
= 1,
103 IB_DEVICE_BAD_PKEY_CNTR
= (1<<1),
104 IB_DEVICE_BAD_QKEY_CNTR
= (1<<2),
105 IB_DEVICE_RAW_MULTI
= (1<<3),
106 IB_DEVICE_AUTO_PATH_MIG
= (1<<4),
107 IB_DEVICE_CHANGE_PHY_PORT
= (1<<5),
108 IB_DEVICE_UD_AV_PORT_ENFORCE
= (1<<6),
109 IB_DEVICE_CURR_QP_STATE_MOD
= (1<<7),
110 IB_DEVICE_SHUTDOWN_PORT
= (1<<8),
111 IB_DEVICE_INIT_TYPE
= (1<<9),
112 IB_DEVICE_PORT_ACTIVE_EVENT
= (1<<10),
113 IB_DEVICE_SYS_IMAGE_GUID
= (1<<11),
114 IB_DEVICE_RC_RNR_NAK_GEN
= (1<<12),
115 IB_DEVICE_SRQ_RESIZE
= (1<<13),
116 IB_DEVICE_N_NOTIFY_CQ
= (1<<14),
117 IB_DEVICE_LOCAL_DMA_LKEY
= (1<<15),
118 IB_DEVICE_RESERVED
= (1<<16), /* old SEND_W_INV */
119 IB_DEVICE_MEM_WINDOW
= (1<<17),
121 * Devices should set IB_DEVICE_UD_IP_SUM if they support
122 * insertion of UDP and TCP checksum on outgoing UD IPoIB
123 * messages and can verify the validity of checksum for
124 * incoming messages. Setting this flag implies that the
125 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
127 IB_DEVICE_UD_IP_CSUM
= (1<<18),
128 IB_DEVICE_UD_TSO
= (1<<19),
129 IB_DEVICE_XRC
= (1<<20),
130 IB_DEVICE_MEM_MGT_EXTENSIONS
= (1<<21),
131 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK
= (1<<22),
132 IB_DEVICE_MEM_WINDOW_TYPE_2A
= (1<<23),
133 IB_DEVICE_MEM_WINDOW_TYPE_2B
= (1<<24),
134 IB_DEVICE_MANAGED_FLOW_STEERING
= (1<<29),
135 IB_DEVICE_SIGNATURE_HANDOVER
= (1<<30),
136 IB_DEVICE_ON_DEMAND_PAGING
= (1<<31),
139 enum ib_signature_prot_cap
{
140 IB_PROT_T10DIF_TYPE_1
= 1,
141 IB_PROT_T10DIF_TYPE_2
= 1 << 1,
142 IB_PROT_T10DIF_TYPE_3
= 1 << 2,
145 enum ib_signature_guard_cap
{
146 IB_GUARD_T10DIF_CRC
= 1,
147 IB_GUARD_T10DIF_CSUM
= 1 << 1,
156 enum ib_odp_general_cap_bits
{
157 IB_ODP_SUPPORT
= 1 << 0,
160 enum ib_odp_transport_cap_bits
{
161 IB_ODP_SUPPORT_SEND
= 1 << 0,
162 IB_ODP_SUPPORT_RECV
= 1 << 1,
163 IB_ODP_SUPPORT_WRITE
= 1 << 2,
164 IB_ODP_SUPPORT_READ
= 1 << 3,
165 IB_ODP_SUPPORT_ATOMIC
= 1 << 4,
169 uint64_t general_caps
;
171 uint32_t rc_odp_caps
;
172 uint32_t uc_odp_caps
;
173 uint32_t ud_odp_caps
;
174 } per_transport_caps
;
177 enum ib_cq_creation_flags
{
178 IB_CQ_FLAGS_TIMESTAMP_COMPLETION
= 1 << 0,
181 struct ib_cq_init_attr
{
187 struct ib_device_attr
{
189 __be64 sys_image_guid
;
197 int device_cap_flags
;
207 int max_qp_init_rd_atom
;
208 int max_ee_init_rd_atom
;
209 enum ib_atomic_cap atomic_cap
;
210 enum ib_atomic_cap masked_atomic_cap
;
217 int max_mcast_qp_attach
;
218 int max_total_mcast_qp_attach
;
225 unsigned int max_fast_reg_page_list_len
;
227 u8 local_ca_ack_delay
;
230 struct ib_odp_caps odp_caps
;
231 uint64_t timestamp_mask
;
232 uint64_t hca_core_clock
; /* in KHZ */
243 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu
)
246 case IB_MTU_256
: return 256;
247 case IB_MTU_512
: return 512;
248 case IB_MTU_1024
: return 1024;
249 case IB_MTU_2048
: return 2048;
250 case IB_MTU_4096
: return 4096;
261 IB_PORT_ACTIVE_DEFER
= 5
264 enum ib_port_cap_flags
{
266 IB_PORT_NOTICE_SUP
= 1 << 2,
267 IB_PORT_TRAP_SUP
= 1 << 3,
268 IB_PORT_OPT_IPD_SUP
= 1 << 4,
269 IB_PORT_AUTO_MIGR_SUP
= 1 << 5,
270 IB_PORT_SL_MAP_SUP
= 1 << 6,
271 IB_PORT_MKEY_NVRAM
= 1 << 7,
272 IB_PORT_PKEY_NVRAM
= 1 << 8,
273 IB_PORT_LED_INFO_SUP
= 1 << 9,
274 IB_PORT_SM_DISABLED
= 1 << 10,
275 IB_PORT_SYS_IMAGE_GUID_SUP
= 1 << 11,
276 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP
= 1 << 12,
277 IB_PORT_EXTENDED_SPEEDS_SUP
= 1 << 14,
278 IB_PORT_CM_SUP
= 1 << 16,
279 IB_PORT_SNMP_TUNNEL_SUP
= 1 << 17,
280 IB_PORT_REINIT_SUP
= 1 << 18,
281 IB_PORT_DEVICE_MGMT_SUP
= 1 << 19,
282 IB_PORT_VENDOR_CLASS_SUP
= 1 << 20,
283 IB_PORT_DR_NOTICE_SUP
= 1 << 21,
284 IB_PORT_CAP_MASK_NOTICE_SUP
= 1 << 22,
285 IB_PORT_BOOT_MGMT_SUP
= 1 << 23,
286 IB_PORT_LINK_LATENCY_SUP
= 1 << 24,
287 IB_PORT_CLIENT_REG_SUP
= 1 << 25,
288 IB_PORT_IP_BASED_GIDS
= 1 << 26
298 static inline int ib_width_enum_to_int(enum ib_port_width width
)
301 case IB_WIDTH_1X
: return 1;
302 case IB_WIDTH_4X
: return 4;
303 case IB_WIDTH_8X
: return 8;
304 case IB_WIDTH_12X
: return 12;
318 struct ib_protocol_stats
{
322 struct iw_protocol_stats
{
325 u64 ipInTooBigErrors
;
328 u64 ipInUnknownProtos
;
329 u64 ipInTruncatedPkts
;
332 u64 ipOutForwDatagrams
;
364 union rdma_protocol_stats
{
365 struct ib_protocol_stats ib
;
366 struct iw_protocol_stats iw
;
369 /* Define bits for the various functionality this port needs to be supported by
372 /* Management 0x00000FFF */
373 #define RDMA_CORE_CAP_IB_MAD 0x00000001
374 #define RDMA_CORE_CAP_IB_SMI 0x00000002
375 #define RDMA_CORE_CAP_IB_CM 0x00000004
376 #define RDMA_CORE_CAP_IW_CM 0x00000008
377 #define RDMA_CORE_CAP_IB_SA 0x00000010
378 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
380 /* Address format 0x000FF000 */
381 #define RDMA_CORE_CAP_AF_IB 0x00001000
382 #define RDMA_CORE_CAP_ETH_AH 0x00002000
384 /* Protocol 0xFFF00000 */
385 #define RDMA_CORE_CAP_PROT_IB 0x00100000
386 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
387 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
389 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
390 | RDMA_CORE_CAP_IB_MAD \
391 | RDMA_CORE_CAP_IB_SMI \
392 | RDMA_CORE_CAP_IB_CM \
393 | RDMA_CORE_CAP_IB_SA \
394 | RDMA_CORE_CAP_AF_IB)
395 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
396 | RDMA_CORE_CAP_IB_MAD \
397 | RDMA_CORE_CAP_IB_CM \
398 | RDMA_CORE_CAP_AF_IB \
399 | RDMA_CORE_CAP_ETH_AH)
400 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
401 | RDMA_CORE_CAP_IW_CM)
402 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
403 | RDMA_CORE_CAP_OPA_MAD)
405 struct ib_port_attr
{
406 enum ib_port_state state
;
408 enum ib_mtu active_mtu
;
427 enum ib_device_modify_flags
{
428 IB_DEVICE_MODIFY_SYS_IMAGE_GUID
= 1 << 0,
429 IB_DEVICE_MODIFY_NODE_DESC
= 1 << 1
432 struct ib_device_modify
{
437 enum ib_port_modify_flags
{
438 IB_PORT_SHUTDOWN
= 1,
439 IB_PORT_INIT_TYPE
= (1<<2),
440 IB_PORT_RESET_QKEY_CNTR
= (1<<3)
443 struct ib_port_modify
{
444 u32 set_port_cap_mask
;
445 u32 clr_port_cap_mask
;
453 IB_EVENT_QP_ACCESS_ERR
,
457 IB_EVENT_PATH_MIG_ERR
,
458 IB_EVENT_DEVICE_FATAL
,
459 IB_EVENT_PORT_ACTIVE
,
462 IB_EVENT_PKEY_CHANGE
,
465 IB_EVENT_SRQ_LIMIT_REACHED
,
466 IB_EVENT_QP_LAST_WQE_REACHED
,
467 IB_EVENT_CLIENT_REREGISTER
,
471 __attribute_const__
const char *ib_event_msg(enum ib_event_type event
);
474 struct ib_device
*device
;
481 enum ib_event_type event
;
484 struct ib_event_handler
{
485 struct ib_device
*device
;
486 void (*handler
)(struct ib_event_handler
*, struct ib_event
*);
487 struct list_head list
;
490 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
492 (_ptr)->device = _device; \
493 (_ptr)->handler = _handler; \
494 INIT_LIST_HEAD(&(_ptr)->list); \
497 struct ib_global_route
{
506 __be32 version_tclass_flow
;
515 IB_MULTICAST_QPN
= 0xffffff
518 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
525 IB_RATE_PORT_CURRENT
= 0,
526 IB_RATE_2_5_GBPS
= 2,
534 IB_RATE_120_GBPS
= 10,
535 IB_RATE_14_GBPS
= 11,
536 IB_RATE_56_GBPS
= 12,
537 IB_RATE_112_GBPS
= 13,
538 IB_RATE_168_GBPS
= 14,
539 IB_RATE_25_GBPS
= 15,
540 IB_RATE_100_GBPS
= 16,
541 IB_RATE_200_GBPS
= 17,
542 IB_RATE_300_GBPS
= 18
546 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
547 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
548 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
549 * @rate: rate to convert.
