2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
39 #if !defined(IB_VERBS_H)
42 #include <linux/types.h>
43 #include <linux/device.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/netdevice.h>
60 #include <linux/if_link.h>
61 #include <linux/atomic.h>
62 #include <linux/mmu_notifier.h>
63 #include <linux/uaccess.h>
64 #include <linux/cgroup_rdma.h>
65 #include <uapi/rdma/ib_user_verbs.h>
67 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
69 extern struct workqueue_struct
*ib_wq
;
70 extern struct workqueue_struct
*ib_comp_wq
;
71 extern struct workqueue_struct
*ib_comp_unbound_wq
;
81 extern union ib_gid zgid
;
84 /* If link layer is Ethernet, this is RoCE V1 */
87 IB_GID_TYPE_ROCE_UDP_ENCAP
= 1,
91 #define ROCE_V2_UDP_DPORT 4791
93 enum ib_gid_type gid_type
;
94 struct net_device
*ndev
;
98 /* IB values map to NodeInfo:NodeType. */
108 /* set the local administered indication */
109 IB_SA_WELL_KNOWN_GUID
= BIT_ULL(57) | 2,
112 enum rdma_transport_type
{
114 RDMA_TRANSPORT_IWARP
,
115 RDMA_TRANSPORT_USNIC
,
116 RDMA_TRANSPORT_USNIC_UDP
119 enum rdma_protocol_type
{
123 RDMA_PROTOCOL_USNIC_UDP
126 __attribute_const__
enum rdma_transport_type
127 rdma_node_get_transport(enum rdma_node_type node_type
);
129 enum rdma_network_type
{
131 RDMA_NETWORK_ROCE_V1
= RDMA_NETWORK_IB
,
136 static inline enum ib_gid_type
ib_network_to_gid_type(enum rdma_network_type network_type
)
138 if (network_type
== RDMA_NETWORK_IPV4
||
139 network_type
== RDMA_NETWORK_IPV6
)
140 return IB_GID_TYPE_ROCE_UDP_ENCAP
;
142 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
143 return IB_GID_TYPE_IB
;
146 static inline enum rdma_network_type
ib_gid_to_network_type(enum ib_gid_type gid_type
,
149 if (gid_type
== IB_GID_TYPE_IB
)
150 return RDMA_NETWORK_IB
;
152 if (ipv6_addr_v4mapped((struct in6_addr
*)gid
))
153 return RDMA_NETWORK_IPV4
;
155 return RDMA_NETWORK_IPV6
;
158 enum rdma_link_layer
{
159 IB_LINK_LAYER_UNSPECIFIED
,
160 IB_LINK_LAYER_INFINIBAND
,
161 IB_LINK_LAYER_ETHERNET
,
164 enum ib_device_cap_flags
{
165 IB_DEVICE_RESIZE_MAX_WR
= (1 << 0),
166 IB_DEVICE_BAD_PKEY_CNTR
= (1 << 1),
167 IB_DEVICE_BAD_QKEY_CNTR
= (1 << 2),
168 IB_DEVICE_RAW_MULTI
= (1 << 3),
169 IB_DEVICE_AUTO_PATH_MIG
= (1 << 4),
170 IB_DEVICE_CHANGE_PHY_PORT
= (1 << 5),
171 IB_DEVICE_UD_AV_PORT_ENFORCE
= (1 << 6),
172 IB_DEVICE_CURR_QP_STATE_MOD
= (1 << 7),
173 IB_DEVICE_SHUTDOWN_PORT
= (1 << 8),
174 /* Not in use, former INIT_TYPE = (1 << 9),*/
175 IB_DEVICE_PORT_ACTIVE_EVENT
= (1 << 10),
176 IB_DEVICE_SYS_IMAGE_GUID
= (1 << 11),
177 IB_DEVICE_RC_RNR_NAK_GEN
= (1 << 12),
178 IB_DEVICE_SRQ_RESIZE
= (1 << 13),
179 IB_DEVICE_N_NOTIFY_CQ
= (1 << 14),
182 * This device supports a per-device lkey or stag that can be
183 * used without performing a memory registration for the local
184 * memory. Note that ULPs should never check this flag, but
185 * instead of use the local_dma_lkey flag in the ib_pd structure,
186 * which will always contain a usable lkey.
