2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
39 #if !defined(IB_VERBS_H)
42 #include <linux/types.h>
43 #include <linux/device.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
59 #include <linux/if_link.h>
60 #include <linux/atomic.h>
61 #include <linux/mmu_notifier.h>
62 #include <asm/uaccess.h>
64 extern struct workqueue_struct
*ib_wq
;
65 extern struct workqueue_struct
*ib_comp_wq
;
75 extern union ib_gid zgid
;
78 /* If link layer is Ethernet, this is RoCE V1 */
81 IB_GID_TYPE_ROCE_UDP_ENCAP
= 1,
85 #define ROCE_V2_UDP_DPORT 4791
87 enum ib_gid_type gid_type
;
88 struct net_device
*ndev
;
92 /* IB values map to NodeInfo:NodeType. */
102 /* set the local administered indication */
103 IB_SA_WELL_KNOWN_GUID
= BIT_ULL(57) | 2,
106 enum rdma_transport_type
{
108 RDMA_TRANSPORT_IWARP
,
109 RDMA_TRANSPORT_USNIC
,
110 RDMA_TRANSPORT_USNIC_UDP
113 enum rdma_protocol_type
{
117 RDMA_PROTOCOL_USNIC_UDP
120 __attribute_const__
enum rdma_transport_type
121 rdma_node_get_transport(enum rdma_node_type node_type
);
123 enum rdma_network_type
{
125 RDMA_NETWORK_ROCE_V1
= RDMA_NETWORK_IB
,
130 static inline enum ib_gid_type
ib_network_to_gid_type(enum rdma_network_type network_type
)
132 if (network_type
== RDMA_NETWORK_IPV4
||
133 network_type
== RDMA_NETWORK_IPV6
)
134 return IB_GID_TYPE_ROCE_UDP_ENCAP
;
136 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
137 return IB_GID_TYPE_IB
;
140 static inline enum rdma_network_type
ib_gid_to_network_type(enum ib_gid_type gid_type
,
143 if (gid_type
== IB_GID_TYPE_IB
)
144 return RDMA_NETWORK_IB
;
146 if (ipv6_addr_v4mapped((struct in6_addr
*)gid
))
147 return RDMA_NETWORK_IPV4
;
149 return RDMA_NETWORK_IPV6
;
152 enum rdma_link_layer
{
153 IB_LINK_LAYER_UNSPECIFIED
,
154 IB_LINK_LAYER_INFINIBAND
,
155 IB_LINK_LAYER_ETHERNET
,
158 enum ib_device_cap_flags
{
159 IB_DEVICE_RESIZE_MAX_WR
= (1 << 0),
160 IB_DEVICE_BAD_PKEY_CNTR
= (1 << 1),
161 IB_DEVICE_BAD_QKEY_CNTR
= (1 << 2),
162 IB_DEVICE_RAW_MULTI
= (1 << 3),
163 IB_DEVICE_AUTO_PATH_MIG
= (1 << 4),
164 IB_DEVICE_CHANGE_PHY_PORT
= (1 << 5),
165 IB_DEVICE_UD_AV_PORT_ENFORCE
= (1 << 6),
166 IB_DEVICE_CURR_QP_STATE_MOD
= (1 << 7),
167 IB_DEVICE_SHUTDOWN_PORT
= (1 << 8),
168 IB_DEVICE_INIT_TYPE
= (1 << 9),
169 IB_DEVICE_PORT_ACTIVE_EVENT
= (1 << 10),
170 IB_DEVICE_SYS_IMAGE_GUID
= (1 << 11),
171 IB_DEVICE_RC_RNR_NAK_GEN
= (1 << 12),
172 IB_DEVICE_SRQ_RESIZE
= (1 << 13),
173 IB_DEVICE_N_NOTIFY_CQ
= (1 << 14),
176 * This device supports a per-device lkey or stag that can be
177 * used without performing a memory registration for the local
178 * memory. Note that ULPs should never check this flag, but
179 * instead of use the local_dma_lkey flag in the ib_pd structure,
180 * which will always contain a usable lkey.
182 IB_DEVICE_LOCAL_DMA_LKEY
= (1 << 15),
183 IB_DEVICE_RESERVED
/* old SEND_W_INV */ = (1 << 16),
184 IB_DEVICE_MEM_WINDOW
= (1 << 17),
186 * Devices should set IB_DEVICE_UD_IP_SUM if they support
187 * insertion of UDP and TCP checksum on outgoing UD IPoIB
188 * messages and can verify the validity of checksum for
189 * incoming messages. Setting this flag implies that the
190 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
192 IB_DEVICE_UD_IP_CSUM
= (1 << 18),
193 IB_DEVICE_UD_TSO
= (1 << 19),
194 IB_DEVICE_XRC
= (1 << 20),
197 * This device supports the IB "base memory management extension",
198 * which includes support for fast registrations (IB_WR_REG_MR,
199 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
200 * also be set by any iWarp device which must support FRs to comply
201 * to the iWarp verbs spec. iWarp devices also support the
202 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
205 IB_DEVICE_MEM_MGT_EXTENSIONS
= (1 << 21),
206 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK
= (1 << 22),
207 IB_DEVICE_MEM_WINDOW_TYPE_2A
= (1 << 23),
208 IB_DEVICE_MEM_WINDOW_TYPE_2B
= (1 << 24),
209 IB_DEVICE_RC_IP_CSUM
= (1 << 25),
210 IB_DEVICE_RAW_IP_CSUM
= (1 << 26),
212 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
213 * support execution of WQEs that involve synchronization
214 * of I/O operations with single completion queue managed
217 IB_DEVICE_CROSS_CHANNEL
= (1 << 27),
218 IB_DEVICE_MANAGED_FLOW_STEERING
= (1 << 29),
219 IB_DEVICE_SIGNATURE_HANDOVER
= (1 << 30),
220 IB_DEVICE_ON_DEMAND_PAGING
= (1ULL << 31),
221 IB_DEVICE_SG_GAPS_REG
= (1ULL << 32),
222 IB_DEVICE_VIRTUAL_FUNCTION
= (1ULL << 33),
223 IB_DEVICE_RAW_SCATTER_FCS
= (1ULL << 34),
226 enum ib_signature_prot_cap
{
227 IB_PROT_T10DIF_TYPE_1
= 1,
228 IB_PROT_T10DIF_TYPE_2
= 1 << 1,
229 IB_PROT_T10DIF_TYPE_3
= 1 << 2,
232 enum ib_signature_guard_cap
{
233 IB_GUARD_T10DIF_CRC
= 1,
234 IB_GUARD_T10DIF_CSUM
= 1 << 1,
243 enum ib_odp_general_cap_bits
{
244 IB_ODP_SUPPORT
= 1 << 0,
247 enum ib_odp_transport_cap_bits
{
248 IB_ODP_SUPPORT_SEND
= 1 << 0,
249 IB_ODP_SUPPORT_RECV
= 1 << 1,
250 IB_ODP_SUPPORT_WRITE
= 1 << 2,
251 IB_ODP_SUPPORT_READ
= 1 << 3,
252 IB_ODP_SUPPORT_ATOMIC
= 1 << 4,
256 uint64_t general_caps
;
258 uint32_t rc_odp_caps
;
259 uint32_t uc_odp_caps
;
260 uint32_t ud_odp_caps
;
261 } per_transport_caps
;
264 enum ib_cq_creation_flags
{
265 IB_CQ_FLAGS_TIMESTAMP_COMPLETION
= 1 << 0,
266 IB_CQ_FLAGS_IGNORE_OVERRUN
= 1 << 1,
269 struct ib_cq_init_attr
{
275 struct ib_device_attr
{
277 __be64 sys_image_guid
;
285 u64 device_cap_flags
;
295 int max_qp_init_rd_atom
;
296 int max_ee_init_rd_atom
;
297 enum ib_atomic_cap atomic_cap
;
298 enum ib_atomic_cap masked_atomic_cap
;
305 int max_mcast_qp_attach
;
306 int max_total_mcast_qp_attach
;
313 unsigned int max_fast_reg_page_list_len
;
315 u8 local_ca_ack_delay
;
318 struct ib_odp_caps odp_caps
;
319 uint64_t timestamp_mask
;
320 uint64_t hca_core_clock
; /* in KHZ */
331 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu
)
334 case IB_MTU_256
: return 256;
335 case IB_MTU_512
: return 512;
336 case IB_MTU_1024
: return 1024;
337 case IB_MTU_2048
: return 2048;
338 case IB_MTU_4096
: return 4096;
349 IB_PORT_ACTIVE_DEFER
= 5
352 enum ib_port_cap_flags
{
354 IB_PORT_NOTICE_SUP
= 1 << 2,
355 IB_PORT_TRAP_SUP
= 1 << 3,
356 IB_PORT_OPT_IPD_SUP
= 1 << 4,
357 IB_PORT_AUTO_MIGR_SUP
= 1 << 5,
358 IB_PORT_SL_MAP_SUP
= 1 << 6,
359 IB_PORT_MKEY_NVRAM
= 1 << 7,
360 IB_PORT_PKEY_NVRAM
= 1 << 8,
361 IB_PORT_LED_INFO_SUP
= 1 << 9,
362 IB_PORT_SM_DISABLED
= 1 << 10,
363 IB_PORT_SYS_IMAGE_GUID_SUP
= 1 << 11,
364 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP
= 1 << 12,
365 IB_PORT_EXTENDED_SPEEDS_SUP
= 1 << 14,
366 IB_PORT_CM_SUP
= 1 << 16,
367 IB_PORT_SNMP_TUNNEL_SUP
= 1 << 17,
368 IB_PORT_REINIT_SUP
= 1 << 18,
369 IB_PORT_DEVICE_MGMT_SUP
= 1 << 19,
370 IB_PORT_VENDOR_CLASS_SUP
= 1 << 20,
371 IB_PORT_DR_NOTICE_SUP
= 1 << 21,
372 IB_PORT_CAP_MASK_NOTICE_SUP
= 1 << 22,
373 IB_PORT_BOOT_MGMT_SUP
= 1 << 23,
374 IB_PORT_LINK_LATENCY_SUP
= 1 << 24,
375 IB_PORT_CLIENT_REG_SUP
= 1 << 25,
376 IB_PORT_IP_BASED_GIDS
= 1 << 26,
386 static inline int ib_width_enum_to_int(enum ib_port_width width
)
389 case IB_WIDTH_1X
: return 1;
390 case IB_WIDTH_4X
: return 4;
391 case IB_WIDTH_8X
: return 8;
392 case IB_WIDTH_12X
: return 12;
407 * struct rdma_hw_stats
408 * @timestamp - Used by the core code to track when the last update was
409 * @lifespan - Used by the core code to determine how old the counters
410 * should be before being updated again. Stored in jiffies, defaults
411 * to 10 milliseconds, drivers can override the default be specifying
412 * their own value during their allocation routine.
413 * @name - Array of pointers to static names used for the counters in
415 * @num_counters - How many hardware counters there are. If name is
416 * shorter than this number, a kernel oops will result. Driver authors
417 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
418 * in their code to prevent this.
