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 <uapi/linux/if_ether.h>
53 #include <linux/atomic.h>
54 #include <linux/mmu_notifier.h>
55 #include <asm/uaccess.h>
57 extern struct workqueue_struct
*ib_wq
;
68 /* IB values map to NodeInfo:NodeType. */
77 enum rdma_transport_type
{
81 RDMA_TRANSPORT_USNIC_UDP
84 enum rdma_protocol_type
{
88 RDMA_PROTOCOL_USNIC_UDP
91 __attribute_const__
enum rdma_transport_type
92 rdma_node_get_transport(enum rdma_node_type node_type
);
94 enum rdma_link_layer
{
95 IB_LINK_LAYER_UNSPECIFIED
,
96 IB_LINK_LAYER_INFINIBAND
,
97 IB_LINK_LAYER_ETHERNET
,
100 enum ib_device_cap_flags
{
101 IB_DEVICE_RESIZE_MAX_WR
= 1,
102 IB_DEVICE_BAD_PKEY_CNTR
= (1<<1),
103 IB_DEVICE_BAD_QKEY_CNTR
= (1<<2),
104 IB_DEVICE_RAW_MULTI
= (1<<3),
105 IB_DEVICE_AUTO_PATH_MIG
= (1<<4),
106 IB_DEVICE_CHANGE_PHY_PORT
= (1<<5),
107 IB_DEVICE_UD_AV_PORT_ENFORCE
= (1<<6),
108 IB_DEVICE_CURR_QP_STATE_MOD
= (1<<7),
109 IB_DEVICE_SHUTDOWN_PORT
= (1<<8),
110 IB_DEVICE_INIT_TYPE
= (1<<9),
111 IB_DEVICE_PORT_ACTIVE_EVENT
= (1<<10),
112 IB_DEVICE_SYS_IMAGE_GUID
= (1<<11),
113 IB_DEVICE_RC_RNR_NAK_GEN
= (1<<12),
114 IB_DEVICE_SRQ_RESIZE
= (1<<13),
115 IB_DEVICE_N_NOTIFY_CQ
= (1<<14),
116 IB_DEVICE_LOCAL_DMA_LKEY
= (1<<15),
117 IB_DEVICE_RESERVED
= (1<<16), /* old SEND_W_INV */
118 IB_DEVICE_MEM_WINDOW
= (1<<17),
120 * Devices should set IB_DEVICE_UD_IP_SUM if they support
121 * insertion of UDP and TCP checksum on outgoing UD IPoIB
122 * messages and can verify the validity of checksum for
123 * incoming messages. Setting this flag implies that the
124 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
126 IB_DEVICE_UD_IP_CSUM
= (1<<18),
127 IB_DEVICE_UD_TSO
= (1<<19),
128 IB_DEVICE_XRC
= (1<<20),
129 IB_DEVICE_MEM_MGT_EXTENSIONS
= (1<<21),
130 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK
= (1<<22),
131 IB_DEVICE_MEM_WINDOW_TYPE_2A
= (1<<23),
132 IB_DEVICE_MEM_WINDOW_TYPE_2B
= (1<<24),
133 IB_DEVICE_MANAGED_FLOW_STEERING
= (1<<29),
134 IB_DEVICE_SIGNATURE_HANDOVER
= (1<<30),
135 IB_DEVICE_ON_DEMAND_PAGING
= (1<<31),
138 enum ib_signature_prot_cap
{
139 IB_PROT_T10DIF_TYPE_1
= 1,
140 IB_PROT_T10DIF_TYPE_2
= 1 << 1,
141 IB_PROT_T10DIF_TYPE_3
= 1 << 2,
144 enum ib_signature_guard_cap
{
145 IB_GUARD_T10DIF_CRC
= 1,
146 IB_GUARD_T10DIF_CSUM
= 1 << 1,
155 enum ib_odp_general_cap_bits
{
156 IB_ODP_SUPPORT
= 1 << 0,
159 enum ib_odp_transport_cap_bits
{
160 IB_ODP_SUPPORT_SEND
= 1 << 0,
161 IB_ODP_SUPPORT_RECV
= 1 << 1,
162 IB_ODP_SUPPORT_WRITE
= 1 << 2,
163 IB_ODP_SUPPORT_READ
= 1 << 3,
164 IB_ODP_SUPPORT_ATOMIC
= 1 << 4,
168 uint64_t general_caps
;
170 uint32_t rc_odp_caps
;
171 uint32_t uc_odp_caps
;
172 uint32_t ud_odp_caps
;
173 } per_transport_caps
;
176 enum ib_cq_creation_flags
{
177 IB_CQ_FLAGS_TIMESTAMP_COMPLETION
= 1 << 0,
180 struct ib_cq_init_attr
{
186 struct ib_device_attr
{
188 __be64 sys_image_guid
;
196 int device_cap_flags
;
206 int max_qp_init_rd_atom
;
207 int max_ee_init_rd_atom
;
208 enum ib_atomic_cap atomic_cap
;
209 enum ib_atomic_cap masked_atomic_cap
;
216 int max_mcast_qp_attach
;
217 int max_total_mcast_qp_attach
;
224 unsigned int max_fast_reg_page_list_len
;
226 u8 local_ca_ack_delay
;
229 struct ib_odp_caps odp_caps
;
230 uint64_t timestamp_mask
;
231 uint64_t hca_core_clock
; /* in KHZ */
242 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu
)
245 case IB_MTU_256
: return 256;
246 case IB_MTU_512
: return 512;
247 case IB_MTU_1024
: return 1024;
248 case IB_MTU_2048
: return 2048;
249 case IB_MTU_4096
: return 4096;
260 IB_PORT_ACTIVE_DEFER
= 5
263 enum ib_port_cap_flags
{
265 IB_PORT_NOTICE_SUP
= 1 << 2,
266 IB_PORT_TRAP_SUP
= 1 << 3,
267 IB_PORT_OPT_IPD_SUP
= 1 << 4,
268 IB_PORT_AUTO_MIGR_SUP
= 1 << 5,
269 IB_PORT_SL_MAP_SUP
= 1 << 6,
270 IB_PORT_MKEY_NVRAM
= 1 << 7,
271 IB_PORT_PKEY_NVRAM
= 1 << 8,
272 IB_PORT_LED_INFO_SUP
= 1 << 9,
273 IB_PORT_SM_DISABLED
= 1 << 10,
274 IB_PORT_SYS_IMAGE_GUID_SUP
= 1 << 11,
275 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP
= 1 << 12,
276 IB_PORT_EXTENDED_SPEEDS_SUP
= 1 << 14,
277 IB_PORT_CM_SUP
= 1 << 16,
278 IB_PORT_SNMP_TUNNEL_SUP
= 1 << 17,
279 IB_PORT_REINIT_SUP
= 1 << 18,
280 IB_PORT_DEVICE_MGMT_SUP
= 1 << 19,
281 IB_PORT_VENDOR_CLASS_SUP
= 1 << 20,
282 IB_PORT_DR_NOTICE_SUP
= 1 << 21,
283 IB_PORT_CAP_MASK_NOTICE_SUP
= 1 << 22,
284 IB_PORT_BOOT_MGMT_SUP
= 1 << 23,
285 IB_PORT_LINK_LATENCY_SUP
= 1 << 24,
286 IB_PORT_CLIENT_REG_SUP
= 1 << 25,
287 IB_PORT_IP_BASED_GIDS
= 1 << 26
297 static inline int ib_width_enum_to_int(enum ib_port_width width
)
300 case IB_WIDTH_1X
: return 1;
301 case IB_WIDTH_4X
: return 4;
302 case IB_WIDTH_8X
: return 8;
303 case IB_WIDTH_12X
: return 12;
317 struct ib_protocol_stats
{
321 struct iw_protocol_stats
{
324 u64 ipInTooBigErrors
;
327 u64 ipInUnknownProtos
;
328 u64 ipInTruncatedPkts
;
331 u64 ipOutForwDatagrams
;
363 union rdma_protocol_stats
{
364 struct ib_protocol_stats ib
;
365 struct iw_protocol_stats iw
;
368 /* Define bits for the various functionality this port needs to be supported by
371 /* Management 0x00000FFF */
372 #define RDMA_CORE_CAP_IB_MAD 0x00000001
373 #define RDMA_CORE_CAP_IB_SMI 0x00000002
374 #define RDMA_CORE_CAP_IB_CM 0x00000004
375 #define RDMA_CORE_CAP_IW_CM 0x00000008
376 #define RDMA_CORE_CAP_IB_SA 0x00000010
378 /* Address format 0x000FF000 */
379 #define RDMA_CORE_CAP_AF_IB 0x00001000
380 #define RDMA_CORE_CAP_ETH_AH 0x00002000
382 /* Protocol 0xFFF00000 */
383 #define RDMA_CORE_CAP_PROT_IB 0x00100000
384 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
385 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
387 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
388 | RDMA_CORE_CAP_IB_MAD \
389 | RDMA_CORE_CAP_IB_SMI \
390 | RDMA_CORE_CAP_IB_CM \
391 | RDMA_CORE_CAP_IB_SA \
392 | RDMA_CORE_CAP_AF_IB)
393 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
394 | RDMA_CORE_CAP_IB_MAD \
395 | RDMA_CORE_CAP_IB_CM \
396 | RDMA_CORE_CAP_AF_IB \
397 | RDMA_CORE_CAP_ETH_AH)
398 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
399 | RDMA_CORE_CAP_IW_CM)
401 struct ib_port_attr
{
402 enum ib_port_state state
;
404 enum ib_mtu active_mtu
;
423 enum ib_device_modify_flags
{
424 IB_DEVICE_MODIFY_SYS_IMAGE_GUID
= 1 << 0,
425 IB_DEVICE_MODIFY_NODE_DESC
= 1 << 1
428 struct ib_device_modify
{
433 enum ib_port_modify_flags
{
434 IB_PORT_SHUTDOWN
= 1,
435 IB_PORT_INIT_TYPE
= (1<<2),
436 IB_PORT_RESET_QKEY_CNTR
= (1<<3)
439 struct ib_port_modify
{
440 u32 set_port_cap_mask
;
441 u32 clr_port_cap_mask
;
449 IB_EVENT_QP_ACCESS_ERR
,
453 IB_EVENT_PATH_MIG_ERR
,
454 IB_EVENT_DEVICE_FATAL
,
455 IB_EVENT_PORT_ACTIVE
,
458 IB_EVENT_PKEY_CHANGE
,
461 IB_EVENT_SRQ_LIMIT_REACHED
,
462 IB_EVENT_QP_LAST_WQE_REACHED
,
463 IB_EVENT_CLIENT_REREGISTER
,
467 __attribute_const__
const char *ib_event_msg(enum ib_event_type event
);
470 struct ib_device
*device
;
477 enum ib_event_type event
;
480 struct ib_event_handler
{
481 struct ib_device
*device
;
482 void (*handler
)(struct ib_event_handler
*, struct ib_event
*);
483 struct list_head list
;
486 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
488 (_ptr)->device = _device; \
489 (_ptr)->handler = _handler; \
490 INIT_LIST_HEAD(&(_ptr)->list); \
493 struct ib_global_route
{
502 __be32 version_tclass_flow
;
511 IB_MULTICAST_QPN
= 0xffffff
514 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
521 IB_RATE_PORT_CURRENT
= 0,
522 IB_RATE_2_5_GBPS
= 2,
530 IB_RATE_120_GBPS
= 10,
531 IB_RATE_14_GBPS
= 11,
532 IB_RATE_56_GBPS
= 12,
533 IB_RATE_112_GBPS
= 13,
534 IB_RATE_168_GBPS
= 14,
535 IB_RATE_25_GBPS
= 15,
536 IB_RATE_100_GBPS
= 16,
537 IB_RATE_200_GBPS
= 17,
538 IB_RATE_300_GBPS
= 18
542 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
543 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
544 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
545 * @rate: rate to convert.
