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
,
568 const char *__attribute_const__
ib_event_msg(enum ib_event_type event
);
571 struct ib_device
*device
;
579 enum ib_event_type event
;
582 struct ib_event_handler
{
583 struct ib_device
*device
;
584 void (*handler
)(struct ib_event_handler
*, struct ib_event
*);
585 struct list_head list
;
588 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
590 (_ptr)->device = _device; \
591 (_ptr)->handler = _handler; \
592 INIT_LIST_HEAD(&(_ptr)->list); \
595 struct ib_global_route
{
604 __be32 version_tclass_flow
;
612 union rdma_network_hdr
{
615 /* The IB spec states that if it's IPv4, the header
616 * is located in the last 20 bytes of the header.
619 struct iphdr roce4grh
;
624 IB_MULTICAST_QPN
= 0xffffff
627 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
628 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
635 IB_RATE_PORT_CURRENT
= 0,
636 IB_RATE_2_5_GBPS
= 2,
644 IB_RATE_120_GBPS
= 10,
645 IB_RATE_14_GBPS
= 11,
646 IB_RATE_56_GBPS
= 12,
647 IB_RATE_112_GBPS
= 13,
648 IB_RATE_168_GBPS
= 14,
649 IB_RATE_25_GBPS
= 15,
650 IB_RATE_100_GBPS
= 16,
651 IB_RATE_200_GBPS
= 17,
652 IB_RATE_300_GBPS
= 18
656 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
657 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
658 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
659 * @rate: rate to convert.
661 __attribute_const__
int ib_rate_to_mult(enum ib_rate rate
);
664 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
665 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
666 * @rate: rate to convert.
668 __attribute_const__
int ib_rate_to_mbps(enum ib_rate rate
);
672 * enum ib_mr_type - memory region type
673 * @IB_MR_TYPE_MEM_REG: memory region that is used for
674 * normal registration
675 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
676 * signature operations (data-integrity
678 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
679 * register any arbitrary sg lists (without
680 * the normal mr constraints - see
685 IB_MR_TYPE_SIGNATURE
,
691 * IB_SIG_TYPE_NONE: Unprotected.
692 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
694 enum ib_signature_type
{
700 * Signature T10-DIF block-guard types
701 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
702 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
704 enum ib_t10_dif_bg_type
{
710 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
712 * @bg_type: T10-DIF block guard type (CRC|CSUM)
713 * @pi_interval: protection information interval.
714 * @bg: seed of guard computation.
715 * @app_tag: application tag of guard block
716 * @ref_tag: initial guard block reference tag.
717 * @ref_remap: Indicate wethear the reftag increments each block
718 * @app_escape: Indicate to skip block check if apptag=0xffff
719 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
720 * @apptag_check_mask: check bitmask of application tag.
722 struct ib_t10_dif_domain
{
723 enum ib_t10_dif_bg_type bg_type
;
731 u16 apptag_check_mask
;
735 * struct ib_sig_domain - Parameters for signature domain
736 * @sig_type: specific signauture type
737 * @sig: union of all signature domain attributes that may
738 * be used to set domain layout.
740 struct ib_sig_domain
{
741 enum ib_signature_type sig_type
;
743 struct ib_t10_dif_domain dif
;
748 * struct ib_sig_attrs - Parameters for signature handover operation
749 * @check_mask: bitmask for signature byte check (8 bytes)
750 * @mem: memory domain layout desciptor.
751 * @wire: wire domain layout desciptor.
753 struct ib_sig_attrs
{
755 struct ib_sig_domain mem
;
756 struct ib_sig_domain wire
;
759 enum ib_sig_err_type
{
766 * struct ib_sig_err - signature error descriptor
769 enum ib_sig_err_type err_type
;
776 enum ib_mr_status_check
{
777 IB_MR_CHECK_SIG_STATUS
= 1,
781 * struct ib_mr_status - Memory region status container
783 * @fail_status: Bitmask of MR checks status. For each
784 * failed check a corresponding status bit is set.
785 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
788 struct ib_mr_status
{
790 struct ib_sig_err sig_err
;
794 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
796 * @mult: multiple to convert.
798 __attribute_const__
enum ib_rate
mult_to_ib_rate(int mult
);
801 struct ib_global_route grh
;
815 IB_WC_LOC_EEC_OP_ERR
,
820 IB_WC_LOC_ACCESS_ERR
,
821 IB_WC_REM_INV_REQ_ERR
,
822 IB_WC_REM_ACCESS_ERR
,
825 IB_WC_RNR_RETRY_EXC_ERR
,
826 IB_WC_LOC_RDD_VIOL_ERR
,
827 IB_WC_REM_INV_RD_REQ_ERR
,
830 IB_WC_INV_EEC_STATE_ERR
,
832 IB_WC_RESP_TIMEOUT_ERR
,
836 const char *__attribute_const__
ib_wc_status_msg(enum ib_wc_status status
);
847 IB_WC_MASKED_COMP_SWAP
,
848 IB_WC_MASKED_FETCH_ADD
,
850 * Set value of IB_WC_RECV so consumers can test if a completion is a
851 * receive by testing (opcode & IB_WC_RECV).
854 IB_WC_RECV_RDMA_WITH_IMM
859 IB_WC_WITH_IMM
= (1<<1),
860 IB_WC_WITH_INVALIDATE
= (1<<2),
861 IB_WC_IP_CSUM_OK
= (1<<3),
862 IB_WC_WITH_SMAC
= (1<<4),
863 IB_WC_WITH_VLAN
= (1<<5),
864 IB_WC_WITH_NETWORK_HDR_TYPE
= (1<<6),
870 struct ib_cqe
*wr_cqe
;
872 enum ib_wc_status status
;
873 enum ib_wc_opcode opcode
;
887 u8 port_num
; /* valid only for DR SMPs on switches */
893 enum ib_cq_notify_flags
{
894 IB_CQ_SOLICITED
= 1 << 0,
895 IB_CQ_NEXT_COMP
= 1 << 1,
896 IB_CQ_SOLICITED_MASK
= IB_CQ_SOLICITED
| IB_CQ_NEXT_COMP
,
897 IB_CQ_REPORT_MISSED_EVENTS
= 1 << 2,
905 enum ib_srq_attr_mask
{
906 IB_SRQ_MAX_WR
= 1 << 0,
907 IB_SRQ_LIMIT
= 1 << 1,
916 struct ib_srq_init_attr
{
917 void (*event_handler
)(struct ib_event
*, void *);
919 struct ib_srq_attr attr
;
920 enum ib_srq_type srq_type
;
924 struct ib_xrcd
*xrcd
;
938 * Maximum number of rdma_rw_ctx structures in flight at a time.
939 * ib_create_qp() will calculate the right amount of neededed WRs
940 * and MRs based on this.
952 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
953 * here (and in that order) since the MAD layer uses them as
954 * indices into a 2-entry table.
963 IB_QPT_RAW_ETHERTYPE
,
964 IB_QPT_RAW_PACKET
= 8,
968 /* Reserve a range for qp types internal to the low level driver.
969 * These qp types will not be visible at the IB core layer, so the
970 * IB_QPT_MAX usages should not be affected in the core layer
972 IB_QPT_RESERVED1
= 0x1000,
984 enum ib_qp_create_flags
{
985 IB_QP_CREATE_IPOIB_UD_LSO
= 1 << 0,
986 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK
= 1 << 1,
987 IB_QP_CREATE_CROSS_CHANNEL
= 1 << 2,
988 IB_QP_CREATE_MANAGED_SEND
= 1 << 3,
989 IB_QP_CREATE_MANAGED_RECV
= 1 << 4,
990 IB_QP_CREATE_NETIF_QP
= 1 << 5,
991 IB_QP_CREATE_SIGNATURE_EN
= 1 << 6,
992 IB_QP_CREATE_USE_GFP_NOIO
= 1 << 7,
993 IB_QP_CREATE_SCATTER_FCS
= 1 << 8,
994 /* reserve bits 26-31 for low level drivers' internal use */
995 IB_QP_CREATE_RESERVED_START
= 1 << 26,
996 IB_QP_CREATE_RESERVED_END
= 1 << 31,
1000 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1001 * callback to destroy the passed in QP.
1004 struct ib_qp_init_attr
{
1005 void (*event_handler
)(struct ib_event
*, void *);
1007 struct ib_cq
*send_cq
;
1008 struct ib_cq
*recv_cq
;
1010 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
1011 struct ib_qp_cap cap
;
1012 enum ib_sig_type sq_sig_type
;
1013 enum ib_qp_type qp_type
;
1014 enum ib_qp_create_flags create_flags
;
1017 * Only needed for special QP types, or when using the RW API.