551 __attribute_const__
int ib_rate_to_mult(enum ib_rate rate
);
554 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
555 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
556 * @rate: rate to convert.
558 __attribute_const__
int ib_rate_to_mbps(enum ib_rate rate
);
560 enum ib_mr_create_flags
{
561 IB_MR_SIGNATURE_EN
= 1,
565 * ib_mr_init_attr - Memory region init attributes passed to routine
567 * @max_reg_descriptors: max number of registration descriptors that
568 * may be used with registration work requests.
569 * @flags: MR creation flags bit mask.
571 struct ib_mr_init_attr
{
572 int max_reg_descriptors
;
578 * IB_SIG_TYPE_NONE: Unprotected.
579 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
581 enum ib_signature_type
{
587 * Signature T10-DIF block-guard types
588 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
589 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
591 enum ib_t10_dif_bg_type
{
597 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
599 * @bg_type: T10-DIF block guard type (CRC|CSUM)
600 * @pi_interval: protection information interval.
601 * @bg: seed of guard computation.
602 * @app_tag: application tag of guard block
603 * @ref_tag: initial guard block reference tag.
604 * @ref_remap: Indicate wethear the reftag increments each block
605 * @app_escape: Indicate to skip block check if apptag=0xffff
606 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
607 * @apptag_check_mask: check bitmask of application tag.
609 struct ib_t10_dif_domain
{
610 enum ib_t10_dif_bg_type bg_type
;
618 u16 apptag_check_mask
;
622 * struct ib_sig_domain - Parameters for signature domain
623 * @sig_type: specific signauture type
624 * @sig: union of all signature domain attributes that may
625 * be used to set domain layout.
627 struct ib_sig_domain
{
628 enum ib_signature_type sig_type
;
630 struct ib_t10_dif_domain dif
;
635 * struct ib_sig_attrs - Parameters for signature handover operation
636 * @check_mask: bitmask for signature byte check (8 bytes)
637 * @mem: memory domain layout desciptor.
638 * @wire: wire domain layout desciptor.
640 struct ib_sig_attrs
{
642 struct ib_sig_domain mem
;
643 struct ib_sig_domain wire
;
646 enum ib_sig_err_type
{
653 * struct ib_sig_err - signature error descriptor
656 enum ib_sig_err_type err_type
;
663 enum ib_mr_status_check
{
664 IB_MR_CHECK_SIG_STATUS
= 1,
668 * struct ib_mr_status - Memory region status container
670 * @fail_status: Bitmask of MR checks status. For each
671 * failed check a corresponding status bit is set.
672 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
675 struct ib_mr_status
{
677 struct ib_sig_err sig_err
;
681 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
683 * @mult: multiple to convert.
685 __attribute_const__
enum ib_rate
mult_to_ib_rate(int mult
);
688 struct ib_global_route grh
;
703 IB_WC_LOC_EEC_OP_ERR
,
708 IB_WC_LOC_ACCESS_ERR
,
709 IB_WC_REM_INV_REQ_ERR
,
710 IB_WC_REM_ACCESS_ERR
,
713 IB_WC_RNR_RETRY_EXC_ERR
,
714 IB_WC_LOC_RDD_VIOL_ERR
,
715 IB_WC_REM_INV_RD_REQ_ERR
,
718 IB_WC_INV_EEC_STATE_ERR
,
720 IB_WC_RESP_TIMEOUT_ERR
,
724 __attribute_const__
const char *ib_wc_status_msg(enum ib_wc_status status
);
736 IB_WC_MASKED_COMP_SWAP
,
737 IB_WC_MASKED_FETCH_ADD
,
739 * Set value of IB_WC_RECV so consumers can test if a completion is a
740 * receive by testing (opcode & IB_WC_RECV).
743 IB_WC_RECV_RDMA_WITH_IMM
748 IB_WC_WITH_IMM
= (1<<1),
749 IB_WC_WITH_INVALIDATE
= (1<<2),
750 IB_WC_IP_CSUM_OK
= (1<<3),
751 IB_WC_WITH_SMAC
= (1<<4),
752 IB_WC_WITH_VLAN
= (1<<5),
757 enum ib_wc_status status
;
758 enum ib_wc_opcode opcode
;
772 u8 port_num
; /* valid only for DR SMPs on switches */
777 enum ib_cq_notify_flags
{
778 IB_CQ_SOLICITED
= 1 << 0,
779 IB_CQ_NEXT_COMP
= 1 << 1,
780 IB_CQ_SOLICITED_MASK
= IB_CQ_SOLICITED
| IB_CQ_NEXT_COMP
,
781 IB_CQ_REPORT_MISSED_EVENTS
= 1 << 2,
789 enum ib_srq_attr_mask
{
790 IB_SRQ_MAX_WR
= 1 << 0,
791 IB_SRQ_LIMIT
= 1 << 1,
800 struct ib_srq_init_attr
{
801 void (*event_handler
)(struct ib_event
*, void *);
803 struct ib_srq_attr attr
;
804 enum ib_srq_type srq_type
;
808 struct ib_xrcd
*xrcd
;
829 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
830 * here (and in that order) since the MAD layer uses them as
831 * indices into a 2-entry table.
840 IB_QPT_RAW_ETHERTYPE
,
841 IB_QPT_RAW_PACKET
= 8,
845 /* Reserve a range for qp types internal to the low level driver.
846 * These qp types will not be visible at the IB core layer, so the
847 * IB_QPT_MAX usages should not be affected in the core layer
849 IB_QPT_RESERVED1
= 0x1000,
861 enum ib_qp_create_flags
{
862 IB_QP_CREATE_IPOIB_UD_LSO
= 1 << 0,
863 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK
= 1 << 1,
864 IB_QP_CREATE_NETIF_QP
= 1 << 5,
865 IB_QP_CREATE_SIGNATURE_EN
= 1 << 6,
866 IB_QP_CREATE_USE_GFP_NOIO
= 1 << 7,
867 /* reserve bits 26-31 for low level drivers' internal use */
868 IB_QP_CREATE_RESERVED_START
= 1 << 26,
869 IB_QP_CREATE_RESERVED_END
= 1 << 31,
874 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
875 * callback to destroy the passed in QP.
878 struct ib_qp_init_attr
{
879 void (*event_handler
)(struct ib_event
*, void *);
881 struct ib_cq
*send_cq
;
882 struct ib_cq
*recv_cq
;
884 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
885 struct ib_qp_cap cap
;
886 enum ib_sig_type sq_sig_type
;
887 enum ib_qp_type qp_type
;
888 enum ib_qp_create_flags create_flags
;
889 u8 port_num
; /* special QP types only */
892 struct ib_qp_open_attr
{
893 void (*event_handler
)(struct ib_event
*, void *);
896 enum ib_qp_type qp_type
;
899 enum ib_rnr_timeout
{
900 IB_RNR_TIMER_655_36
= 0,
901 IB_RNR_TIMER_000_01
= 1,
902 IB_RNR_TIMER_000_02
= 2,
903 IB_RNR_TIMER_000_03
= 3,
904 IB_RNR_TIMER_000_04
= 4,
905 IB_RNR_TIMER_000_06
= 5,
906 IB_RNR_TIMER_000_08
= 6,
907 IB_RNR_TIMER_000_12
= 7,
908 IB_RNR_TIMER_000_16
= 8,
909 IB_RNR_TIMER_000_24
= 9,
910 IB_RNR_TIMER_000_32
= 10,
911 IB_RNR_TIMER_000_48
= 11,
912 IB_RNR_TIMER_000_64
= 12,
913 IB_RNR_TIMER_000_96
= 13,
914 IB_RNR_TIMER_001_28
= 14,
915 IB_RNR_TIMER_001_92
= 15,
916 IB_RNR_TIMER_002_56
= 16,
917 IB_RNR_TIMER_003_84
= 17,
918 IB_RNR_TIMER_005_12
= 18,
919 IB_RNR_TIMER_007_68
= 19,
920 IB_RNR_TIMER_010_24
= 20,
921 IB_RNR_TIMER_015_36
= 21,
922 IB_RNR_TIMER_020_48
= 22,
923 IB_RNR_TIMER_030_72
= 23,
924 IB_RNR_TIMER_040_96
= 24,
925 IB_RNR_TIMER_061_44
= 25,
926 IB_RNR_TIMER_081_92
= 26,
927 IB_RNR_TIMER_122_88
= 27,
928 IB_RNR_TIMER_163_84
= 28,
929 IB_RNR_TIMER_245_76
= 29,
930 IB_RNR_TIMER_327_68
= 30,
931 IB_RNR_TIMER_491_52
= 31
934 enum ib_qp_attr_mask
{
936 IB_QP_CUR_STATE
= (1<<1),
937 IB_QP_EN_SQD_ASYNC_NOTIFY
= (1<<2),
938 IB_QP_ACCESS_FLAGS
= (1<<3),
939 IB_QP_PKEY_INDEX
= (1<<4),
943 IB_QP_PATH_MTU
= (1<<8),
944 IB_QP_TIMEOUT
= (1<<9),
945 IB_QP_RETRY_CNT
= (1<<10),
946 IB_QP_RNR_RETRY
= (1<<11),
947 IB_QP_RQ_PSN
= (1<<12),
948 IB_QP_MAX_QP_RD_ATOMIC
= (1<<13),
949 IB_QP_ALT_PATH
= (1<<14),
950 IB_QP_MIN_RNR_TIMER
= (1<<15),
951 IB_QP_SQ_PSN
= (1<<16),
952 IB_QP_MAX_DEST_RD_ATOMIC
= (1<<17),
953 IB_QP_PATH_MIG_STATE
= (1<<18),
955 IB_QP_DEST_QPN
= (1<<20),
956 IB_QP_SMAC
= (1<<21),
957 IB_QP_ALT_SMAC
= (1<<22),
959 IB_QP_ALT_VID
= (1<<24),
984 enum ib_qp_state qp_state
;
985 enum ib_qp_state cur_qp_state
;
986 enum ib_mtu path_mtu
;
987 enum ib_mig_state path_mig_state
;
993 struct ib_qp_cap cap
;
994 struct ib_ah_attr ah_attr
;
995 struct ib_ah_attr alt_ah_attr
;
998 u8 en_sqd_async_notify
;
1001 u8 max_dest_rd_atomic
;
1010 u8 alt_smac
[ETH_ALEN
];
1017 IB_WR_RDMA_WRITE_WITH_IMM
,
1019 IB_WR_SEND_WITH_IMM
,
1021 IB_WR_ATOMIC_CMP_AND_SWP
,
1022 IB_WR_ATOMIC_FETCH_AND_ADD
,
1024 IB_WR_SEND_WITH_INV
,
1025 IB_WR_RDMA_READ_WITH_INV
,
1028 IB_WR_MASKED_ATOMIC_CMP_AND_SWP
,
1029 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD
,
1032 /* reserve values for low level drivers' internal use.