547 __attribute_const__
int ib_rate_to_mult(enum ib_rate rate
);
550 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
551 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
552 * @rate: rate to convert.
554 __attribute_const__
int ib_rate_to_mbps(enum ib_rate rate
);
556 enum ib_mr_create_flags
{
557 IB_MR_SIGNATURE_EN
= 1,
561 * ib_mr_init_attr - Memory region init attributes passed to routine
563 * @max_reg_descriptors: max number of registration descriptors that
564 * may be used with registration work requests.
565 * @flags: MR creation flags bit mask.
567 struct ib_mr_init_attr
{
568 int max_reg_descriptors
;
574 * IB_SIG_TYPE_NONE: Unprotected.
575 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
577 enum ib_signature_type
{
583 * Signature T10-DIF block-guard types
584 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
585 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
587 enum ib_t10_dif_bg_type
{
593 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
595 * @bg_type: T10-DIF block guard type (CRC|CSUM)
596 * @pi_interval: protection information interval.
597 * @bg: seed of guard computation.
598 * @app_tag: application tag of guard block
599 * @ref_tag: initial guard block reference tag.
600 * @ref_remap: Indicate wethear the reftag increments each block
601 * @app_escape: Indicate to skip block check if apptag=0xffff
602 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
603 * @apptag_check_mask: check bitmask of application tag.
605 struct ib_t10_dif_domain
{
606 enum ib_t10_dif_bg_type bg_type
;
614 u16 apptag_check_mask
;
618 * struct ib_sig_domain - Parameters for signature domain
619 * @sig_type: specific signauture type
620 * @sig: union of all signature domain attributes that may
621 * be used to set domain layout.
623 struct ib_sig_domain
{
624 enum ib_signature_type sig_type
;
626 struct ib_t10_dif_domain dif
;
631 * struct ib_sig_attrs - Parameters for signature handover operation
632 * @check_mask: bitmask for signature byte check (8 bytes)
633 * @mem: memory domain layout desciptor.
634 * @wire: wire domain layout desciptor.
636 struct ib_sig_attrs
{
638 struct ib_sig_domain mem
;
639 struct ib_sig_domain wire
;
642 enum ib_sig_err_type
{
649 * struct ib_sig_err - signature error descriptor
652 enum ib_sig_err_type err_type
;
659 enum ib_mr_status_check
{
660 IB_MR_CHECK_SIG_STATUS
= 1,
664 * struct ib_mr_status - Memory region status container
666 * @fail_status: Bitmask of MR checks status. For each
667 * failed check a corresponding status bit is set.
668 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
671 struct ib_mr_status
{
673 struct ib_sig_err sig_err
;
677 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
679 * @mult: multiple to convert.
681 __attribute_const__
enum ib_rate
mult_to_ib_rate(int mult
);
684 struct ib_global_route grh
;
699 IB_WC_LOC_EEC_OP_ERR
,
704 IB_WC_LOC_ACCESS_ERR
,
705 IB_WC_REM_INV_REQ_ERR
,
706 IB_WC_REM_ACCESS_ERR
,
709 IB_WC_RNR_RETRY_EXC_ERR
,
710 IB_WC_LOC_RDD_VIOL_ERR
,
711 IB_WC_REM_INV_RD_REQ_ERR
,
714 IB_WC_INV_EEC_STATE_ERR
,
716 IB_WC_RESP_TIMEOUT_ERR
,
720 __attribute_const__
const char *ib_wc_status_msg(enum ib_wc_status status
);
732 IB_WC_MASKED_COMP_SWAP
,
733 IB_WC_MASKED_FETCH_ADD
,
735 * Set value of IB_WC_RECV so consumers can test if a completion is a
736 * receive by testing (opcode & IB_WC_RECV).
739 IB_WC_RECV_RDMA_WITH_IMM
744 IB_WC_WITH_IMM
= (1<<1),
745 IB_WC_WITH_INVALIDATE
= (1<<2),
746 IB_WC_IP_CSUM_OK
= (1<<3),
747 IB_WC_WITH_SMAC
= (1<<4),
748 IB_WC_WITH_VLAN
= (1<<5),
753 enum ib_wc_status status
;
754 enum ib_wc_opcode opcode
;
768 u8 port_num
; /* valid only for DR SMPs on switches */
773 enum ib_cq_notify_flags
{
774 IB_CQ_SOLICITED
= 1 << 0,
775 IB_CQ_NEXT_COMP
= 1 << 1,
776 IB_CQ_SOLICITED_MASK
= IB_CQ_SOLICITED
| IB_CQ_NEXT_COMP
,
777 IB_CQ_REPORT_MISSED_EVENTS
= 1 << 2,
785 enum ib_srq_attr_mask
{
786 IB_SRQ_MAX_WR
= 1 << 0,
787 IB_SRQ_LIMIT
= 1 << 1,
796 struct ib_srq_init_attr
{
797 void (*event_handler
)(struct ib_event
*, void *);
799 struct ib_srq_attr attr
;
800 enum ib_srq_type srq_type
;
804 struct ib_xrcd
*xrcd
;
825 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
826 * here (and in that order) since the MAD layer uses them as
827 * indices into a 2-entry table.
836 IB_QPT_RAW_ETHERTYPE
,
837 IB_QPT_RAW_PACKET
= 8,
841 /* Reserve a range for qp types internal to the low level driver.
842 * These qp types will not be visible at the IB core layer, so the
843 * IB_QPT_MAX usages should not be affected in the core layer
845 IB_QPT_RESERVED1
= 0x1000,
857 enum ib_qp_create_flags
{
858 IB_QP_CREATE_IPOIB_UD_LSO
= 1 << 0,
859 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK
= 1 << 1,
860 IB_QP_CREATE_NETIF_QP
= 1 << 5,
861 IB_QP_CREATE_SIGNATURE_EN
= 1 << 6,
862 IB_QP_CREATE_USE_GFP_NOIO
= 1 << 7,
863 /* reserve bits 26-31 for low level drivers' internal use */
864 IB_QP_CREATE_RESERVED_START
= 1 << 26,
865 IB_QP_CREATE_RESERVED_END
= 1 << 31,
870 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
871 * callback to destroy the passed in QP.
874 struct ib_qp_init_attr
{
875 void (*event_handler
)(struct ib_event
*, void *);
877 struct ib_cq
*send_cq
;
878 struct ib_cq
*recv_cq
;
880 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
881 struct ib_qp_cap cap
;
882 enum ib_sig_type sq_sig_type
;
883 enum ib_qp_type qp_type
;
884 enum ib_qp_create_flags create_flags
;
885 u8 port_num
; /* special QP types only */
888 struct ib_qp_open_attr
{
889 void (*event_handler
)(struct ib_event
*, void *);
892 enum ib_qp_type qp_type
;
895 enum ib_rnr_timeout
{
896 IB_RNR_TIMER_655_36
= 0,
897 IB_RNR_TIMER_000_01
= 1,
898 IB_RNR_TIMER_000_02
= 2,
899 IB_RNR_TIMER_000_03
= 3,
900 IB_RNR_TIMER_000_04
= 4,
901 IB_RNR_TIMER_000_06
= 5,
902 IB_RNR_TIMER_000_08
= 6,
903 IB_RNR_TIMER_000_12
= 7,
904 IB_RNR_TIMER_000_16
= 8,
905 IB_RNR_TIMER_000_24
= 9,
906 IB_RNR_TIMER_000_32
= 10,
907 IB_RNR_TIMER_000_48
= 11,
908 IB_RNR_TIMER_000_64
= 12,
909 IB_RNR_TIMER_000_96
= 13,
910 IB_RNR_TIMER_001_28
= 14,
911 IB_RNR_TIMER_001_92
= 15,
912 IB_RNR_TIMER_002_56
= 16,
913 IB_RNR_TIMER_003_84
= 17,
914 IB_RNR_TIMER_005_12
= 18,
915 IB_RNR_TIMER_007_68
= 19,
916 IB_RNR_TIMER_010_24
= 20,
917 IB_RNR_TIMER_015_36
= 21,
918 IB_RNR_TIMER_020_48
= 22,
919 IB_RNR_TIMER_030_72
= 23,
920 IB_RNR_TIMER_040_96
= 24,
921 IB_RNR_TIMER_061_44
= 25,
922 IB_RNR_TIMER_081_92
= 26,
923 IB_RNR_TIMER_122_88
= 27,
924 IB_RNR_TIMER_163_84
= 28,
925 IB_RNR_TIMER_245_76
= 29,
926 IB_RNR_TIMER_327_68
= 30,
927 IB_RNR_TIMER_491_52
= 31
930 enum ib_qp_attr_mask
{
932 IB_QP_CUR_STATE
= (1<<1),
933 IB_QP_EN_SQD_ASYNC_NOTIFY
= (1<<2),
934 IB_QP_ACCESS_FLAGS
= (1<<3),
935 IB_QP_PKEY_INDEX
= (1<<4),
939 IB_QP_PATH_MTU
= (1<<8),
940 IB_QP_TIMEOUT
= (1<<9),
941 IB_QP_RETRY_CNT
= (1<<10),
942 IB_QP_RNR_RETRY
= (1<<11),
943 IB_QP_RQ_PSN
= (1<<12),
944 IB_QP_MAX_QP_RD_ATOMIC
= (1<<13),
945 IB_QP_ALT_PATH
= (1<<14),
946 IB_QP_MIN_RNR_TIMER
= (1<<15),
947 IB_QP_SQ_PSN
= (1<<16),
948 IB_QP_MAX_DEST_RD_ATOMIC
= (1<<17),
949 IB_QP_PATH_MIG_STATE
= (1<<18),
951 IB_QP_DEST_QPN
= (1<<20),
952 IB_QP_SMAC
= (1<<21),
953 IB_QP_ALT_SMAC
= (1<<22),
955 IB_QP_ALT_VID
= (1<<24),
980 enum ib_qp_state qp_state
;
981 enum ib_qp_state cur_qp_state
;
982 enum ib_mtu path_mtu
;
983 enum ib_mig_state path_mig_state
;
989 struct ib_qp_cap cap
;
990 struct ib_ah_attr ah_attr
;
991 struct ib_ah_attr alt_ah_attr
;
994 u8 en_sqd_async_notify
;
997 u8 max_dest_rd_atomic
;
1006 u8 alt_smac
[ETH_ALEN
];
1013 IB_WR_RDMA_WRITE_WITH_IMM
,
1015 IB_WR_SEND_WITH_IMM
,
1017 IB_WR_ATOMIC_CMP_AND_SWP
,
1018 IB_WR_ATOMIC_FETCH_AND_ADD
,
1020 IB_WR_SEND_WITH_INV
,
1021 IB_WR_RDMA_READ_WITH_INV
,
1024 IB_WR_MASKED_ATOMIC_CMP_AND_SWP
,
1025 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD
,
1028 /* reserve values for low level drivers' internal use.