419 * @value - Array of u64 counters that are accessed by the sysfs code and
420 * filled in by the drivers get_stats routine
422 struct rdma_hw_stats
{
423 unsigned long timestamp
;
424 unsigned long lifespan
;
425 const char * const *names
;
430 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
432 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
434 * @names - Array of static const char *
435 * @num_counters - How many elements in array
436 * @lifespan - How many milliseconds between updates
438 static inline struct rdma_hw_stats
*rdma_alloc_hw_stats_struct(
439 const char * const *names
, int num_counters
,
440 unsigned long lifespan
)
442 struct rdma_hw_stats
*stats
;
444 stats
= kzalloc(sizeof(*stats
) + num_counters
* sizeof(u64
),
448 stats
->names
= names
;
449 stats
->num_counters
= num_counters
;
450 stats
->lifespan
= msecs_to_jiffies(lifespan
);
456 /* Define bits for the various functionality this port needs to be supported by
459 /* Management 0x00000FFF */
460 #define RDMA_CORE_CAP_IB_MAD 0x00000001
461 #define RDMA_CORE_CAP_IB_SMI 0x00000002
462 #define RDMA_CORE_CAP_IB_CM 0x00000004
463 #define RDMA_CORE_CAP_IW_CM 0x00000008
464 #define RDMA_CORE_CAP_IB_SA 0x00000010
465 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
467 /* Address format 0x000FF000 */
468 #define RDMA_CORE_CAP_AF_IB 0x00001000
469 #define RDMA_CORE_CAP_ETH_AH 0x00002000
471 /* Protocol 0xFFF00000 */
472 #define RDMA_CORE_CAP_PROT_IB 0x00100000
473 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
474 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
475 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
477 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
478 | RDMA_CORE_CAP_IB_MAD \
479 | RDMA_CORE_CAP_IB_SMI \
480 | RDMA_CORE_CAP_IB_CM \
481 | RDMA_CORE_CAP_IB_SA \
482 | RDMA_CORE_CAP_AF_IB)
483 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
484 | RDMA_CORE_CAP_IB_MAD \
485 | RDMA_CORE_CAP_IB_CM \
486 | RDMA_CORE_CAP_AF_IB \
487 | RDMA_CORE_CAP_ETH_AH)
488 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
489 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
490 | RDMA_CORE_CAP_IB_MAD \
491 | RDMA_CORE_CAP_IB_CM \
492 | RDMA_CORE_CAP_AF_IB \
493 | RDMA_CORE_CAP_ETH_AH)
494 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
495 | RDMA_CORE_CAP_IW_CM)
496 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
497 | RDMA_CORE_CAP_OPA_MAD)
499 struct ib_port_attr
{
501 enum ib_port_state state
;
503 enum ib_mtu active_mtu
;
523 enum ib_device_modify_flags
{
524 IB_DEVICE_MODIFY_SYS_IMAGE_GUID
= 1 << 0,
525 IB_DEVICE_MODIFY_NODE_DESC
= 1 << 1
528 struct ib_device_modify
{
533 enum ib_port_modify_flags
{
534 IB_PORT_SHUTDOWN
= 1,
535 IB_PORT_INIT_TYPE
= (1<<2),
536 IB_PORT_RESET_QKEY_CNTR
= (1<<3)
539 struct ib_port_modify
{
540 u32 set_port_cap_mask
;
541 u32 clr_port_cap_mask
;
549 IB_EVENT_QP_ACCESS_ERR
,
553 IB_EVENT_PATH_MIG_ERR
,
554 IB_EVENT_DEVICE_FATAL
,
555 IB_EVENT_PORT_ACTIVE
,
558 IB_EVENT_PKEY_CHANGE
,
561 IB_EVENT_SRQ_LIMIT_REACHED
,
562 IB_EVENT_QP_LAST_WQE_REACHED
,
563 IB_EVENT_CLIENT_REREGISTER
,
567 const char *__attribute_const__
ib_event_msg(enum ib_event_type event
);
570 struct ib_device
*device
;
577 enum ib_event_type event
;
580 struct ib_event_handler
{
581 struct ib_device
*device
;
582 void (*handler
)(struct ib_event_handler
*, struct ib_event
*);
583 struct list_head list
;
586 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
588 (_ptr)->device = _device; \
589 (_ptr)->handler = _handler; \
590 INIT_LIST_HEAD(&(_ptr)->list); \
593 struct ib_global_route
{
602 __be32 version_tclass_flow
;
610 union rdma_network_hdr
{
613 /* The IB spec states that if it's IPv4, the header
614 * is located in the last 20 bytes of the header.
617 struct iphdr roce4grh
;
622 IB_MULTICAST_QPN
= 0xffffff
625 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
626 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
633 IB_RATE_PORT_CURRENT
= 0,
634 IB_RATE_2_5_GBPS
= 2,
642 IB_RATE_120_GBPS
= 10,
643 IB_RATE_14_GBPS
= 11,
644 IB_RATE_56_GBPS
= 12,
645 IB_RATE_112_GBPS
= 13,
646 IB_RATE_168_GBPS
= 14,
647 IB_RATE_25_GBPS
= 15,
648 IB_RATE_100_GBPS
= 16,
649 IB_RATE_200_GBPS
= 17,
650 IB_RATE_300_GBPS
= 18
654 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
655 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
656 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
657 * @rate: rate to convert.
659 __attribute_const__
int ib_rate_to_mult(enum ib_rate rate
);
662 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
663 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
664 * @rate: rate to convert.
666 __attribute_const__
int ib_rate_to_mbps(enum ib_rate rate
);
670 * enum ib_mr_type - memory region type
671 * @IB_MR_TYPE_MEM_REG: memory region that is used for
672 * normal registration
673 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
674 * signature operations (data-integrity
676 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
677 * register any arbitrary sg lists (without
678 * the normal mr constraints - see
683 IB_MR_TYPE_SIGNATURE
,
689 * IB_SIG_TYPE_NONE: Unprotected.
690 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
692 enum ib_signature_type
{
698 * Signature T10-DIF block-guard types
699 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
700 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
702 enum ib_t10_dif_bg_type
{
708 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
710 * @bg_type: T10-DIF block guard type (CRC|CSUM)
711 * @pi_interval: protection information interval.
712 * @bg: seed of guard computation.
713 * @app_tag: application tag of guard block
714 * @ref_tag: initial guard block reference tag.
715 * @ref_remap: Indicate wethear the reftag increments each block
716 * @app_escape: Indicate to skip block check if apptag=0xffff
717 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
718 * @apptag_check_mask: check bitmask of application tag.
720 struct ib_t10_dif_domain
{
721 enum ib_t10_dif_bg_type bg_type
;
729 u16 apptag_check_mask
;
733 * struct ib_sig_domain - Parameters for signature domain
734 * @sig_type: specific signauture type
735 * @sig: union of all signature domain attributes that may
736 * be used to set domain layout.
738 struct ib_sig_domain
{
739 enum ib_signature_type sig_type
;
741 struct ib_t10_dif_domain dif
;
746 * struct ib_sig_attrs - Parameters for signature handover operation
747 * @check_mask: bitmask for signature byte check (8 bytes)
748 * @mem: memory domain layout desciptor.
749 * @wire: wire domain layout desciptor.
751 struct ib_sig_attrs
{
753 struct ib_sig_domain mem
;
754 struct ib_sig_domain wire
;
757 enum ib_sig_err_type
{
764 * struct ib_sig_err - signature error descriptor
767 enum ib_sig_err_type err_type
;
774 enum ib_mr_status_check
{
775 IB_MR_CHECK_SIG_STATUS
= 1,
779 * struct ib_mr_status - Memory region status container
781 * @fail_status: Bitmask of MR checks status. For each
782 * failed check a corresponding status bit is set.
783 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
786 struct ib_mr_status
{
788 struct ib_sig_err sig_err
;
792 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
794 * @mult: multiple to convert.
796 __attribute_const__
enum ib_rate
mult_to_ib_rate(int mult
);
799 struct ib_global_route grh
;
813 IB_WC_LOC_EEC_OP_ERR
,
818 IB_WC_LOC_ACCESS_ERR
,
819 IB_WC_REM_INV_REQ_ERR
,
820 IB_WC_REM_ACCESS_ERR
,
823 IB_WC_RNR_RETRY_EXC_ERR
,
824 IB_WC_LOC_RDD_VIOL_ERR
,
825 IB_WC_REM_INV_RD_REQ_ERR
,
828 IB_WC_INV_EEC_STATE_ERR
,
830 IB_WC_RESP_TIMEOUT_ERR
,
834 const char *__attribute_const__
ib_wc_status_msg(enum ib_wc_status status
);
845 IB_WC_MASKED_COMP_SWAP
,
846 IB_WC_MASKED_FETCH_ADD
,
848 * Set value of IB_WC_RECV so consumers can test if a completion is a
849 * receive by testing (opcode & IB_WC_RECV).
852 IB_WC_RECV_RDMA_WITH_IMM
857 IB_WC_WITH_IMM
= (1<<1),
858 IB_WC_WITH_INVALIDATE
= (1<<2),
859 IB_WC_IP_CSUM_OK
= (1<<3),
860 IB_WC_WITH_SMAC
= (1<<4),
861 IB_WC_WITH_VLAN
= (1<<5),
862 IB_WC_WITH_NETWORK_HDR_TYPE
= (1<<6),
868 struct ib_cqe
*wr_cqe
;
870 enum ib_wc_status status
;
871 enum ib_wc_opcode opcode
;
885 u8 port_num
; /* valid only for DR SMPs on switches */
891 enum ib_cq_notify_flags
{
892 IB_CQ_SOLICITED
= 1 << 0,
893 IB_CQ_NEXT_COMP
= 1 << 1,
894 IB_CQ_SOLICITED_MASK
= IB_CQ_SOLICITED
| IB_CQ_NEXT_COMP
,
895 IB_CQ_REPORT_MISSED_EVENTS
= 1 << 2,
903 enum ib_srq_attr_mask
{
904 IB_SRQ_MAX_WR
= 1 << 0,
905 IB_SRQ_LIMIT
= 1 << 1,
914 struct ib_srq_init_attr
{
915 void (*event_handler
)(struct ib_event
*, void *);
917 struct ib_srq_attr attr
;
918 enum ib_srq_type srq_type
;
922 struct ib_xrcd
*xrcd
;
936 * Maximum number of rdma_rw_ctx structures in flight at a time.
937 * ib_create_qp() will calculate the right amount of neededed WRs
938 * and MRs based on this.
950 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
951 * here (and in that order) since the MAD layer uses them as
952 * indices into a 2-entry table.
961 IB_QPT_RAW_ETHERTYPE
,
962 IB_QPT_RAW_PACKET
= 8,
966 /* Reserve a range for qp types internal to the low level driver.
967 * These qp types will not be visible at the IB core layer, so the
968 * IB_QPT_MAX usages should not be affected in the core layer
970 IB_QPT_RESERVED1
= 0x1000,
982 enum ib_qp_create_flags
{
983 IB_QP_CREATE_IPOIB_UD_LSO
= 1 << 0,
984 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK
= 1 << 1,
985 IB_QP_CREATE_CROSS_CHANNEL
= 1 << 2,
986 IB_QP_CREATE_MANAGED_SEND
= 1 << 3,
987 IB_QP_CREATE_MANAGED_RECV
= 1 << 4,
988 IB_QP_CREATE_NETIF_QP
= 1 << 5,
989 IB_QP_CREATE_SIGNATURE_EN
= 1 << 6,
990 IB_QP_CREATE_USE_GFP_NOIO
= 1 << 7,
991 IB_QP_CREATE_SCATTER_FCS
= 1 << 8,
992 /* reserve bits 26-31 for low level drivers' internal use */
993 IB_QP_CREATE_RESERVED_START
= 1 << 26,
994 IB_QP_CREATE_RESERVED_END
= 1 << 31,
998 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
999 * callback to destroy the passed in QP.
1002 struct ib_qp_init_attr
{
1003 void (*event_handler
)(struct ib_event
*, void *);
1005 struct ib_cq
*send_cq
;
1006 struct ib_cq
*recv_cq
;
1008 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
1009 struct ib_qp_cap cap
;
1010 enum ib_sig_type sq_sig_type
;
1011 enum ib_qp_type qp_type
;
1012 enum ib_qp_create_flags create_flags
;
1015 * Only needed for special QP types, or when using the RW API.