1033 * These values will not be used at all in the ib core layer.
1035 IB_WR_RESERVED1
= 0xf0,
1047 enum ib_send_flags
{
1049 IB_SEND_SIGNALED
= (1<<1),
1050 IB_SEND_SOLICITED
= (1<<2),
1051 IB_SEND_INLINE
= (1<<3),
1052 IB_SEND_IP_CSUM
= (1<<4),
1054 /* reserve bits 26-31 for low level drivers' internal use */
1055 IB_SEND_RESERVED_START
= (1 << 26),
1056 IB_SEND_RESERVED_END
= (1 << 31),
1065 struct ib_fast_reg_page_list
{
1066 struct ib_device
*device
;
1068 unsigned int max_page_list_len
;
1072 * struct ib_mw_bind_info - Parameters for a memory window bind operation.
1073 * @mr: A memory region to bind the memory window to.
1074 * @addr: The address where the memory window should begin.
1075 * @length: The length of the memory window, in bytes.
1076 * @mw_access_flags: Access flags from enum ib_access_flags for the window.
1078 * This struct contains the shared parameters for type 1 and type 2
1079 * memory window bind operations.
1081 struct ib_mw_bind_info
{
1085 int mw_access_flags
;
1089 struct ib_send_wr
*next
;
1091 struct ib_sge
*sg_list
;
1093 enum ib_wr_opcode opcode
;
1097 u32 invalidate_rkey
;
1108 u64 compare_add_mask
;
1119 u16 pkey_index
; /* valid for GSI only */
1120 u8 port_num
; /* valid for DR SMPs on switch only */
1124 struct ib_fast_reg_page_list
*page_list
;
1125 unsigned int page_shift
;
1126 unsigned int page_list_len
;
1133 /* The new rkey for the memory window. */
1135 struct ib_mw_bind_info bind_info
;
1138 struct ib_sig_attrs
*sig_attrs
;
1139 struct ib_mr
*sig_mr
;
1141 struct ib_sge
*prot
;
1144 u32 xrc_remote_srq_num
; /* XRC TGT QPs only */
1148 struct ib_recv_wr
*next
;
1150 struct ib_sge
*sg_list
;
1154 enum ib_access_flags
{
1155 IB_ACCESS_LOCAL_WRITE
= 1,
1156 IB_ACCESS_REMOTE_WRITE
= (1<<1),
1157 IB_ACCESS_REMOTE_READ
= (1<<2),
1158 IB_ACCESS_REMOTE_ATOMIC
= (1<<3),
1159 IB_ACCESS_MW_BIND
= (1<<4),
1160 IB_ZERO_BASED
= (1<<5),
1161 IB_ACCESS_ON_DEMAND
= (1<<6),
1164 struct ib_phys_buf
{
1171 u64 device_virt_addr
;
1173 int mr_access_flags
;
1178 enum ib_mr_rereg_flags
{
1179 IB_MR_REREG_TRANS
= 1,
1180 IB_MR_REREG_PD
= (1<<1),
1181 IB_MR_REREG_ACCESS
= (1<<2),
1182 IB_MR_REREG_SUPPORTED
= ((IB_MR_REREG_ACCESS
<< 1) - 1)
1186 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation.
1187 * @wr_id: Work request id.
1188 * @send_flags: Flags from ib_send_flags enum.
1189 * @bind_info: More parameters of the bind operation.
1194 struct ib_mw_bind_info bind_info
;
1197 struct ib_fmr_attr
{
1205 struct ib_ucontext
{
1206 struct ib_device
*device
;
1207 struct list_head pd_list
;
1208 struct list_head mr_list
;
1209 struct list_head mw_list
;
1210 struct list_head cq_list
;
1211 struct list_head qp_list
;
1212 struct list_head srq_list
;
1213 struct list_head ah_list
;
1214 struct list_head xrcd_list
;
1215 struct list_head rule_list
;
1219 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1220 struct rb_root umem_tree
;
1222 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1223 * mmu notifiers registration.
1225 struct rw_semaphore umem_rwsem
;
1226 void (*invalidate_range
)(struct ib_umem
*umem
,
1227 unsigned long start
, unsigned long end
);
1229 struct mmu_notifier mn
;
1230 atomic_t notifier_count
;
1231 /* A list of umems that don't have private mmu notifier counters yet. */
1232 struct list_head no_private_counters
;
1238 u64 user_handle
; /* handle given to us by userspace */
1239 struct ib_ucontext
*context
; /* associated user context */
1240 void *object
; /* containing object */
1241 struct list_head list
; /* link to context's list */
1242 int id
; /* index into kernel idr */
1244 struct rw_semaphore mutex
; /* protects .live */
1249 const void __user
*inbuf
;
1250 void __user
*outbuf
;
1256 struct ib_device
*device
;
1257 struct ib_uobject
*uobject
;
1258 atomic_t usecnt
; /* count all resources */
1262 struct ib_device
*device
;
1263 atomic_t usecnt
; /* count all exposed resources */
1264 struct inode
*inode
;
1266 struct mutex tgt_qp_mutex
;
1267 struct list_head tgt_qp_list
;
1271 struct ib_device
*device
;
1273 struct ib_uobject
*uobject
;
1276 typedef void (*ib_comp_handler
)(struct ib_cq
*cq
, void *cq_context
);
1279 struct ib_device
*device
;
1280 struct ib_uobject
*uobject
;
1281 ib_comp_handler comp_handler
;
1282 void (*event_handler
)(struct ib_event
*, void *);
1285 atomic_t usecnt
; /* count number of work queues */
1289 struct ib_device
*device
;
1291 struct ib_uobject
*uobject
;
1292 void (*event_handler
)(struct ib_event
*, void *);
1294 enum ib_srq_type srq_type
;
1299 struct ib_xrcd
*xrcd
;
1307 struct ib_device
*device
;
1309 struct ib_cq
*send_cq
;
1310 struct ib_cq
*recv_cq
;
1312 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
1313 struct list_head xrcd_list
;
1314 /* count times opened, mcast attaches, flow attaches */
1316 struct list_head open_list
;
1317 struct ib_qp
*real_qp
;
1318 struct ib_uobject
*uobject
;
1319 void (*event_handler
)(struct ib_event
*, void *);
1322 enum ib_qp_type qp_type
;
1326 struct ib_device
*device
;
1328 struct ib_uobject
*uobject
;
1331 atomic_t usecnt
; /* count number of MWs */
1335 struct ib_device
*device
;
1337 struct ib_uobject
*uobject
;
1339 enum ib_mw_type type
;
1343 struct ib_device
*device
;
1345 struct list_head list
;
1350 /* Supported steering options */
1351 enum ib_flow_attr_type
{
1352 /* steering according to rule specifications */
1353 IB_FLOW_ATTR_NORMAL
= 0x0,
1354 /* default unicast and multicast rule -
1355 * receive all Eth traffic which isn't steered to any QP
1357 IB_FLOW_ATTR_ALL_DEFAULT
= 0x1,
1358 /* default multicast rule -
1359 * receive all Eth multicast traffic which isn't steered to any QP
1361 IB_FLOW_ATTR_MC_DEFAULT
= 0x2,
1362 /* sniffer rule - receive all port traffic */
1363 IB_FLOW_ATTR_SNIFFER
= 0x3
1366 /* Supported steering header types */
1367 enum ib_flow_spec_type
{
1369 IB_FLOW_SPEC_ETH
= 0x20,
1370 IB_FLOW_SPEC_IB
= 0x22,
1372 IB_FLOW_SPEC_IPV4
= 0x30,
1374 IB_FLOW_SPEC_TCP
= 0x40,
1375 IB_FLOW_SPEC_UDP
= 0x41
1377 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1378 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1380 /* Flow steering rule priority is set according to it's domain.
1381 * Lower domain value means higher priority.