1029 * These values will not be used at all in the ib core layer.
1031 IB_WR_RESERVED1
= 0xf0,
1043 enum ib_send_flags
{
1045 IB_SEND_SIGNALED
= (1<<1),
1046 IB_SEND_SOLICITED
= (1<<2),
1047 IB_SEND_INLINE
= (1<<3),
1048 IB_SEND_IP_CSUM
= (1<<4),
1050 /* reserve bits 26-31 for low level drivers' internal use */
1051 IB_SEND_RESERVED_START
= (1 << 26),
1052 IB_SEND_RESERVED_END
= (1 << 31),
1061 struct ib_fast_reg_page_list
{
1062 struct ib_device
*device
;
1064 unsigned int max_page_list_len
;
1068 * struct ib_mw_bind_info - Parameters for a memory window bind operation.
1069 * @mr: A memory region to bind the memory window to.
1070 * @addr: The address where the memory window should begin.
1071 * @length: The length of the memory window, in bytes.
1072 * @mw_access_flags: Access flags from enum ib_access_flags for the window.
1074 * This struct contains the shared parameters for type 1 and type 2
1075 * memory window bind operations.
1077 struct ib_mw_bind_info
{
1081 int mw_access_flags
;
1085 struct ib_send_wr
*next
;
1087 struct ib_sge
*sg_list
;
1089 enum ib_wr_opcode opcode
;
1093 u32 invalidate_rkey
;
1104 u64 compare_add_mask
;
1115 u16 pkey_index
; /* valid for GSI only */
1116 u8 port_num
; /* valid for DR SMPs on switch only */
1120 struct ib_fast_reg_page_list
*page_list
;
1121 unsigned int page_shift
;
1122 unsigned int page_list_len
;
1129 /* The new rkey for the memory window. */
1131 struct ib_mw_bind_info bind_info
;
1134 struct ib_sig_attrs
*sig_attrs
;
1135 struct ib_mr
*sig_mr
;
1137 struct ib_sge
*prot
;
1140 u32 xrc_remote_srq_num
; /* XRC TGT QPs only */
1144 struct ib_recv_wr
*next
;
1146 struct ib_sge
*sg_list
;
1150 enum ib_access_flags
{
1151 IB_ACCESS_LOCAL_WRITE
= 1,
1152 IB_ACCESS_REMOTE_WRITE
= (1<<1),
1153 IB_ACCESS_REMOTE_READ
= (1<<2),
1154 IB_ACCESS_REMOTE_ATOMIC
= (1<<3),
1155 IB_ACCESS_MW_BIND
= (1<<4),
1156 IB_ZERO_BASED
= (1<<5),
1157 IB_ACCESS_ON_DEMAND
= (1<<6),
1160 struct ib_phys_buf
{
1167 u64 device_virt_addr
;
1169 int mr_access_flags
;
1174 enum ib_mr_rereg_flags
{
1175 IB_MR_REREG_TRANS
= 1,
1176 IB_MR_REREG_PD
= (1<<1),
1177 IB_MR_REREG_ACCESS
= (1<<2),
1178 IB_MR_REREG_SUPPORTED
= ((IB_MR_REREG_ACCESS
<< 1) - 1)
1182 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation.
1183 * @wr_id: Work request id.
1184 * @send_flags: Flags from ib_send_flags enum.
1185 * @bind_info: More parameters of the bind operation.
1190 struct ib_mw_bind_info bind_info
;
1193 struct ib_fmr_attr
{
1201 struct ib_ucontext
{
1202 struct ib_device
*device
;
1203 struct list_head pd_list
;
1204 struct list_head mr_list
;
1205 struct list_head mw_list
;
1206 struct list_head cq_list
;
1207 struct list_head qp_list
;
1208 struct list_head srq_list
;
1209 struct list_head ah_list
;
1210 struct list_head xrcd_list
;
1211 struct list_head rule_list
;
1215 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1216 struct rb_root umem_tree
;
1218 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1219 * mmu notifiers registration.
1221 struct rw_semaphore umem_rwsem
;
1222 void (*invalidate_range
)(struct ib_umem
*umem
,
1223 unsigned long start
, unsigned long end
);
1225 struct mmu_notifier mn
;
1226 atomic_t notifier_count
;
1227 /* A list of umems that don't have private mmu notifier counters yet. */
1228 struct list_head no_private_counters
;
1234 u64 user_handle
; /* handle given to us by userspace */
1235 struct ib_ucontext
*context
; /* associated user context */
1236 void *object
; /* containing object */
1237 struct list_head list
; /* link to context's list */
1238 int id
; /* index into kernel idr */
1240 struct rw_semaphore mutex
; /* protects .live */
1245 const void __user
*inbuf
;
1246 void __user
*outbuf
;
1252 struct ib_device
*device
;
1253 struct ib_uobject
*uobject
;
1254 atomic_t usecnt
; /* count all resources */
1258 struct ib_device
*device
;
1259 atomic_t usecnt
; /* count all exposed resources */
1260 struct inode
*inode
;
1262 struct mutex tgt_qp_mutex
;
1263 struct list_head tgt_qp_list
;
1267 struct ib_device
*device
;
1269 struct ib_uobject
*uobject
;
1272 typedef void (*ib_comp_handler
)(struct ib_cq
*cq
, void *cq_context
);
1275 struct ib_device
*device
;
1276 struct ib_uobject
*uobject
;
1277 ib_comp_handler comp_handler
;
1278 void (*event_handler
)(struct ib_event
*, void *);
1281 atomic_t usecnt
; /* count number of work queues */
1285 struct ib_device
*device
;
1287 struct ib_uobject
*uobject
;
1288 void (*event_handler
)(struct ib_event
*, void *);
1290 enum ib_srq_type srq_type
;
1295 struct ib_xrcd
*xrcd
;
1303 struct ib_device
*device
;
1305 struct ib_cq
*send_cq
;
1306 struct ib_cq
*recv_cq
;
1308 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
1309 struct list_head xrcd_list
;
1310 /* count times opened, mcast attaches, flow attaches */
1312 struct list_head open_list
;
1313 struct ib_qp
*real_qp
;
1314 struct ib_uobject
*uobject
;
1315 void (*event_handler
)(struct ib_event
*, void *);
1318 enum ib_qp_type qp_type
;
1322 struct ib_device
*device
;
1324 struct ib_uobject
*uobject
;
1327 atomic_t usecnt
; /* count number of MWs */
1331 struct ib_device
*device
;
1333 struct ib_uobject
*uobject
;
1335 enum ib_mw_type type
;
1339 struct ib_device
*device
;
1341 struct list_head list
;
1346 /* Supported steering options */
1347 enum ib_flow_attr_type
{
1348 /* steering according to rule specifications */
1349 IB_FLOW_ATTR_NORMAL
= 0x0,
1350 /* default unicast and multicast rule -
1351 * receive all Eth traffic which isn't steered to any QP
1353 IB_FLOW_ATTR_ALL_DEFAULT
= 0x1,
1354 /* default multicast rule -
1355 * receive all Eth multicast traffic which isn't steered to any QP
1357 IB_FLOW_ATTR_MC_DEFAULT
= 0x2,
1358 /* sniffer rule - receive all port traffic */
1359 IB_FLOW_ATTR_SNIFFER
= 0x3
1362 /* Supported steering header types */
1363 enum ib_flow_spec_type
{
1365 IB_FLOW_SPEC_ETH
= 0x20,
1366 IB_FLOW_SPEC_IB
= 0x22,
1368 IB_FLOW_SPEC_IPV4
= 0x30,
1370 IB_FLOW_SPEC_TCP
= 0x40,
1371 IB_FLOW_SPEC_UDP
= 0x41
1373 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1374 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1376 /* Flow steering rule priority is set according to it's domain.
1377 * Lower domain value means higher priority.