1020 struct ib_rwq_ind_table
*rwq_ind_tbl
;
1023 struct ib_qp_open_attr
{
1024 void (*event_handler
)(struct ib_event
*, void *);
1027 enum ib_qp_type qp_type
;
1030 enum ib_rnr_timeout
{
1031 IB_RNR_TIMER_655_36
= 0,
1032 IB_RNR_TIMER_000_01
= 1,
1033 IB_RNR_TIMER_000_02
= 2,
1034 IB_RNR_TIMER_000_03
= 3,
1035 IB_RNR_TIMER_000_04
= 4,
1036 IB_RNR_TIMER_000_06
= 5,
1037 IB_RNR_TIMER_000_08
= 6,
1038 IB_RNR_TIMER_000_12
= 7,
1039 IB_RNR_TIMER_000_16
= 8,
1040 IB_RNR_TIMER_000_24
= 9,
1041 IB_RNR_TIMER_000_32
= 10,
1042 IB_RNR_TIMER_000_48
= 11,
1043 IB_RNR_TIMER_000_64
= 12,
1044 IB_RNR_TIMER_000_96
= 13,
1045 IB_RNR_TIMER_001_28
= 14,
1046 IB_RNR_TIMER_001_92
= 15,
1047 IB_RNR_TIMER_002_56
= 16,
1048 IB_RNR_TIMER_003_84
= 17,
1049 IB_RNR_TIMER_005_12
= 18,
1050 IB_RNR_TIMER_007_68
= 19,
1051 IB_RNR_TIMER_010_24
= 20,
1052 IB_RNR_TIMER_015_36
= 21,
1053 IB_RNR_TIMER_020_48
= 22,
1054 IB_RNR_TIMER_030_72
= 23,
1055 IB_RNR_TIMER_040_96
= 24,
1056 IB_RNR_TIMER_061_44
= 25,
1057 IB_RNR_TIMER_081_92
= 26,
1058 IB_RNR_TIMER_122_88
= 27,
1059 IB_RNR_TIMER_163_84
= 28,
1060 IB_RNR_TIMER_245_76
= 29,
1061 IB_RNR_TIMER_327_68
= 30,
1062 IB_RNR_TIMER_491_52
= 31
1065 enum ib_qp_attr_mask
{
1067 IB_QP_CUR_STATE
= (1<<1),
1068 IB_QP_EN_SQD_ASYNC_NOTIFY
= (1<<2),
1069 IB_QP_ACCESS_FLAGS
= (1<<3),
1070 IB_QP_PKEY_INDEX
= (1<<4),
1071 IB_QP_PORT
= (1<<5),
1072 IB_QP_QKEY
= (1<<6),
1074 IB_QP_PATH_MTU
= (1<<8),
1075 IB_QP_TIMEOUT
= (1<<9),
1076 IB_QP_RETRY_CNT
= (1<<10),
1077 IB_QP_RNR_RETRY
= (1<<11),
1078 IB_QP_RQ_PSN
= (1<<12),
1079 IB_QP_MAX_QP_RD_ATOMIC
= (1<<13),
1080 IB_QP_ALT_PATH
= (1<<14),
1081 IB_QP_MIN_RNR_TIMER
= (1<<15),
1082 IB_QP_SQ_PSN
= (1<<16),
1083 IB_QP_MAX_DEST_RD_ATOMIC
= (1<<17),
1084 IB_QP_PATH_MIG_STATE
= (1<<18),
1085 IB_QP_CAP
= (1<<19),
1086 IB_QP_DEST_QPN
= (1<<20),
1087 IB_QP_RESERVED1
= (1<<21),
1088 IB_QP_RESERVED2
= (1<<22),
1089 IB_QP_RESERVED3
= (1<<23),
1090 IB_QP_RESERVED4
= (1<<24),
1115 enum ib_qp_state qp_state
;
1116 enum ib_qp_state cur_qp_state
;
1117 enum ib_mtu path_mtu
;
1118 enum ib_mig_state path_mig_state
;
1123 int qp_access_flags
;
1124 struct ib_qp_cap cap
;
1125 struct ib_ah_attr ah_attr
;
1126 struct ib_ah_attr alt_ah_attr
;
1129 u8 en_sqd_async_notify
;
1132 u8 max_dest_rd_atomic
;
1144 IB_WR_RDMA_WRITE_WITH_IMM
,
1146 IB_WR_SEND_WITH_IMM
,
1148 IB_WR_ATOMIC_CMP_AND_SWP
,
1149 IB_WR_ATOMIC_FETCH_AND_ADD
,
1151 IB_WR_SEND_WITH_INV
,
1152 IB_WR_RDMA_READ_WITH_INV
,
1155 IB_WR_MASKED_ATOMIC_CMP_AND_SWP
,
1156 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD
,
1158 /* reserve values for low level drivers' internal use.
1159 * These values will not be used at all in the ib core layer.
1161 IB_WR_RESERVED1
= 0xf0,
1173 enum ib_send_flags
{
1175 IB_SEND_SIGNALED
= (1<<1),
1176 IB_SEND_SOLICITED
= (1<<2),
1177 IB_SEND_INLINE
= (1<<3),
1178 IB_SEND_IP_CSUM
= (1<<4),
1180 /* reserve bits 26-31 for low level drivers' internal use */
1181 IB_SEND_RESERVED_START
= (1 << 26),
1182 IB_SEND_RESERVED_END
= (1 << 31),
1192 void (*done
)(struct ib_cq
*cq
, struct ib_wc
*wc
);
1196 struct ib_send_wr
*next
;
1199 struct ib_cqe
*wr_cqe
;
1201 struct ib_sge
*sg_list
;
1203 enum ib_wr_opcode opcode
;
1207 u32 invalidate_rkey
;
1212 struct ib_send_wr wr
;
1217 static inline struct ib_rdma_wr
*rdma_wr(struct ib_send_wr
*wr
)
1219 return container_of(wr
, struct ib_rdma_wr
, wr
);
1222 struct ib_atomic_wr
{
1223 struct ib_send_wr wr
;
1227 u64 compare_add_mask
;
1232 static inline struct ib_atomic_wr
*atomic_wr(struct ib_send_wr
*wr
)
1234 return container_of(wr
, struct ib_atomic_wr
, wr
);
1238 struct ib_send_wr wr
;
1245 u16 pkey_index
; /* valid for GSI only */
1246 u8 port_num
; /* valid for DR SMPs on switch only */
1249 static inline struct ib_ud_wr
*ud_wr(struct ib_send_wr
*wr
)
1251 return container_of(wr
, struct ib_ud_wr
, wr
);
1255 struct ib_send_wr wr
;
1261 static inline struct ib_reg_wr
*reg_wr(struct ib_send_wr
*wr
)
1263 return container_of(wr
, struct ib_reg_wr
, wr
);
1266 struct ib_sig_handover_wr
{
1267 struct ib_send_wr wr
;
1268 struct ib_sig_attrs
*sig_attrs
;
1269 struct ib_mr
*sig_mr
;
1271 struct ib_sge
*prot
;
1274 static inline struct ib_sig_handover_wr
*sig_handover_wr(struct ib_send_wr
*wr
)
1276 return container_of(wr
, struct ib_sig_handover_wr
, wr
);
1280 struct ib_recv_wr
*next
;
1283 struct ib_cqe
*wr_cqe
;
1285 struct ib_sge
*sg_list
;
1289 enum ib_access_flags
{
1290 IB_ACCESS_LOCAL_WRITE
= 1,
1291 IB_ACCESS_REMOTE_WRITE
= (1<<1),
1292 IB_ACCESS_REMOTE_READ
= (1<<2),
1293 IB_ACCESS_REMOTE_ATOMIC
= (1<<3),
1294 IB_ACCESS_MW_BIND
= (1<<4),
1295 IB_ZERO_BASED
= (1<<5),
1296 IB_ACCESS_ON_DEMAND
= (1<<6),
1300 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1301 * are hidden here instead of a uapi header!
1303 enum ib_mr_rereg_flags
{
1304 IB_MR_REREG_TRANS
= 1,
1305 IB_MR_REREG_PD
= (1<<1),
1306 IB_MR_REREG_ACCESS
= (1<<2),
1307 IB_MR_REREG_SUPPORTED
= ((IB_MR_REREG_ACCESS
<< 1) - 1)
1310 struct ib_fmr_attr
{
1318 struct ib_ucontext
{
1319 struct ib_device
*device
;
1320 struct list_head pd_list
;
1321 struct list_head mr_list
;
1322 struct list_head mw_list
;
1323 struct list_head cq_list
;
1324 struct list_head qp_list
;
1325 struct list_head srq_list
;
1326 struct list_head ah_list
;
1327 struct list_head xrcd_list
;
1328 struct list_head rule_list
;
1329 struct list_head wq_list
;
1330 struct list_head rwq_ind_tbl_list
;
1334 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1335 struct rb_root umem_tree
;
1337 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1338 * mmu notifiers registration.
1340 struct rw_semaphore umem_rwsem
;
1341 void (*invalidate_range
)(struct ib_umem
*umem
,
1342 unsigned long start
, unsigned long end
);
1344 struct mmu_notifier mn
;
1345 atomic_t notifier_count
;
1346 /* A list of umems that don't have private mmu notifier counters yet. */
1347 struct list_head no_private_counters
;
1353 u64 user_handle
; /* handle given to us by userspace */
1354 struct ib_ucontext
*context
; /* associated user context */
1355 void *object
; /* containing object */
1356 struct list_head list
; /* link to context's list */
1357 int id
; /* index into kernel idr */
1359 struct rw_semaphore mutex
; /* protects .live */
1360 struct rcu_head rcu
; /* kfree_rcu() overhead */
1365 const void __user
*inbuf
;
1366 void __user
*outbuf
;
1373 struct ib_device
*device
;
1374 struct ib_uobject
*uobject
;
1375 atomic_t usecnt
; /* count all resources */
1376 struct ib_mr
*local_mr
;
1380 struct ib_device
*device
;
1381 atomic_t usecnt
; /* count all exposed resources */
1382 struct inode
*inode
;
1384 struct mutex tgt_qp_mutex
;
1385 struct list_head tgt_qp_list
;
1389 struct ib_device
*device
;
1391 struct ib_uobject
*uobject
;
1394 typedef void (*ib_comp_handler
)(struct ib_cq
*cq
, void *cq_context
);
1396 enum ib_poll_context
{
1397 IB_POLL_DIRECT
, /* caller context, no hw completions */
1398 IB_POLL_SOFTIRQ
, /* poll from softirq context */
1399 IB_POLL_WORKQUEUE
, /* poll from workqueue */
1403 struct ib_device
*device
;
1404 struct ib_uobject
*uobject
;
1405 ib_comp_handler comp_handler
;
1406 void (*event_handler
)(struct ib_event
*, void *);
1409 atomic_t usecnt
; /* count number of work queues */
1410 enum ib_poll_context poll_ctx
;
1413 struct irq_poll iop
;
1414 struct work_struct work
;
1419 struct ib_device
*device
;
1421 struct ib_uobject
*uobject
;
1422 void (*event_handler
)(struct ib_event
*, void *);
1424 enum ib_srq_type srq_type
;
1429 struct ib_xrcd
*xrcd
;
1447 struct ib_device
*device
;
1448 struct ib_uobject
*uobject
;
1450 void (*event_handler
)(struct ib_event
*, void *);
1454 enum ib_wq_state state
;
1455 enum ib_wq_type wq_type
;
1459 struct ib_wq_init_attr
{
1461 enum ib_wq_type wq_type
;
1465 void (*event_handler
)(struct ib_event
*, void *);
1468 enum ib_wq_attr_mask
{
1469 IB_WQ_STATE
= 1 << 0,
1470 IB_WQ_CUR_STATE
= 1 << 1,
1474 enum ib_wq_state wq_state
;
1475 enum ib_wq_state curr_wq_state
;
1478 struct ib_rwq_ind_table
{
1479 struct ib_device
*device
;
1480 struct ib_uobject
*uobject
;
1483 u32 log_ind_tbl_size
;
1484 struct ib_wq
**ind_tbl
;
1487 struct ib_rwq_ind_table_init_attr
{
1488 u32 log_ind_tbl_size
;
1489 /* Each entry is a pointer to Receive Work Queue */
1490 struct ib_wq
**ind_tbl
;
1494 struct ib_device
*device
;
1496 struct ib_cq
*send_cq
;
1497 struct ib_cq
*recv_cq
;
1500 struct list_head rdma_mrs
;
1501 struct list_head sig_mrs
;
1503 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
1504 struct list_head xrcd_list
;
1506 /* count times opened, mcast attaches, flow attaches */
1508 struct list_head open_list
;
1509 struct ib_qp
*real_qp
;
1510 struct ib_uobject
*uobject
;
1511 void (*event_handler
)(struct ib_event
*, void *);
1514 enum ib_qp_type qp_type
;
1515 struct ib_rwq_ind_table
*rwq_ind_tbl
;
1519 struct ib_device
*device
;
1525 unsigned int page_size
;
1528 struct ib_uobject
*uobject
; /* user */
1529 struct list_head qp_entry
; /* FR */
1534 struct ib_device
*device
;
1536 struct ib_uobject
*uobject
;
1538 enum ib_mw_type type
;
1542 struct ib_device
*device
;
1544 struct list_head list
;
1549 /* Supported steering options */
1550 enum ib_flow_attr_type
{
1551 /* steering according to rule specifications */
1552 IB_FLOW_ATTR_NORMAL
= 0x0,
1553 /* default unicast and multicast rule -
1554 * receive all Eth traffic which isn't steered to any QP
1556 IB_FLOW_ATTR_ALL_DEFAULT
= 0x1,
1557 /* default multicast rule -
1558 * receive all Eth multicast traffic which isn't steered to any QP
1560 IB_FLOW_ATTR_MC_DEFAULT
= 0x2,
1561 /* sniffer rule - receive all port traffic */
1562 IB_FLOW_ATTR_SNIFFER
= 0x3
1565 /* Supported steering header types */
1566 enum ib_flow_spec_type
{
1568 IB_FLOW_SPEC_ETH
= 0x20,
1569 IB_FLOW_SPEC_IB
= 0x22,
1571 IB_FLOW_SPEC_IPV4
= 0x30,
1573 IB_FLOW_SPEC_TCP
= 0x40,
1574 IB_FLOW_SPEC_UDP
= 0x41
1576 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1577 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1579 /* Flow steering rule priority is set according to it's domain.