1020 struct ib_qp_open_attr
{
1021 void (*event_handler
)(struct ib_event
*, void *);
1024 enum ib_qp_type qp_type
;
1027 enum ib_rnr_timeout
{
1028 IB_RNR_TIMER_655_36
= 0,
1029 IB_RNR_TIMER_000_01
= 1,
1030 IB_RNR_TIMER_000_02
= 2,
1031 IB_RNR_TIMER_000_03
= 3,
1032 IB_RNR_TIMER_000_04
= 4,
1033 IB_RNR_TIMER_000_06
= 5,
1034 IB_RNR_TIMER_000_08
= 6,
1035 IB_RNR_TIMER_000_12
= 7,
1036 IB_RNR_TIMER_000_16
= 8,
1037 IB_RNR_TIMER_000_24
= 9,
1038 IB_RNR_TIMER_000_32
= 10,
1039 IB_RNR_TIMER_000_48
= 11,
1040 IB_RNR_TIMER_000_64
= 12,
1041 IB_RNR_TIMER_000_96
= 13,
1042 IB_RNR_TIMER_001_28
= 14,
1043 IB_RNR_TIMER_001_92
= 15,
1044 IB_RNR_TIMER_002_56
= 16,
1045 IB_RNR_TIMER_003_84
= 17,
1046 IB_RNR_TIMER_005_12
= 18,
1047 IB_RNR_TIMER_007_68
= 19,
1048 IB_RNR_TIMER_010_24
= 20,
1049 IB_RNR_TIMER_015_36
= 21,
1050 IB_RNR_TIMER_020_48
= 22,
1051 IB_RNR_TIMER_030_72
= 23,
1052 IB_RNR_TIMER_040_96
= 24,
1053 IB_RNR_TIMER_061_44
= 25,
1054 IB_RNR_TIMER_081_92
= 26,
1055 IB_RNR_TIMER_122_88
= 27,
1056 IB_RNR_TIMER_163_84
= 28,
1057 IB_RNR_TIMER_245_76
= 29,
1058 IB_RNR_TIMER_327_68
= 30,
1059 IB_RNR_TIMER_491_52
= 31
1062 enum ib_qp_attr_mask
{
1064 IB_QP_CUR_STATE
= (1<<1),
1065 IB_QP_EN_SQD_ASYNC_NOTIFY
= (1<<2),
1066 IB_QP_ACCESS_FLAGS
= (1<<3),
1067 IB_QP_PKEY_INDEX
= (1<<4),
1068 IB_QP_PORT
= (1<<5),
1069 IB_QP_QKEY
= (1<<6),
1071 IB_QP_PATH_MTU
= (1<<8),
1072 IB_QP_TIMEOUT
= (1<<9),
1073 IB_QP_RETRY_CNT
= (1<<10),
1074 IB_QP_RNR_RETRY
= (1<<11),
1075 IB_QP_RQ_PSN
= (1<<12),
1076 IB_QP_MAX_QP_RD_ATOMIC
= (1<<13),
1077 IB_QP_ALT_PATH
= (1<<14),
1078 IB_QP_MIN_RNR_TIMER
= (1<<15),
1079 IB_QP_SQ_PSN
= (1<<16),
1080 IB_QP_MAX_DEST_RD_ATOMIC
= (1<<17),
1081 IB_QP_PATH_MIG_STATE
= (1<<18),
1082 IB_QP_CAP
= (1<<19),
1083 IB_QP_DEST_QPN
= (1<<20),
1084 IB_QP_RESERVED1
= (1<<21),
1085 IB_QP_RESERVED2
= (1<<22),
1086 IB_QP_RESERVED3
= (1<<23),
1087 IB_QP_RESERVED4
= (1<<24),
1112 enum ib_qp_state qp_state
;
1113 enum ib_qp_state cur_qp_state
;
1114 enum ib_mtu path_mtu
;
1115 enum ib_mig_state path_mig_state
;
1120 int qp_access_flags
;
1121 struct ib_qp_cap cap
;
1122 struct ib_ah_attr ah_attr
;
1123 struct ib_ah_attr alt_ah_attr
;
1126 u8 en_sqd_async_notify
;
1129 u8 max_dest_rd_atomic
;
1141 IB_WR_RDMA_WRITE_WITH_IMM
,
1143 IB_WR_SEND_WITH_IMM
,
1145 IB_WR_ATOMIC_CMP_AND_SWP
,
1146 IB_WR_ATOMIC_FETCH_AND_ADD
,
1148 IB_WR_SEND_WITH_INV
,
1149 IB_WR_RDMA_READ_WITH_INV
,
1152 IB_WR_MASKED_ATOMIC_CMP_AND_SWP
,
1153 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD
,
1155 /* reserve values for low level drivers' internal use.
1156 * These values will not be used at all in the ib core layer.
1158 IB_WR_RESERVED1
= 0xf0,
1170 enum ib_send_flags
{
1172 IB_SEND_SIGNALED
= (1<<1),
1173 IB_SEND_SOLICITED
= (1<<2),
1174 IB_SEND_INLINE
= (1<<3),
1175 IB_SEND_IP_CSUM
= (1<<4),
1177 /* reserve bits 26-31 for low level drivers' internal use */
1178 IB_SEND_RESERVED_START
= (1 << 26),
1179 IB_SEND_RESERVED_END
= (1 << 31),
1189 void (*done
)(struct ib_cq
*cq
, struct ib_wc
*wc
);
1193 struct ib_send_wr
*next
;
1196 struct ib_cqe
*wr_cqe
;
1198 struct ib_sge
*sg_list
;
1200 enum ib_wr_opcode opcode
;
1204 u32 invalidate_rkey
;
1209 struct ib_send_wr wr
;
1214 static inline struct ib_rdma_wr
*rdma_wr(struct ib_send_wr
*wr
)
1216 return container_of(wr
, struct ib_rdma_wr
, wr
);
1219 struct ib_atomic_wr
{
1220 struct ib_send_wr wr
;
1224 u64 compare_add_mask
;
1229 static inline struct ib_atomic_wr
*atomic_wr(struct ib_send_wr
*wr
)
1231 return container_of(wr
, struct ib_atomic_wr
, wr
);
1235 struct ib_send_wr wr
;
1242 u16 pkey_index
; /* valid for GSI only */
1243 u8 port_num
; /* valid for DR SMPs on switch only */
1246 static inline struct ib_ud_wr
*ud_wr(struct ib_send_wr
*wr
)
1248 return container_of(wr
, struct ib_ud_wr
, wr
);
1252 struct ib_send_wr wr
;
1258 static inline struct ib_reg_wr
*reg_wr(struct ib_send_wr
*wr
)
1260 return container_of(wr
, struct ib_reg_wr
, wr
);
1263 struct ib_sig_handover_wr
{
1264 struct ib_send_wr wr
;
1265 struct ib_sig_attrs
*sig_attrs
;
1266 struct ib_mr
*sig_mr
;
1268 struct ib_sge
*prot
;
1271 static inline struct ib_sig_handover_wr
*sig_handover_wr(struct ib_send_wr
*wr
)
1273 return container_of(wr
, struct ib_sig_handover_wr
, wr
);
1277 struct ib_recv_wr
*next
;
1280 struct ib_cqe
*wr_cqe
;
1282 struct ib_sge
*sg_list
;
1286 enum ib_access_flags
{
1287 IB_ACCESS_LOCAL_WRITE
= 1,
1288 IB_ACCESS_REMOTE_WRITE
= (1<<1),
1289 IB_ACCESS_REMOTE_READ
= (1<<2),
1290 IB_ACCESS_REMOTE_ATOMIC
= (1<<3),
1291 IB_ACCESS_MW_BIND
= (1<<4),
1292 IB_ZERO_BASED
= (1<<5),
1293 IB_ACCESS_ON_DEMAND
= (1<<6),
1297 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1298 * are hidden here instead of a uapi header!
1300 enum ib_mr_rereg_flags
{
1301 IB_MR_REREG_TRANS
= 1,
1302 IB_MR_REREG_PD
= (1<<1),
1303 IB_MR_REREG_ACCESS
= (1<<2),
1304 IB_MR_REREG_SUPPORTED
= ((IB_MR_REREG_ACCESS
<< 1) - 1)
1307 struct ib_fmr_attr
{
1315 struct ib_ucontext
{
1316 struct ib_device
*device
;
1317 struct list_head pd_list
;
1318 struct list_head mr_list
;
1319 struct list_head mw_list
;
1320 struct list_head cq_list
;
1321 struct list_head qp_list
;
1322 struct list_head srq_list
;
1323 struct list_head ah_list
;
1324 struct list_head xrcd_list
;
1325 struct list_head rule_list
;
1329 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1330 struct rb_root umem_tree
;
1332 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1333 * mmu notifiers registration.
1335 struct rw_semaphore umem_rwsem
;
1336 void (*invalidate_range
)(struct ib_umem
*umem
,
1337 unsigned long start
, unsigned long end
);
1339 struct mmu_notifier mn
;
1340 atomic_t notifier_count
;
1341 /* A list of umems that don't have private mmu notifier counters yet. */
1342 struct list_head no_private_counters
;
1348 u64 user_handle
; /* handle given to us by userspace */
1349 struct ib_ucontext
*context
; /* associated user context */
1350 void *object
; /* containing object */
1351 struct list_head list
; /* link to context's list */
1352 int id
; /* index into kernel idr */
1354 struct rw_semaphore mutex
; /* protects .live */
1355 struct rcu_head rcu
; /* kfree_rcu() overhead */
1360 const void __user
*inbuf
;
1361 void __user
*outbuf
;
1368 struct ib_device
*device
;
1369 struct ib_uobject
*uobject
;
1370 atomic_t usecnt
; /* count all resources */
1371 struct ib_mr
*local_mr
;
1375 struct ib_device
*device
;
1376 atomic_t usecnt
; /* count all exposed resources */
1377 struct inode
*inode
;
1379 struct mutex tgt_qp_mutex
;
1380 struct list_head tgt_qp_list
;
1384 struct ib_device
*device
;
1386 struct ib_uobject
*uobject
;
1389 typedef void (*ib_comp_handler
)(struct ib_cq
*cq
, void *cq_context
);
1391 enum ib_poll_context
{
1392 IB_POLL_DIRECT
, /* caller context, no hw completions */
1393 IB_POLL_SOFTIRQ
, /* poll from softirq context */
1394 IB_POLL_WORKQUEUE
, /* poll from workqueue */
1398 struct ib_device
*device
;
1399 struct ib_uobject
*uobject
;
1400 ib_comp_handler comp_handler
;
1401 void (*event_handler
)(struct ib_event
*, void *);
1404 atomic_t usecnt
; /* count number of work queues */
1405 enum ib_poll_context poll_ctx
;
1408 struct irq_poll iop
;
1409 struct work_struct work
;
1414 struct ib_device
*device
;
1416 struct ib_uobject
*uobject
;
1417 void (*event_handler
)(struct ib_event
*, void *);
1419 enum ib_srq_type srq_type
;
1424 struct ib_xrcd
*xrcd
;
1432 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1433 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1436 struct ib_device
*device
;
1438 struct ib_cq
*send_cq
;
1439 struct ib_cq
*recv_cq
;
1442 struct list_head rdma_mrs
;
1443 struct list_head sig_mrs
;
1445 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
1446 struct list_head xrcd_list
;
1448 /* count times opened, mcast attaches, flow attaches */
1450 struct list_head open_list
;
1451 struct ib_qp
*real_qp
;
1452 struct ib_uobject
*uobject
;
1453 void (*event_handler
)(struct ib_event
*, void *);
1458 enum ib_qp_type qp_type
;
1462 struct ib_device
*device
;
1468 unsigned int page_size
;
1471 struct ib_uobject
*uobject
; /* user */
1472 struct list_head qp_entry
; /* FR */
1477 struct ib_device
*device
;
1479 struct ib_uobject
*uobject
;
1481 enum ib_mw_type type
;
1485 struct ib_device
*device
;
1487 struct list_head list
;
1492 /* Supported steering options */
1493 enum ib_flow_attr_type
{
1494 /* steering according to rule specifications */
1495 IB_FLOW_ATTR_NORMAL
= 0x0,
1496 /* default unicast and multicast rule -
1497 * receive all Eth traffic which isn't steered to any QP
1499 IB_FLOW_ATTR_ALL_DEFAULT
= 0x1,
1500 /* default multicast rule -
1501 * receive all Eth multicast traffic which isn't steered to any QP
1503 IB_FLOW_ATTR_MC_DEFAULT
= 0x2,
1504 /* sniffer rule - receive all port traffic */
1505 IB_FLOW_ATTR_SNIFFER
= 0x3
1508 /* Supported steering header types */
1509 enum ib_flow_spec_type
{
1511 IB_FLOW_SPEC_ETH
= 0x20,
1512 IB_FLOW_SPEC_IB
= 0x22,
1514 IB_FLOW_SPEC_IPV4
= 0x30,
1516 IB_FLOW_SPEC_TCP
= 0x40,
1517 IB_FLOW_SPEC_UDP
= 0x41
1519 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1520 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1522 /* Flow steering rule priority is set according to it's domain.