1379 enum ib_flow_domain
{
1380 IB_FLOW_DOMAIN_USER
,
1381 IB_FLOW_DOMAIN_ETHTOOL
,
1384 IB_FLOW_DOMAIN_NUM
/* Must be last */
1387 struct ib_flow_eth_filter
{
1394 struct ib_flow_spec_eth
{
1395 enum ib_flow_spec_type type
;
1397 struct ib_flow_eth_filter val
;
1398 struct ib_flow_eth_filter mask
;
1401 struct ib_flow_ib_filter
{
1406 struct ib_flow_spec_ib
{
1407 enum ib_flow_spec_type type
;
1409 struct ib_flow_ib_filter val
;
1410 struct ib_flow_ib_filter mask
;
1413 struct ib_flow_ipv4_filter
{
1418 struct ib_flow_spec_ipv4
{
1419 enum ib_flow_spec_type type
;
1421 struct ib_flow_ipv4_filter val
;
1422 struct ib_flow_ipv4_filter mask
;
1425 struct ib_flow_tcp_udp_filter
{
1430 struct ib_flow_spec_tcp_udp
{
1431 enum ib_flow_spec_type type
;
1433 struct ib_flow_tcp_udp_filter val
;
1434 struct ib_flow_tcp_udp_filter mask
;
1437 union ib_flow_spec
{
1439 enum ib_flow_spec_type type
;
1442 struct ib_flow_spec_eth eth
;
1443 struct ib_flow_spec_ib ib
;
1444 struct ib_flow_spec_ipv4 ipv4
;
1445 struct ib_flow_spec_tcp_udp tcp_udp
;
1448 struct ib_flow_attr
{
1449 enum ib_flow_attr_type type
;
1455 /* Following are the optional layers according to user request
1456 * struct ib_flow_spec_xxx
1457 * struct ib_flow_spec_yyy
1463 struct ib_uobject
*uobject
;
1469 enum ib_process_mad_flags
{
1470 IB_MAD_IGNORE_MKEY
= 1,
1471 IB_MAD_IGNORE_BKEY
= 2,
1472 IB_MAD_IGNORE_ALL
= IB_MAD_IGNORE_MKEY
| IB_MAD_IGNORE_BKEY
1475 enum ib_mad_result
{
1476 IB_MAD_RESULT_FAILURE
= 0, /* (!SUCCESS is the important flag) */
1477 IB_MAD_RESULT_SUCCESS
= 1 << 0, /* MAD was successfully processed */
1478 IB_MAD_RESULT_REPLY
= 1 << 1, /* Reply packet needs to be sent */
1479 IB_MAD_RESULT_CONSUMED
= 1 << 2 /* Packet consumed: stop processing */
1482 #define IB_DEVICE_NAME_MAX 64
1486 struct ib_event_handler event_handler
;
1487 struct ib_pkey_cache
**pkey_cache
;
1488 struct ib_gid_cache
**gid_cache
;
1492 struct ib_dma_mapping_ops
{
1493 int (*mapping_error
)(struct ib_device
*dev
,
1495 u64 (*map_single
)(struct ib_device
*dev
,
1496 void *ptr
, size_t size
,
1497 enum dma_data_direction direction
);
1498 void (*unmap_single
)(struct ib_device
*dev
,
1499 u64 addr
, size_t size
,
1500 enum dma_data_direction direction
);
1501 u64 (*map_page
)(struct ib_device
*dev
,
1502 struct page
*page
, unsigned long offset
,
1504 enum dma_data_direction direction
);
1505 void (*unmap_page
)(struct ib_device
*dev
,
1506 u64 addr
, size_t size
,
1507 enum dma_data_direction direction
);
1508 int (*map_sg
)(struct ib_device
*dev
,
1509 struct scatterlist
*sg
, int nents
,
1510 enum dma_data_direction direction
);
1511 void (*unmap_sg
)(struct ib_device
*dev
,
1512 struct scatterlist
*sg
, int nents
,
1513 enum dma_data_direction direction
);
1514 void (*sync_single_for_cpu
)(struct ib_device
*dev
,
1517 enum dma_data_direction dir
);
1518 void (*sync_single_for_device
)(struct ib_device
*dev
,
1521 enum dma_data_direction dir
);
1522 void *(*alloc_coherent
)(struct ib_device
*dev
,
1526 void (*free_coherent
)(struct ib_device
*dev
,
1527 size_t size
, void *cpu_addr
,
1533 struct ib_port_immutable
{
1541 struct device
*dma_device
;
1543 char name
[IB_DEVICE_NAME_MAX
];
1545 struct list_head event_handler_list
;
1546 spinlock_t event_handler_lock
;
1548 spinlock_t client_data_lock
;
1549 struct list_head core_list
;
1550 struct list_head client_data_list
;
1552 struct ib_cache cache
;
1554 * port_immutable is indexed by port number
1556 struct ib_port_immutable
*port_immutable
;
1558 int num_comp_vectors
;
1560 struct iw_cm_verbs
*iwcm
;
1562 int (*get_protocol_stats
)(struct ib_device
*device
,
1563 union rdma_protocol_stats
*stats
);
1564 int (*query_device
)(struct ib_device
*device
,
1565 struct ib_device_attr
*device_attr
,
1566 struct ib_udata
*udata
);
1567 int (*query_port
)(struct ib_device
*device
,
1569 struct ib_port_attr
*port_attr
);
1570 enum rdma_link_layer (*get_link_layer
)(struct ib_device
*device
,
1572 int (*query_gid
)(struct ib_device
*device
,
1573 u8 port_num
, int index
,
1575 int (*query_pkey
)(struct ib_device
*device
,
1576 u8 port_num
, u16 index
, u16
*pkey
);
1577 int (*modify_device
)(struct ib_device
*device
,
1578 int device_modify_mask
,
1579 struct ib_device_modify
*device_modify
);
1580 int (*modify_port
)(struct ib_device
*device
,
1581 u8 port_num
, int port_modify_mask
,
1582 struct ib_port_modify
*port_modify
);
1583 struct ib_ucontext
* (*alloc_ucontext
)(struct ib_device
*device
,
1584 struct ib_udata
*udata
);
1585 int (*dealloc_ucontext
)(struct ib_ucontext
*context
);
1586 int (*mmap
)(struct ib_ucontext
*context
,
1587 struct vm_area_struct
*vma
);
1588 struct ib_pd
* (*alloc_pd
)(struct ib_device
*device
,
1589 struct ib_ucontext
*context
,
1590 struct ib_udata
*udata
);
1591 int (*dealloc_pd
)(struct ib_pd
*pd
);
1592 struct ib_ah
* (*create_ah
)(struct ib_pd
*pd
,
1593 struct ib_ah_attr
*ah_attr
);
1594 int (*modify_ah
)(struct ib_ah
*ah
,
1595 struct ib_ah_attr
*ah_attr
);
1596 int (*query_ah
)(struct ib_ah
*ah
,
1597 struct ib_ah_attr
*ah_attr
);
1598 int (*destroy_ah
)(struct ib_ah
*ah
);
1599 struct ib_srq
* (*create_srq
)(struct ib_pd
*pd
,
1600 struct ib_srq_init_attr
*srq_init_attr
,
1601 struct ib_udata
*udata
);
1602 int (*modify_srq
)(struct ib_srq
*srq
,
1603 struct ib_srq_attr
*srq_attr
,
1604 enum ib_srq_attr_mask srq_attr_mask
,
1605 struct ib_udata
*udata
);
1606 int (*query_srq
)(struct ib_srq
*srq
,
1607 struct ib_srq_attr
*srq_attr
);
1608 int (*destroy_srq
)(struct ib_srq
*srq
);
1609 int (*post_srq_recv
)(struct ib_srq
*srq
,
1610 struct ib_recv_wr
*recv_wr
,
1611 struct ib_recv_wr
**bad_recv_wr
);
1612 struct ib_qp
* (*create_qp
)(struct ib_pd
*pd
,
1613 struct ib_qp_init_attr
*qp_init_attr
,
1614 struct ib_udata
*udata
);
1615 int (*modify_qp
)(struct ib_qp
*qp
,
1616 struct ib_qp_attr
*qp_attr
,
1618 struct ib_udata
*udata
);
1619 int (*query_qp
)(struct ib_qp
*qp
,
1620 struct ib_qp_attr
*qp_attr
,
1622 struct ib_qp_init_attr
*qp_init_attr
);
1623 int (*destroy_qp
)(struct ib_qp
*qp
);
1624 int (*post_send
)(struct ib_qp
*qp
,
1625 struct ib_send_wr
*send_wr
,
1626 struct ib_send_wr
**bad_send_wr
);
1627 int (*post_recv
)(struct ib_qp
*qp
,
1628 struct ib_recv_wr
*recv_wr
,
1629 struct ib_recv_wr
**bad_recv_wr
);
1630 struct ib_cq
* (*create_cq
)(struct ib_device
*device
,
1631 const struct ib_cq_init_attr
*attr
,
1632 struct ib_ucontext
*context
,
1633 struct ib_udata
*udata
);
1634 int (*modify_cq
)(struct ib_cq
*cq
, u16 cq_count
,
1636 int (*destroy_cq
)(struct ib_cq
*cq
);
1637 int (*resize_cq
)(struct ib_cq
*cq
, int cqe
,
1638 struct ib_udata
*udata
);
1639 int (*poll_cq
)(struct ib_cq
*cq
, int num_entries
,
1641 int (*peek_cq
)(struct ib_cq
*cq
, int wc_cnt
);
1642 int (*req_notify_cq
)(struct ib_cq
*cq
,
1643 enum ib_cq_notify_flags flags
);
1644 int (*req_ncomp_notif
)(struct ib_cq
*cq
,
1646 struct ib_mr
* (*get_dma_mr
)(struct ib_pd
*pd
,
1647 int mr_access_flags
);
1648 struct ib_mr
* (*reg_phys_mr
)(struct ib_pd
*pd
,
1649 struct ib_phys_buf
*phys_buf_array
,
1651 int mr_access_flags
,
1653 struct ib_mr
* (*reg_user_mr
)(struct ib_pd
*pd
,
1654 u64 start
, u64 length
,
1656 int mr_access_flags
,
1657 struct ib_udata
*udata
);
1658 int (*rereg_user_mr
)(struct ib_mr
*mr
,
1660 u64 start
, u64 length
,
1662 int mr_access_flags
,
1664 struct ib_udata
*udata
);
1665 int (*query_mr
)(struct ib_mr
*mr
,
1666 struct ib_mr_attr
*mr_attr
);
1667 int (*dereg_mr
)(struct ib_mr
*mr
);
1668 int (*destroy_mr
)(struct ib_mr
*mr
);
1669 struct ib_mr
* (*create_mr
)(struct ib_pd
*pd
,
1670 struct ib_mr_init_attr
*mr_init_attr
);
1671 struct ib_mr
* (*alloc_fast_reg_mr
)(struct ib_pd
*pd
,
1672 int max_page_list_len
);
1673 struct ib_fast_reg_page_list
* (*alloc_fast_reg_page_list
)(struct ib_device
*device
,
1675 void (*free_fast_reg_page_list
)(struct ib_fast_reg_page_list
*page_list
);
1676 int (*rereg_phys_mr
)(struct ib_mr
*mr
,
1679 struct ib_phys_buf
*phys_buf_array
,
1681 int mr_access_flags
,
1683 struct ib_mw
* (*alloc_mw
)(struct ib_pd
*pd
,
1684 enum ib_mw_type type
);
1685 int (*bind_mw
)(struct ib_qp
*qp
,
1687 struct ib_mw_bind
*mw_bind
);
1688 int (*dealloc_mw
)(struct ib_mw
*mw
);
1689 struct ib_fmr
* (*alloc_fmr
)(struct ib_pd
*pd
,
1690 int mr_access_flags
,
1691 struct ib_fmr_attr
*fmr_attr
);
1692 int (*map_phys_fmr
)(struct ib_fmr
*fmr
,
1693 u64
*page_list
, int list_len
,
1695 int (*unmap_fmr
)(struct list_head
*fmr_list
);
1696 int (*dealloc_fmr
)(struct ib_fmr
*fmr
);
1697 int (*attach_mcast
)(struct ib_qp
*qp
,
1700 int (*detach_mcast
)(struct ib_qp
*qp
,
1703 int (*process_mad
)(struct ib_device
*device
,
1704 int process_mad_flags
,
1706 const struct ib_wc
*in_wc
,
1707 const struct ib_grh
*in_grh
,
1708 const struct ib_mad_hdr
*in_mad
,
1710 struct ib_mad_hdr
*out_mad
,
1711 size_t *out_mad_size
,
1712 u16
*out_mad_pkey_index
);
1713 struct ib_xrcd
* (*alloc_xrcd
)(struct ib_device
*device
,
1714 struct ib_ucontext
*ucontext
,
1715 struct ib_udata
*udata
);
1716 int (*dealloc_xrcd
)(struct ib_xrcd
*xrcd
);
1717 struct ib_flow
* (*create_flow
)(struct ib_qp
*qp
,
1721 int (*destroy_flow
)(struct ib_flow
*flow_id
);
1722 int (*check_mr_status
)(struct ib_mr
*mr
, u32 check_mask
,
1723 struct ib_mr_status
*mr_status
);
1725 struct ib_dma_mapping_ops
*dma_ops
;
1727 struct module
*owner
;
1729 struct kobject
*ports_parent
;
1730 struct list_head port_list
;
1733 IB_DEV_UNINITIALIZED
,
1739 u64 uverbs_cmd_mask
;
1740 u64 uverbs_ex_cmd_mask
;
1749 * The following mandatory functions are used only at device
1750 * registration. Keep functions such as these at the end of this
1751 * structure to avoid cache line misses when accessing struct ib_device
1754 int (*get_port_immutable
)(struct ib_device
*, u8
, struct ib_port_immutable
*);
1759 void (*add
) (struct ib_device
*);
1760 void (*remove
)(struct ib_device
*);
1762 struct list_head list
;
1765 struct ib_device
*ib_alloc_device(size_t size
);
1766 void ib_dealloc_device(struct ib_device
*device
);
1768 int ib_register_device(struct ib_device
*device
,
1769 int (*port_callback
)(struct ib_device
*,
1770 u8
, struct kobject
*));
1771 void ib_unregister_device(struct ib_device
*device
);
1773 int ib_register_client (struct ib_client
*client
);
1774 void ib_unregister_client(struct ib_client
*client
);
1776 void *ib_get_client_data(struct ib_device
*device
, struct ib_client
*client
);
1777 void ib_set_client_data(struct ib_device
*device
, struct ib_client
*client
,
1780 static inline int ib_copy_from_udata(void *dest
, struct ib_udata
*udata
, size_t len
)
1782 return copy_from_user(dest
, udata
->inbuf
, len
) ? -EFAULT
: 0;
1785 static inline int ib_copy_to_udata(struct ib_udata
*udata
, void *src
, size_t len
)
1787 return copy_to_user(udata
->outbuf
, src
, len
) ? -EFAULT
: 0;
1791 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1792 * contains all required attributes and no attributes not allowed for
1793 * the given QP state transition.