1580 * Lower domain value means higher priority.
1582 enum ib_flow_domain
{
1583 IB_FLOW_DOMAIN_USER
,
1584 IB_FLOW_DOMAIN_ETHTOOL
,
1587 IB_FLOW_DOMAIN_NUM
/* Must be last */
1590 enum ib_flow_flags
{
1591 IB_FLOW_ATTR_FLAGS_DONT_TRAP
= 1UL << 1, /* Continue match, no steal */
1592 IB_FLOW_ATTR_FLAGS_RESERVED
= 1UL << 2 /* Must be last */
1595 struct ib_flow_eth_filter
{
1602 struct ib_flow_spec_eth
{
1603 enum ib_flow_spec_type type
;
1605 struct ib_flow_eth_filter val
;
1606 struct ib_flow_eth_filter mask
;
1609 struct ib_flow_ib_filter
{
1614 struct ib_flow_spec_ib
{
1615 enum ib_flow_spec_type type
;
1617 struct ib_flow_ib_filter val
;
1618 struct ib_flow_ib_filter mask
;
1621 struct ib_flow_ipv4_filter
{
1626 struct ib_flow_spec_ipv4
{
1627 enum ib_flow_spec_type type
;
1629 struct ib_flow_ipv4_filter val
;
1630 struct ib_flow_ipv4_filter mask
;
1633 struct ib_flow_tcp_udp_filter
{
1638 struct ib_flow_spec_tcp_udp
{
1639 enum ib_flow_spec_type type
;
1641 struct ib_flow_tcp_udp_filter val
;
1642 struct ib_flow_tcp_udp_filter mask
;
1645 union ib_flow_spec
{
1647 enum ib_flow_spec_type type
;
1650 struct ib_flow_spec_eth eth
;
1651 struct ib_flow_spec_ib ib
;
1652 struct ib_flow_spec_ipv4 ipv4
;
1653 struct ib_flow_spec_tcp_udp tcp_udp
;
1656 struct ib_flow_attr
{
1657 enum ib_flow_attr_type type
;
1663 /* Following are the optional layers according to user request
1664 * struct ib_flow_spec_xxx
1665 * struct ib_flow_spec_yyy
1671 struct ib_uobject
*uobject
;
1677 enum ib_process_mad_flags
{
1678 IB_MAD_IGNORE_MKEY
= 1,
1679 IB_MAD_IGNORE_BKEY
= 2,
1680 IB_MAD_IGNORE_ALL
= IB_MAD_IGNORE_MKEY
| IB_MAD_IGNORE_BKEY
1683 enum ib_mad_result
{
1684 IB_MAD_RESULT_FAILURE
= 0, /* (!SUCCESS is the important flag) */
1685 IB_MAD_RESULT_SUCCESS
= 1 << 0, /* MAD was successfully processed */
1686 IB_MAD_RESULT_REPLY
= 1 << 1, /* Reply packet needs to be sent */
1687 IB_MAD_RESULT_CONSUMED
= 1 << 2 /* Packet consumed: stop processing */
1690 #define IB_DEVICE_NAME_MAX 64
1694 struct ib_event_handler event_handler
;
1695 struct ib_pkey_cache
**pkey_cache
;
1696 struct ib_gid_table
**gid_cache
;
1700 struct ib_dma_mapping_ops
{
1701 int (*mapping_error
)(struct ib_device
*dev
,
1703 u64 (*map_single
)(struct ib_device
*dev
,
1704 void *ptr
, size_t size
,
1705 enum dma_data_direction direction
);
1706 void (*unmap_single
)(struct ib_device
*dev
,
1707 u64 addr
, size_t size
,
1708 enum dma_data_direction direction
);
1709 u64 (*map_page
)(struct ib_device
*dev
,
1710 struct page
*page
, unsigned long offset
,
1712 enum dma_data_direction direction
);
1713 void (*unmap_page
)(struct ib_device
*dev
,
1714 u64 addr
, size_t size
,
1715 enum dma_data_direction direction
);
1716 int (*map_sg
)(struct ib_device
*dev
,
1717 struct scatterlist
*sg
, int nents
,
1718 enum dma_data_direction direction
);
1719 void (*unmap_sg
)(struct ib_device
*dev
,
1720 struct scatterlist
*sg
, int nents
,
1721 enum dma_data_direction direction
);
1722 void (*sync_single_for_cpu
)(struct ib_device
*dev
,
1725 enum dma_data_direction dir
);
1726 void (*sync_single_for_device
)(struct ib_device
*dev
,
1729 enum dma_data_direction dir
);
1730 void *(*alloc_coherent
)(struct ib_device
*dev
,
1734 void (*free_coherent
)(struct ib_device
*dev
,
1735 size_t size
, void *cpu_addr
,
1741 struct ib_port_immutable
{
1749 struct device
*dma_device
;
1751 char name
[IB_DEVICE_NAME_MAX
];
1753 struct list_head event_handler_list
;
1754 spinlock_t event_handler_lock
;
1756 spinlock_t client_data_lock
;
1757 struct list_head core_list
;
1758 /* Access to the client_data_list is protected by the client_data_lock
1759 * spinlock and the lists_rwsem read-write semaphore */
1760 struct list_head client_data_list
;
1762 struct ib_cache cache
;
1764 * port_immutable is indexed by port number
1766 struct ib_port_immutable
*port_immutable
;
1768 int num_comp_vectors
;
1770 struct iw_cm_verbs
*iwcm
;
1773 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1774 * driver initialized data. The struct is kfree()'ed by the sysfs
1775 * core when the device is removed. A lifespan of -1 in the return
1776 * struct tells the core to set a default lifespan.
1778 struct rdma_hw_stats
*(*alloc_hw_stats
)(struct ib_device
*device
,
1781 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1782 * @index - The index in the value array we wish to have updated, or
1783 * num_counters if we want all stats updated
1785 * < 0 - Error, no counters updated
1786 * index - Updated the single counter pointed to by index
1787 * num_counters - Updated all counters (will reset the timestamp
1788 * and prevent further calls for lifespan milliseconds)
1789 * Drivers are allowed to update all counters in leiu of just the
1790 * one given in index at their option
1792 int (*get_hw_stats
)(struct ib_device
*device
,
1793 struct rdma_hw_stats
*stats
,
1794 u8 port
, int index
);
1795 int (*query_device
)(struct ib_device
*device
,
1796 struct ib_device_attr
*device_attr
,
1797 struct ib_udata
*udata
);
1798 int (*query_port
)(struct ib_device
*device
,
1800 struct ib_port_attr
*port_attr
);
1801 enum rdma_link_layer (*get_link_layer
)(struct ib_device
*device
,
1803 /* When calling get_netdev, the HW vendor's driver should return the
1804 * net device of device @device at port @port_num or NULL if such
1805 * a net device doesn't exist. The vendor driver should call dev_hold
1806 * on this net device. The HW vendor's device driver must guarantee
1807 * that this function returns NULL before the net device reaches
1808 * NETDEV_UNREGISTER_FINAL state.
1810 struct net_device
*(*get_netdev
)(struct ib_device
*device
,
1812 int (*query_gid
)(struct ib_device
*device
,
1813 u8 port_num
, int index
,
1815 /* When calling add_gid, the HW vendor's driver should
1816 * add the gid of device @device at gid index @index of
1817 * port @port_num to be @gid. Meta-info of that gid (for example,
1818 * the network device related to this gid is available
1819 * at @attr. @context allows the HW vendor driver to store extra
1820 * information together with a GID entry. The HW vendor may allocate
1821 * memory to contain this information and store it in @context when a
1822 * new GID entry is written to. Params are consistent until the next
1823 * call of add_gid or delete_gid. The function should return 0 on
1824 * success or error otherwise. The function could be called
1825 * concurrently for different ports. This function is only called
1826 * when roce_gid_table is used.