1383 enum ib_flow_domain
{
1384 IB_FLOW_DOMAIN_USER
,
1385 IB_FLOW_DOMAIN_ETHTOOL
,
1388 IB_FLOW_DOMAIN_NUM
/* Must be last */
1391 struct ib_flow_eth_filter
{
1398 struct ib_flow_spec_eth
{
1399 enum ib_flow_spec_type type
;
1401 struct ib_flow_eth_filter val
;
1402 struct ib_flow_eth_filter mask
;
1405 struct ib_flow_ib_filter
{
1410 struct ib_flow_spec_ib
{
1411 enum ib_flow_spec_type type
;
1413 struct ib_flow_ib_filter val
;
1414 struct ib_flow_ib_filter mask
;
1417 struct ib_flow_ipv4_filter
{
1422 struct ib_flow_spec_ipv4
{
1423 enum ib_flow_spec_type type
;
1425 struct ib_flow_ipv4_filter val
;
1426 struct ib_flow_ipv4_filter mask
;
1429 struct ib_flow_tcp_udp_filter
{
1434 struct ib_flow_spec_tcp_udp
{
1435 enum ib_flow_spec_type type
;
1437 struct ib_flow_tcp_udp_filter val
;
1438 struct ib_flow_tcp_udp_filter mask
;
1441 union ib_flow_spec
{
1443 enum ib_flow_spec_type type
;
1446 struct ib_flow_spec_eth eth
;
1447 struct ib_flow_spec_ib ib
;
1448 struct ib_flow_spec_ipv4 ipv4
;
1449 struct ib_flow_spec_tcp_udp tcp_udp
;
1452 struct ib_flow_attr
{
1453 enum ib_flow_attr_type type
;
1459 /* Following are the optional layers according to user request
1460 * struct ib_flow_spec_xxx
1461 * struct ib_flow_spec_yyy
1467 struct ib_uobject
*uobject
;
1473 enum ib_process_mad_flags
{
1474 IB_MAD_IGNORE_MKEY
= 1,
1475 IB_MAD_IGNORE_BKEY
= 2,
1476 IB_MAD_IGNORE_ALL
= IB_MAD_IGNORE_MKEY
| IB_MAD_IGNORE_BKEY
1479 enum ib_mad_result
{
1480 IB_MAD_RESULT_FAILURE
= 0, /* (!SUCCESS is the important flag) */
1481 IB_MAD_RESULT_SUCCESS
= 1 << 0, /* MAD was successfully processed */
1482 IB_MAD_RESULT_REPLY
= 1 << 1, /* Reply packet needs to be sent */
1483 IB_MAD_RESULT_CONSUMED
= 1 << 2 /* Packet consumed: stop processing */
1486 #define IB_DEVICE_NAME_MAX 64
1490 struct ib_event_handler event_handler
;
1491 struct ib_pkey_cache
**pkey_cache
;
1492 struct ib_gid_cache
**gid_cache
;
1496 struct ib_dma_mapping_ops
{
1497 int (*mapping_error
)(struct ib_device
*dev
,
1499 u64 (*map_single
)(struct ib_device
*dev
,
1500 void *ptr
, size_t size
,
1501 enum dma_data_direction direction
);
1502 void (*unmap_single
)(struct ib_device
*dev
,
1503 u64 addr
, size_t size
,
1504 enum dma_data_direction direction
);
1505 u64 (*map_page
)(struct ib_device
*dev
,
1506 struct page
*page
, unsigned long offset
,
1508 enum dma_data_direction direction
);
1509 void (*unmap_page
)(struct ib_device
*dev
,
1510 u64 addr
, size_t size
,
1511 enum dma_data_direction direction
);
1512 int (*map_sg
)(struct ib_device
*dev
,
1513 struct scatterlist
*sg
, int nents
,
1514 enum dma_data_direction direction
);
1515 void (*unmap_sg
)(struct ib_device
*dev
,
1516 struct scatterlist
*sg
, int nents
,
1517 enum dma_data_direction direction
);
1518 void (*sync_single_for_cpu
)(struct ib_device
*dev
,
1521 enum dma_data_direction dir
);
1522 void (*sync_single_for_device
)(struct ib_device
*dev
,
1525 enum dma_data_direction dir
);
1526 void *(*alloc_coherent
)(struct ib_device
*dev
,
1530 void (*free_coherent
)(struct ib_device
*dev
,
1531 size_t size
, void *cpu_addr
,
1537 struct ib_port_immutable
{
1545 struct device
*dma_device
;
1547 char name
[IB_DEVICE_NAME_MAX
];
1549 struct list_head event_handler_list
;
1550 spinlock_t event_handler_lock
;
1552 spinlock_t client_data_lock
;
1553 struct list_head core_list
;
1554 /* Access to the client_data_list is protected by the client_data_lock
1555 * spinlock and the lists_rwsem read-write semaphore */
1556 struct list_head client_data_list
;
1558 struct ib_cache cache
;
1560 * port_immutable is indexed by port number
1562 struct ib_port_immutable
*port_immutable
;
1564 int num_comp_vectors
;
1566 struct iw_cm_verbs
*iwcm
;
1568 int (*get_protocol_stats
)(struct ib_device
*device
,
1569 union rdma_protocol_stats
*stats
);
1570 int (*query_device
)(struct ib_device
*device
,
1571 struct ib_device_attr
*device_attr
,
1572 struct ib_udata
*udata
);
1573 int (*query_port
)(struct ib_device
*device
,
1575 struct ib_port_attr
*port_attr
);
1576 enum rdma_link_layer (*get_link_layer
)(struct ib_device
*device
,
1578 int (*query_gid
)(struct ib_device
*device
,
1579 u8 port_num
, int index
,
1581 int (*query_pkey
)(struct ib_device
*device
,
1582 u8 port_num
, u16 index
, u16
*pkey
);
1583 int (*modify_device
)(struct ib_device
*device
,
1584 int device_modify_mask
,
1585 struct ib_device_modify
*device_modify
);
1586 int (*modify_port
)(struct ib_device
*device
,
1587 u8 port_num
, int port_modify_mask
,
1588 struct ib_port_modify
*port_modify
);
1589 struct ib_ucontext
* (*alloc_ucontext
)(struct ib_device
*device
,
1590 struct ib_udata
*udata
);
1591 int (*dealloc_ucontext
)(struct ib_ucontext
*context
);
1592 int (*mmap
)(struct ib_ucontext
*context
,
1593 struct vm_area_struct
*vma
);
1594 struct ib_pd
* (*alloc_pd
)(struct ib_device
*device
,
1595 struct ib_ucontext
*context
,
1596 struct ib_udata
*udata
);
1597 int (*dealloc_pd
)(struct ib_pd
*pd
);
1598 struct ib_ah
* (*create_ah
)(struct ib_pd
*pd
,
1599 struct ib_ah_attr
*ah_attr
);
1600 int (*modify_ah
)(struct ib_ah
*ah
,
1601 struct ib_ah_attr
*ah_attr
);
1602 int (*query_ah
)(struct ib_ah
*ah
,
1603 struct ib_ah_attr
*ah_attr
);
1604 int (*destroy_ah
)(struct ib_ah
*ah
);
1605 struct ib_srq
* (*create_srq
)(struct ib_pd
*pd
,
1606 struct ib_srq_init_attr
*srq_init_attr
,
1607 struct ib_udata
*udata
);
1608 int (*modify_srq
)(struct ib_srq
*srq
,
1609 struct ib_srq_attr
*srq_attr
,
1610 enum ib_srq_attr_mask srq_attr_mask
,
1611 struct ib_udata
*udata
);
1612 int (*query_srq
)(struct ib_srq
*srq
,
1613 struct ib_srq_attr
*srq_attr
);
1614 int (*destroy_srq
)(struct ib_srq
*srq
);
1615 int (*post_srq_recv
)(struct ib_srq
*srq
,
1616 struct ib_recv_wr
*recv_wr
,
1617 struct ib_recv_wr
**bad_recv_wr
);
1618 struct ib_qp
* (*create_qp
)(struct ib_pd
*pd
,
1619 struct ib_qp_init_attr
*qp_init_attr
,
1620 struct ib_udata
*udata
);
1621 int (*modify_qp
)(struct ib_qp
*qp
,
1622 struct ib_qp_attr
*qp_attr
,
1624 struct ib_udata
*udata
);
1625 int (*query_qp
)(struct ib_qp
*qp
,
1626 struct ib_qp_attr
*qp_attr
,
1628 struct ib_qp_init_attr
*qp_init_attr
);
1629 int (*destroy_qp
)(struct ib_qp
*qp
);
1630 int (*post_send
)(struct ib_qp
*qp
,
1631 struct ib_send_wr
*send_wr
,
1632 struct ib_send_wr
**bad_send_wr
);
1633 int (*post_recv
)(struct ib_qp
*qp
,
1634 struct ib_recv_wr
*recv_wr
,
1635 struct ib_recv_wr
**bad_recv_wr
);
1636 struct ib_cq
* (*create_cq
)(struct ib_device
*device
,
1637 const struct ib_cq_init_attr
*attr
,
1638 struct ib_ucontext
*context
,
1639 struct ib_udata
*udata
);
1640 int (*modify_cq
)(struct ib_cq
*cq
, u16 cq_count
,
1642 int (*destroy_cq
)(struct ib_cq
*cq
);
1643 int (*resize_cq
)(struct ib_cq
*cq
, int cqe
,
1644 struct ib_udata
*udata
);
1645 int (*poll_cq
)(struct ib_cq
*cq
, int num_entries
,
1647 int (*peek_cq
)(struct ib_cq
*cq
, int wc_cnt
);
1648 int (*req_notify_cq
)(struct ib_cq
*cq
,
1649 enum ib_cq_notify_flags flags
);
1650 int (*req_ncomp_notif
)(struct ib_cq
*cq
,
1652 struct ib_mr
* (*get_dma_mr
)(struct ib_pd
*pd
,
1653 int mr_access_flags
);
1654 struct ib_mr
* (*reg_phys_mr
)(struct ib_pd
*pd
,
1655 struct ib_phys_buf
*phys_buf_array
,
1657 int mr_access_flags
,
1659 struct ib_mr
* (*reg_user_mr
)(struct ib_pd
*pd
,
1660 u64 start
, u64 length
,
1662 int mr_access_flags
,
1663 struct ib_udata
*udata
);
1664 int (*rereg_user_mr
)(struct ib_mr
*mr
,
1666 u64 start
, u64 length
,
1668 int mr_access_flags
,
1670 struct ib_udata
*udata
);
1671 int (*query_mr
)(struct ib_mr
*mr
,
1672 struct ib_mr_attr
*mr_attr
);
1673 int (*dereg_mr
)(struct ib_mr
*mr
);
1674 int (*destroy_mr
)(struct ib_mr
*mr
);
1675 struct ib_mr
* (*create_mr
)(struct ib_pd
*pd
,
1676 struct ib_mr_init_attr
*mr_init_attr
);
1677 struct ib_mr
* (*alloc_fast_reg_mr
)(struct ib_pd
*pd
,
1678 int max_page_list_len
);
1679 struct ib_fast_reg_page_list
* (*alloc_fast_reg_page_list
)(struct ib_device
*device
,
1681 void (*free_fast_reg_page_list
)(struct ib_fast_reg_page_list
*page_list
);
1682 int (*rereg_phys_mr
)(struct ib_mr
*mr
,
1685 struct ib_phys_buf
*phys_buf_array
,
1687 int mr_access_flags
,
1689 struct ib_mw
* (*alloc_mw
)(struct ib_pd
*pd
,
1690 enum ib_mw_type type
);
1691 int (*bind_mw
)(struct ib_qp
*qp
,
1693 struct ib_mw_bind
*mw_bind
);
1694 int (*dealloc_mw
)(struct ib_mw
*mw
);
1695 struct ib_fmr
* (*alloc_fmr
)(struct ib_pd
*pd
,
1696 int mr_access_flags
,
1697 struct ib_fmr_attr
*fmr_attr
);
1698 int (*map_phys_fmr
)(struct ib_fmr
*fmr
,
1699 u64
*page_list
, int list_len
,
1701 int (*unmap_fmr
)(struct list_head
*fmr_list
);
1702 int (*dealloc_fmr
)(struct ib_fmr
*fmr
);
1703 int (*attach_mcast
)(struct ib_qp
*qp
,
1706 int (*detach_mcast
)(struct ib_qp
*qp
,
1709 int (*process_mad
)(struct ib_device
*device
,
1710 int process_mad_flags
,
1712 const struct ib_wc
*in_wc
,
1713 const struct ib_grh
*in_grh
,
1714 const struct ib_mad_hdr
*in_mad
,
1716 struct ib_mad_hdr
*out_mad
,
1717 size_t *out_mad_size
,
1718 u16
*out_mad_pkey_index
);
1719 struct ib_xrcd
* (*alloc_xrcd
)(struct ib_device
*device
,
1720 struct ib_ucontext
*ucontext
,
1721 struct ib_udata
*udata
);
1722 int (*dealloc_xrcd
)(struct ib_xrcd
*xrcd
);
1723 struct ib_flow
* (*create_flow
)(struct ib_qp
*qp
,
1727 int (*destroy_flow
)(struct ib_flow
*flow_id
);
1728 int (*check_mr_status
)(struct ib_mr
*mr
, u32 check_mask
,
1729 struct ib_mr_status
*mr_status
);
1731 struct ib_dma_mapping_ops
*dma_ops
;
1733 struct module
*owner
;
1735 struct kobject
*ports_parent
;
1736 struct list_head port_list
;
1739 IB_DEV_UNINITIALIZED
,
1745 u64 uverbs_cmd_mask
;
1746 u64 uverbs_ex_cmd_mask
;
1756 * The following mandatory functions are used only at device
1757 * registration. Keep functions such as these at the end of this
1758 * structure to avoid cache line misses when accessing struct ib_device
1761 int (*get_port_immutable
)(struct ib_device
*, u8
, struct ib_port_immutable
*);
1766 void (*add
) (struct ib_device
*);
1767 void (*remove
)(struct ib_device
*, void *client_data
);
1769 /* Returns the net_dev belonging to this ib_client and matching the
1771 * @dev: An RDMA device that the net_dev use for communication.