1794 * @cur_state: Current QP state
1795 * @next_state: Next QP state
1797 * @mask: Mask of supplied QP attributes
1798 * @ll : link layer of port
1800 * This function is a helper function that a low-level driver's
1801 * modify_qp method can use to validate the consumer's input. It
1802 * checks that cur_state and next_state are valid QP states, that a
1803 * transition from cur_state to next_state is allowed by the IB spec,
1804 * and that the attribute mask supplied is allowed for the transition.
1806 int ib_modify_qp_is_ok(enum ib_qp_state cur_state
, enum ib_qp_state next_state
,
1807 enum ib_qp_type type
, enum ib_qp_attr_mask mask
,
1808 enum rdma_link_layer ll
);
1810 int ib_register_event_handler (struct ib_event_handler
*event_handler
);
1811 int ib_unregister_event_handler(struct ib_event_handler
*event_handler
);
1812 void ib_dispatch_event(struct ib_event
*event
);
1814 int ib_query_device(struct ib_device
*device
,
1815 struct ib_device_attr
*device_attr
);
1817 int ib_query_port(struct ib_device
*device
,
1818 u8 port_num
, struct ib_port_attr
*port_attr
);
1820 enum rdma_link_layer
rdma_port_get_link_layer(struct ib_device
*device
,
1824 * rdma_start_port - Return the first valid port number for the device
1827 * @device: Device to be checked
1829 * Return start port number
1831 static inline u8
rdma_start_port(const struct ib_device
*device
)
1833 return (device
->node_type
== RDMA_NODE_IB_SWITCH
) ? 0 : 1;
1837 * rdma_end_port - Return the last valid port number for the device
1840 * @device: Device to be checked
1842 * Return last port number
1844 static inline u8
rdma_end_port(const struct ib_device
*device
)
1846 return (device
->node_type
== RDMA_NODE_IB_SWITCH
) ?
1847 0 : device
->phys_port_cnt
;
1850 static inline bool rdma_protocol_ib(const struct ib_device
*device
, u8 port_num
)
1852 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IB
;
1855 static inline bool rdma_protocol_roce(const struct ib_device
*device
, u8 port_num
)
1857 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_ROCE
;
1860 static inline bool rdma_protocol_iwarp(const struct ib_device
*device
, u8 port_num
)
1862 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IWARP
;
1865 static inline bool rdma_ib_or_roce(const struct ib_device
*device
, u8 port_num
)
1867 return device
->port_immutable
[port_num
].core_cap_flags
&
1868 (RDMA_CORE_CAP_PROT_IB
| RDMA_CORE_CAP_PROT_ROCE
);
1872 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
1873 * Management Datagrams.
1874 * @device: Device to check
1875 * @port_num: Port number to check
1877 * Management Datagrams (MAD) are a required part of the InfiniBand
1878 * specification and are supported on all InfiniBand devices. A slightly
1879 * extended version are also supported on OPA interfaces.
1881 * Return: true if the port supports sending/receiving of MAD packets.
1883 static inline bool rdma_cap_ib_mad(const struct ib_device
*device
, u8 port_num
)
1885 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_MAD
;
1889 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
1890 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
1891 * @device: Device to check
1892 * @port_num: Port number to check
1894 * Each InfiniBand node is required to provide a Subnet Management Agent
1895 * that the subnet manager can access. Prior to the fabric being fully
1896 * configured by the subnet manager, the SMA is accessed via a well known
1897 * interface called the Subnet Management Interface (SMI). This interface
1898 * uses directed route packets to communicate with the SM to get around the
1899 * chicken and egg problem of the SM needing to know what's on the fabric
1900 * in order to configure the fabric, and needing to configure the fabric in
1901 * order to send packets to the devices on the fabric. These directed
1902 * route packets do not need the fabric fully configured in order to reach
1903 * their destination. The SMI is the only method allowed to send
1904 * directed route packets on an InfiniBand fabric.
1906 * Return: true if the port provides an SMI.
1908 static inline bool rdma_cap_ib_smi(const struct ib_device
*device
, u8 port_num
)
1910 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SMI
;
1914 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
1915 * Communication Manager.
1916 * @device: Device to check
1917 * @port_num: Port number to check
1919 * The InfiniBand Communication Manager is one of many pre-defined General
1920 * Service Agents (GSA) that are accessed via the General Service
1921 * Interface (GSI). It's role is to facilitate establishment of connections
1922 * between nodes as well as other management related tasks for established
1925 * Return: true if the port supports an IB CM (this does not guarantee that
1926 * a CM is actually running however).
1928 static inline bool rdma_cap_ib_cm(const struct ib_device
*device
, u8 port_num
)
1930 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_CM
;
1934 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
1935 * Communication Manager.
1936 * @device: Device to check
1937 * @port_num: Port number to check
1939 * Similar to above, but specific to iWARP connections which have a different
1940 * managment protocol than InfiniBand.
1942 * Return: true if the port supports an iWARP CM (this does not guarantee that
1943 * a CM is actually running however).
1945 static inline bool rdma_cap_iw_cm(const struct ib_device
*device
, u8 port_num
)
1947 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IW_CM
;
1951 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
1952 * Subnet Administration.
1953 * @device: Device to check
1954 * @port_num: Port number to check
1956 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
1957 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
1958 * fabrics, devices should resolve routes to other hosts by contacting the
1959 * SA to query the proper route.
1961 * Return: true if the port should act as a client to the fabric Subnet
1962 * Administration interface. This does not imply that the SA service is
1965 static inline bool rdma_cap_ib_sa(const struct ib_device
*device
, u8 port_num
)
1967 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SA
;
1971 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
1973 * @device: Device to check
1974 * @port_num: Port number to check
1976 * InfiniBand multicast registration is more complex than normal IPv4 or
1977 * IPv6 multicast registration. Each Host Channel Adapter must register
1978 * with the Subnet Manager when it wishes to join a multicast group. It
1979 * should do so only once regardless of how many queue pairs it subscribes
1980 * to this group. And it should leave the group only after all queue pairs
1981 * attached to the group have been detached.
1983 * Return: true if the port must undertake the additional adminstrative
1984 * overhead of registering/unregistering with the SM and tracking of the
1985 * total number of queue pairs attached to the multicast group.
1987 static inline bool rdma_cap_ib_mcast(const struct ib_device
*device
, u8 port_num
)
1989 return rdma_cap_ib_sa(device
, port_num
);
1993 * rdma_cap_af_ib - Check if the port of device has the capability
1994 * Native Infiniband Address.
1995 * @device: Device to check
1996 * @port_num: Port number to check
1998 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
1999 * GID. RoCE uses a different mechanism, but still generates a GID via
2000 * a prescribed mechanism and port specific data.
2002 * Return: true if the port uses a GID address to identify devices on the
2005 static inline bool rdma_cap_af_ib(const struct ib_device
*device
, u8 port_num
)
2007 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_AF_IB
;
2011 * rdma_cap_eth_ah - Check if the port of device has the capability
2012 * Ethernet Address Handle.
2013 * @device: Device to check
2014 * @port_num: Port number to check
2016 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2017 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2018 * port. Normally, packet headers are generated by the sending host
2019 * adapter, but when sending connectionless datagrams, we must manually
2020 * inject the proper headers for the fabric we are communicating over.
2022 * Return: true if we are running as a RoCE port and must force the
2023 * addition of a Global Route Header built from our Ethernet Address
2024 * Handle into our header list for connectionless packets.
2026 static inline bool rdma_cap_eth_ah(const struct ib_device
*device
, u8 port_num
)
2028 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_ETH_AH
;
2032 * rdma_cap_read_multi_sge - Check if the port of device has the capability
2033 * RDMA Read Multiple Scatter-Gather Entries.
2034 * @device: Device to check
2035 * @port_num: Port number to check
2037 * iWARP has a restriction that RDMA READ requests may only have a single
2038 * Scatter/Gather Entry (SGE) in the work request.
2040 * NOTE: although the linux kernel currently assumes all devices are either
2041 * single SGE RDMA READ devices or identical SGE maximums for RDMA READs and
2042 * WRITEs, according to Tom Talpey, this is not accurate. There are some
2043 * devices out there that support more than a single SGE on RDMA READ
2044 * requests, but do not support the same number of SGEs as they do on
2045 * RDMA WRITE requests. The linux kernel would need rearchitecting to
2046 * support these imbalanced READ/WRITE SGEs allowed devices. So, for now,
2047 * suffice with either the device supports the same READ/WRITE SGEs, or
2048 * it only gets one READ sge.