1828 int (*add_gid
)(struct ib_device
*device
,
1831 const union ib_gid
*gid
,
1832 const struct ib_gid_attr
*attr
,
1834 /* When calling del_gid, the HW vendor's driver should delete the
1835 * gid of device @device at gid index @index of port @port_num.
1836 * Upon the deletion of a GID entry, the HW vendor must free any
1837 * allocated memory. The caller will clear @context afterwards.
1838 * This function is only called when roce_gid_table is used.
1840 int (*del_gid
)(struct ib_device
*device
,
1844 int (*query_pkey
)(struct ib_device
*device
,
1845 u8 port_num
, u16 index
, u16
*pkey
);
1846 int (*modify_device
)(struct ib_device
*device
,
1847 int device_modify_mask
,
1848 struct ib_device_modify
*device_modify
);
1849 int (*modify_port
)(struct ib_device
*device
,
1850 u8 port_num
, int port_modify_mask
,
1851 struct ib_port_modify
*port_modify
);
1852 struct ib_ucontext
* (*alloc_ucontext
)(struct ib_device
*device
,
1853 struct ib_udata
*udata
);
1854 int (*dealloc_ucontext
)(struct ib_ucontext
*context
);
1855 int (*mmap
)(struct ib_ucontext
*context
,
1856 struct vm_area_struct
*vma
);
1857 struct ib_pd
* (*alloc_pd
)(struct ib_device
*device
,
1858 struct ib_ucontext
*context
,
1859 struct ib_udata
*udata
);
1860 int (*dealloc_pd
)(struct ib_pd
*pd
);
1861 struct ib_ah
* (*create_ah
)(struct ib_pd
*pd
,
1862 struct ib_ah_attr
*ah_attr
);
1863 int (*modify_ah
)(struct ib_ah
*ah
,
1864 struct ib_ah_attr
*ah_attr
);
1865 int (*query_ah
)(struct ib_ah
*ah
,
1866 struct ib_ah_attr
*ah_attr
);
1867 int (*destroy_ah
)(struct ib_ah
*ah
);
1868 struct ib_srq
* (*create_srq
)(struct ib_pd
*pd
,
1869 struct ib_srq_init_attr
*srq_init_attr
,
1870 struct ib_udata
*udata
);
1871 int (*modify_srq
)(struct ib_srq
*srq
,
1872 struct ib_srq_attr
*srq_attr
,
1873 enum ib_srq_attr_mask srq_attr_mask
,
1874 struct ib_udata
*udata
);
1875 int (*query_srq
)(struct ib_srq
*srq
,
1876 struct ib_srq_attr
*srq_attr
);
1877 int (*destroy_srq
)(struct ib_srq
*srq
);
1878 int (*post_srq_recv
)(struct ib_srq
*srq
,
1879 struct ib_recv_wr
*recv_wr
,
1880 struct ib_recv_wr
**bad_recv_wr
);
1881 struct ib_qp
* (*create_qp
)(struct ib_pd
*pd
,
1882 struct ib_qp_init_attr
*qp_init_attr
,
1883 struct ib_udata
*udata
);
1884 int (*modify_qp
)(struct ib_qp
*qp
,
1885 struct ib_qp_attr
*qp_attr
,
1887 struct ib_udata
*udata
);
1888 int (*query_qp
)(struct ib_qp
*qp
,
1889 struct ib_qp_attr
*qp_attr
,
1891 struct ib_qp_init_attr
*qp_init_attr
);
1892 int (*destroy_qp
)(struct ib_qp
*qp
);
1893 int (*post_send
)(struct ib_qp
*qp
,
1894 struct ib_send_wr
*send_wr
,
1895 struct ib_send_wr
**bad_send_wr
);
1896 int (*post_recv
)(struct ib_qp
*qp
,
1897 struct ib_recv_wr
*recv_wr
,
1898 struct ib_recv_wr
**bad_recv_wr
);
1899 struct ib_cq
* (*create_cq
)(struct ib_device
*device
,
1900 const struct ib_cq_init_attr
*attr
,
1901 struct ib_ucontext
*context
,
1902 struct ib_udata
*udata
);
1903 int (*modify_cq
)(struct ib_cq
*cq
, u16 cq_count
,
1905 int (*destroy_cq
)(struct ib_cq
*cq
);
1906 int (*resize_cq
)(struct ib_cq
*cq
, int cqe
,
1907 struct ib_udata
*udata
);
1908 int (*poll_cq
)(struct ib_cq
*cq
, int num_entries
,
1910 int (*peek_cq
)(struct ib_cq
*cq
, int wc_cnt
);
1911 int (*req_notify_cq
)(struct ib_cq
*cq
,
1912 enum ib_cq_notify_flags flags
);
1913 int (*req_ncomp_notif
)(struct ib_cq
*cq
,
1915 struct ib_mr
* (*get_dma_mr
)(struct ib_pd
*pd
,
1916 int mr_access_flags
);
1917 struct ib_mr
* (*reg_user_mr
)(struct ib_pd
*pd
,
1918 u64 start
, u64 length
,
1920 int mr_access_flags
,
1921 struct ib_udata
*udata
);
1922 int (*rereg_user_mr
)(struct ib_mr
*mr
,
1924 u64 start
, u64 length
,
1926 int mr_access_flags
,
1928 struct ib_udata
*udata
);
1929 int (*dereg_mr
)(struct ib_mr
*mr
);
1930 struct ib_mr
* (*alloc_mr
)(struct ib_pd
*pd
,
1931 enum ib_mr_type mr_type
,
1933 int (*map_mr_sg
)(struct ib_mr
*mr
,
1934 struct scatterlist
*sg
,
1936 unsigned int *sg_offset
);
1937 struct ib_mw
* (*alloc_mw
)(struct ib_pd
*pd
,
1938 enum ib_mw_type type
,
1939 struct ib_udata
*udata
);
1940 int (*dealloc_mw
)(struct ib_mw
*mw
);
1941 struct ib_fmr
* (*alloc_fmr
)(struct ib_pd
*pd
,
1942 int mr_access_flags
,
1943 struct ib_fmr_attr
*fmr_attr
);
1944 int (*map_phys_fmr
)(struct ib_fmr
*fmr
,
1945 u64
*page_list
, int list_len
,
1947 int (*unmap_fmr
)(struct list_head
*fmr_list
);
1948 int (*dealloc_fmr
)(struct ib_fmr
*fmr
);
1949 int (*attach_mcast
)(struct ib_qp
*qp
,
1952 int (*detach_mcast
)(struct ib_qp
*qp
,
1955 int (*process_mad
)(struct ib_device
*device
,
1956 int process_mad_flags
,
1958 const struct ib_wc
*in_wc
,
1959 const struct ib_grh
*in_grh
,
1960 const struct ib_mad_hdr
*in_mad
,
1962 struct ib_mad_hdr
*out_mad
,
1963 size_t *out_mad_size
,
1964 u16
*out_mad_pkey_index
);
1965 struct ib_xrcd
* (*alloc_xrcd
)(struct ib_device
*device
,
1966 struct ib_ucontext
*ucontext
,
1967 struct ib_udata
*udata
);
1968 int (*dealloc_xrcd
)(struct ib_xrcd
*xrcd
);
1969 struct ib_flow
* (*create_flow
)(struct ib_qp
*qp
,
1973 int (*destroy_flow
)(struct ib_flow
*flow_id
);
1974 int (*check_mr_status
)(struct ib_mr
*mr
, u32 check_mask
,
1975 struct ib_mr_status
*mr_status
);
1976 void (*disassociate_ucontext
)(struct ib_ucontext
*ibcontext
);
1977 void (*drain_rq
)(struct ib_qp
*qp
);
1978 void (*drain_sq
)(struct ib_qp
*qp
);
1979 int (*set_vf_link_state
)(struct ib_device
*device
, int vf
, u8 port
,
1981 int (*get_vf_config
)(struct ib_device
*device
, int vf
, u8 port
,
1982 struct ifla_vf_info
*ivf
);
1983 int (*get_vf_stats
)(struct ib_device
*device
, int vf
, u8 port
,
1984 struct ifla_vf_stats
*stats
);
1985 int (*set_vf_guid
)(struct ib_device
*device
, int vf
, u8 port
, u64 guid
,
1987 struct ib_wq
* (*create_wq
)(struct ib_pd
*pd
,
1988 struct ib_wq_init_attr
*init_attr
,
1989 struct ib_udata
*udata
);
1990 int (*destroy_wq
)(struct ib_wq
*wq
);
1991 int (*modify_wq
)(struct ib_wq
*wq
,
1992 struct ib_wq_attr
*attr
,
1994 struct ib_udata
*udata
);
1995 struct ib_rwq_ind_table
* (*create_rwq_ind_table
)(struct ib_device
*device
,
1996 struct ib_rwq_ind_table_init_attr
*init_attr
,
1997 struct ib_udata
*udata
);
1998 int (*destroy_rwq_ind_table
)(struct ib_rwq_ind_table
*wq_ind_table
);
1999 struct ib_dma_mapping_ops
*dma_ops
;
2001 struct module
*owner
;
2003 struct kobject
*ports_parent
;
2004 struct list_head port_list
;
2007 IB_DEV_UNINITIALIZED
,
2013 u64 uverbs_cmd_mask
;
2014 u64 uverbs_ex_cmd_mask
;
2022 struct ib_device_attr attrs
;
2023 struct attribute_group
*hw_stats_ag
;
2024 struct rdma_hw_stats
*hw_stats
;
2027 * The following mandatory functions are used only at device
2028 * registration. Keep functions such as these at the end of this
2029 * structure to avoid cache line misses when accessing struct ib_device
2032 int (*get_port_immutable
)(struct ib_device
*, u8
, struct ib_port_immutable
*);
2037 void (*add
) (struct ib_device
*);
2038 void (*remove
)(struct ib_device
*, void *client_data
);
2040 /* Returns the net_dev belonging to this ib_client and matching the
2042 * @dev: An RDMA device that the net_dev use for communication.
2043 * @port: A physical port number on the RDMA device.
2044 * @pkey: P_Key that the net_dev uses if applicable.