1772 * @port: A physical port number on the RDMA device.
1773 * @pkey: P_Key that the net_dev uses if applicable.
1774 * @gid: A GID that the net_dev uses to communicate.
1775 * @addr: An IP address the net_dev is configured with.
1776 * @client_data: The device's client data set by ib_set_client_data().
1778 * An ib_client that implements a net_dev on top of RDMA devices
1779 * (such as IP over IB) should implement this callback, allowing the
1780 * rdma_cm module to find the right net_dev for a given request.
1782 * The caller is responsible for calling dev_put on the returned
1784 struct net_device
*(*get_net_dev_by_params
)(
1785 struct ib_device
*dev
,
1788 const union ib_gid
*gid
,
1789 const struct sockaddr
*addr
,
1791 struct list_head list
;
1794 struct ib_device
*ib_alloc_device(size_t size
);
1795 void ib_dealloc_device(struct ib_device
*device
);
1797 int ib_register_device(struct ib_device
*device
,
1798 int (*port_callback
)(struct ib_device
*,
1799 u8
, struct kobject
*));
1800 void ib_unregister_device(struct ib_device
*device
);
1802 int ib_register_client (struct ib_client
*client
);
1803 void ib_unregister_client(struct ib_client
*client
);
1805 void *ib_get_client_data(struct ib_device
*device
, struct ib_client
*client
);
1806 void ib_set_client_data(struct ib_device
*device
, struct ib_client
*client
,
1809 static inline int ib_copy_from_udata(void *dest
, struct ib_udata
*udata
, size_t len
)
1811 return copy_from_user(dest
, udata
->inbuf
, len
) ? -EFAULT
: 0;
1814 static inline int ib_copy_to_udata(struct ib_udata
*udata
, void *src
, size_t len
)
1816 return copy_to_user(udata
->outbuf
, src
, len
) ? -EFAULT
: 0;
1820 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1821 * contains all required attributes and no attributes not allowed for
1822 * the given QP state transition.
1823 * @cur_state: Current QP state
1824 * @next_state: Next QP state
1826 * @mask: Mask of supplied QP attributes
1827 * @ll : link layer of port
1829 * This function is a helper function that a low-level driver's
1830 * modify_qp method can use to validate the consumer's input. It
1831 * checks that cur_state and next_state are valid QP states, that a
1832 * transition from cur_state to next_state is allowed by the IB spec,
1833 * and that the attribute mask supplied is allowed for the transition.
1835 int ib_modify_qp_is_ok(enum ib_qp_state cur_state
, enum ib_qp_state next_state
,
1836 enum ib_qp_type type
, enum ib_qp_attr_mask mask
,
1837 enum rdma_link_layer ll
);
1839 int ib_register_event_handler (struct ib_event_handler
*event_handler
);
1840 int ib_unregister_event_handler(struct ib_event_handler
*event_handler
);
1841 void ib_dispatch_event(struct ib_event
*event
);
1843 int ib_query_device(struct ib_device
*device
,
1844 struct ib_device_attr
*device_attr
);
1846 int ib_query_port(struct ib_device
*device
,
1847 u8 port_num
, struct ib_port_attr
*port_attr
);
1849 enum rdma_link_layer
rdma_port_get_link_layer(struct ib_device
*device
,
1853 * rdma_cap_ib_switch - Check if the device is IB switch
1854 * @device: Device to check
1856 * Device driver is responsible for setting is_switch bit on
1857 * in ib_device structure at init time.
1859 * Return: true if the device is IB switch.
1861 static inline bool rdma_cap_ib_switch(const struct ib_device
*device
)
1863 return device
->is_switch
;
1867 * rdma_start_port - Return the first valid port number for the device
1870 * @device: Device to be checked
1872 * Return start port number
1874 static inline u8
rdma_start_port(const struct ib_device
*device
)
1876 return rdma_cap_ib_switch(device
) ? 0 : 1;
1880 * rdma_end_port - Return the last valid port number for the device
1883 * @device: Device to be checked
1885 * Return last port number
1887 static inline u8
rdma_end_port(const struct ib_device
*device
)
1889 return rdma_cap_ib_switch(device
) ? 0 : device
->phys_port_cnt
;
1892 static inline bool rdma_protocol_ib(const struct ib_device
*device
, u8 port_num
)
1894 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IB
;
1897 static inline bool rdma_protocol_roce(const struct ib_device
*device
, u8 port_num
)
1899 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_ROCE
;
1902 static inline bool rdma_protocol_iwarp(const struct ib_device
*device
, u8 port_num
)
1904 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IWARP
;
1907 static inline bool rdma_ib_or_roce(const struct ib_device
*device
, u8 port_num
)
1909 return device
->port_immutable
[port_num
].core_cap_flags
&
1910 (RDMA_CORE_CAP_PROT_IB
| RDMA_CORE_CAP_PROT_ROCE
);
1914 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
1915 * Management Datagrams.
1916 * @device: Device to check
1917 * @port_num: Port number to check
1919 * Management Datagrams (MAD) are a required part of the InfiniBand
1920 * specification and are supported on all InfiniBand devices. A slightly
1921 * extended version are also supported on OPA interfaces.
1923 * Return: true if the port supports sending/receiving of MAD packets.
1925 static inline bool rdma_cap_ib_mad(const struct ib_device
*device
, u8 port_num
)
1927 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_MAD
;
1931 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
1932 * Management Datagrams.
1933 * @device: Device to check
1934 * @port_num: Port number to check
1936 * Intel OmniPath devices extend and/or replace the InfiniBand Management
1937 * datagrams with their own versions. These OPA MADs share many but not all of
1938 * the characteristics of InfiniBand MADs.
1940 * OPA MADs differ in the following ways:
1942 * 1) MADs are variable size up to 2K
1943 * IBTA defined MADs remain fixed at 256 bytes
1944 * 2) OPA SMPs must carry valid PKeys
1945 * 3) OPA SMP packets are a different format
1947 * Return: true if the port supports OPA MAD packet formats.
1949 static inline bool rdma_cap_opa_mad(struct ib_device
*device
, u8 port_num
)
1951 return (device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_OPA_MAD
)
1952 == RDMA_CORE_CAP_OPA_MAD
;
1956 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
1957 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
1958 * @device: Device to check
1959 * @port_num: Port number to check
1961 * Each InfiniBand node is required to provide a Subnet Management Agent
1962 * that the subnet manager can access. Prior to the fabric being fully
1963 * configured by the subnet manager, the SMA is accessed via a well known
1964 * interface called the Subnet Management Interface (SMI). This interface
1965 * uses directed route packets to communicate with the SM to get around the
1966 * chicken and egg problem of the SM needing to know what's on the fabric
1967 * in order to configure the fabric, and needing to configure the fabric in
1968 * order to send packets to the devices on the fabric. These directed
1969 * route packets do not need the fabric fully configured in order to reach
1970 * their destination. The SMI is the only method allowed to send
1971 * directed route packets on an InfiniBand fabric.
1973 * Return: true if the port provides an SMI.
1975 static inline bool rdma_cap_ib_smi(const struct ib_device
*device
, u8 port_num
)
1977 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SMI
;
1981 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
1982 * Communication Manager.
1983 * @device: Device to check
1984 * @port_num: Port number to check
1986 * The InfiniBand Communication Manager is one of many pre-defined General
1987 * Service Agents (GSA) that are accessed via the General Service
1988 * Interface (GSI). It's role is to facilitate establishment of connections
1989 * between nodes as well as other management related tasks for established
1992 * Return: true if the port supports an IB CM (this does not guarantee that
1993 * a CM is actually running however).
1995 static inline bool rdma_cap_ib_cm(const struct ib_device
*device
, u8 port_num
)
1997 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_CM
;
2001 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2002 * Communication Manager.
2003 * @device: Device to check
2004 * @port_num: Port number to check
2006 * Similar to above, but specific to iWARP connections which have a different
2007 * managment protocol than InfiniBand.
2009 * Return: true if the port supports an iWARP CM (this does not guarantee that
2010 * a CM is actually running however).
2012 static inline bool rdma_cap_iw_cm(const struct ib_device
*device
, u8 port_num
)
2014 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IW_CM
;
2018 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2019 * Subnet Administration.
2020 * @device: Device to check
2021 * @port_num: Port number to check
2023 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2024 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2025 * fabrics, devices should resolve routes to other hosts by contacting the
2026 * SA to query the proper route.
2028 * Return: true if the port should act as a client to the fabric Subnet
2029 * Administration interface. This does not imply that the SA service is
2032 static inline bool rdma_cap_ib_sa(const struct ib_device
*device
, u8 port_num
)
2034 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SA
;
2038 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2040 * @device: Device to check
2041 * @port_num: Port number to check
2043 * InfiniBand multicast registration is more complex than normal IPv4 or
2044 * IPv6 multicast registration. Each Host Channel Adapter must register
2045 * with the Subnet Manager when it wishes to join a multicast group. It
2046 * should do so only once regardless of how many queue pairs it subscribes
2047 * to this group. And it should leave the group only after all queue pairs
2048 * attached to the group have been detached.
2050 * Return: true if the port must undertake the additional adminstrative
2051 * overhead of registering/unregistering with the SM and tracking of the
2052 * total number of queue pairs attached to the multicast group.
2054 static inline bool rdma_cap_ib_mcast(const struct ib_device
*device
, u8 port_num
)
2056 return rdma_cap_ib_sa(device
, port_num
);
2060 * rdma_cap_af_ib - Check if the port of device has the capability
2061 * Native Infiniband Address.
2062 * @device: Device to check
2063 * @port_num: Port number to check
2065 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2066 * GID. RoCE uses a different mechanism, but still generates a GID via
2067 * a prescribed mechanism and port specific data.
2069 * Return: true if the port uses a GID address to identify devices on the
2072 static inline bool rdma_cap_af_ib(const struct ib_device
*device
, u8 port_num
)
2074 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_AF_IB
;
2078 * rdma_cap_eth_ah - Check if the port of device has the capability
2079 * Ethernet Address Handle.
2080 * @device: Device to check
2081 * @port_num: Port number to check
2083 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2084 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2085 * port. Normally, packet headers are generated by the sending host
2086 * adapter, but when sending connectionless datagrams, we must manually
2087 * inject the proper headers for the fabric we are communicating over.