1523 * Lower domain value means higher priority.
1525 enum ib_flow_domain
{
1526 IB_FLOW_DOMAIN_USER
,
1527 IB_FLOW_DOMAIN_ETHTOOL
,
1530 IB_FLOW_DOMAIN_NUM
/* Must be last */
1533 enum ib_flow_flags
{
1534 IB_FLOW_ATTR_FLAGS_DONT_TRAP
= 1UL << 1, /* Continue match, no steal */
1535 IB_FLOW_ATTR_FLAGS_RESERVED
= 1UL << 2 /* Must be last */
1538 struct ib_flow_eth_filter
{
1545 struct ib_flow_spec_eth
{
1546 enum ib_flow_spec_type type
;
1548 struct ib_flow_eth_filter val
;
1549 struct ib_flow_eth_filter mask
;
1552 struct ib_flow_ib_filter
{
1557 struct ib_flow_spec_ib
{
1558 enum ib_flow_spec_type type
;
1560 struct ib_flow_ib_filter val
;
1561 struct ib_flow_ib_filter mask
;
1564 struct ib_flow_ipv4_filter
{
1569 struct ib_flow_spec_ipv4
{
1570 enum ib_flow_spec_type type
;
1572 struct ib_flow_ipv4_filter val
;
1573 struct ib_flow_ipv4_filter mask
;
1576 struct ib_flow_tcp_udp_filter
{
1581 struct ib_flow_spec_tcp_udp
{
1582 enum ib_flow_spec_type type
;
1584 struct ib_flow_tcp_udp_filter val
;
1585 struct ib_flow_tcp_udp_filter mask
;
1588 union ib_flow_spec
{
1590 enum ib_flow_spec_type type
;
1593 struct ib_flow_spec_eth eth
;
1594 struct ib_flow_spec_ib ib
;
1595 struct ib_flow_spec_ipv4 ipv4
;
1596 struct ib_flow_spec_tcp_udp tcp_udp
;
1599 struct ib_flow_attr
{
1600 enum ib_flow_attr_type type
;
1606 /* Following are the optional layers according to user request
1607 * struct ib_flow_spec_xxx
1608 * struct ib_flow_spec_yyy
1614 struct ib_uobject
*uobject
;
1620 enum ib_process_mad_flags
{
1621 IB_MAD_IGNORE_MKEY
= 1,
1622 IB_MAD_IGNORE_BKEY
= 2,
1623 IB_MAD_IGNORE_ALL
= IB_MAD_IGNORE_MKEY
| IB_MAD_IGNORE_BKEY
1626 enum ib_mad_result
{
1627 IB_MAD_RESULT_FAILURE
= 0, /* (!SUCCESS is the important flag) */
1628 IB_MAD_RESULT_SUCCESS
= 1 << 0, /* MAD was successfully processed */
1629 IB_MAD_RESULT_REPLY
= 1 << 1, /* Reply packet needs to be sent */
1630 IB_MAD_RESULT_CONSUMED
= 1 << 2 /* Packet consumed: stop processing */
1633 #define IB_DEVICE_NAME_MAX 64
1637 struct ib_event_handler event_handler
;
1638 struct ib_pkey_cache
**pkey_cache
;
1639 struct ib_gid_table
**gid_cache
;
1643 struct ib_dma_mapping_ops
{
1644 int (*mapping_error
)(struct ib_device
*dev
,
1646 u64 (*map_single
)(struct ib_device
*dev
,
1647 void *ptr
, size_t size
,
1648 enum dma_data_direction direction
);
1649 void (*unmap_single
)(struct ib_device
*dev
,
1650 u64 addr
, size_t size
,
1651 enum dma_data_direction direction
);
1652 u64 (*map_page
)(struct ib_device
*dev
,
1653 struct page
*page
, unsigned long offset
,
1655 enum dma_data_direction direction
);
1656 void (*unmap_page
)(struct ib_device
*dev
,
1657 u64 addr
, size_t size
,
1658 enum dma_data_direction direction
);
1659 int (*map_sg
)(struct ib_device
*dev
,
1660 struct scatterlist
*sg
, int nents
,
1661 enum dma_data_direction direction
);
1662 void (*unmap_sg
)(struct ib_device
*dev
,
1663 struct scatterlist
*sg
, int nents
,
1664 enum dma_data_direction direction
);
1665 void (*sync_single_for_cpu
)(struct ib_device
*dev
,
1668 enum dma_data_direction dir
);
1669 void (*sync_single_for_device
)(struct ib_device
*dev
,
1672 enum dma_data_direction dir
);
1673 void *(*alloc_coherent
)(struct ib_device
*dev
,
1677 void (*free_coherent
)(struct ib_device
*dev
,
1678 size_t size
, void *cpu_addr
,
1684 struct ib_port_immutable
{
1692 struct device
*dma_device
;
1694 char name
[IB_DEVICE_NAME_MAX
];
1696 struct list_head event_handler_list
;
1697 spinlock_t event_handler_lock
;
1699 spinlock_t client_data_lock
;
1700 struct list_head core_list
;
1701 /* Access to the client_data_list is protected by the client_data_lock
1702 * spinlock and the lists_rwsem read-write semaphore */
1703 struct list_head client_data_list
;
1705 struct ib_cache cache
;
1707 * port_immutable is indexed by port number
1709 struct ib_port_immutable
*port_immutable
;
1711 int num_comp_vectors
;
1713 struct iw_cm_verbs
*iwcm
;
1716 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1717 * driver initialized data. The struct is kfree()'ed by the sysfs
1718 * core when the device is removed. A lifespan of -1 in the return
1719 * struct tells the core to set a default lifespan.
1721 struct rdma_hw_stats
*(*alloc_hw_stats
)(struct ib_device
*device
,
1724 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1725 * @index - The index in the value array we wish to have updated, or
1726 * num_counters if we want all stats updated
1728 * < 0 - Error, no counters updated
1729 * index - Updated the single counter pointed to by index
1730 * num_counters - Updated all counters (will reset the timestamp
1731 * and prevent further calls for lifespan milliseconds)
1732 * Drivers are allowed to update all counters in leiu of just the
1733 * one given in index at their option
1735 int (*get_hw_stats
)(struct ib_device
*device
,
1736 struct rdma_hw_stats
*stats
,
1737 u8 port
, int index
);
1738 int (*query_device
)(struct ib_device
*device
,
1739 struct ib_device_attr
*device_attr
,
1740 struct ib_udata
*udata
);
1741 int (*query_port
)(struct ib_device
*device
,
1743 struct ib_port_attr
*port_attr
);
1744 enum rdma_link_layer (*get_link_layer
)(struct ib_device
*device
,
1746 /* When calling get_netdev, the HW vendor's driver should return the
1747 * net device of device @device at port @port_num or NULL if such
1748 * a net device doesn't exist. The vendor driver should call dev_hold
1749 * on this net device. The HW vendor's device driver must guarantee
1750 * that this function returns NULL before the net device reaches
1751 * NETDEV_UNREGISTER_FINAL state.
1753 struct net_device
*(*get_netdev
)(struct ib_device
*device
,
1755 int (*query_gid
)(struct ib_device
*device
,
1756 u8 port_num
, int index
,
1758 /* When calling add_gid, the HW vendor's driver should
1759 * add the gid of device @device at gid index @index of
1760 * port @port_num to be @gid. Meta-info of that gid (for example,
1761 * the network device related to this gid is available
1762 * at @attr. @context allows the HW vendor driver to store extra
1763 * information together with a GID entry. The HW vendor may allocate
1764 * memory to contain this information and store it in @context when a
1765 * new GID entry is written to. Params are consistent until the next
1766 * call of add_gid or delete_gid. The function should return 0 on
1767 * success or error otherwise. The function could be called
1768 * concurrently for different ports. This function is only called
1769 * when roce_gid_table is used.
1771 int (*add_gid
)(struct ib_device
*device
,
1774 const union ib_gid
*gid
,
1775 const struct ib_gid_attr
*attr
,
1777 /* When calling del_gid, the HW vendor's driver should delete the
1778 * gid of device @device at gid index @index of port @port_num.
1779 * Upon the deletion of a GID entry, the HW vendor must free any
1780 * allocated memory. The caller will clear @context afterwards.
1781 * This function is only called when roce_gid_table is used.