2050 * Return: true for any device that allows more than one SGE in RDMA READ
2053 static inline bool rdma_cap_read_multi_sge(struct ib_device
*device
,
2056 return !(device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IWARP
);
2060 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2063 * @port_num: Port number
2065 * This MAD size includes the MAD headers and MAD payload. No other headers
2068 * Return the max MAD size required by the Port. Will return 0 if the port
2069 * does not support MADs
2071 static inline size_t rdma_max_mad_size(const struct ib_device
*device
, u8 port_num
)
2073 return device
->port_immutable
[port_num
].max_mad_size
;
2076 int ib_query_gid(struct ib_device
*device
,
2077 u8 port_num
, int index
, union ib_gid
*gid
);
2079 int ib_query_pkey(struct ib_device
*device
,
2080 u8 port_num
, u16 index
, u16
*pkey
);
2082 int ib_modify_device(struct ib_device
*device
,
2083 int device_modify_mask
,
2084 struct ib_device_modify
*device_modify
);
2086 int ib_modify_port(struct ib_device
*device
,
2087 u8 port_num
, int port_modify_mask
,
2088 struct ib_port_modify
*port_modify
);
2090 int ib_find_gid(struct ib_device
*device
, union ib_gid
*gid
,
2091 u8
*port_num
, u16
*index
);
2093 int ib_find_pkey(struct ib_device
*device
,
2094 u8 port_num
, u16 pkey
, u16
*index
);
2097 * ib_alloc_pd - Allocates an unused protection domain.
2098 * @device: The device on which to allocate the protection domain.
2100 * A protection domain object provides an association between QPs, shared
2101 * receive queues, address handles, memory regions, and memory windows.
2103 struct ib_pd
*ib_alloc_pd(struct ib_device
*device
);
2106 * ib_dealloc_pd - Deallocates a protection domain.
2107 * @pd: The protection domain to deallocate.
2109 int ib_dealloc_pd(struct ib_pd
*pd
);
2112 * ib_create_ah - Creates an address handle for the given address vector.
2113 * @pd: The protection domain associated with the address handle.
2114 * @ah_attr: The attributes of the address vector.
2116 * The address handle is used to reference a local or global destination
2117 * in all UD QP post sends.
2119 struct ib_ah
*ib_create_ah(struct ib_pd
*pd
, struct ib_ah_attr
*ah_attr
);
2122 * ib_init_ah_from_wc - Initializes address handle attributes from a
2124 * @device: Device on which the received message arrived.
2125 * @port_num: Port on which the received message arrived.
2126 * @wc: Work completion associated with the received message.
2127 * @grh: References the received global route header. This parameter is
2128 * ignored unless the work completion indicates that the GRH is valid.
2129 * @ah_attr: Returned attributes that can be used when creating an address
2130 * handle for replying to the message.
2132 int ib_init_ah_from_wc(struct ib_device
*device
, u8 port_num
,
2133 const struct ib_wc
*wc
, const struct ib_grh
*grh
,
2134 struct ib_ah_attr
*ah_attr
);
2137 * ib_create_ah_from_wc - Creates an address handle associated with the
2138 * sender of the specified work completion.
2139 * @pd: The protection domain associated with the address handle.
2140 * @wc: Work completion information associated with a received message.
2141 * @grh: References the received global route header. This parameter is
2142 * ignored unless the work completion indicates that the GRH is valid.
2143 * @port_num: The outbound port number to associate with the address.
2145 * The address handle is used to reference a local or global destination
2146 * in all UD QP post sends.
2148 struct ib_ah
*ib_create_ah_from_wc(struct ib_pd
*pd
, const struct ib_wc
*wc
,
2149 const struct ib_grh
*grh
, u8 port_num
);
2152 * ib_modify_ah - Modifies the address vector associated with an address
2154 * @ah: The address handle to modify.
2155 * @ah_attr: The new address vector attributes to associate with the
2158 int ib_modify_ah(struct ib_ah
*ah
, struct ib_ah_attr
*ah_attr
);
2161 * ib_query_ah - Queries the address vector associated with an address
2163 * @ah: The address handle to query.
2164 * @ah_attr: The address vector attributes associated with the address
2167 int ib_query_ah(struct ib_ah
*ah
, struct ib_ah_attr
*ah_attr
);
2170 * ib_destroy_ah - Destroys an address handle.
2171 * @ah: The address handle to destroy.
2173 int ib_destroy_ah(struct ib_ah
*ah
);
2176 * ib_create_srq - Creates a SRQ associated with the specified protection
2178 * @pd: The protection domain associated with the SRQ.
2179 * @srq_init_attr: A list of initial attributes required to create the
2180 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2181 * the actual capabilities of the created SRQ.
2183 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2184 * requested size of the SRQ, and set to the actual values allocated
2185 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2186 * will always be at least as large as the requested values.
2188 struct ib_srq
*ib_create_srq(struct ib_pd
*pd
,
2189 struct ib_srq_init_attr
*srq_init_attr
);
2192 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2193 * @srq: The SRQ to modify.
2194 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2195 * the current values of selected SRQ attributes are returned.
2196 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2197 * are being modified.
2199 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2200 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2201 * the number of receives queued drops below the limit.
2203 int ib_modify_srq(struct ib_srq
*srq
,
2204 struct ib_srq_attr
*srq_attr
,
2205 enum ib_srq_attr_mask srq_attr_mask
);
2208 * ib_query_srq - Returns the attribute list and current values for the
2210 * @srq: The SRQ to query.
2211 * @srq_attr: The attributes of the specified SRQ.
2213 int ib_query_srq(struct ib_srq
*srq
,
2214 struct ib_srq_attr
*srq_attr
);
2217 * ib_destroy_srq - Destroys the specified SRQ.
2218 * @srq: The SRQ to destroy.
2220 int ib_destroy_srq(struct ib_srq
*srq
);
2223 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2224 * @srq: The SRQ to post the work request on.
2225 * @recv_wr: A list of work requests to post on the receive queue.
2226 * @bad_recv_wr: On an immediate failure, this parameter will reference
2227 * the work request that failed to be posted on the QP.
2229 static inline int ib_post_srq_recv(struct ib_srq
*srq
,
2230 struct ib_recv_wr
*recv_wr
,
2231 struct ib_recv_wr
**bad_recv_wr
)
2233 return srq
->device
->post_srq_recv(srq
, recv_wr
, bad_recv_wr
);
2237 * ib_create_qp - Creates a QP associated with the specified protection
2239 * @pd: The protection domain associated with the QP.
2240 * @qp_init_attr: A list of initial attributes required to create the
2241 * QP. If QP creation succeeds, then the attributes are updated to
2242 * the actual capabilities of the created QP.
2244 struct ib_qp
*ib_create_qp(struct ib_pd
*pd
,
2245 struct ib_qp_init_attr
*qp_init_attr
);
2248 * ib_modify_qp - Modifies the attributes for the specified QP and then
2249 * transitions the QP to the given state.
2250 * @qp: The QP to modify.
2251 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2252 * the current values of selected QP attributes are returned.
2253 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2254 * are being modified.
2256 int ib_modify_qp(struct ib_qp
*qp
,
2257 struct ib_qp_attr
*qp_attr
,
2261 * ib_query_qp - Returns the attribute list and current values for the
2263 * @qp: The QP to query.
2264 * @qp_attr: The attributes of the specified QP.
2265 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2266 * @qp_init_attr: Additional attributes of the selected QP.
2268 * The qp_attr_mask may be used to limit the query to gathering only the
2269 * selected attributes.
2271 int ib_query_qp(struct ib_qp
*qp
,
2272 struct ib_qp_attr
*qp_attr
,
2274 struct ib_qp_init_attr
*qp_init_attr
);
2277 * ib_destroy_qp - Destroys the specified QP.
2278 * @qp: The QP to destroy.
2280 int ib_destroy_qp(struct ib_qp
*qp
);
2283 * ib_open_qp - Obtain a reference to an existing sharable QP.
2284 * @xrcd - XRC domain
2285 * @qp_open_attr: Attributes identifying the QP to open.
2287 * Returns a reference to a sharable QP.
2289 struct ib_qp
*ib_open_qp(struct ib_xrcd
*xrcd
,
2290 struct ib_qp_open_attr
*qp_open_attr
);
2293 * ib_close_qp - Release an external reference to a QP.
2294 * @qp: The QP handle to release
2296 * The opened QP handle is released by the caller. The underlying
2297 * shared QP is not destroyed until all internal references are released.
2299 int ib_close_qp(struct ib_qp
*qp
);
2302 * ib_post_send - Posts a list of work requests to the send queue of
2304 * @qp: The QP to post the work request on.
2305 * @send_wr: A list of work requests to post on the send queue.
2306 * @bad_send_wr: On an immediate failure, this parameter will reference
2307 * the work request that failed to be posted on the QP.
2309 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2310 * error is returned, the QP state shall not be affected,
2311 * ib_post_send() will return an immediate error after queueing any
2312 * earlier work requests in the list.
2314 static inline int ib_post_send(struct ib_qp
*qp
,
2315 struct ib_send_wr
*send_wr
,
2316 struct ib_send_wr
**bad_send_wr
)
2318 return qp
->device
->post_send(qp
, send_wr
, bad_send_wr
);
2322 * ib_post_recv - Posts a list of work requests to the receive queue of
2324 * @qp: The QP to post the work request on.
2325 * @recv_wr: A list of work requests to post on the receive queue.
2326 * @bad_recv_wr: On an immediate failure, this parameter will reference
2327 * the work request that failed to be posted on the QP.
2329 static inline int ib_post_recv(struct ib_qp
*qp
,
2330 struct ib_recv_wr
*recv_wr
,
2331 struct ib_recv_wr
**bad_recv_wr
)
2333 return qp
->device
->post_recv(qp
, recv_wr
, bad_recv_wr
);
2337 * ib_create_cq - Creates a CQ on the specified device.
2338 * @device: The device on which to create the CQ.
2339 * @comp_handler: A user-specified callback that is invoked when a
2340 * completion event occurs on the CQ.
2341 * @event_handler: A user-specified callback that is invoked when an
2342 * asynchronous event not associated with a completion occurs on the CQ.
2343 * @cq_context: Context associated with the CQ returned to the user via
2344 * the associated completion and event handlers.
2345 * @cq_attr: The attributes the CQ should be created upon.
2347 * Users can examine the cq structure to determine the actual CQ size.
2349 struct ib_cq
*ib_create_cq(struct ib_device
*device
,
2350 ib_comp_handler comp_handler
,
2351 void (*event_handler
)(struct ib_event
*, void *),
2353 const struct ib_cq_init_attr
*cq_attr
);
2356 * ib_resize_cq - Modifies the capacity of the CQ.