2089 * Return: true if we are running as a RoCE port and must force the
2090 * addition of a Global Route Header built from our Ethernet Address
2091 * Handle into our header list for connectionless packets.
2093 static inline bool rdma_cap_eth_ah(const struct ib_device
*device
, u8 port_num
)
2095 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_ETH_AH
;
2099 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2102 * @port_num: Port number
2104 * This MAD size includes the MAD headers and MAD payload. No other headers
2107 * Return the max MAD size required by the Port. Will return 0 if the port
2108 * does not support MADs
2110 static inline size_t rdma_max_mad_size(const struct ib_device
*device
, u8 port_num
)
2112 return device
->port_immutable
[port_num
].max_mad_size
;
2115 int ib_query_gid(struct ib_device
*device
,
2116 u8 port_num
, int index
, union ib_gid
*gid
);
2118 int ib_query_pkey(struct ib_device
*device
,
2119 u8 port_num
, u16 index
, u16
*pkey
);
2121 int ib_modify_device(struct ib_device
*device
,
2122 int device_modify_mask
,
2123 struct ib_device_modify
*device_modify
);
2125 int ib_modify_port(struct ib_device
*device
,
2126 u8 port_num
, int port_modify_mask
,
2127 struct ib_port_modify
*port_modify
);
2129 int ib_find_gid(struct ib_device
*device
, union ib_gid
*gid
,
2130 u8
*port_num
, u16
*index
);
2132 int ib_find_pkey(struct ib_device
*device
,
2133 u8 port_num
, u16 pkey
, u16
*index
);
2136 * ib_alloc_pd - Allocates an unused protection domain.
2137 * @device: The device on which to allocate the protection domain.
2139 * A protection domain object provides an association between QPs, shared
2140 * receive queues, address handles, memory regions, and memory windows.
2142 struct ib_pd
*ib_alloc_pd(struct ib_device
*device
);
2145 * ib_dealloc_pd - Deallocates a protection domain.
2146 * @pd: The protection domain to deallocate.
2148 int ib_dealloc_pd(struct ib_pd
*pd
);
2151 * ib_create_ah - Creates an address handle for the given address vector.
2152 * @pd: The protection domain associated with the address handle.
2153 * @ah_attr: The attributes of the address vector.
2155 * The address handle is used to reference a local or global destination
2156 * in all UD QP post sends.
2158 struct ib_ah
*ib_create_ah(struct ib_pd
*pd
, struct ib_ah_attr
*ah_attr
);
2161 * ib_init_ah_from_wc - Initializes address handle attributes from a
2163 * @device: Device on which the received message arrived.
2164 * @port_num: Port on which the received message arrived.
2165 * @wc: Work completion associated with the received message.
2166 * @grh: References the received global route header. This parameter is
2167 * ignored unless the work completion indicates that the GRH is valid.
2168 * @ah_attr: Returned attributes that can be used when creating an address
2169 * handle for replying to the message.
2171 int ib_init_ah_from_wc(struct ib_device
*device
, u8 port_num
,
2172 const struct ib_wc
*wc
, const struct ib_grh
*grh
,
2173 struct ib_ah_attr
*ah_attr
);
2176 * ib_create_ah_from_wc - Creates an address handle associated with the
2177 * sender of the specified work completion.
2178 * @pd: The protection domain associated with the address handle.
2179 * @wc: Work completion information associated with a received message.
2180 * @grh: References the received global route header. This parameter is
2181 * ignored unless the work completion indicates that the GRH is valid.
2182 * @port_num: The outbound port number to associate with the address.
2184 * The address handle is used to reference a local or global destination
2185 * in all UD QP post sends.
2187 struct ib_ah
*ib_create_ah_from_wc(struct ib_pd
*pd
, const struct ib_wc
*wc
,
2188 const struct ib_grh
*grh
, u8 port_num
);
2191 * ib_modify_ah - Modifies the address vector associated with an address
2193 * @ah: The address handle to modify.
2194 * @ah_attr: The new address vector attributes to associate with the
2197 int ib_modify_ah(struct ib_ah
*ah
, struct ib_ah_attr
*ah_attr
);
2200 * ib_query_ah - Queries the address vector associated with an address
2202 * @ah: The address handle to query.
2203 * @ah_attr: The address vector attributes associated with the address
2206 int ib_query_ah(struct ib_ah
*ah
, struct ib_ah_attr
*ah_attr
);
2209 * ib_destroy_ah - Destroys an address handle.
2210 * @ah: The address handle to destroy.
2212 int ib_destroy_ah(struct ib_ah
*ah
);
2215 * ib_create_srq - Creates a SRQ associated with the specified protection
2217 * @pd: The protection domain associated with the SRQ.
2218 * @srq_init_attr: A list of initial attributes required to create the
2219 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2220 * the actual capabilities of the created SRQ.
2222 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2223 * requested size of the SRQ, and set to the actual values allocated
2224 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2225 * will always be at least as large as the requested values.
2227 struct ib_srq
*ib_create_srq(struct ib_pd
*pd
,
2228 struct ib_srq_init_attr
*srq_init_attr
);
2231 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2232 * @srq: The SRQ to modify.
2233 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2234 * the current values of selected SRQ attributes are returned.
2235 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2236 * are being modified.
2238 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2239 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2240 * the number of receives queued drops below the limit.
2242 int ib_modify_srq(struct ib_srq
*srq
,
2243 struct ib_srq_attr
*srq_attr
,
2244 enum ib_srq_attr_mask srq_attr_mask
);
2247 * ib_query_srq - Returns the attribute list and current values for the
2249 * @srq: The SRQ to query.
2250 * @srq_attr: The attributes of the specified SRQ.
2252 int ib_query_srq(struct ib_srq
*srq
,
2253 struct ib_srq_attr
*srq_attr
);
2256 * ib_destroy_srq - Destroys the specified SRQ.
2257 * @srq: The SRQ to destroy.
2259 int ib_destroy_srq(struct ib_srq
*srq
);
2262 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2263 * @srq: The SRQ to post the work request on.
2264 * @recv_wr: A list of work requests to post on the receive queue.
2265 * @bad_recv_wr: On an immediate failure, this parameter will reference
2266 * the work request that failed to be posted on the QP.
2268 static inline int ib_post_srq_recv(struct ib_srq
*srq
,
2269 struct ib_recv_wr
*recv_wr
,
2270 struct ib_recv_wr
**bad_recv_wr
)
2272 return srq
->device
->post_srq_recv(srq
, recv_wr
, bad_recv_wr
);
2276 * ib_create_qp - Creates a QP associated with the specified protection
2278 * @pd: The protection domain associated with the QP.
2279 * @qp_init_attr: A list of initial attributes required to create the
2280 * QP. If QP creation succeeds, then the attributes are updated to
2281 * the actual capabilities of the created QP.
2283 struct ib_qp
*ib_create_qp(struct ib_pd
*pd
,
2284 struct ib_qp_init_attr
*qp_init_attr
);
2287 * ib_modify_qp - Modifies the attributes for the specified QP and then
2288 * transitions the QP to the given state.
2289 * @qp: The QP to modify.
2290 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2291 * the current values of selected QP attributes are returned.
2292 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2293 * are being modified.
2295 int ib_modify_qp(struct ib_qp
*qp
,
2296 struct ib_qp_attr
*qp_attr
,
2300 * ib_query_qp - Returns the attribute list and current values for the
2302 * @qp: The QP to query.
2303 * @qp_attr: The attributes of the specified QP.
2304 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2305 * @qp_init_attr: Additional attributes of the selected QP.
2307 * The qp_attr_mask may be used to limit the query to gathering only the
2308 * selected attributes.
2310 int ib_query_qp(struct ib_qp
*qp
,
2311 struct ib_qp_attr
*qp_attr
,
2313 struct ib_qp_init_attr
*qp_init_attr
);
2316 * ib_destroy_qp - Destroys the specified QP.
2317 * @qp: The QP to destroy.
2319 int ib_destroy_qp(struct ib_qp
*qp
);
2322 * ib_open_qp - Obtain a reference to an existing sharable QP.
2323 * @xrcd - XRC domain
2324 * @qp_open_attr: Attributes identifying the QP to open.
2326 * Returns a reference to a sharable QP.
2328 struct ib_qp
*ib_open_qp(struct ib_xrcd
*xrcd
,
2329 struct ib_qp_open_attr
*qp_open_attr
);
2332 * ib_close_qp - Release an external reference to a QP.
2333 * @qp: The QP handle to release
2335 * The opened QP handle is released by the caller. The underlying
2336 * shared QP is not destroyed until all internal references are released.
2338 int ib_close_qp(struct ib_qp
*qp
);
2341 * ib_post_send - Posts a list of work requests to the send queue of
2343 * @qp: The QP to post the work request on.
2344 * @send_wr: A list of work requests to post on the send queue.
2345 * @bad_send_wr: On an immediate failure, this parameter will reference
2346 * the work request that failed to be posted on the QP.
2348 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2349 * error is returned, the QP state shall not be affected,
2350 * ib_post_send() will return an immediate error after queueing any
2351 * earlier work requests in the list.
2353 static inline int ib_post_send(struct ib_qp
*qp
,
2354 struct ib_send_wr
*send_wr
,
2355 struct ib_send_wr
**bad_send_wr
)
2357 return qp
->device
->post_send(qp
, send_wr
, bad_send_wr
);
2361 * ib_post_recv - Posts a list of work requests to the receive queue of
2363 * @qp: The QP to post the work request on.
2364 * @recv_wr: A list of work requests to post on the receive queue.
2365 * @bad_recv_wr: On an immediate failure, this parameter will reference
2366 * the work request that failed to be posted on the QP.
2368 static inline int ib_post_recv(struct ib_qp
*qp
,
2369 struct ib_recv_wr
*recv_wr
,
2370 struct ib_recv_wr
**bad_recv_wr
)
2372 return qp
->device
->post_recv(qp
, recv_wr
, bad_recv_wr
);
2376 * ib_create_cq - Creates a CQ on the specified device.
2377 * @device: The device on which to create the CQ.
2378 * @comp_handler: A user-specified callback that is invoked when a
2379 * completion event occurs on the CQ.
2380 * @event_handler: A user-specified callback that is invoked when an
2381 * asynchronous event not associated with a completion occurs on the CQ.
2382 * @cq_context: Context associated with the CQ returned to the user via
2383 * the associated completion and event handlers.
2384 * @cq_attr: The attributes the CQ should be created upon.
2386 * Users can examine the cq structure to determine the actual CQ size.
2388 struct ib_cq
*ib_create_cq(struct ib_device
*device
,
2389 ib_comp_handler comp_handler
,
2390 void (*event_handler
)(struct ib_event
*, void *),
2392 const struct ib_cq_init_attr
*cq_attr
);
2395 * ib_resize_cq - Modifies the capacity of the CQ.
2396 * @cq: The CQ to resize.