1783 int (*del_gid
)(struct ib_device
*device
,
1787 int (*query_pkey
)(struct ib_device
*device
,
1788 u8 port_num
, u16 index
, u16
*pkey
);
1789 int (*modify_device
)(struct ib_device
*device
,
1790 int device_modify_mask
,
1791 struct ib_device_modify
*device_modify
);
1792 int (*modify_port
)(struct ib_device
*device
,
1793 u8 port_num
, int port_modify_mask
,
1794 struct ib_port_modify
*port_modify
);
1795 struct ib_ucontext
* (*alloc_ucontext
)(struct ib_device
*device
,
1796 struct ib_udata
*udata
);
1797 int (*dealloc_ucontext
)(struct ib_ucontext
*context
);
1798 int (*mmap
)(struct ib_ucontext
*context
,
1799 struct vm_area_struct
*vma
);
1800 struct ib_pd
* (*alloc_pd
)(struct ib_device
*device
,
1801 struct ib_ucontext
*context
,
1802 struct ib_udata
*udata
);
1803 int (*dealloc_pd
)(struct ib_pd
*pd
);
1804 struct ib_ah
* (*create_ah
)(struct ib_pd
*pd
,
1805 struct ib_ah_attr
*ah_attr
);
1806 int (*modify_ah
)(struct ib_ah
*ah
,
1807 struct ib_ah_attr
*ah_attr
);
1808 int (*query_ah
)(struct ib_ah
*ah
,
1809 struct ib_ah_attr
*ah_attr
);
1810 int (*destroy_ah
)(struct ib_ah
*ah
);
1811 struct ib_srq
* (*create_srq
)(struct ib_pd
*pd
,
1812 struct ib_srq_init_attr
*srq_init_attr
,
1813 struct ib_udata
*udata
);
1814 int (*modify_srq
)(struct ib_srq
*srq
,
1815 struct ib_srq_attr
*srq_attr
,
1816 enum ib_srq_attr_mask srq_attr_mask
,
1817 struct ib_udata
*udata
);
1818 int (*query_srq
)(struct ib_srq
*srq
,
1819 struct ib_srq_attr
*srq_attr
);
1820 int (*destroy_srq
)(struct ib_srq
*srq
);
1821 int (*post_srq_recv
)(struct ib_srq
*srq
,
1822 struct ib_recv_wr
*recv_wr
,
1823 struct ib_recv_wr
**bad_recv_wr
);
1824 struct ib_qp
* (*create_qp
)(struct ib_pd
*pd
,
1825 struct ib_qp_init_attr
*qp_init_attr
,
1826 struct ib_udata
*udata
);
1827 int (*modify_qp
)(struct ib_qp
*qp
,
1828 struct ib_qp_attr
*qp_attr
,
1830 struct ib_udata
*udata
);
1831 int (*query_qp
)(struct ib_qp
*qp
,
1832 struct ib_qp_attr
*qp_attr
,
1834 struct ib_qp_init_attr
*qp_init_attr
);
1835 int (*destroy_qp
)(struct ib_qp
*qp
);
1836 int (*post_send
)(struct ib_qp
*qp
,
1837 struct ib_send_wr
*send_wr
,
1838 struct ib_send_wr
**bad_send_wr
);
1839 int (*post_recv
)(struct ib_qp
*qp
,
1840 struct ib_recv_wr
*recv_wr
,
1841 struct ib_recv_wr
**bad_recv_wr
);
1842 struct ib_cq
* (*create_cq
)(struct ib_device
*device
,
1843 const struct ib_cq_init_attr
*attr
,
1844 struct ib_ucontext
*context
,
1845 struct ib_udata
*udata
);
1846 int (*modify_cq
)(struct ib_cq
*cq
, u16 cq_count
,
1848 int (*destroy_cq
)(struct ib_cq
*cq
);
1849 int (*resize_cq
)(struct ib_cq
*cq
, int cqe
,
1850 struct ib_udata
*udata
);
1851 int (*poll_cq
)(struct ib_cq
*cq
, int num_entries
,
1853 int (*peek_cq
)(struct ib_cq
*cq
, int wc_cnt
);
1854 int (*req_notify_cq
)(struct ib_cq
*cq
,
1855 enum ib_cq_notify_flags flags
);
1856 int (*req_ncomp_notif
)(struct ib_cq
*cq
,
1858 struct ib_mr
* (*get_dma_mr
)(struct ib_pd
*pd
,
1859 int mr_access_flags
);
1860 struct ib_mr
* (*reg_user_mr
)(struct ib_pd
*pd
,
1861 u64 start
, u64 length
,
1863 int mr_access_flags
,
1864 struct ib_udata
*udata
);
1865 int (*rereg_user_mr
)(struct ib_mr
*mr
,
1867 u64 start
, u64 length
,
1869 int mr_access_flags
,
1871 struct ib_udata
*udata
);
1872 int (*dereg_mr
)(struct ib_mr
*mr
);
1873 struct ib_mr
* (*alloc_mr
)(struct ib_pd
*pd
,
1874 enum ib_mr_type mr_type
,
1876 int (*map_mr_sg
)(struct ib_mr
*mr
,
1877 struct scatterlist
*sg
,
1879 unsigned int *sg_offset
);
1880 struct ib_mw
* (*alloc_mw
)(struct ib_pd
*pd
,
1881 enum ib_mw_type type
,
1882 struct ib_udata
*udata
);
1883 int (*dealloc_mw
)(struct ib_mw
*mw
);
1884 struct ib_fmr
* (*alloc_fmr
)(struct ib_pd
*pd
,
1885 int mr_access_flags
,
1886 struct ib_fmr_attr
*fmr_attr
);
1887 int (*map_phys_fmr
)(struct ib_fmr
*fmr
,
1888 u64
*page_list
, int list_len
,
1890 int (*unmap_fmr
)(struct list_head
*fmr_list
);
1891 int (*dealloc_fmr
)(struct ib_fmr
*fmr
);
1892 int (*attach_mcast
)(struct ib_qp
*qp
,
1895 int (*detach_mcast
)(struct ib_qp
*qp
,
1898 int (*process_mad
)(struct ib_device
*device
,
1899 int process_mad_flags
,
1901 const struct ib_wc
*in_wc
,
1902 const struct ib_grh
*in_grh
,
1903 const struct ib_mad_hdr
*in_mad
,
1905 struct ib_mad_hdr
*out_mad
,
1906 size_t *out_mad_size
,
1907 u16
*out_mad_pkey_index
);
1908 struct ib_xrcd
* (*alloc_xrcd
)(struct ib_device
*device
,
1909 struct ib_ucontext
*ucontext
,
1910 struct ib_udata
*udata
);
1911 int (*dealloc_xrcd
)(struct ib_xrcd
*xrcd
);
1912 struct ib_flow
* (*create_flow
)(struct ib_qp
*qp
,
1916 int (*destroy_flow
)(struct ib_flow
*flow_id
);
1917 int (*check_mr_status
)(struct ib_mr
*mr
, u32 check_mask
,
1918 struct ib_mr_status
*mr_status
);
1919 void (*disassociate_ucontext
)(struct ib_ucontext
*ibcontext
);
1920 void (*drain_rq
)(struct ib_qp
*qp
);
1921 void (*drain_sq
)(struct ib_qp
*qp
);
1922 int (*set_vf_link_state
)(struct ib_device
*device
, int vf
, u8 port
,
1924 int (*get_vf_config
)(struct ib_device
*device
, int vf
, u8 port
,
1925 struct ifla_vf_info
*ivf
);
1926 int (*get_vf_stats
)(struct ib_device
*device
, int vf
, u8 port
,
1927 struct ifla_vf_stats
*stats
);
1928 int (*set_vf_guid
)(struct ib_device
*device
, int vf
, u8 port
, u64 guid
,
1931 struct ib_dma_mapping_ops
*dma_ops
;
1933 struct module
*owner
;
1935 struct kobject
*ports_parent
;
1936 struct list_head port_list
;
1939 IB_DEV_UNINITIALIZED
,
1945 u64 uverbs_cmd_mask
;
1946 u64 uverbs_ex_cmd_mask
;
1954 struct ib_device_attr attrs
;
1955 struct attribute_group
*hw_stats_ag
;
1956 struct rdma_hw_stats
*hw_stats
;
1959 * The following mandatory functions are used only at device
1960 * registration. Keep functions such as these at the end of this
1961 * structure to avoid cache line misses when accessing struct ib_device
1964 int (*get_port_immutable
)(struct ib_device
*, u8
, struct ib_port_immutable
*);
1969 void (*add
) (struct ib_device
*);
1970 void (*remove
)(struct ib_device
*, void *client_data
);
1972 /* Returns the net_dev belonging to this ib_client and matching the
1974 * @dev: An RDMA device that the net_dev use for communication.
1975 * @port: A physical port number on the RDMA device.
1976 * @pkey: P_Key that the net_dev uses if applicable.
1977 * @gid: A GID that the net_dev uses to communicate.
1978 * @addr: An IP address the net_dev is configured with.
1979 * @client_data: The device's client data set by ib_set_client_data().
1981 * An ib_client that implements a net_dev on top of RDMA devices
1982 * (such as IP over IB) should implement this callback, allowing the
1983 * rdma_cm module to find the right net_dev for a given request.
1985 * The caller is responsible for calling dev_put on the returned
1987 struct net_device
*(*get_net_dev_by_params
)(
1988 struct ib_device
*dev
,
1991 const union ib_gid
*gid
,
1992 const struct sockaddr
*addr
,
1994 struct list_head list
;
1997 struct ib_device
*ib_alloc_device(size_t size
);
1998 void ib_dealloc_device(struct ib_device
*device
);
2000 int ib_register_device(struct ib_device
*device
,
2001 int (*port_callback
)(struct ib_device
*,
2002 u8
, struct kobject
*));
2003 void ib_unregister_device(struct ib_device
*device
);
2005 int ib_register_client (struct ib_client
*client
);
2006 void ib_unregister_client(struct ib_client
*client
);
2008 void *ib_get_client_data(struct ib_device
*device
, struct ib_client
*client
);
2009 void ib_set_client_data(struct ib_device
*device
, struct ib_client
*client
,
2012 static inline int ib_copy_from_udata(void *dest
, struct ib_udata
*udata
, size_t len
)
2014 return copy_from_user(dest
, udata
->inbuf
, len
) ? -EFAULT
: 0;
2017 static inline int ib_copy_to_udata(struct ib_udata
*udata
, void *src
, size_t len
)
2019 return copy_to_user(udata
->outbuf
, src
, len
) ? -EFAULT
: 0;
2022 static inline bool ib_is_udata_cleared(struct ib_udata
*udata
,
2026 const void __user
*p
= udata
->inbuf
+ offset
;
2030 if (len
> USHRT_MAX
)
2033 buf
= kmalloc(len
, GFP_KERNEL
);
2037 if (copy_from_user(buf
, p
, len
))
2040 ret
= !memchr_inv(buf
, 0, len
);
2048 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2049 * contains all required attributes and no attributes not allowed for
2050 * the given QP state transition.
2051 * @cur_state: Current QP state
2052 * @next_state: Next QP state
2054 * @mask: Mask of supplied QP attributes
2055 * @ll : link layer of port
2057 * This function is a helper function that a low-level driver's
2058 * modify_qp method can use to validate the consumer's input. It
2059 * checks that cur_state and next_state are valid QP states, that a
2060 * transition from cur_state to next_state is allowed by the IB spec,
2061 * and that the attribute mask supplied is allowed for the transition.
2063 int ib_modify_qp_is_ok(enum ib_qp_state cur_state
, enum ib_qp_state next_state
,
2064 enum ib_qp_type type
, enum ib_qp_attr_mask mask
,
2065 enum rdma_link_layer ll
);
2067 int ib_register_event_handler (struct ib_event_handler
*event_handler
);
2068 int ib_unregister_event_handler(struct ib_event_handler
*event_handler
);
2069 void ib_dispatch_event(struct ib_event
*event
);
2071 int ib_query_port(struct ib_device
*device
,
2072 u8 port_num
, struct ib_port_attr
*port_attr
);
2074 enum rdma_link_layer
rdma_port_get_link_layer(struct ib_device
*device
,
2078 * rdma_cap_ib_switch - Check if the device is IB switch
2079 * @device: Device to check
2081 * Device driver is responsible for setting is_switch bit on
2082 * in ib_device structure at init time.
2084 * Return: true if the device is IB switch.
2086 static inline bool rdma_cap_ib_switch(const struct ib_device
*device
)
2088 return device
->is_switch
;
2092 * rdma_start_port - Return the first valid port number for the device
2095 * @device: Device to be checked
2097 * Return start port number
2099 static inline u8
rdma_start_port(const struct ib_device
*device
)
2101 return rdma_cap_ib_switch(device
) ? 0 : 1;
2105 * rdma_end_port - Return the last valid port number for the device
2108 * @device: Device to be checked
2110 * Return last port number
2112 static inline u8
rdma_end_port(const struct ib_device
*device
)
2114 return rdma_cap_ib_switch(device
) ? 0 : device
->phys_port_cnt
;
2117 static inline bool rdma_protocol_ib(const struct ib_device
*device
, u8 port_num
)
2119 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IB
;
2122 static inline bool rdma_protocol_roce(const struct ib_device
*device
, u8 port_num
)
2124 return device
->port_immutable
[port_num
].core_cap_flags
&
2125 (RDMA_CORE_CAP_PROT_ROCE
| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP
);
2128 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device
*device
, u8 port_num
)
2130 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP
;
2133 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device
*device
, u8 port_num
)
2135 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_ROCE
;
2138 static inline bool rdma_protocol_iwarp(const struct ib_device
*device
, u8 port_num
)
2140 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IWARP
;
2143 static inline bool rdma_ib_or_roce(const struct ib_device
*device
, u8 port_num
)
2145 return rdma_protocol_ib(device
, port_num
) ||
2146 rdma_protocol_roce(device
, port_num
);
2150 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2151 * Management Datagrams.