2045 * @gid: A GID that the net_dev uses to communicate.
2046 * @addr: An IP address the net_dev is configured with.
2047 * @client_data: The device's client data set by ib_set_client_data().
2049 * An ib_client that implements a net_dev on top of RDMA devices
2050 * (such as IP over IB) should implement this callback, allowing the
2051 * rdma_cm module to find the right net_dev for a given request.
2053 * The caller is responsible for calling dev_put on the returned
2055 struct net_device
*(*get_net_dev_by_params
)(
2056 struct ib_device
*dev
,
2059 const union ib_gid
*gid
,
2060 const struct sockaddr
*addr
,
2062 struct list_head list
;
2065 struct ib_device
*ib_alloc_device(size_t size
);
2066 void ib_dealloc_device(struct ib_device
*device
);
2068 int ib_register_device(struct ib_device
*device
,
2069 int (*port_callback
)(struct ib_device
*,
2070 u8
, struct kobject
*));
2071 void ib_unregister_device(struct ib_device
*device
);
2073 int ib_register_client (struct ib_client
*client
);
2074 void ib_unregister_client(struct ib_client
*client
);
2076 void *ib_get_client_data(struct ib_device
*device
, struct ib_client
*client
);
2077 void ib_set_client_data(struct ib_device
*device
, struct ib_client
*client
,
2080 static inline int ib_copy_from_udata(void *dest
, struct ib_udata
*udata
, size_t len
)
2082 return copy_from_user(dest
, udata
->inbuf
, len
) ? -EFAULT
: 0;
2085 static inline int ib_copy_to_udata(struct ib_udata
*udata
, void *src
, size_t len
)
2087 return copy_to_user(udata
->outbuf
, src
, len
) ? -EFAULT
: 0;
2090 static inline bool ib_is_udata_cleared(struct ib_udata
*udata
,
2094 const void __user
*p
= udata
->inbuf
+ offset
;
2098 if (len
> USHRT_MAX
)
2101 buf
= kmalloc(len
, GFP_KERNEL
);
2105 if (copy_from_user(buf
, p
, len
))
2108 ret
= !memchr_inv(buf
, 0, len
);
2116 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2117 * contains all required attributes and no attributes not allowed for
2118 * the given QP state transition.
2119 * @cur_state: Current QP state
2120 * @next_state: Next QP state
2122 * @mask: Mask of supplied QP attributes
2123 * @ll : link layer of port
2125 * This function is a helper function that a low-level driver's
2126 * modify_qp method can use to validate the consumer's input. It
2127 * checks that cur_state and next_state are valid QP states, that a
2128 * transition from cur_state to next_state is allowed by the IB spec,
2129 * and that the attribute mask supplied is allowed for the transition.
2131 int ib_modify_qp_is_ok(enum ib_qp_state cur_state
, enum ib_qp_state next_state
,
2132 enum ib_qp_type type
, enum ib_qp_attr_mask mask
,
2133 enum rdma_link_layer ll
);
2135 int ib_register_event_handler (struct ib_event_handler
*event_handler
);
2136 int ib_unregister_event_handler(struct ib_event_handler
*event_handler
);
2137 void ib_dispatch_event(struct ib_event
*event
);
2139 int ib_query_port(struct ib_device
*device
,
2140 u8 port_num
, struct ib_port_attr
*port_attr
);
2142 enum rdma_link_layer
rdma_port_get_link_layer(struct ib_device
*device
,
2146 * rdma_cap_ib_switch - Check if the device is IB switch
2147 * @device: Device to check
2149 * Device driver is responsible for setting is_switch bit on
2150 * in ib_device structure at init time.
2152 * Return: true if the device is IB switch.
2154 static inline bool rdma_cap_ib_switch(const struct ib_device
*device
)
2156 return device
->is_switch
;
2160 * rdma_start_port - Return the first valid port number for the device
2163 * @device: Device to be checked
2165 * Return start port number
2167 static inline u8
rdma_start_port(const struct ib_device
*device
)
2169 return rdma_cap_ib_switch(device
) ? 0 : 1;
2173 * rdma_end_port - Return the last valid port number for the device
2176 * @device: Device to be checked
2178 * Return last port number
2180 static inline u8
rdma_end_port(const struct ib_device
*device
)
2182 return rdma_cap_ib_switch(device
) ? 0 : device
->phys_port_cnt
;
2185 static inline bool rdma_protocol_ib(const struct ib_device
*device
, u8 port_num
)
2187 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IB
;
2190 static inline bool rdma_protocol_roce(const struct ib_device
*device
, u8 port_num
)
2192 return device
->port_immutable
[port_num
].core_cap_flags
&
2193 (RDMA_CORE_CAP_PROT_ROCE
| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP
);
2196 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device
*device
, u8 port_num
)
2198 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP
;
2201 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device
*device
, u8 port_num
)
2203 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_ROCE
;
2206 static inline bool rdma_protocol_iwarp(const struct ib_device
*device
, u8 port_num
)
2208 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IWARP
;
2211 static inline bool rdma_ib_or_roce(const struct ib_device
*device
, u8 port_num
)
2213 return rdma_protocol_ib(device
, port_num
) ||
2214 rdma_protocol_roce(device
, port_num
);
2218 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2219 * Management Datagrams.
2220 * @device: Device to check
2221 * @port_num: Port number to check
2223 * Management Datagrams (MAD) are a required part of the InfiniBand
2224 * specification and are supported on all InfiniBand devices. A slightly
2225 * extended version are also supported on OPA interfaces.
2227 * Return: true if the port supports sending/receiving of MAD packets.
2229 static inline bool rdma_cap_ib_mad(const struct ib_device
*device
, u8 port_num
)
2231 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_MAD
;
2235 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2236 * Management Datagrams.
2237 * @device: Device to check
2238 * @port_num: Port number to check
2240 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2241 * datagrams with their own versions. These OPA MADs share many but not all of
2242 * the characteristics of InfiniBand MADs.
2244 * OPA MADs differ in the following ways:
2246 * 1) MADs are variable size up to 2K
2247 * IBTA defined MADs remain fixed at 256 bytes
2248 * 2) OPA SMPs must carry valid PKeys
2249 * 3) OPA SMP packets are a different format
2251 * Return: true if the port supports OPA MAD packet formats.
2253 static inline bool rdma_cap_opa_mad(struct ib_device
*device
, u8 port_num
)
2255 return (device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_OPA_MAD
)
2256 == RDMA_CORE_CAP_OPA_MAD
;
2260 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2261 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2262 * @device: Device to check
2263 * @port_num: Port number to check
2265 * Each InfiniBand node is required to provide a Subnet Management Agent
2266 * that the subnet manager can access. Prior to the fabric being fully
2267 * configured by the subnet manager, the SMA is accessed via a well known
2268 * interface called the Subnet Management Interface (SMI). This interface
2269 * uses directed route packets to communicate with the SM to get around the
2270 * chicken and egg problem of the SM needing to know what's on the fabric
2271 * in order to configure the fabric, and needing to configure the fabric in
2272 * order to send packets to the devices on the fabric. These directed
2273 * route packets do not need the fabric fully configured in order to reach
2274 * their destination. The SMI is the only method allowed to send
2275 * directed route packets on an InfiniBand fabric.
2277 * Return: true if the port provides an SMI.
2279 static inline bool rdma_cap_ib_smi(const struct ib_device
*device
, u8 port_num
)
2281 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SMI
;
2285 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2286 * Communication Manager.
2287 * @device: Device to check
2288 * @port_num: Port number to check
2290 * The InfiniBand Communication Manager is one of many pre-defined General
2291 * Service Agents (GSA) that are accessed via the General Service
2292 * Interface (GSI). It's role is to facilitate establishment of connections
2293 * between nodes as well as other management related tasks for established
2296 * Return: true if the port supports an IB CM (this does not guarantee that
2297 * a CM is actually running however).
2299 static inline bool rdma_cap_ib_cm(const struct ib_device
*device
, u8 port_num
)
2301 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_CM
;
2305 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2306 * Communication Manager.
2307 * @device: Device to check
2308 * @port_num: Port number to check
2310 * Similar to above, but specific to iWARP connections which have a different
2311 * managment protocol than InfiniBand.
2313 * Return: true if the port supports an iWARP CM (this does not guarantee that
2314 * a CM is actually running however).
2316 static inline bool rdma_cap_iw_cm(const struct ib_device
*device
, u8 port_num
)
2318 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IW_CM
;
2322 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2323 * Subnet Administration.
2324 * @device: Device to check
2325 * @port_num: Port number to check
2327 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2328 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2329 * fabrics, devices should resolve routes to other hosts by contacting the
2330 * SA to query the proper route.
2332 * Return: true if the port should act as a client to the fabric Subnet
2333 * Administration interface. This does not imply that the SA service is
2336 static inline bool rdma_cap_ib_sa(const struct ib_device
*device
, u8 port_num
)
2338 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SA
;
2342 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2344 * @device: Device to check
2345 * @port_num: Port number to check
2347 * InfiniBand multicast registration is more complex than normal IPv4 or
2348 * IPv6 multicast registration. Each Host Channel Adapter must register
2349 * with the Subnet Manager when it wishes to join a multicast group. It
2350 * should do so only once regardless of how many queue pairs it subscribes
2351 * to this group. And it should leave the group only after all queue pairs
2352 * attached to the group have been detached.
2354 * Return: true if the port must undertake the additional adminstrative
2355 * overhead of registering/unregistering with the SM and tracking of the
2356 * total number of queue pairs attached to the multicast group.
2358 static inline bool rdma_cap_ib_mcast(const struct ib_device
*device
, u8 port_num
)
2360 return rdma_cap_ib_sa(device
, port_num
);
2364 * rdma_cap_af_ib - Check if the port of device has the capability
2365 * Native Infiniband Address.
2366 * @device: Device to check
2367 * @port_num: Port number to check
2369 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2370 * GID. RoCE uses a different mechanism, but still generates a GID via
2371 * a prescribed mechanism and port specific data.
2373 * Return: true if the port uses a GID address to identify devices on the
2376 static inline bool rdma_cap_af_ib(const struct ib_device
*device
, u8 port_num
)
2378 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_AF_IB
;
2382 * rdma_cap_eth_ah - Check if the port of device has the capability
2383 * Ethernet Address Handle.