2397 * @cqe: The minimum size of the CQ.
2399 * Users can examine the cq structure to determine the actual CQ size.
2401 int ib_resize_cq(struct ib_cq
*cq
, int cqe
);
2404 * ib_modify_cq - Modifies moderation params of the CQ
2405 * @cq: The CQ to modify.
2406 * @cq_count: number of CQEs that will trigger an event
2407 * @cq_period: max period of time in usec before triggering an event
2410 int ib_modify_cq(struct ib_cq
*cq
, u16 cq_count
, u16 cq_period
);
2413 * ib_destroy_cq - Destroys the specified CQ.
2414 * @cq: The CQ to destroy.
2416 int ib_destroy_cq(struct ib_cq
*cq
);
2419 * ib_poll_cq - poll a CQ for completion(s)
2420 * @cq:the CQ being polled
2421 * @num_entries:maximum number of completions to return
2422 * @wc:array of at least @num_entries &struct ib_wc where completions
2425 * Poll a CQ for (possibly multiple) completions. If the return value
2426 * is < 0, an error occurred. If the return value is >= 0, it is the
2427 * number of completions returned. If the return value is
2428 * non-negative and < num_entries, then the CQ was emptied.
2430 static inline int ib_poll_cq(struct ib_cq
*cq
, int num_entries
,
2433 return cq
->device
->poll_cq(cq
, num_entries
, wc
);
2437 * ib_peek_cq - Returns the number of unreaped completions currently
2438 * on the specified CQ.
2439 * @cq: The CQ to peek.
2440 * @wc_cnt: A minimum number of unreaped completions to check for.
2442 * If the number of unreaped completions is greater than or equal to wc_cnt,
2443 * this function returns wc_cnt, otherwise, it returns the actual number of
2444 * unreaped completions.
2446 int ib_peek_cq(struct ib_cq
*cq
, int wc_cnt
);
2449 * ib_req_notify_cq - Request completion notification on a CQ.
2450 * @cq: The CQ to generate an event for.
2452 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2453 * to request an event on the next solicited event or next work
2454 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2455 * may also be |ed in to request a hint about missed events, as
2459 * < 0 means an error occurred while requesting notification
2460 * == 0 means notification was requested successfully, and if
2461 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2462 * were missed and it is safe to wait for another event. In
2463 * this case is it guaranteed that any work completions added
2464 * to the CQ since the last CQ poll will trigger a completion
2465 * notification event.
2466 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2467 * in. It means that the consumer must poll the CQ again to
2468 * make sure it is empty to avoid missing an event because of a
2469 * race between requesting notification and an entry being
2470 * added to the CQ. This return value means it is possible
2471 * (but not guaranteed) that a work completion has been added
2472 * to the CQ since the last poll without triggering a
2473 * completion notification event.
2475 static inline int ib_req_notify_cq(struct ib_cq
*cq
,
2476 enum ib_cq_notify_flags flags
)
2478 return cq
->device
->req_notify_cq(cq
, flags
);
2482 * ib_req_ncomp_notif - Request completion notification when there are
2483 * at least the specified number of unreaped completions on the CQ.
2484 * @cq: The CQ to generate an event for.
2485 * @wc_cnt: The number of unreaped completions that should be on the
2486 * CQ before an event is generated.
2488 static inline int ib_req_ncomp_notif(struct ib_cq
*cq
, int wc_cnt
)
2490 return cq
->device
->req_ncomp_notif
?
2491 cq
->device
->req_ncomp_notif(cq
, wc_cnt
) :
2496 * ib_get_dma_mr - Returns a memory region for system memory that is
2498 * @pd: The protection domain associated with the memory region.
2499 * @mr_access_flags: Specifies the memory access rights.
2501 * Note that the ib_dma_*() functions defined below must be used
2502 * to create/destroy addresses used with the Lkey or Rkey returned
2503 * by ib_get_dma_mr().
2505 struct ib_mr
*ib_get_dma_mr(struct ib_pd
*pd
, int mr_access_flags
);
2508 * ib_dma_mapping_error - check a DMA addr for error
2509 * @dev: The device for which the dma_addr was created
2510 * @dma_addr: The DMA address to check
2512 static inline int ib_dma_mapping_error(struct ib_device
*dev
, u64 dma_addr
)
2515 return dev
->dma_ops
->mapping_error(dev
, dma_addr
);
2516 return dma_mapping_error(dev
->dma_device
, dma_addr
);
2520 * ib_dma_map_single - Map a kernel virtual address to DMA address
2521 * @dev: The device for which the dma_addr is to be created
2522 * @cpu_addr: The kernel virtual address
2523 * @size: The size of the region in bytes
2524 * @direction: The direction of the DMA
2526 static inline u64
ib_dma_map_single(struct ib_device
*dev
,
2527 void *cpu_addr
, size_t size
,
2528 enum dma_data_direction direction
)
2531 return dev
->dma_ops
->map_single(dev
, cpu_addr
, size
, direction
);
2532 return dma_map_single(dev
->dma_device
, cpu_addr
, size
, direction
);
2536 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2537 * @dev: The device for which the DMA address was created
2538 * @addr: The DMA address
2539 * @size: The size of the region in bytes
2540 * @direction: The direction of the DMA
2542 static inline void ib_dma_unmap_single(struct ib_device
*dev
,
2543 u64 addr
, size_t size
,
2544 enum dma_data_direction direction
)
2547 dev
->dma_ops
->unmap_single(dev
, addr
, size
, direction
);
2549 dma_unmap_single(dev
->dma_device
, addr
, size
, direction
);
2552 static inline u64
ib_dma_map_single_attrs(struct ib_device
*dev
,
2553 void *cpu_addr
, size_t size
,
2554 enum dma_data_direction direction
,
2555 struct dma_attrs
*attrs
)
2557 return dma_map_single_attrs(dev
->dma_device
, cpu_addr
, size
,
2561 static inline void ib_dma_unmap_single_attrs(struct ib_device
*dev
,
2562 u64 addr
, size_t size
,
2563 enum dma_data_direction direction
,
2564 struct dma_attrs
*attrs
)
2566 return dma_unmap_single_attrs(dev
->dma_device
, addr
, size
,
2571 * ib_dma_map_page - Map a physical page to DMA address
2572 * @dev: The device for which the dma_addr is to be created
2573 * @page: The page to be mapped
2574 * @offset: The offset within the page
2575 * @size: The size of the region in bytes
2576 * @direction: The direction of the DMA
2578 static inline u64
ib_dma_map_page(struct ib_device
*dev
,
2580 unsigned long offset
,
2582 enum dma_data_direction direction
)
2585 return dev
->dma_ops
->map_page(dev
, page
, offset
, size
, direction
);
2586 return dma_map_page(dev
->dma_device
, page
, offset
, size
, direction
);
2590 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2591 * @dev: The device for which the DMA address was created
2592 * @addr: The DMA address
2593 * @size: The size of the region in bytes
2594 * @direction: The direction of the DMA
2596 static inline void ib_dma_unmap_page(struct ib_device
*dev
,
2597 u64 addr
, size_t size
,
2598 enum dma_data_direction direction
)
2601 dev
->dma_ops
->unmap_page(dev
, addr
, size
, direction
);
2603 dma_unmap_page(dev
->dma_device
, addr
, size
, direction
);
2607 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2608 * @dev: The device for which the DMA addresses are to be created
2609 * @sg: The array of scatter/gather entries
2610 * @nents: The number of scatter/gather entries
2611 * @direction: The direction of the DMA
2613 static inline int ib_dma_map_sg(struct ib_device
*dev
,
2614 struct scatterlist
*sg
, int nents
,
2615 enum dma_data_direction direction
)
2618 return dev
->dma_ops
->map_sg(dev
, sg
, nents
, direction
);
2619 return dma_map_sg(dev
->dma_device
, sg
, nents
, direction
);
2623 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2624 * @dev: The device for which the DMA addresses were created
2625 * @sg: The array of scatter/gather entries
2626 * @nents: The number of scatter/gather entries
2627 * @direction: The direction of the DMA
2629 static inline void ib_dma_unmap_sg(struct ib_device
*dev
,
2630 struct scatterlist
*sg
, int nents
,
2631 enum dma_data_direction direction
)
2634 dev
->dma_ops
->unmap_sg(dev
, sg
, nents
, direction
);
2636 dma_unmap_sg(dev
->dma_device
, sg
, nents
, direction
);
2639 static inline int ib_dma_map_sg_attrs(struct ib_device
*dev
,
2640 struct scatterlist
*sg
, int nents
,
2641 enum dma_data_direction direction
,
2642 struct dma_attrs
*attrs
)
2644 return dma_map_sg_attrs(dev
->dma_device
, sg
, nents
, direction
, attrs
);
2647 static inline void ib_dma_unmap_sg_attrs(struct ib_device
*dev
,
2648 struct scatterlist
*sg
, int nents
,
2649 enum dma_data_direction direction
,
2650 struct dma_attrs
*attrs
)
2652 dma_unmap_sg_attrs(dev
->dma_device
, sg
, nents
, direction
, attrs
);
2655 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2656 * @dev: The device for which the DMA addresses were created
2657 * @sg: The scatter/gather entry
2659 * Note: this function is obsolete. To do: change all occurrences of
2660 * ib_sg_dma_address() into sg_dma_address().
2662 static inline u64
ib_sg_dma_address(struct ib_device
*dev
,
2663 struct scatterlist
*sg
)
2665 return sg_dma_address(sg
);
2669 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2670 * @dev: The device for which the DMA addresses were created
2671 * @sg: The scatter/gather entry
2673 * Note: this function is obsolete. To do: change all occurrences of
2674 * ib_sg_dma_len() into sg_dma_len().
2676 static inline unsigned int ib_sg_dma_len(struct ib_device
*dev
,
2677 struct scatterlist
*sg
)
2679 return sg_dma_len(sg
);
2683 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2684 * @dev: The device for which the DMA address was created
2685 * @addr: The DMA address
2686 * @size: The size of the region in bytes
2687 * @dir: The direction of the DMA
2689 static inline void ib_dma_sync_single_for_cpu(struct ib_device
*dev
,
2692 enum dma_data_direction dir
)
2695 dev
->dma_ops
->sync_single_for_cpu(dev
, addr
, size
, dir
);
2697 dma_sync_single_for_cpu(dev
->dma_device
, addr
, size
, dir
);
2701 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2702 * @dev: The device for which the DMA address was created
2703 * @addr: The DMA address
2704 * @size: The size of the region in bytes
2705 * @dir: The direction of the DMA
2707 static inline void ib_dma_sync_single_for_device(struct ib_device
*dev
,
2710 enum dma_data_direction dir
)
2713 dev
->dma_ops
->sync_single_for_device(dev
, addr
, size
, dir
);
2715 dma_sync_single_for_device(dev
->dma_device
, addr
, size
, dir
);
2719 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2720 * @dev: The device for which the DMA address is requested
2721 * @size: The size of the region to allocate in bytes
2722 * @dma_handle: A pointer for returning the DMA address of the region
2723 * @flag: memory allocator flags
2725 static inline void *ib_dma_alloc_coherent(struct ib_device
*dev
,
2731 return dev
->dma_ops
->alloc_coherent(dev
, size
, dma_handle
, flag
);
2736 ret
= dma_alloc_coherent(dev
->dma_device
, size
, &handle
, flag
);
2737 *dma_handle
= handle
;
2743 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2744 * @dev: The device for which the DMA addresses were allocated
2745 * @size: The size of the region
2746 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2747 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2749 static inline void ib_dma_free_coherent(struct ib_device
*dev
,
2750 size_t size
, void *cpu_addr
,
2754 dev
->dma_ops
->free_coherent(dev
, size
, cpu_addr
, dma_handle
);
2756 dma_free_coherent(dev
->dma_device
, size
, cpu_addr
, dma_handle
);
2760 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
2762 * @pd: The protection domain associated assigned to the registered region.