2357 * @cq: The CQ to resize.
2358 * @cqe: The minimum size of the CQ.
2360 * Users can examine the cq structure to determine the actual CQ size.
2362 int ib_resize_cq(struct ib_cq
*cq
, int cqe
);
2365 * ib_modify_cq - Modifies moderation params of the CQ
2366 * @cq: The CQ to modify.
2367 * @cq_count: number of CQEs that will trigger an event
2368 * @cq_period: max period of time in usec before triggering an event
2371 int ib_modify_cq(struct ib_cq
*cq
, u16 cq_count
, u16 cq_period
);
2374 * ib_destroy_cq - Destroys the specified CQ.
2375 * @cq: The CQ to destroy.
2377 int ib_destroy_cq(struct ib_cq
*cq
);
2380 * ib_poll_cq - poll a CQ for completion(s)
2381 * @cq:the CQ being polled
2382 * @num_entries:maximum number of completions to return
2383 * @wc:array of at least @num_entries &struct ib_wc where completions
2386 * Poll a CQ for (possibly multiple) completions. If the return value
2387 * is < 0, an error occurred. If the return value is >= 0, it is the
2388 * number of completions returned. If the return value is
2389 * non-negative and < num_entries, then the CQ was emptied.
2391 static inline int ib_poll_cq(struct ib_cq
*cq
, int num_entries
,
2394 return cq
->device
->poll_cq(cq
, num_entries
, wc
);
2398 * ib_peek_cq - Returns the number of unreaped completions currently
2399 * on the specified CQ.
2400 * @cq: The CQ to peek.
2401 * @wc_cnt: A minimum number of unreaped completions to check for.
2403 * If the number of unreaped completions is greater than or equal to wc_cnt,
2404 * this function returns wc_cnt, otherwise, it returns the actual number of
2405 * unreaped completions.
2407 int ib_peek_cq(struct ib_cq
*cq
, int wc_cnt
);
2410 * ib_req_notify_cq - Request completion notification on a CQ.
2411 * @cq: The CQ to generate an event for.
2413 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2414 * to request an event on the next solicited event or next work
2415 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2416 * may also be |ed in to request a hint about missed events, as
2420 * < 0 means an error occurred while requesting notification
2421 * == 0 means notification was requested successfully, and if
2422 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2423 * were missed and it is safe to wait for another event. In
2424 * this case is it guaranteed that any work completions added
2425 * to the CQ since the last CQ poll will trigger a completion
2426 * notification event.
2427 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2428 * in. It means that the consumer must poll the CQ again to
2429 * make sure it is empty to avoid missing an event because of a
2430 * race between requesting notification and an entry being
2431 * added to the CQ. This return value means it is possible
2432 * (but not guaranteed) that a work completion has been added
2433 * to the CQ since the last poll without triggering a
2434 * completion notification event.
2436 static inline int ib_req_notify_cq(struct ib_cq
*cq
,
2437 enum ib_cq_notify_flags flags
)
2439 return cq
->device
->req_notify_cq(cq
, flags
);
2443 * ib_req_ncomp_notif - Request completion notification when there are
2444 * at least the specified number of unreaped completions on the CQ.
2445 * @cq: The CQ to generate an event for.
2446 * @wc_cnt: The number of unreaped completions that should be on the
2447 * CQ before an event is generated.
2449 static inline int ib_req_ncomp_notif(struct ib_cq
*cq
, int wc_cnt
)
2451 return cq
->device
->req_ncomp_notif
?
2452 cq
->device
->req_ncomp_notif(cq
, wc_cnt
) :
2457 * ib_get_dma_mr - Returns a memory region for system memory that is
2459 * @pd: The protection domain associated with the memory region.
2460 * @mr_access_flags: Specifies the memory access rights.
2462 * Note that the ib_dma_*() functions defined below must be used
2463 * to create/destroy addresses used with the Lkey or Rkey returned
2464 * by ib_get_dma_mr().
2466 struct ib_mr
*ib_get_dma_mr(struct ib_pd
*pd
, int mr_access_flags
);
2469 * ib_dma_mapping_error - check a DMA addr for error
2470 * @dev: The device for which the dma_addr was created
2471 * @dma_addr: The DMA address to check
2473 static inline int ib_dma_mapping_error(struct ib_device
*dev
, u64 dma_addr
)
2476 return dev
->dma_ops
->mapping_error(dev
, dma_addr
);
2477 return dma_mapping_error(dev
->dma_device
, dma_addr
);
2481 * ib_dma_map_single - Map a kernel virtual address to DMA address
2482 * @dev: The device for which the dma_addr is to be created
2483 * @cpu_addr: The kernel virtual address
2484 * @size: The size of the region in bytes
2485 * @direction: The direction of the DMA
2487 static inline u64
ib_dma_map_single(struct ib_device
*dev
,
2488 void *cpu_addr
, size_t size
,
2489 enum dma_data_direction direction
)
2492 return dev
->dma_ops
->map_single(dev
, cpu_addr
, size
, direction
);
2493 return dma_map_single(dev
->dma_device
, cpu_addr
, size
, direction
);
2497 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2498 * @dev: The device for which the DMA address was created
2499 * @addr: The DMA address
2500 * @size: The size of the region in bytes
2501 * @direction: The direction of the DMA
2503 static inline void ib_dma_unmap_single(struct ib_device
*dev
,
2504 u64 addr
, size_t size
,
2505 enum dma_data_direction direction
)
2508 dev
->dma_ops
->unmap_single(dev
, addr
, size
, direction
);
2510 dma_unmap_single(dev
->dma_device
, addr
, size
, direction
);
2513 static inline u64
ib_dma_map_single_attrs(struct ib_device
*dev
,
2514 void *cpu_addr
, size_t size
,
2515 enum dma_data_direction direction
,
2516 struct dma_attrs
*attrs
)
2518 return dma_map_single_attrs(dev
->dma_device
, cpu_addr
, size
,
2522 static inline void ib_dma_unmap_single_attrs(struct ib_device
*dev
,
2523 u64 addr
, size_t size
,
2524 enum dma_data_direction direction
,
2525 struct dma_attrs
*attrs
)
2527 return dma_unmap_single_attrs(dev
->dma_device
, addr
, size
,
2532 * ib_dma_map_page - Map a physical page to DMA address
2533 * @dev: The device for which the dma_addr is to be created
2534 * @page: The page to be mapped
2535 * @offset: The offset within the page
2536 * @size: The size of the region in bytes
2537 * @direction: The direction of the DMA
2539 static inline u64
ib_dma_map_page(struct ib_device
*dev
,
2541 unsigned long offset
,
2543 enum dma_data_direction direction
)
2546 return dev
->dma_ops
->map_page(dev
, page
, offset
, size
, direction
);
2547 return dma_map_page(dev
->dma_device
, page
, offset
, size
, direction
);
2551 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2552 * @dev: The device for which the DMA address was created
2553 * @addr: The DMA address
2554 * @size: The size of the region in bytes
2555 * @direction: The direction of the DMA
2557 static inline void ib_dma_unmap_page(struct ib_device
*dev
,
2558 u64 addr
, size_t size
,
2559 enum dma_data_direction direction
)
2562 dev
->dma_ops
->unmap_page(dev
, addr
, size
, direction
);
2564 dma_unmap_page(dev
->dma_device
, addr
, size
, direction
);
2568 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2569 * @dev: The device for which the DMA addresses are to be created
2570 * @sg: The array of scatter/gather entries
2571 * @nents: The number of scatter/gather entries
2572 * @direction: The direction of the DMA
2574 static inline int ib_dma_map_sg(struct ib_device
*dev
,
2575 struct scatterlist
*sg
, int nents
,
2576 enum dma_data_direction direction
)
2579 return dev
->dma_ops
->map_sg(dev
, sg
, nents
, direction
);
2580 return dma_map_sg(dev
->dma_device
, sg
, nents
, direction
);
2584 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2585 * @dev: The device for which the DMA addresses were created
2586 * @sg: The array of scatter/gather entries
2587 * @nents: The number of scatter/gather entries
2588 * @direction: The direction of the DMA
2590 static inline void ib_dma_unmap_sg(struct ib_device
*dev
,
2591 struct scatterlist
*sg
, int nents
,
2592 enum dma_data_direction direction
)
2595 dev
->dma_ops
->unmap_sg(dev
, sg
, nents
, direction
);
2597 dma_unmap_sg(dev
->dma_device
, sg
, nents
, direction
);
2600 static inline int ib_dma_map_sg_attrs(struct ib_device
*dev
,
2601 struct scatterlist
*sg
, int nents
,
2602 enum dma_data_direction direction
,
2603 struct dma_attrs
*attrs
)
2605 return dma_map_sg_attrs(dev
->dma_device
, sg
, nents
, direction
, attrs
);
2608 static inline void ib_dma_unmap_sg_attrs(struct ib_device
*dev
,
2609 struct scatterlist
*sg
, int nents
,
2610 enum dma_data_direction direction
,
2611 struct dma_attrs
*attrs
)
2613 dma_unmap_sg_attrs(dev
->dma_device
, sg
, nents
, direction
, attrs
);
2616 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2617 * @dev: The device for which the DMA addresses were created
2618 * @sg: The scatter/gather entry
2620 * Note: this function is obsolete. To do: change all occurrences of
2621 * ib_sg_dma_address() into sg_dma_address().
2623 static inline u64
ib_sg_dma_address(struct ib_device
*dev
,
2624 struct scatterlist
*sg
)
2626 return sg_dma_address(sg
);
2630 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2631 * @dev: The device for which the DMA addresses were created
2632 * @sg: The scatter/gather entry
2634 * Note: this function is obsolete. To do: change all occurrences of
2635 * ib_sg_dma_len() into sg_dma_len().