2384 * @device: Device to check
2385 * @port_num: Port number to check
2387 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2388 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2389 * port. Normally, packet headers are generated by the sending host
2390 * adapter, but when sending connectionless datagrams, we must manually
2391 * inject the proper headers for the fabric we are communicating over.
2393 * Return: true if we are running as a RoCE port and must force the
2394 * addition of a Global Route Header built from our Ethernet Address
2395 * Handle into our header list for connectionless packets.
2397 static inline bool rdma_cap_eth_ah(const struct ib_device
*device
, u8 port_num
)
2399 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_ETH_AH
;
2403 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2406 * @port_num: Port number
2408 * This MAD size includes the MAD headers and MAD payload. No other headers
2411 * Return the max MAD size required by the Port. Will return 0 if the port
2412 * does not support MADs
2414 static inline size_t rdma_max_mad_size(const struct ib_device
*device
, u8 port_num
)
2416 return device
->port_immutable
[port_num
].max_mad_size
;
2420 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2421 * @device: Device to check
2422 * @port_num: Port number to check
2424 * RoCE GID table mechanism manages the various GIDs for a device.
2426 * NOTE: if allocating the port's GID table has failed, this call will still
2427 * return true, but any RoCE GID table API will fail.
2429 * Return: true if the port uses RoCE GID table mechanism in order to manage
2432 static inline bool rdma_cap_roce_gid_table(const struct ib_device
*device
,
2435 return rdma_protocol_roce(device
, port_num
) &&
2436 device
->add_gid
&& device
->del_gid
;
2440 * Check if the device supports READ W/ INVALIDATE.
2442 static inline bool rdma_cap_read_inv(struct ib_device
*dev
, u32 port_num
)
2445 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2446 * has support for it yet.
2448 return rdma_protocol_iwarp(dev
, port_num
);
2451 int ib_query_gid(struct ib_device
*device
,
2452 u8 port_num
, int index
, union ib_gid
*gid
,
2453 struct ib_gid_attr
*attr
);
2455 int ib_set_vf_link_state(struct ib_device
*device
, int vf
, u8 port
,
2457 int ib_get_vf_config(struct ib_device
*device
, int vf
, u8 port
,
2458 struct ifla_vf_info
*info
);
2459 int ib_get_vf_stats(struct ib_device
*device
, int vf
, u8 port
,
2460 struct ifla_vf_stats
*stats
);
2461 int ib_set_vf_guid(struct ib_device
*device
, int vf
, u8 port
, u64 guid
,
2464 int ib_query_pkey(struct ib_device
*device
,
2465 u8 port_num
, u16 index
, u16
*pkey
);
2467 int ib_modify_device(struct ib_device
*device
,
2468 int device_modify_mask
,
2469 struct ib_device_modify
*device_modify
);
2471 int ib_modify_port(struct ib_device
*device
,
2472 u8 port_num
, int port_modify_mask
,
2473 struct ib_port_modify
*port_modify
);
2475 int ib_find_gid(struct ib_device
*device
, union ib_gid
*gid
,
2476 enum ib_gid_type gid_type
, struct net_device
*ndev
,
2477 u8
*port_num
, u16
*index
);
2479 int ib_find_pkey(struct ib_device
*device
,
2480 u8 port_num
, u16 pkey
, u16
*index
);
2482 struct ib_pd
*ib_alloc_pd(struct ib_device
*device
);
2484 void ib_dealloc_pd(struct ib_pd
*pd
);
2487 * ib_create_ah - Creates an address handle for the given address vector.
2488 * @pd: The protection domain associated with the address handle.
2489 * @ah_attr: The attributes of the address vector.
2491 * The address handle is used to reference a local or global destination
2492 * in all UD QP post sends.
2494 struct ib_ah
*ib_create_ah(struct ib_pd
*pd
, struct ib_ah_attr
*ah_attr
);
2497 * ib_init_ah_from_wc - Initializes address handle attributes from a
2499 * @device: Device on which the received message arrived.
2500 * @port_num: Port on which the received message arrived.
2501 * @wc: Work completion associated with the received message.
2502 * @grh: References the received global route header. This parameter is
2503 * ignored unless the work completion indicates that the GRH is valid.
2504 * @ah_attr: Returned attributes that can be used when creating an address
2505 * handle for replying to the message.
2507 int ib_init_ah_from_wc(struct ib_device
*device
, u8 port_num
,
2508 const struct ib_wc
*wc
, const struct ib_grh
*grh
,
2509 struct ib_ah_attr
*ah_attr
);
2512 * ib_create_ah_from_wc - Creates an address handle associated with the
2513 * sender of the specified work completion.
2514 * @pd: The protection domain associated with the address handle.
2515 * @wc: Work completion information associated with a received message.
2516 * @grh: References the received global route header. This parameter is
2517 * ignored unless the work completion indicates that the GRH is valid.
2518 * @port_num: The outbound port number to associate with the address.
2520 * The address handle is used to reference a local or global destination
2521 * in all UD QP post sends.
2523 struct ib_ah
*ib_create_ah_from_wc(struct ib_pd
*pd
, const struct ib_wc
*wc
,
2524 const struct ib_grh
*grh
, u8 port_num
);
2527 * ib_modify_ah - Modifies the address vector associated with an address
2529 * @ah: The address handle to modify.
2530 * @ah_attr: The new address vector attributes to associate with the
2533 int ib_modify_ah(struct ib_ah
*ah
, struct ib_ah_attr
*ah_attr
);
2536 * ib_query_ah - Queries the address vector associated with an address
2538 * @ah: The address handle to query.
2539 * @ah_attr: The address vector attributes associated with the address
2542 int ib_query_ah(struct ib_ah
*ah
, struct ib_ah_attr
*ah_attr
);
2545 * ib_destroy_ah - Destroys an address handle.
2546 * @ah: The address handle to destroy.
2548 int ib_destroy_ah(struct ib_ah
*ah
);
2551 * ib_create_srq - Creates a SRQ associated with the specified protection
2553 * @pd: The protection domain associated with the SRQ.
2554 * @srq_init_attr: A list of initial attributes required to create the
2555 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2556 * the actual capabilities of the created SRQ.
2558 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2559 * requested size of the SRQ, and set to the actual values allocated
2560 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2561 * will always be at least as large as the requested values.
2563 struct ib_srq
*ib_create_srq(struct ib_pd
*pd
,
2564 struct ib_srq_init_attr
*srq_init_attr
);
2567 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2568 * @srq: The SRQ to modify.
2569 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2570 * the current values of selected SRQ attributes are returned.
2571 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2572 * are being modified.
2574 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2575 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2576 * the number of receives queued drops below the limit.
2578 int ib_modify_srq(struct ib_srq
*srq
,
2579 struct ib_srq_attr
*srq_attr
,
2580 enum ib_srq_attr_mask srq_attr_mask
);
2583 * ib_query_srq - Returns the attribute list and current values for the
2585 * @srq: The SRQ to query.
2586 * @srq_attr: The attributes of the specified SRQ.
2588 int ib_query_srq(struct ib_srq
*srq
,
2589 struct ib_srq_attr
*srq_attr
);
2592 * ib_destroy_srq - Destroys the specified SRQ.
2593 * @srq: The SRQ to destroy.
2595 int ib_destroy_srq(struct ib_srq
*srq
);
2598 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2599 * @srq: The SRQ to post the work request on.
2600 * @recv_wr: A list of work requests to post on the receive queue.
2601 * @bad_recv_wr: On an immediate failure, this parameter will reference
2602 * the work request that failed to be posted on the QP.
2604 static inline int ib_post_srq_recv(struct ib_srq
*srq
,
2605 struct ib_recv_wr
*recv_wr
,
2606 struct ib_recv_wr
**bad_recv_wr
)
2608 return srq
->device
->post_srq_recv(srq
, recv_wr
, bad_recv_wr
);
2612 * ib_create_qp - Creates a QP associated with the specified protection
2614 * @pd: The protection domain associated with the QP.
2615 * @qp_init_attr: A list of initial attributes required to create the
2616 * QP. If QP creation succeeds, then the attributes are updated to
2617 * the actual capabilities of the created QP.
2619 struct ib_qp
*ib_create_qp(struct ib_pd
*pd
,
2620 struct ib_qp_init_attr
*qp_init_attr
);
2623 * ib_modify_qp - Modifies the attributes for the specified QP and then
2624 * transitions the QP to the given state.
2625 * @qp: The QP to modify.
2626 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2627 * the current values of selected QP attributes are returned.
2628 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2629 * are being modified.
2631 int ib_modify_qp(struct ib_qp
*qp
,
2632 struct ib_qp_attr
*qp_attr
,
2636 * ib_query_qp - Returns the attribute list and current values for the
2638 * @qp: The QP to query.
2639 * @qp_attr: The attributes of the specified QP.
2640 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2641 * @qp_init_attr: Additional attributes of the selected QP.
2643 * The qp_attr_mask may be used to limit the query to gathering only the
2644 * selected attributes.
2646 int ib_query_qp(struct ib_qp
*qp
,
2647 struct ib_qp_attr
*qp_attr
,
2649 struct ib_qp_init_attr
*qp_init_attr
);
2652 * ib_destroy_qp - Destroys the specified QP.
2653 * @qp: The QP to destroy.
2655 int ib_destroy_qp(struct ib_qp
*qp
);
2658 * ib_open_qp - Obtain a reference to an existing sharable QP.
2659 * @xrcd - XRC domain
2660 * @qp_open_attr: Attributes identifying the QP to open.
2662 * Returns a reference to a sharable QP.
2664 struct ib_qp
*ib_open_qp(struct ib_xrcd
*xrcd
,
2665 struct ib_qp_open_attr
*qp_open_attr
);
2668 * ib_close_qp - Release an external reference to a QP.
2669 * @qp: The QP handle to release
2671 * The opened QP handle is released by the caller. The underlying
2672 * shared QP is not destroyed until all internal references are released.
2674 int ib_close_qp(struct ib_qp
*qp
);
2677 * ib_post_send - Posts a list of work requests to the send queue of
2679 * @qp: The QP to post the work request on.
2680 * @send_wr: A list of work requests to post on the send queue.
2681 * @bad_send_wr: On an immediate failure, this parameter will reference
2682 * the work request that failed to be posted on the QP.