2152 * @device: Device to check
2153 * @port_num: Port number to check
2155 * Management Datagrams (MAD) are a required part of the InfiniBand
2156 * specification and are supported on all InfiniBand devices. A slightly
2157 * extended version are also supported on OPA interfaces.
2159 * Return: true if the port supports sending/receiving of MAD packets.
2161 static inline bool rdma_cap_ib_mad(const struct ib_device
*device
, u8 port_num
)
2163 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_MAD
;
2167 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2168 * Management Datagrams.
2169 * @device: Device to check
2170 * @port_num: Port number to check
2172 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2173 * datagrams with their own versions. These OPA MADs share many but not all of
2174 * the characteristics of InfiniBand MADs.
2176 * OPA MADs differ in the following ways:
2178 * 1) MADs are variable size up to 2K
2179 * IBTA defined MADs remain fixed at 256 bytes
2180 * 2) OPA SMPs must carry valid PKeys
2181 * 3) OPA SMP packets are a different format
2183 * Return: true if the port supports OPA MAD packet formats.
2185 static inline bool rdma_cap_opa_mad(struct ib_device
*device
, u8 port_num
)
2187 return (device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_OPA_MAD
)
2188 == RDMA_CORE_CAP_OPA_MAD
;
2192 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2193 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2194 * @device: Device to check
2195 * @port_num: Port number to check
2197 * Each InfiniBand node is required to provide a Subnet Management Agent
2198 * that the subnet manager can access. Prior to the fabric being fully
2199 * configured by the subnet manager, the SMA is accessed via a well known
2200 * interface called the Subnet Management Interface (SMI). This interface
2201 * uses directed route packets to communicate with the SM to get around the
2202 * chicken and egg problem of the SM needing to know what's on the fabric
2203 * in order to configure the fabric, and needing to configure the fabric in
2204 * order to send packets to the devices on the fabric. These directed
2205 * route packets do not need the fabric fully configured in order to reach
2206 * their destination. The SMI is the only method allowed to send
2207 * directed route packets on an InfiniBand fabric.
2209 * Return: true if the port provides an SMI.
2211 static inline bool rdma_cap_ib_smi(const struct ib_device
*device
, u8 port_num
)
2213 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SMI
;
2217 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2218 * Communication Manager.
2219 * @device: Device to check
2220 * @port_num: Port number to check
2222 * The InfiniBand Communication Manager is one of many pre-defined General
2223 * Service Agents (GSA) that are accessed via the General Service
2224 * Interface (GSI). It's role is to facilitate establishment of connections
2225 * between nodes as well as other management related tasks for established
2228 * Return: true if the port supports an IB CM (this does not guarantee that
2229 * a CM is actually running however).
2231 static inline bool rdma_cap_ib_cm(const struct ib_device
*device
, u8 port_num
)
2233 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_CM
;
2237 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2238 * Communication Manager.
2239 * @device: Device to check
2240 * @port_num: Port number to check
2242 * Similar to above, but specific to iWARP connections which have a different
2243 * managment protocol than InfiniBand.
2245 * Return: true if the port supports an iWARP CM (this does not guarantee that
2246 * a CM is actually running however).
2248 static inline bool rdma_cap_iw_cm(const struct ib_device
*device
, u8 port_num
)
2250 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IW_CM
;
2254 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2255 * Subnet Administration.
2256 * @device: Device to check
2257 * @port_num: Port number to check
2259 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2260 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2261 * fabrics, devices should resolve routes to other hosts by contacting the
2262 * SA to query the proper route.
2264 * Return: true if the port should act as a client to the fabric Subnet
2265 * Administration interface. This does not imply that the SA service is
2268 static inline bool rdma_cap_ib_sa(const struct ib_device
*device
, u8 port_num
)
2270 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SA
;
2274 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2276 * @device: Device to check
2277 * @port_num: Port number to check
2279 * InfiniBand multicast registration is more complex than normal IPv4 or
2280 * IPv6 multicast registration. Each Host Channel Adapter must register
2281 * with the Subnet Manager when it wishes to join a multicast group. It
2282 * should do so only once regardless of how many queue pairs it subscribes
2283 * to this group. And it should leave the group only after all queue pairs
2284 * attached to the group have been detached.
2286 * Return: true if the port must undertake the additional adminstrative
2287 * overhead of registering/unregistering with the SM and tracking of the
2288 * total number of queue pairs attached to the multicast group.
2290 static inline bool rdma_cap_ib_mcast(const struct ib_device
*device
, u8 port_num
)
2292 return rdma_cap_ib_sa(device
, port_num
);
2296 * rdma_cap_af_ib - Check if the port of device has the capability
2297 * Native Infiniband Address.
2298 * @device: Device to check
2299 * @port_num: Port number to check
2301 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2302 * GID. RoCE uses a different mechanism, but still generates a GID via
2303 * a prescribed mechanism and port specific data.
2305 * Return: true if the port uses a GID address to identify devices on the
2308 static inline bool rdma_cap_af_ib(const struct ib_device
*device
, u8 port_num
)
2310 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_AF_IB
;
2314 * rdma_cap_eth_ah - Check if the port of device has the capability
2315 * Ethernet Address Handle.
2316 * @device: Device to check
2317 * @port_num: Port number to check
2319 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2320 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2321 * port. Normally, packet headers are generated by the sending host
2322 * adapter, but when sending connectionless datagrams, we must manually
2323 * inject the proper headers for the fabric we are communicating over.
2325 * Return: true if we are running as a RoCE port and must force the
2326 * addition of a Global Route Header built from our Ethernet Address
2327 * Handle into our header list for connectionless packets.
2329 static inline bool rdma_cap_eth_ah(const struct ib_device
*device
, u8 port_num
)
2331 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_ETH_AH
;
2335 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2338 * @port_num: Port number
2340 * This MAD size includes the MAD headers and MAD payload. No other headers
2343 * Return the max MAD size required by the Port. Will return 0 if the port
2344 * does not support MADs
2346 static inline size_t rdma_max_mad_size(const struct ib_device
*device
, u8 port_num
)
2348 return device
->port_immutable
[port_num
].max_mad_size
;
2352 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2353 * @device: Device to check
2354 * @port_num: Port number to check
2356 * RoCE GID table mechanism manages the various GIDs for a device.
2358 * NOTE: if allocating the port's GID table has failed, this call will still
2359 * return true, but any RoCE GID table API will fail.
2361 * Return: true if the port uses RoCE GID table mechanism in order to manage
2364 static inline bool rdma_cap_roce_gid_table(const struct ib_device
*device
,
2367 return rdma_protocol_roce(device
, port_num
) &&
2368 device
->add_gid
&& device
->del_gid
;
2372 * Check if the device supports READ W/ INVALIDATE.
2374 static inline bool rdma_cap_read_inv(struct ib_device
*dev
, u32 port_num
)
2377 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2378 * has support for it yet.
2380 return rdma_protocol_iwarp(dev
, port_num
);
2383 int ib_query_gid(struct ib_device
*device
,
2384 u8 port_num
, int index
, union ib_gid
*gid
,
2385 struct ib_gid_attr
*attr
);
2387 int ib_set_vf_link_state(struct ib_device
*device
, int vf
, u8 port
,
2389 int ib_get_vf_config(struct ib_device
*device
, int vf
, u8 port
,
2390 struct ifla_vf_info
*info
);
2391 int ib_get_vf_stats(struct ib_device
*device
, int vf
, u8 port
,
2392 struct ifla_vf_stats
*stats
);
2393 int ib_set_vf_guid(struct ib_device
*device
, int vf
, u8 port
, u64 guid
,
2396 int ib_query_pkey(struct ib_device
*device
,
2397 u8 port_num
, u16 index
, u16
*pkey
);
2399 int ib_modify_device(struct ib_device
*device
,
2400 int device_modify_mask
,
2401 struct ib_device_modify
*device_modify
);
2403 int ib_modify_port(struct ib_device
*device
,
2404 u8 port_num
, int port_modify_mask
,
2405 struct ib_port_modify
*port_modify
);
2407 int ib_find_gid(struct ib_device
*device
, union ib_gid
*gid
,
2408 enum ib_gid_type gid_type
, struct net_device
*ndev
,
2409 u8
*port_num
, u16
*index
);
2411 int ib_find_pkey(struct ib_device
*device
,
2412 u8 port_num
, u16 pkey
, u16
*index
);
2414 struct ib_pd
*ib_alloc_pd(struct ib_device
*device
);
2416 void ib_dealloc_pd(struct ib_pd
*pd
);
2419 * ib_create_ah - Creates an address handle for the given address vector.
2420 * @pd: The protection domain associated with the address handle.
2421 * @ah_attr: The attributes of the address vector.
2423 * The address handle is used to reference a local or global destination
2424 * in all UD QP post sends.
2426 struct ib_ah
*ib_create_ah(struct ib_pd
*pd
, struct ib_ah_attr
*ah_attr
);
2429 * ib_init_ah_from_wc - Initializes address handle attributes from a
2431 * @device: Device on which the received message arrived.
2432 * @port_num: Port on which the received message arrived.
2433 * @wc: Work completion associated with the received message.
2434 * @grh: References the received global route header. This parameter is
2435 * ignored unless the work completion indicates that the GRH is valid.
2436 * @ah_attr: Returned attributes that can be used when creating an address
2437 * handle for replying to the message.
2439 int ib_init_ah_from_wc(struct ib_device
*device
, u8 port_num
,
2440 const struct ib_wc
*wc
, const struct ib_grh
*grh
,
2441 struct ib_ah_attr
*ah_attr
);
2444 * ib_create_ah_from_wc - Creates an address handle associated with the
2445 * sender of the specified work completion.
2446 * @pd: The protection domain associated with the address handle.
2447 * @wc: Work completion information associated with a received message.
2448 * @grh: References the received global route header. This parameter is
2449 * ignored unless the work completion indicates that the GRH is valid.
2450 * @port_num: The outbound port number to associate with the address.
2452 * The address handle is used to reference a local or global destination
2453 * in all UD QP post sends.
2455 struct ib_ah
*ib_create_ah_from_wc(struct ib_pd
*pd
, const struct ib_wc
*wc
,
2456 const struct ib_grh
*grh
, u8 port_num
);
2459 * ib_modify_ah - Modifies the address vector associated with an address
2461 * @ah: The address handle to modify.
2462 * @ah_attr: The new address vector attributes to associate with the
2465 int ib_modify_ah(struct ib_ah
*ah
, struct ib_ah_attr
*ah_attr
);
2468 * ib_query_ah - Queries the address vector associated with an address
2470 * @ah: The address handle to query.
2471 * @ah_attr: The address vector attributes associated with the address
2474 int ib_query_ah(struct ib_ah
*ah
, struct ib_ah_attr
*ah_attr
);
2477 * ib_destroy_ah - Destroys an address handle.
2478 * @ah: The address handle to destroy.
2480 int ib_destroy_ah(struct ib_ah
*ah
);
2483 * ib_create_srq - Creates a SRQ associated with the specified protection
2485 * @pd: The protection domain associated with the SRQ.
2486 * @srq_init_attr: A list of initial attributes required to create the
2487 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2488 * the actual capabilities of the created SRQ.
2490 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2491 * requested size of the SRQ, and set to the actual values allocated
2492 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2493 * will always be at least as large as the requested values.
2495 struct ib_srq
*ib_create_srq(struct ib_pd
*pd
,
2496 struct ib_srq_init_attr
*srq_init_attr
);
2499 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2500 * @srq: The SRQ to modify.
2501 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2502 * the current values of selected SRQ attributes are returned.
2503 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2504 * are being modified.
2506 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2507 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2508 * the number of receives queued drops below the limit.