2637 static inline unsigned int ib_sg_dma_len(struct ib_device
*dev
,
2638 struct scatterlist
*sg
)
2640 return sg_dma_len(sg
);
2644 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2645 * @dev: The device for which the DMA address was created
2646 * @addr: The DMA address
2647 * @size: The size of the region in bytes
2648 * @dir: The direction of the DMA
2650 static inline void ib_dma_sync_single_for_cpu(struct ib_device
*dev
,
2653 enum dma_data_direction dir
)
2656 dev
->dma_ops
->sync_single_for_cpu(dev
, addr
, size
, dir
);
2658 dma_sync_single_for_cpu(dev
->dma_device
, addr
, size
, dir
);
2662 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2663 * @dev: The device for which the DMA address was created
2664 * @addr: The DMA address
2665 * @size: The size of the region in bytes
2666 * @dir: The direction of the DMA
2668 static inline void ib_dma_sync_single_for_device(struct ib_device
*dev
,
2671 enum dma_data_direction dir
)
2674 dev
->dma_ops
->sync_single_for_device(dev
, addr
, size
, dir
);
2676 dma_sync_single_for_device(dev
->dma_device
, addr
, size
, dir
);
2680 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2681 * @dev: The device for which the DMA address is requested
2682 * @size: The size of the region to allocate in bytes
2683 * @dma_handle: A pointer for returning the DMA address of the region
2684 * @flag: memory allocator flags
2686 static inline void *ib_dma_alloc_coherent(struct ib_device
*dev
,
2692 return dev
->dma_ops
->alloc_coherent(dev
, size
, dma_handle
, flag
);
2697 ret
= dma_alloc_coherent(dev
->dma_device
, size
, &handle
, flag
);
2698 *dma_handle
= handle
;
2704 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2705 * @dev: The device for which the DMA addresses were allocated
2706 * @size: The size of the region
2707 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2708 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2710 static inline void ib_dma_free_coherent(struct ib_device
*dev
,
2711 size_t size
, void *cpu_addr
,
2715 dev
->dma_ops
->free_coherent(dev
, size
, cpu_addr
, dma_handle
);
2717 dma_free_coherent(dev
->dma_device
, size
, cpu_addr
, dma_handle
);
2721 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
2723 * @pd: The protection domain associated assigned to the registered region.
2724 * @phys_buf_array: Specifies a list of physical buffers to use in the
2726 * @num_phys_buf: Specifies the size of the phys_buf_array.
2727 * @mr_access_flags: Specifies the memory access rights.
2728 * @iova_start: The offset of the region's starting I/O virtual address.
2730 struct ib_mr
*ib_reg_phys_mr(struct ib_pd
*pd
,
2731 struct ib_phys_buf
*phys_buf_array
,
2733 int mr_access_flags
,
2737 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
2738 * Conceptually, this call performs the functions deregister memory region
2739 * followed by register physical memory region. Where possible,
2740 * resources are reused instead of deallocated and reallocated.
2741 * @mr: The memory region to modify.
2742 * @mr_rereg_mask: A bit-mask used to indicate which of the following
2743 * properties of the memory region are being modified.
2744 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
2745 * the new protection domain to associated with the memory region,
2746 * otherwise, this parameter is ignored.
2747 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2748 * field specifies a list of physical buffers to use in the new
2749 * translation, otherwise, this parameter is ignored.
2750 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2751 * field specifies the size of the phys_buf_array, otherwise, this
2752 * parameter is ignored.
2753 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
2754 * field specifies the new memory access rights, otherwise, this
2755 * parameter is ignored.
2756 * @iova_start: The offset of the region's starting I/O virtual address.
2758 int ib_rereg_phys_mr(struct ib_mr
*mr
,
2761 struct ib_phys_buf
*phys_buf_array
,
2763 int mr_access_flags
,
2767 * ib_query_mr - Retrieves information about a specific memory region.
2768 * @mr: The memory region to retrieve information about.
2769 * @mr_attr: The attributes of the specified memory region.
2771 int ib_query_mr(struct ib_mr
*mr
, struct ib_mr_attr
*mr_attr
);
2774 * ib_dereg_mr - Deregisters a memory region and removes it from the
2775 * HCA translation table.
2776 * @mr: The memory region to deregister.
2778 * This function can fail, if the memory region has memory windows bound to it.
2780 int ib_dereg_mr(struct ib_mr
*mr
);
2784 * ib_create_mr - Allocates a memory region that may be used for
2785 * signature handover operations.
2786 * @pd: The protection domain associated with the region.
2787 * @mr_init_attr: memory region init attributes.
2789 struct ib_mr
*ib_create_mr(struct ib_pd
*pd
,
2790 struct ib_mr_init_attr
*mr_init_attr
);
2793 * ib_destroy_mr - Destroys a memory region that was created using
2794 * ib_create_mr and removes it from HW translation tables.
2795 * @mr: The memory region to destroy.
2797 * This function can fail, if the memory region has memory windows bound to it.
2799 int ib_destroy_mr(struct ib_mr
*mr
);
2802 * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2803 * IB_WR_FAST_REG_MR send work request.
2804 * @pd: The protection domain associated with the region.
2805 * @max_page_list_len: requested max physical buffer list length to be
2806 * used with fast register work requests for this MR.
2808 struct ib_mr
*ib_alloc_fast_reg_mr(struct ib_pd
*pd
, int max_page_list_len
);
2811 * ib_alloc_fast_reg_page_list - Allocates a page list array
2812 * @device - ib device pointer.
2813 * @page_list_len - size of the page list array to be allocated.
2815 * This allocates and returns a struct ib_fast_reg_page_list * and a
2816 * page_list array that is at least page_list_len in size. The actual
2817 * size is returned in max_page_list_len. The caller is responsible
2818 * for initializing the contents of the page_list array before posting
2819 * a send work request with the IB_WC_FAST_REG_MR opcode.
2821 * The page_list array entries must be translated using one of the
2822 * ib_dma_*() functions just like the addresses passed to
2823 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct
2824 * ib_fast_reg_page_list must not be modified by the caller until the
2825 * IB_WC_FAST_REG_MR work request completes.
2827 struct ib_fast_reg_page_list
*ib_alloc_fast_reg_page_list(
2828 struct ib_device
*device
, int page_list_len
);
2831 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2833 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2835 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list
*page_list
);
2838 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2840 * @mr - struct ib_mr pointer to be updated.
2841 * @newkey - new key to be used.
2843 static inline void ib_update_fast_reg_key(struct ib_mr
*mr
, u8 newkey
)
2845 mr
->lkey
= (mr
->lkey
& 0xffffff00) | newkey
;
2846 mr
->rkey
= (mr
->rkey
& 0xffffff00) | newkey
;
2850 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2851 * for calculating a new rkey for type 2 memory windows.
2852 * @rkey - the rkey to increment.
2854 static inline u32
ib_inc_rkey(u32 rkey
)
2856 const u32 mask
= 0x000000ff;
2857 return ((rkey
+ 1) & mask
) | (rkey
& ~mask
);
2861 * ib_alloc_mw - Allocates a memory window.
2862 * @pd: The protection domain associated with the memory window.
2863 * @type: The type of the memory window (1 or 2).
2865 struct ib_mw
*ib_alloc_mw(struct ib_pd
*pd
, enum ib_mw_type type
);
2868 * ib_bind_mw - Posts a work request to the send queue of the specified
2869 * QP, which binds the memory window to the given address range and
2870 * remote access attributes.
2871 * @qp: QP to post the bind work request on.
2872 * @mw: The memory window to bind.
2873 * @mw_bind: Specifies information about the memory window, including
2874 * its address range, remote access rights, and associated memory region.
2876 * If there is no immediate error, the function will update the rkey member
2877 * of the mw parameter to its new value. The bind operation can still fail
2880 static inline int ib_bind_mw(struct ib_qp
*qp
,
2882 struct ib_mw_bind
*mw_bind
)
2884 /* XXX reference counting in corresponding MR? */
2885 return mw
->device
->bind_mw
?
2886 mw
->device
->bind_mw(qp
, mw
, mw_bind
) :
2891 * ib_dealloc_mw - Deallocates a memory window.
2892 * @mw: The memory window to deallocate.
2894 int ib_dealloc_mw(struct ib_mw
*mw
);
2897 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2898 * @pd: The protection domain associated with the unmapped region.
2899 * @mr_access_flags: Specifies the memory access rights.
2900 * @fmr_attr: Attributes of the unmapped region.
2902 * A fast memory region must be mapped before it can be used as part of
2905 struct ib_fmr
*ib_alloc_fmr(struct ib_pd
*pd
,
2906 int mr_access_flags
,
2907 struct ib_fmr_attr
*fmr_attr
);
2910 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2911 * @fmr: The fast memory region to associate with the pages.
2912 * @page_list: An array of physical pages to map to the fast memory region.
2913 * @list_len: The number of pages in page_list.
2914 * @iova: The I/O virtual address to use with the mapped region.
2916 static inline int ib_map_phys_fmr(struct ib_fmr
*fmr
,
2917 u64
*page_list
, int list_len
,
2920 return fmr
->device
->map_phys_fmr(fmr
, page_list
, list_len
, iova
);
2924 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2925 * @fmr_list: A linked list of fast memory regions to unmap.
2927 int ib_unmap_fmr(struct list_head
*fmr_list
);
2930 * ib_dealloc_fmr - Deallocates a fast memory region.
2931 * @fmr: The fast memory region to deallocate.
2933 int ib_dealloc_fmr(struct ib_fmr
*fmr
);
2936 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2937 * @qp: QP to attach to the multicast group. The QP must be type
2939 * @gid: Multicast group GID.
2940 * @lid: Multicast group LID in host byte order.
2942 * In order to send and receive multicast packets, subnet
2943 * administration must have created the multicast group and configured
2944 * the fabric appropriately. The port associated with the specified
2945 * QP must also be a member of the multicast group.
2947 int ib_attach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
2950 * ib_detach_mcast - Detaches the specified QP from a multicast group.
2951 * @qp: QP to detach from the multicast group.
2952 * @gid: Multicast group GID.
2953 * @lid: Multicast group LID in host byte order.
2955 int ib_detach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
2958 * ib_alloc_xrcd - Allocates an XRC domain.
2959 * @device: The device on which to allocate the XRC domain.
2961 struct ib_xrcd
*ib_alloc_xrcd(struct ib_device
*device
);
2964 * ib_dealloc_xrcd - Deallocates an XRC domain.
2965 * @xrcd: The XRC domain to deallocate.
2967 int ib_dealloc_xrcd(struct ib_xrcd
*xrcd
);
2969 struct ib_flow
*ib_create_flow(struct ib_qp
*qp
,
2970 struct ib_flow_attr
*flow_attr
, int domain
);
2971 int ib_destroy_flow(struct ib_flow
*flow_id
);
2973 static inline int ib_check_mr_access(int flags
)
2976 * Local write permission is required if remote write or
2977 * remote atomic permission is also requested.
2979 if (flags
& (IB_ACCESS_REMOTE_ATOMIC
| IB_ACCESS_REMOTE_WRITE
) &&
2980 !(flags
& IB_ACCESS_LOCAL_WRITE
))
2987 * ib_check_mr_status: lightweight check of MR status.
2988 * This routine may provide status checks on a selected
2989 * ib_mr. first use is for signature status check.
2991 * @mr: A memory region.
2992 * @check_mask: Bitmask of which checks to perform from
2993 * ib_mr_status_check enumeration.
2994 * @mr_status: The container of relevant status checks.
2995 * failed checks will be indicated in the status bitmask
2996 * and the relevant info shall be in the error item.
2998 int ib_check_mr_status(struct ib_mr
*mr
, u32 check_mask
,
2999 struct ib_mr_status
*mr_status
);
3001 #endif /* IB_VERBS_H */