2684 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2685 * error is returned, the QP state shall not be affected,
2686 * ib_post_send() will return an immediate error after queueing any
2687 * earlier work requests in the list.
2689 static inline int ib_post_send(struct ib_qp
*qp
,
2690 struct ib_send_wr
*send_wr
,
2691 struct ib_send_wr
**bad_send_wr
)
2693 return qp
->device
->post_send(qp
, send_wr
, bad_send_wr
);
2697 * ib_post_recv - Posts a list of work requests to the receive queue of
2699 * @qp: The QP to post the work request on.
2700 * @recv_wr: A list of work requests to post on the receive queue.
2701 * @bad_recv_wr: On an immediate failure, this parameter will reference
2702 * the work request that failed to be posted on the QP.
2704 static inline int ib_post_recv(struct ib_qp
*qp
,
2705 struct ib_recv_wr
*recv_wr
,
2706 struct ib_recv_wr
**bad_recv_wr
)
2708 return qp
->device
->post_recv(qp
, recv_wr
, bad_recv_wr
);
2711 struct ib_cq
*ib_alloc_cq(struct ib_device
*dev
, void *private,
2712 int nr_cqe
, int comp_vector
, enum ib_poll_context poll_ctx
);
2713 void ib_free_cq(struct ib_cq
*cq
);
2714 int ib_process_cq_direct(struct ib_cq
*cq
, int budget
);
2717 * ib_create_cq - Creates a CQ on the specified device.
2718 * @device: The device on which to create the CQ.
2719 * @comp_handler: A user-specified callback that is invoked when a
2720 * completion event occurs on the CQ.
2721 * @event_handler: A user-specified callback that is invoked when an
2722 * asynchronous event not associated with a completion occurs on the CQ.
2723 * @cq_context: Context associated with the CQ returned to the user via
2724 * the associated completion and event handlers.
2725 * @cq_attr: The attributes the CQ should be created upon.
2727 * Users can examine the cq structure to determine the actual CQ size.
2729 struct ib_cq
*ib_create_cq(struct ib_device
*device
,
2730 ib_comp_handler comp_handler
,
2731 void (*event_handler
)(struct ib_event
*, void *),
2733 const struct ib_cq_init_attr
*cq_attr
);
2736 * ib_resize_cq - Modifies the capacity of the CQ.
2737 * @cq: The CQ to resize.
2738 * @cqe: The minimum size of the CQ.
2740 * Users can examine the cq structure to determine the actual CQ size.
2742 int ib_resize_cq(struct ib_cq
*cq
, int cqe
);
2745 * ib_modify_cq - Modifies moderation params of the CQ
2746 * @cq: The CQ to modify.
2747 * @cq_count: number of CQEs that will trigger an event
2748 * @cq_period: max period of time in usec before triggering an event
2751 int ib_modify_cq(struct ib_cq
*cq
, u16 cq_count
, u16 cq_period
);
2754 * ib_destroy_cq - Destroys the specified CQ.
2755 * @cq: The CQ to destroy.
2757 int ib_destroy_cq(struct ib_cq
*cq
);
2760 * ib_poll_cq - poll a CQ for completion(s)
2761 * @cq:the CQ being polled
2762 * @num_entries:maximum number of completions to return
2763 * @wc:array of at least @num_entries &struct ib_wc where completions
2766 * Poll a CQ for (possibly multiple) completions. If the return value
2767 * is < 0, an error occurred. If the return value is >= 0, it is the
2768 * number of completions returned. If the return value is
2769 * non-negative and < num_entries, then the CQ was emptied.
2771 static inline int ib_poll_cq(struct ib_cq
*cq
, int num_entries
,
2774 return cq
->device
->poll_cq(cq
, num_entries
, wc
);
2778 * ib_peek_cq - Returns the number of unreaped completions currently
2779 * on the specified CQ.
2780 * @cq: The CQ to peek.
2781 * @wc_cnt: A minimum number of unreaped completions to check for.
2783 * If the number of unreaped completions is greater than or equal to wc_cnt,
2784 * this function returns wc_cnt, otherwise, it returns the actual number of
2785 * unreaped completions.
2787 int ib_peek_cq(struct ib_cq
*cq
, int wc_cnt
);
2790 * ib_req_notify_cq - Request completion notification on a CQ.
2791 * @cq: The CQ to generate an event for.
2793 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2794 * to request an event on the next solicited event or next work
2795 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2796 * may also be |ed in to request a hint about missed events, as
2800 * < 0 means an error occurred while requesting notification
2801 * == 0 means notification was requested successfully, and if
2802 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2803 * were missed and it is safe to wait for another event. In
2804 * this case is it guaranteed that any work completions added
2805 * to the CQ since the last CQ poll will trigger a completion
2806 * notification event.
2807 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2808 * in. It means that the consumer must poll the CQ again to
2809 * make sure it is empty to avoid missing an event because of a
2810 * race between requesting notification and an entry being
2811 * added to the CQ. This return value means it is possible
2812 * (but not guaranteed) that a work completion has been added
2813 * to the CQ since the last poll without triggering a
2814 * completion notification event.
2816 static inline int ib_req_notify_cq(struct ib_cq
*cq
,
2817 enum ib_cq_notify_flags flags
)
2819 return cq
->device
->req_notify_cq(cq
, flags
);
2823 * ib_req_ncomp_notif - Request completion notification when there are
2824 * at least the specified number of unreaped completions on the CQ.
2825 * @cq: The CQ to generate an event for.
2826 * @wc_cnt: The number of unreaped completions that should be on the
2827 * CQ before an event is generated.
2829 static inline int ib_req_ncomp_notif(struct ib_cq
*cq
, int wc_cnt
)
2831 return cq
->device
->req_ncomp_notif
?
2832 cq
->device
->req_ncomp_notif(cq
, wc_cnt
) :
2837 * ib_get_dma_mr - Returns a memory region for system memory that is
2839 * @pd: The protection domain associated with the memory region.
2840 * @mr_access_flags: Specifies the memory access rights.
2842 * Note that the ib_dma_*() functions defined below must be used
2843 * to create/destroy addresses used with the Lkey or Rkey returned
2844 * by ib_get_dma_mr().
2846 struct ib_mr
*ib_get_dma_mr(struct ib_pd
*pd
, int mr_access_flags
);
2849 * ib_dma_mapping_error - check a DMA addr for error
2850 * @dev: The device for which the dma_addr was created
2851 * @dma_addr: The DMA address to check
2853 static inline int ib_dma_mapping_error(struct ib_device
*dev
, u64 dma_addr
)
2856 return dev
->dma_ops
->mapping_error(dev
, dma_addr
);
2857 return dma_mapping_error(dev
->dma_device
, dma_addr
);
2861 * ib_dma_map_single - Map a kernel virtual address to DMA address
2862 * @dev: The device for which the dma_addr is to be created
2863 * @cpu_addr: The kernel virtual address
2864 * @size: The size of the region in bytes
2865 * @direction: The direction of the DMA
2867 static inline u64
ib_dma_map_single(struct ib_device
*dev
,
2868 void *cpu_addr
, size_t size
,
2869 enum dma_data_direction direction
)
2872 return dev
->dma_ops
->map_single(dev
, cpu_addr
, size
, direction
);
2873 return dma_map_single(dev
->dma_device
, cpu_addr
, size
, direction
);
2877 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2878 * @dev: The device for which the DMA address was created
2879 * @addr: The DMA address
2880 * @size: The size of the region in bytes
2881 * @direction: The direction of the DMA
2883 static inline void ib_dma_unmap_single(struct ib_device
*dev
,
2884 u64 addr
, size_t size
,
2885 enum dma_data_direction direction
)
2888 dev
->dma_ops
->unmap_single(dev
, addr
, size
, direction
);
2890 dma_unmap_single(dev
->dma_device
, addr
, size
, direction
);
2893 static inline u64
ib_dma_map_single_attrs(struct ib_device
*dev
,
2894 void *cpu_addr
, size_t size
,
2895 enum dma_data_direction direction
,
2896 struct dma_attrs
*attrs
)
2898 return dma_map_single_attrs(dev
->dma_device
, cpu_addr
, size
,
2902 static inline void ib_dma_unmap_single_attrs(struct ib_device
*dev
,
2903 u64 addr
, size_t size
,
2904 enum dma_data_direction direction
,
2905 struct dma_attrs
*attrs
)
2907 return dma_unmap_single_attrs(dev
->dma_device
, addr
, size
,
2912 * ib_dma_map_page - Map a physical page to DMA address
2913 * @dev: The device for which the dma_addr is to be created
2914 * @page: The page to be mapped
2915 * @offset: The offset within the page
2916 * @size: The size of the region in bytes
2917 * @direction: The direction of the DMA
2919 static inline u64
ib_dma_map_page(struct ib_device
*dev
,
2921 unsigned long offset
,
2923 enum dma_data_direction direction
)
2926 return dev
->dma_ops
->map_page(dev
, page
, offset
, size
, direction
);
2927 return dma_map_page(dev
->dma_device
, page
, offset
, size
, direction
);
2931 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2932 * @dev: The device for which the DMA address was created
2933 * @addr: The DMA address
2934 * @size: The size of the region in bytes
2935 * @direction: The direction of the DMA
2937 static inline void ib_dma_unmap_page(struct ib_device
*dev
,
2938 u64 addr
, size_t size
,
2939 enum dma_data_direction direction
)
2942 dev
->dma_ops
->unmap_page(dev
, addr
, size
, direction
);
2944 dma_unmap_page(dev
->dma_device
, addr
, size
, direction
);
2948 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2949 * @dev: The device for which the DMA addresses are to be created
2950 * @sg: The array of scatter/gather entries
2951 * @nents: The number of scatter/gather entries
2952 * @direction: The direction of the DMA
2954 static inline int ib_dma_map_sg(struct ib_device
*dev
,
2955 struct scatterlist
*sg
, int nents
,
2956 enum dma_data_direction direction
)
2959 return dev
->dma_ops
->map_sg(dev
, sg
, nents
, direction
);
2960 return dma_map_sg(dev
->dma_device
, sg
, nents
, direction
);
2964 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2965 * @dev: The device for which the DMA addresses were created
2966 * @sg: The array of scatter/gather entries
2967 * @nents: The number of scatter/gather entries
2968 * @direction: The direction of the DMA
2970 static inline void ib_dma_unmap_sg(struct ib_device
*dev
,
2971 struct scatterlist
*sg
, int nents
,
2972 enum dma_data_direction direction
)
2975 dev
->dma_ops
->unmap_sg(dev
, sg
, nents
, direction
);
2977 dma_unmap_sg(dev
->dma_device
, sg
, nents
, direction
);
2980 static inline int ib_dma_map_sg_attrs(struct ib_device
*dev
,
2981 struct scatterlist
*sg
, int nents
,
2982 enum dma_data_direction direction
,
2983 struct dma_attrs
*attrs
)
2985 return dma_map_sg_attrs(dev
->dma_device
, sg
, nents
, direction
, attrs
);
2988 static inline void ib_dma_unmap_sg_attrs(struct ib_device
*dev
,
2989 struct scatterlist
*sg
, int nents
,
2990 enum dma_data_direction direction
,
2991 struct dma_attrs
*attrs
)
2993 dma_unmap_sg_attrs(dev
->dma_device
, sg
, nents
, direction
, attrs
);
2996 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2997 * @dev: The device for which the DMA addresses were created
2998 * @sg: The scatter/gather entry
3000 * Note: this function is obsolete. To do: change all occurrences of
3001 * ib_sg_dma_address() into sg_dma_address().