2763 * @phys_buf_array: Specifies a list of physical buffers to use in the
2765 * @num_phys_buf: Specifies the size of the phys_buf_array.
2766 * @mr_access_flags: Specifies the memory access rights.
2767 * @iova_start: The offset of the region's starting I/O virtual address.
2769 struct ib_mr
*ib_reg_phys_mr(struct ib_pd
*pd
,
2770 struct ib_phys_buf
*phys_buf_array
,
2772 int mr_access_flags
,
2776 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
2777 * Conceptually, this call performs the functions deregister memory region
2778 * followed by register physical memory region. Where possible,
2779 * resources are reused instead of deallocated and reallocated.
2780 * @mr: The memory region to modify.
2781 * @mr_rereg_mask: A bit-mask used to indicate which of the following
2782 * properties of the memory region are being modified.
2783 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
2784 * the new protection domain to associated with the memory region,
2785 * otherwise, this parameter is ignored.
2786 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2787 * field specifies a list of physical buffers to use in the new
2788 * translation, otherwise, this parameter is ignored.
2789 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2790 * field specifies the size of the phys_buf_array, otherwise, this
2791 * parameter is ignored.
2792 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
2793 * field specifies the new memory access rights, otherwise, this
2794 * parameter is ignored.
2795 * @iova_start: The offset of the region's starting I/O virtual address.
2797 int ib_rereg_phys_mr(struct ib_mr
*mr
,
2800 struct ib_phys_buf
*phys_buf_array
,
2802 int mr_access_flags
,
2806 * ib_query_mr - Retrieves information about a specific memory region.
2807 * @mr: The memory region to retrieve information about.
2808 * @mr_attr: The attributes of the specified memory region.
2810 int ib_query_mr(struct ib_mr
*mr
, struct ib_mr_attr
*mr_attr
);
2813 * ib_dereg_mr - Deregisters a memory region and removes it from the
2814 * HCA translation table.
2815 * @mr: The memory region to deregister.
2817 * This function can fail, if the memory region has memory windows bound to it.
2819 int ib_dereg_mr(struct ib_mr
*mr
);
2823 * ib_create_mr - Allocates a memory region that may be used for
2824 * signature handover operations.
2825 * @pd: The protection domain associated with the region.
2826 * @mr_init_attr: memory region init attributes.
2828 struct ib_mr
*ib_create_mr(struct ib_pd
*pd
,
2829 struct ib_mr_init_attr
*mr_init_attr
);
2832 * ib_destroy_mr - Destroys a memory region that was created using
2833 * ib_create_mr and removes it from HW translation tables.
2834 * @mr: The memory region to destroy.
2836 * This function can fail, if the memory region has memory windows bound to it.
2838 int ib_destroy_mr(struct ib_mr
*mr
);
2841 * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2842 * IB_WR_FAST_REG_MR send work request.
2843 * @pd: The protection domain associated with the region.
2844 * @max_page_list_len: requested max physical buffer list length to be
2845 * used with fast register work requests for this MR.
2847 struct ib_mr
*ib_alloc_fast_reg_mr(struct ib_pd
*pd
, int max_page_list_len
);
2850 * ib_alloc_fast_reg_page_list - Allocates a page list array
2851 * @device - ib device pointer.
2852 * @page_list_len - size of the page list array to be allocated.
2854 * This allocates and returns a struct ib_fast_reg_page_list * and a
2855 * page_list array that is at least page_list_len in size. The actual
2856 * size is returned in max_page_list_len. The caller is responsible
2857 * for initializing the contents of the page_list array before posting
2858 * a send work request with the IB_WC_FAST_REG_MR opcode.
2860 * The page_list array entries must be translated using one of the
2861 * ib_dma_*() functions just like the addresses passed to
2862 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct
2863 * ib_fast_reg_page_list must not be modified by the caller until the
2864 * IB_WC_FAST_REG_MR work request completes.
2866 struct ib_fast_reg_page_list
*ib_alloc_fast_reg_page_list(
2867 struct ib_device
*device
, int page_list_len
);
2870 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2872 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2874 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list
*page_list
);
2877 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2879 * @mr - struct ib_mr pointer to be updated.
2880 * @newkey - new key to be used.
2882 static inline void ib_update_fast_reg_key(struct ib_mr
*mr
, u8 newkey
)
2884 mr
->lkey
= (mr
->lkey
& 0xffffff00) | newkey
;
2885 mr
->rkey
= (mr
->rkey
& 0xffffff00) | newkey
;
2889 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2890 * for calculating a new rkey for type 2 memory windows.
2891 * @rkey - the rkey to increment.
2893 static inline u32
ib_inc_rkey(u32 rkey
)
2895 const u32 mask
= 0x000000ff;
2896 return ((rkey
+ 1) & mask
) | (rkey
& ~mask
);
2900 * ib_alloc_mw - Allocates a memory window.
2901 * @pd: The protection domain associated with the memory window.
2902 * @type: The type of the memory window (1 or 2).
2904 struct ib_mw
*ib_alloc_mw(struct ib_pd
*pd
, enum ib_mw_type type
);
2907 * ib_bind_mw - Posts a work request to the send queue of the specified
2908 * QP, which binds the memory window to the given address range and
2909 * remote access attributes.
2910 * @qp: QP to post the bind work request on.
2911 * @mw: The memory window to bind.
2912 * @mw_bind: Specifies information about the memory window, including
2913 * its address range, remote access rights, and associated memory region.
2915 * If there is no immediate error, the function will update the rkey member
2916 * of the mw parameter to its new value. The bind operation can still fail
2919 static inline int ib_bind_mw(struct ib_qp
*qp
,
2921 struct ib_mw_bind
*mw_bind
)
2923 /* XXX reference counting in corresponding MR? */
2924 return mw
->device
->bind_mw
?
2925 mw
->device
->bind_mw(qp
, mw
, mw_bind
) :
2930 * ib_dealloc_mw - Deallocates a memory window.
2931 * @mw: The memory window to deallocate.
2933 int ib_dealloc_mw(struct ib_mw
*mw
);
2936 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2937 * @pd: The protection domain associated with the unmapped region.
2938 * @mr_access_flags: Specifies the memory access rights.
2939 * @fmr_attr: Attributes of the unmapped region.
2941 * A fast memory region must be mapped before it can be used as part of
2944 struct ib_fmr
*ib_alloc_fmr(struct ib_pd
*pd
,
2945 int mr_access_flags
,
2946 struct ib_fmr_attr
*fmr_attr
);
2949 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2950 * @fmr: The fast memory region to associate with the pages.
2951 * @page_list: An array of physical pages to map to the fast memory region.
2952 * @list_len: The number of pages in page_list.
2953 * @iova: The I/O virtual address to use with the mapped region.
2955 static inline int ib_map_phys_fmr(struct ib_fmr
*fmr
,
2956 u64
*page_list
, int list_len
,
2959 return fmr
->device
->map_phys_fmr(fmr
, page_list
, list_len
, iova
);
2963 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2964 * @fmr_list: A linked list of fast memory regions to unmap.
2966 int ib_unmap_fmr(struct list_head
*fmr_list
);
2969 * ib_dealloc_fmr - Deallocates a fast memory region.
2970 * @fmr: The fast memory region to deallocate.
2972 int ib_dealloc_fmr(struct ib_fmr
*fmr
);
2975 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2976 * @qp: QP to attach to the multicast group. The QP must be type
2978 * @gid: Multicast group GID.
2979 * @lid: Multicast group LID in host byte order.
2981 * In order to send and receive multicast packets, subnet
2982 * administration must have created the multicast group and configured
2983 * the fabric appropriately. The port associated with the specified
2984 * QP must also be a member of the multicast group.
2986 int ib_attach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
2989 * ib_detach_mcast - Detaches the specified QP from a multicast group.
2990 * @qp: QP to detach from the multicast group.
2991 * @gid: Multicast group GID.
2992 * @lid: Multicast group LID in host byte order.
2994 int ib_detach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
2997 * ib_alloc_xrcd - Allocates an XRC domain.
2998 * @device: The device on which to allocate the XRC domain.
3000 struct ib_xrcd
*ib_alloc_xrcd(struct ib_device
*device
);
3003 * ib_dealloc_xrcd - Deallocates an XRC domain.
3004 * @xrcd: The XRC domain to deallocate.
3006 int ib_dealloc_xrcd(struct ib_xrcd
*xrcd
);
3008 struct ib_flow
*ib_create_flow(struct ib_qp
*qp
,
3009 struct ib_flow_attr
*flow_attr
, int domain
);
3010 int ib_destroy_flow(struct ib_flow
*flow_id
);
3012 static inline int ib_check_mr_access(int flags
)
3015 * Local write permission is required if remote write or
3016 * remote atomic permission is also requested.
3018 if (flags
& (IB_ACCESS_REMOTE_ATOMIC
| IB_ACCESS_REMOTE_WRITE
) &&
3019 !(flags
& IB_ACCESS_LOCAL_WRITE
))
3026 * ib_check_mr_status: lightweight check of MR status.
3027 * This routine may provide status checks on a selected
3028 * ib_mr. first use is for signature status check.
3030 * @mr: A memory region.
3031 * @check_mask: Bitmask of which checks to perform from
3032 * ib_mr_status_check enumeration.
3033 * @mr_status: The container of relevant status checks.
3034 * failed checks will be indicated in the status bitmask
3035 * and the relevant info shall be in the error item.
3037 int ib_check_mr_status(struct ib_mr
*mr
, u32 check_mask
,
3038 struct ib_mr_status
*mr_status
);
3040 struct net_device
*ib_get_net_dev_by_params(struct ib_device
*dev
, u8 port
,
3041 u16 pkey
, const union ib_gid
*gid
,
3042 const struct sockaddr
*addr
);
3044 #endif /* IB_VERBS_H */