2510 int ib_modify_srq(struct ib_srq
*srq
,
2511 struct ib_srq_attr
*srq_attr
,
2512 enum ib_srq_attr_mask srq_attr_mask
);
2515 * ib_query_srq - Returns the attribute list and current values for the
2517 * @srq: The SRQ to query.
2518 * @srq_attr: The attributes of the specified SRQ.
2520 int ib_query_srq(struct ib_srq
*srq
,
2521 struct ib_srq_attr
*srq_attr
);
2524 * ib_destroy_srq - Destroys the specified SRQ.
2525 * @srq: The SRQ to destroy.
2527 int ib_destroy_srq(struct ib_srq
*srq
);
2530 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2531 * @srq: The SRQ to post the work request on.
2532 * @recv_wr: A list of work requests to post on the receive queue.
2533 * @bad_recv_wr: On an immediate failure, this parameter will reference
2534 * the work request that failed to be posted on the QP.
2536 static inline int ib_post_srq_recv(struct ib_srq
*srq
,
2537 struct ib_recv_wr
*recv_wr
,
2538 struct ib_recv_wr
**bad_recv_wr
)
2540 return srq
->device
->post_srq_recv(srq
, recv_wr
, bad_recv_wr
);
2544 * ib_create_qp - Creates a QP associated with the specified protection
2546 * @pd: The protection domain associated with the QP.
2547 * @qp_init_attr: A list of initial attributes required to create the
2548 * QP. If QP creation succeeds, then the attributes are updated to
2549 * the actual capabilities of the created QP.
2551 struct ib_qp
*ib_create_qp(struct ib_pd
*pd
,
2552 struct ib_qp_init_attr
*qp_init_attr
);
2555 * ib_modify_qp - Modifies the attributes for the specified QP and then
2556 * transitions the QP to the given state.
2557 * @qp: The QP to modify.
2558 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2559 * the current values of selected QP attributes are returned.
2560 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2561 * are being modified.
2563 int ib_modify_qp(struct ib_qp
*qp
,
2564 struct ib_qp_attr
*qp_attr
,
2568 * ib_query_qp - Returns the attribute list and current values for the
2570 * @qp: The QP to query.
2571 * @qp_attr: The attributes of the specified QP.
2572 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2573 * @qp_init_attr: Additional attributes of the selected QP.
2575 * The qp_attr_mask may be used to limit the query to gathering only the
2576 * selected attributes.
2578 int ib_query_qp(struct ib_qp
*qp
,
2579 struct ib_qp_attr
*qp_attr
,
2581 struct ib_qp_init_attr
*qp_init_attr
);
2584 * ib_destroy_qp - Destroys the specified QP.
2585 * @qp: The QP to destroy.
2587 int ib_destroy_qp(struct ib_qp
*qp
);
2590 * ib_open_qp - Obtain a reference to an existing sharable QP.
2591 * @xrcd - XRC domain
2592 * @qp_open_attr: Attributes identifying the QP to open.
2594 * Returns a reference to a sharable QP.
2596 struct ib_qp
*ib_open_qp(struct ib_xrcd
*xrcd
,
2597 struct ib_qp_open_attr
*qp_open_attr
);
2600 * ib_close_qp - Release an external reference to a QP.
2601 * @qp: The QP handle to release
2603 * The opened QP handle is released by the caller. The underlying
2604 * shared QP is not destroyed until all internal references are released.
2606 int ib_close_qp(struct ib_qp
*qp
);
2609 * ib_post_send - Posts a list of work requests to the send queue of
2611 * @qp: The QP to post the work request on.
2612 * @send_wr: A list of work requests to post on the send queue.
2613 * @bad_send_wr: On an immediate failure, this parameter will reference
2614 * the work request that failed to be posted on the QP.
2616 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2617 * error is returned, the QP state shall not be affected,
2618 * ib_post_send() will return an immediate error after queueing any
2619 * earlier work requests in the list.
2621 static inline int ib_post_send(struct ib_qp
*qp
,
2622 struct ib_send_wr
*send_wr
,
2623 struct ib_send_wr
**bad_send_wr
)
2625 return qp
->device
->post_send(qp
, send_wr
, bad_send_wr
);
2629 * ib_post_recv - Posts a list of work requests to the receive queue of
2631 * @qp: The QP to post the work request on.
2632 * @recv_wr: A list of work requests to post on the receive queue.
2633 * @bad_recv_wr: On an immediate failure, this parameter will reference
2634 * the work request that failed to be posted on the QP.
2636 static inline int ib_post_recv(struct ib_qp
*qp
,
2637 struct ib_recv_wr
*recv_wr
,
2638 struct ib_recv_wr
**bad_recv_wr
)
2640 return qp
->device
->post_recv(qp
, recv_wr
, bad_recv_wr
);
2643 struct ib_cq
*ib_alloc_cq(struct ib_device
*dev
, void *private,
2644 int nr_cqe
, int comp_vector
, enum ib_poll_context poll_ctx
);
2645 void ib_free_cq(struct ib_cq
*cq
);
2646 int ib_process_cq_direct(struct ib_cq
*cq
, int budget
);
2649 * ib_create_cq - Creates a CQ on the specified device.
2650 * @device: The device on which to create the CQ.
2651 * @comp_handler: A user-specified callback that is invoked when a
2652 * completion event occurs on the CQ.
2653 * @event_handler: A user-specified callback that is invoked when an
2654 * asynchronous event not associated with a completion occurs on the CQ.
2655 * @cq_context: Context associated with the CQ returned to the user via
2656 * the associated completion and event handlers.
2657 * @cq_attr: The attributes the CQ should be created upon.
2659 * Users can examine the cq structure to determine the actual CQ size.
2661 struct ib_cq
*ib_create_cq(struct ib_device
*device
,
2662 ib_comp_handler comp_handler
,
2663 void (*event_handler
)(struct ib_event
*, void *),
2665 const struct ib_cq_init_attr
*cq_attr
);
2668 * ib_resize_cq - Modifies the capacity of the CQ.
2669 * @cq: The CQ to resize.
2670 * @cqe: The minimum size of the CQ.
2672 * Users can examine the cq structure to determine the actual CQ size.
2674 int ib_resize_cq(struct ib_cq
*cq
, int cqe
);
2677 * ib_modify_cq - Modifies moderation params of the CQ
2678 * @cq: The CQ to modify.
2679 * @cq_count: number of CQEs that will trigger an event
2680 * @cq_period: max period of time in usec before triggering an event
2683 int ib_modify_cq(struct ib_cq
*cq
, u16 cq_count
, u16 cq_period
);
2686 * ib_destroy_cq - Destroys the specified CQ.
2687 * @cq: The CQ to destroy.
2689 int ib_destroy_cq(struct ib_cq
*cq
);
2692 * ib_poll_cq - poll a CQ for completion(s)
2693 * @cq:the CQ being polled
2694 * @num_entries:maximum number of completions to return
2695 * @wc:array of at least @num_entries &struct ib_wc where completions
2698 * Poll a CQ for (possibly multiple) completions. If the return value
2699 * is < 0, an error occurred. If the return value is >= 0, it is the
2700 * number of completions returned. If the return value is
2701 * non-negative and < num_entries, then the CQ was emptied.
2703 static inline int ib_poll_cq(struct ib_cq
*cq
, int num_entries
,
2706 return cq
->device
->poll_cq(cq
, num_entries
, wc
);
2710 * ib_peek_cq - Returns the number of unreaped completions currently
2711 * on the specified CQ.
2712 * @cq: The CQ to peek.
2713 * @wc_cnt: A minimum number of unreaped completions to check for.
2715 * If the number of unreaped completions is greater than or equal to wc_cnt,
2716 * this function returns wc_cnt, otherwise, it returns the actual number of
2717 * unreaped completions.
2719 int ib_peek_cq(struct ib_cq
*cq
, int wc_cnt
);
2722 * ib_req_notify_cq - Request completion notification on a CQ.
2723 * @cq: The CQ to generate an event for.
2725 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2726 * to request an event on the next solicited event or next work
2727 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2728 * may also be |ed in to request a hint about missed events, as
2732 * < 0 means an error occurred while requesting notification
2733 * == 0 means notification was requested successfully, and if
2734 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2735 * were missed and it is safe to wait for another event. In
2736 * this case is it guaranteed that any work completions added
2737 * to the CQ since the last CQ poll will trigger a completion
2738 * notification event.
2739 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2740 * in. It means that the consumer must poll the CQ again to
2741 * make sure it is empty to avoid missing an event because of a
2742 * race between requesting notification and an entry being
2743 * added to the CQ. This return value means it is possible
2744 * (but not guaranteed) that a work completion has been added
2745 * to the CQ since the last poll without triggering a
2746 * completion notification event.
2748 static inline int ib_req_notify_cq(struct ib_cq
*cq
,
2749 enum ib_cq_notify_flags flags
)
2751 return cq
->device
->req_notify_cq(cq
, flags
);
2755 * ib_req_ncomp_notif - Request completion notification when there are
2756 * at least the specified number of unreaped completions on the CQ.
2757 * @cq: The CQ to generate an event for.
2758 * @wc_cnt: The number of unreaped completions that should be on the
2759 * CQ before an event is generated.
2761 static inline int ib_req_ncomp_notif(struct ib_cq
*cq
, int wc_cnt
)
2763 return cq
->device
->req_ncomp_notif
?
2764 cq
->device
->req_ncomp_notif(cq
, wc_cnt
) :
2769 * ib_get_dma_mr - Returns a memory region for system memory that is
2771 * @pd: The protection domain associated with the memory region.
2772 * @mr_access_flags: Specifies the memory access rights.
2774 * Note that the ib_dma_*() functions defined below must be used
2775 * to create/destroy addresses used with the Lkey or Rkey returned
2776 * by ib_get_dma_mr().