3003 static inline u64
ib_sg_dma_address(struct ib_device
*dev
,
3004 struct scatterlist
*sg
)
3006 return sg_dma_address(sg
);
3010 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3011 * @dev: The device for which the DMA addresses were created
3012 * @sg: The scatter/gather entry
3014 * Note: this function is obsolete. To do: change all occurrences of
3015 * ib_sg_dma_len() into sg_dma_len().
3017 static inline unsigned int ib_sg_dma_len(struct ib_device
*dev
,
3018 struct scatterlist
*sg
)
3020 return sg_dma_len(sg
);
3024 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3025 * @dev: The device for which the DMA address was created
3026 * @addr: The DMA address
3027 * @size: The size of the region in bytes
3028 * @dir: The direction of the DMA
3030 static inline void ib_dma_sync_single_for_cpu(struct ib_device
*dev
,
3033 enum dma_data_direction dir
)
3036 dev
->dma_ops
->sync_single_for_cpu(dev
, addr
, size
, dir
);
3038 dma_sync_single_for_cpu(dev
->dma_device
, addr
, size
, dir
);
3042 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3043 * @dev: The device for which the DMA address was created
3044 * @addr: The DMA address
3045 * @size: The size of the region in bytes
3046 * @dir: The direction of the DMA
3048 static inline void ib_dma_sync_single_for_device(struct ib_device
*dev
,
3051 enum dma_data_direction dir
)
3054 dev
->dma_ops
->sync_single_for_device(dev
, addr
, size
, dir
);
3056 dma_sync_single_for_device(dev
->dma_device
, addr
, size
, dir
);
3060 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3061 * @dev: The device for which the DMA address is requested
3062 * @size: The size of the region to allocate in bytes
3063 * @dma_handle: A pointer for returning the DMA address of the region
3064 * @flag: memory allocator flags
3066 static inline void *ib_dma_alloc_coherent(struct ib_device
*dev
,
3072 return dev
->dma_ops
->alloc_coherent(dev
, size
, dma_handle
, flag
);
3077 ret
= dma_alloc_coherent(dev
->dma_device
, size
, &handle
, flag
);
3078 *dma_handle
= handle
;
3084 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3085 * @dev: The device for which the DMA addresses were allocated
3086 * @size: The size of the region
3087 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3088 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3090 static inline void ib_dma_free_coherent(struct ib_device
*dev
,
3091 size_t size
, void *cpu_addr
,
3095 dev
->dma_ops
->free_coherent(dev
, size
, cpu_addr
, dma_handle
);
3097 dma_free_coherent(dev
->dma_device
, size
, cpu_addr
, dma_handle
);
3101 * ib_dereg_mr - Deregisters a memory region and removes it from the
3102 * HCA translation table.
3103 * @mr: The memory region to deregister.
3105 * This function can fail, if the memory region has memory windows bound to it.
3107 int ib_dereg_mr(struct ib_mr
*mr
);
3109 struct ib_mr
*ib_alloc_mr(struct ib_pd
*pd
,
3110 enum ib_mr_type mr_type
,
3114 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3116 * @mr - struct ib_mr pointer to be updated.
3117 * @newkey - new key to be used.
3119 static inline void ib_update_fast_reg_key(struct ib_mr
*mr
, u8 newkey
)
3121 mr
->lkey
= (mr
->lkey
& 0xffffff00) | newkey
;
3122 mr
->rkey
= (mr
->rkey
& 0xffffff00) | newkey
;
3126 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3127 * for calculating a new rkey for type 2 memory windows.
3128 * @rkey - the rkey to increment.
3130 static inline u32
ib_inc_rkey(u32 rkey
)
3132 const u32 mask
= 0x000000ff;
3133 return ((rkey
+ 1) & mask
) | (rkey
& ~mask
);
3137 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3138 * @pd: The protection domain associated with the unmapped region.
3139 * @mr_access_flags: Specifies the memory access rights.
3140 * @fmr_attr: Attributes of the unmapped region.
3142 * A fast memory region must be mapped before it can be used as part of
3145 struct ib_fmr
*ib_alloc_fmr(struct ib_pd
*pd
,
3146 int mr_access_flags
,
3147 struct ib_fmr_attr
*fmr_attr
);
3150 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3151 * @fmr: The fast memory region to associate with the pages.
3152 * @page_list: An array of physical pages to map to the fast memory region.
3153 * @list_len: The number of pages in page_list.
3154 * @iova: The I/O virtual address to use with the mapped region.
3156 static inline int ib_map_phys_fmr(struct ib_fmr
*fmr
,
3157 u64
*page_list
, int list_len
,
3160 return fmr
->device
->map_phys_fmr(fmr
, page_list
, list_len
, iova
);
3164 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3165 * @fmr_list: A linked list of fast memory regions to unmap.
3167 int ib_unmap_fmr(struct list_head
*fmr_list
);
3170 * ib_dealloc_fmr - Deallocates a fast memory region.
3171 * @fmr: The fast memory region to deallocate.
3173 int ib_dealloc_fmr(struct ib_fmr
*fmr
);
3176 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3177 * @qp: QP to attach to the multicast group. The QP must be type
3179 * @gid: Multicast group GID.
3180 * @lid: Multicast group LID in host byte order.
3182 * In order to send and receive multicast packets, subnet
3183 * administration must have created the multicast group and configured
3184 * the fabric appropriately. The port associated with the specified
3185 * QP must also be a member of the multicast group.
3187 int ib_attach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
3190 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3191 * @qp: QP to detach from the multicast group.
3192 * @gid: Multicast group GID.
3193 * @lid: Multicast group LID in host byte order.
3195 int ib_detach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
3198 * ib_alloc_xrcd - Allocates an XRC domain.
3199 * @device: The device on which to allocate the XRC domain.
3201 struct ib_xrcd
*ib_alloc_xrcd(struct ib_device
*device
);
3204 * ib_dealloc_xrcd - Deallocates an XRC domain.
3205 * @xrcd: The XRC domain to deallocate.
3207 int ib_dealloc_xrcd(struct ib_xrcd
*xrcd
);
3209 struct ib_flow
*ib_create_flow(struct ib_qp
*qp
,
3210 struct ib_flow_attr
*flow_attr
, int domain
);
3211 int ib_destroy_flow(struct ib_flow
*flow_id
);
3213 static inline int ib_check_mr_access(int flags
)
3216 * Local write permission is required if remote write or
3217 * remote atomic permission is also requested.
3219 if (flags
& (IB_ACCESS_REMOTE_ATOMIC
| IB_ACCESS_REMOTE_WRITE
) &&
3220 !(flags
& IB_ACCESS_LOCAL_WRITE
))
3227 * ib_check_mr_status: lightweight check of MR status.
3228 * This routine may provide status checks on a selected
3229 * ib_mr. first use is for signature status check.
3231 * @mr: A memory region.
3232 * @check_mask: Bitmask of which checks to perform from
3233 * ib_mr_status_check enumeration.
3234 * @mr_status: The container of relevant status checks.
3235 * failed checks will be indicated in the status bitmask
3236 * and the relevant info shall be in the error item.
3238 int ib_check_mr_status(struct ib_mr
*mr
, u32 check_mask
,
3239 struct ib_mr_status
*mr_status
);
3241 struct net_device
*ib_get_net_dev_by_params(struct ib_device
*dev
, u8 port
,
3242 u16 pkey
, const union ib_gid
*gid
,
3243 const struct sockaddr
*addr
);
3244 struct ib_wq
*ib_create_wq(struct ib_pd
*pd
,
3245 struct ib_wq_init_attr
*init_attr
);
3246 int ib_destroy_wq(struct ib_wq
*wq
);
3247 int ib_modify_wq(struct ib_wq
*wq
, struct ib_wq_attr
*attr
,
3249 struct ib_rwq_ind_table
*ib_create_rwq_ind_table(struct ib_device
*device
,
3250 struct ib_rwq_ind_table_init_attr
*
3251 wq_ind_table_init_attr
);
3252 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table
*wq_ind_table
);
3254 int ib_map_mr_sg(struct ib_mr
*mr
, struct scatterlist
*sg
, int sg_nents
,
3255 unsigned int *sg_offset
, unsigned int page_size
);
3258 ib_map_mr_sg_zbva(struct ib_mr
*mr
, struct scatterlist
*sg
, int sg_nents
,
3259 unsigned int *sg_offset
, unsigned int page_size
)
3263 n
= ib_map_mr_sg(mr
, sg
, sg_nents
, sg_offset
, page_size
);
3269 int ib_sg_to_pages(struct ib_mr
*mr
, struct scatterlist
*sgl
, int sg_nents
,
3270 unsigned int *sg_offset
, int (*set_page
)(struct ib_mr
*, u64
));
3272 void ib_drain_rq(struct ib_qp
*qp
);
3273 void ib_drain_sq(struct ib_qp
*qp
);
3274 void ib_drain_qp(struct ib_qp
*qp
);
3275 #endif /* IB_VERBS_H */