2778 struct ib_mr
*ib_get_dma_mr(struct ib_pd
*pd
, int mr_access_flags
);
2781 * ib_dma_mapping_error - check a DMA addr for error
2782 * @dev: The device for which the dma_addr was created
2783 * @dma_addr: The DMA address to check
2785 static inline int ib_dma_mapping_error(struct ib_device
*dev
, u64 dma_addr
)
2788 return dev
->dma_ops
->mapping_error(dev
, dma_addr
);
2789 return dma_mapping_error(dev
->dma_device
, dma_addr
);
2793 * ib_dma_map_single - Map a kernel virtual address to DMA address
2794 * @dev: The device for which the dma_addr is to be created
2795 * @cpu_addr: The kernel virtual address
2796 * @size: The size of the region in bytes
2797 * @direction: The direction of the DMA
2799 static inline u64
ib_dma_map_single(struct ib_device
*dev
,
2800 void *cpu_addr
, size_t size
,
2801 enum dma_data_direction direction
)
2804 return dev
->dma_ops
->map_single(dev
, cpu_addr
, size
, direction
);
2805 return dma_map_single(dev
->dma_device
, cpu_addr
, size
, direction
);
2809 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2810 * @dev: The device for which the DMA address was created
2811 * @addr: The DMA address
2812 * @size: The size of the region in bytes
2813 * @direction: The direction of the DMA
2815 static inline void ib_dma_unmap_single(struct ib_device
*dev
,
2816 u64 addr
, size_t size
,
2817 enum dma_data_direction direction
)
2820 dev
->dma_ops
->unmap_single(dev
, addr
, size
, direction
);
2822 dma_unmap_single(dev
->dma_device
, addr
, size
, direction
);
2825 static inline u64
ib_dma_map_single_attrs(struct ib_device
*dev
,
2826 void *cpu_addr
, size_t size
,
2827 enum dma_data_direction direction
,
2828 struct dma_attrs
*attrs
)
2830 return dma_map_single_attrs(dev
->dma_device
, cpu_addr
, size
,
2834 static inline void ib_dma_unmap_single_attrs(struct ib_device
*dev
,
2835 u64 addr
, size_t size
,
2836 enum dma_data_direction direction
,
2837 struct dma_attrs
*attrs
)
2839 return dma_unmap_single_attrs(dev
->dma_device
, addr
, size
,
2844 * ib_dma_map_page - Map a physical page to DMA address
2845 * @dev: The device for which the dma_addr is to be created
2846 * @page: The page to be mapped
2847 * @offset: The offset within the page
2848 * @size: The size of the region in bytes
2849 * @direction: The direction of the DMA
2851 static inline u64
ib_dma_map_page(struct ib_device
*dev
,
2853 unsigned long offset
,
2855 enum dma_data_direction direction
)
2858 return dev
->dma_ops
->map_page(dev
, page
, offset
, size
, direction
);
2859 return dma_map_page(dev
->dma_device
, page
, offset
, size
, direction
);
2863 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2864 * @dev: The device for which the DMA address was created
2865 * @addr: The DMA address
2866 * @size: The size of the region in bytes
2867 * @direction: The direction of the DMA
2869 static inline void ib_dma_unmap_page(struct ib_device
*dev
,
2870 u64 addr
, size_t size
,
2871 enum dma_data_direction direction
)
2874 dev
->dma_ops
->unmap_page(dev
, addr
, size
, direction
);
2876 dma_unmap_page(dev
->dma_device
, addr
, size
, direction
);
2880 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2881 * @dev: The device for which the DMA addresses are to be created
2882 * @sg: The array of scatter/gather entries
2883 * @nents: The number of scatter/gather entries
2884 * @direction: The direction of the DMA
2886 static inline int ib_dma_map_sg(struct ib_device
*dev
,
2887 struct scatterlist
*sg
, int nents
,
2888 enum dma_data_direction direction
)
2891 return dev
->dma_ops
->map_sg(dev
, sg
, nents
, direction
);
2892 return dma_map_sg(dev
->dma_device
, sg
, nents
, direction
);
2896 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2897 * @dev: The device for which the DMA addresses were created
2898 * @sg: The array of scatter/gather entries
2899 * @nents: The number of scatter/gather entries
2900 * @direction: The direction of the DMA
2902 static inline void ib_dma_unmap_sg(struct ib_device
*dev
,
2903 struct scatterlist
*sg
, int nents
,
2904 enum dma_data_direction direction
)
2907 dev
->dma_ops
->unmap_sg(dev
, sg
, nents
, direction
);
2909 dma_unmap_sg(dev
->dma_device
, sg
, nents
, direction
);
2912 static inline int ib_dma_map_sg_attrs(struct ib_device
*dev
,
2913 struct scatterlist
*sg
, int nents
,
2914 enum dma_data_direction direction
,
2915 struct dma_attrs
*attrs
)
2917 return dma_map_sg_attrs(dev
->dma_device
, sg
, nents
, direction
, attrs
);
2920 static inline void ib_dma_unmap_sg_attrs(struct ib_device
*dev
,
2921 struct scatterlist
*sg
, int nents
,
2922 enum dma_data_direction direction
,
2923 struct dma_attrs
*attrs
)
2925 dma_unmap_sg_attrs(dev
->dma_device
, sg
, nents
, direction
, attrs
);
2928 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2929 * @dev: The device for which the DMA addresses were created
2930 * @sg: The scatter/gather entry
2932 * Note: this function is obsolete. To do: change all occurrences of
2933 * ib_sg_dma_address() into sg_dma_address().
2935 static inline u64
ib_sg_dma_address(struct ib_device
*dev
,
2936 struct scatterlist
*sg
)
2938 return sg_dma_address(sg
);
2942 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2943 * @dev: The device for which the DMA addresses were created
2944 * @sg: The scatter/gather entry
2946 * Note: this function is obsolete. To do: change all occurrences of
2947 * ib_sg_dma_len() into sg_dma_len().
2949 static inline unsigned int ib_sg_dma_len(struct ib_device
*dev
,
2950 struct scatterlist
*sg
)
2952 return sg_dma_len(sg
);
2956 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2957 * @dev: The device for which the DMA address was created
2958 * @addr: The DMA address
2959 * @size: The size of the region in bytes
2960 * @dir: The direction of the DMA
2962 static inline void ib_dma_sync_single_for_cpu(struct ib_device
*dev
,
2965 enum dma_data_direction dir
)
2968 dev
->dma_ops
->sync_single_for_cpu(dev
, addr
, size
, dir
);
2970 dma_sync_single_for_cpu(dev
->dma_device
, addr
, size
, dir
);
2974 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2975 * @dev: The device for which the DMA address was created
2976 * @addr: The DMA address
2977 * @size: The size of the region in bytes
2978 * @dir: The direction of the DMA
2980 static inline void ib_dma_sync_single_for_device(struct ib_device
*dev
,
2983 enum dma_data_direction dir
)
2986 dev
->dma_ops
->sync_single_for_device(dev
, addr
, size
, dir
);
2988 dma_sync_single_for_device(dev
->dma_device
, addr
, size
, dir
);
2992 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2993 * @dev: The device for which the DMA address is requested
2994 * @size: The size of the region to allocate in bytes
2995 * @dma_handle: A pointer for returning the DMA address of the region
2996 * @flag: memory allocator flags
2998 static inline void *ib_dma_alloc_coherent(struct ib_device
*dev
,
3004 return dev
->dma_ops
->alloc_coherent(dev
, size
, dma_handle
, flag
);
3009 ret
= dma_alloc_coherent(dev
->dma_device
, size
, &handle
, flag
);
3010 *dma_handle
= handle
;
3016 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3017 * @dev: The device for which the DMA addresses were allocated
3018 * @size: The size of the region
3019 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3020 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3022 static inline void ib_dma_free_coherent(struct ib_device
*dev
,
3023 size_t size
, void *cpu_addr
,
3027 dev
->dma_ops
->free_coherent(dev
, size
, cpu_addr
, dma_handle
);
3029 dma_free_coherent(dev
->dma_device
, size
, cpu_addr
, dma_handle
);
3033 * ib_dereg_mr - Deregisters a memory region and removes it from the
3034 * HCA translation table.
3035 * @mr: The memory region to deregister.
3037 * This function can fail, if the memory region has memory windows bound to it.
3039 int ib_dereg_mr(struct ib_mr
*mr
);
3041 struct ib_mr
*ib_alloc_mr(struct ib_pd
*pd
,
3042 enum ib_mr_type mr_type
,
3046 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3048 * @mr - struct ib_mr pointer to be updated.
3049 * @newkey - new key to be used.
3051 static inline void ib_update_fast_reg_key(struct ib_mr
*mr
, u8 newkey
)
3053 mr
->lkey
= (mr
->lkey
& 0xffffff00) | newkey
;
3054 mr
->rkey
= (mr
->rkey
& 0xffffff00) | newkey
;
3058 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3059 * for calculating a new rkey for type 2 memory windows.
3060 * @rkey - the rkey to increment.
3062 static inline u32
ib_inc_rkey(u32 rkey
)
3064 const u32 mask
= 0x000000ff;
3065 return ((rkey
+ 1) & mask
) | (rkey
& ~mask
);
3069 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3070 * @pd: The protection domain associated with the unmapped region.
3071 * @mr_access_flags: Specifies the memory access rights.
3072 * @fmr_attr: Attributes of the unmapped region.
3074 * A fast memory region must be mapped before it can be used as part of
3077 struct ib_fmr
*ib_alloc_fmr(struct ib_pd
*pd
,
3078 int mr_access_flags
,
3079 struct ib_fmr_attr
*fmr_attr
);
3082 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3083 * @fmr: The fast memory region to associate with the pages.
3084 * @page_list: An array of physical pages to map to the fast memory region.
3085 * @list_len: The number of pages in page_list.
3086 * @iova: The I/O virtual address to use with the mapped region.
3088 static inline int ib_map_phys_fmr(struct ib_fmr
*fmr
,
3089 u64
*page_list
, int list_len
,
3092 return fmr
->device
->map_phys_fmr(fmr
, page_list
, list_len
, iova
);
3096 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3097 * @fmr_list: A linked list of fast memory regions to unmap.
3099 int ib_unmap_fmr(struct list_head
*fmr_list
);
3102 * ib_dealloc_fmr - Deallocates a fast memory region.
3103 * @fmr: The fast memory region to deallocate.
3105 int ib_dealloc_fmr(struct ib_fmr
*fmr
);
3108 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3109 * @qp: QP to attach to the multicast group. The QP must be type
3111 * @gid: Multicast group GID.
3112 * @lid: Multicast group LID in host byte order.
3114 * In order to send and receive multicast packets, subnet
3115 * administration must have created the multicast group and configured
3116 * the fabric appropriately. The port associated with the specified
3117 * QP must also be a member of the multicast group.
3119 int ib_attach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
3122 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3123 * @qp: QP to detach from the multicast group.
3124 * @gid: Multicast group GID.
3125 * @lid: Multicast group LID in host byte order.
3127 int ib_detach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
3130 * ib_alloc_xrcd - Allocates an XRC domain.
3131 * @device: The device on which to allocate the XRC domain.
3133 struct ib_xrcd
*ib_alloc_xrcd(struct ib_device
*device
);
3136 * ib_dealloc_xrcd - Deallocates an XRC domain.
3137 * @xrcd: The XRC domain to deallocate.
3139 int ib_dealloc_xrcd(struct ib_xrcd
*xrcd
);
3141 struct ib_flow
*ib_create_flow(struct ib_qp
*qp
,
3142 struct ib_flow_attr
*flow_attr
, int domain
);
3143 int ib_destroy_flow(struct ib_flow
*flow_id
);
3145 static inline int ib_check_mr_access(int flags
)
3148 * Local write permission is required if remote write or
3149 * remote atomic permission is also requested.
3151 if (flags
& (IB_ACCESS_REMOTE_ATOMIC
| IB_ACCESS_REMOTE_WRITE
) &&
3152 !(flags
& IB_ACCESS_LOCAL_WRITE
))
3159 * ib_check_mr_status: lightweight check of MR status.
3160 * This routine may provide status checks on a selected
3161 * ib_mr. first use is for signature status check.
3163 * @mr: A memory region.
3164 * @check_mask: Bitmask of which checks to perform from
3165 * ib_mr_status_check enumeration.
3166 * @mr_status: The container of relevant status checks.
3167 * failed checks will be indicated in the status bitmask
3168 * and the relevant info shall be in the error item.
3170 int ib_check_mr_status(struct ib_mr
*mr
, u32 check_mask
,
3171 struct ib_mr_status
*mr_status
);
3173 struct net_device
*ib_get_net_dev_by_params(struct ib_device
*dev
, u8 port
,
3174 u16 pkey
, const union ib_gid
*gid
,
3175 const struct sockaddr
*addr
);
3177 int ib_map_mr_sg(struct ib_mr
*mr
, struct scatterlist
*sg
, int sg_nents
,
3178 unsigned int *sg_offset
, unsigned int page_size
);
3181 ib_map_mr_sg_zbva(struct ib_mr
*mr
, struct scatterlist
*sg
, int sg_nents
,
3182 unsigned int *sg_offset
, unsigned int page_size
)
3186 n
= ib_map_mr_sg(mr
, sg
, sg_nents
, sg_offset
, page_size
);
3192 int ib_sg_to_pages(struct ib_mr
*mr
, struct scatterlist
*sgl
, int sg_nents
,
3193 unsigned int *sg_offset
, int (*set_page
)(struct ib_mr
*, u64
));
3195 void ib_drain_rq(struct ib_qp
*qp
);
3196 void ib_drain_sq(struct ib_qp
*qp
);
3197 void ib_drain_qp(struct ib_qp
*qp
);
3198 #endif /* IB_VERBS_H */