188 IB_DEVICE_LOCAL_DMA_LKEY
= (1 << 15),
189 /* Reserved, old SEND_W_INV = (1 << 16),*/
190 IB_DEVICE_MEM_WINDOW
= (1 << 17),
192 * Devices should set IB_DEVICE_UD_IP_SUM if they support
193 * insertion of UDP and TCP checksum on outgoing UD IPoIB
194 * messages and can verify the validity of checksum for
195 * incoming messages. Setting this flag implies that the
196 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
198 IB_DEVICE_UD_IP_CSUM
= (1 << 18),
199 IB_DEVICE_UD_TSO
= (1 << 19),
200 IB_DEVICE_XRC
= (1 << 20),
203 * This device supports the IB "base memory management extension",
204 * which includes support for fast registrations (IB_WR_REG_MR,
205 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
206 * also be set by any iWarp device which must support FRs to comply
207 * to the iWarp verbs spec. iWarp devices also support the
208 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
211 IB_DEVICE_MEM_MGT_EXTENSIONS
= (1 << 21),
212 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK
= (1 << 22),
213 IB_DEVICE_MEM_WINDOW_TYPE_2A
= (1 << 23),
214 IB_DEVICE_MEM_WINDOW_TYPE_2B
= (1 << 24),
215 IB_DEVICE_RC_IP_CSUM
= (1 << 25),
216 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
217 IB_DEVICE_RAW_IP_CSUM
= (1 << 26),
219 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
220 * support execution of WQEs that involve synchronization
221 * of I/O operations with single completion queue managed
224 IB_DEVICE_CROSS_CHANNEL
= (1 << 27),
225 IB_DEVICE_MANAGED_FLOW_STEERING
= (1 << 29),
226 IB_DEVICE_SIGNATURE_HANDOVER
= (1 << 30),
227 IB_DEVICE_ON_DEMAND_PAGING
= (1ULL << 31),
228 IB_DEVICE_SG_GAPS_REG
= (1ULL << 32),
229 IB_DEVICE_VIRTUAL_FUNCTION
= (1ULL << 33),
230 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
231 IB_DEVICE_RAW_SCATTER_FCS
= (1ULL << 34),
232 IB_DEVICE_RDMA_NETDEV_OPA_VNIC
= (1ULL << 35),
233 /* The device supports padding incoming writes to cacheline. */
234 IB_DEVICE_PCI_WRITE_END_PADDING
= (1ULL << 36),
237 enum ib_signature_prot_cap
{
238 IB_PROT_T10DIF_TYPE_1
= 1,
239 IB_PROT_T10DIF_TYPE_2
= 1 << 1,
240 IB_PROT_T10DIF_TYPE_3
= 1 << 2,
243 enum ib_signature_guard_cap
{
244 IB_GUARD_T10DIF_CRC
= 1,
245 IB_GUARD_T10DIF_CSUM
= 1 << 1,
254 enum ib_odp_general_cap_bits
{
255 IB_ODP_SUPPORT
= 1 << 0,
256 IB_ODP_SUPPORT_IMPLICIT
= 1 << 1,
259 enum ib_odp_transport_cap_bits
{
260 IB_ODP_SUPPORT_SEND
= 1 << 0,
261 IB_ODP_SUPPORT_RECV
= 1 << 1,
262 IB_ODP_SUPPORT_WRITE
= 1 << 2,
263 IB_ODP_SUPPORT_READ
= 1 << 3,
264 IB_ODP_SUPPORT_ATOMIC
= 1 << 4,
268 uint64_t general_caps
;
270 uint32_t rc_odp_caps
;
271 uint32_t uc_odp_caps
;
272 uint32_t ud_odp_caps
;
273 } per_transport_caps
;
277 /* Corresponding bit will be set if qp type from
278 * 'enum ib_qp_type' is supported, e.g.
279 * supported_qpts |= 1 << IB_QPT_UD
282 u32 max_rwq_indirection_tables
;
283 u32 max_rwq_indirection_table_size
;
286 enum ib_tm_cap_flags
{
287 /* Support tag matching with rendezvous offload for RC transport */
288 IB_TM_CAP_RNDV_RC
= 1 << 0,
292 /* Max size of RNDV header */
293 u32 max_rndv_hdr_size
;
294 /* Max number of entries in tag matching list */
296 /* From enum ib_tm_cap_flags */
298 /* Max number of outstanding list operations */
300 /* Max number of SGE in tag matching entry */
304 enum ib_cq_creation_flags
{
305 IB_CQ_FLAGS_TIMESTAMP_COMPLETION
= 1 << 0,
306 IB_CQ_FLAGS_IGNORE_OVERRUN
= 1 << 1,
309 struct ib_cq_init_attr
{
315 enum ib_cq_attr_mask
{
316 IB_CQ_MODERATE
= 1 << 0,
320 u16 max_cq_moderation_count
;
321 u16 max_cq_moderation_period
;
324 struct ib_device_attr
{
326 __be64 sys_image_guid
;
334 u64 device_cap_flags
;
344 int max_qp_init_rd_atom
;
345 int max_ee_init_rd_atom
;
346 enum ib_atomic_cap atomic_cap
;
347 enum ib_atomic_cap masked_atomic_cap
;
354 int max_mcast_qp_attach
;
355 int max_total_mcast_qp_attach
;
362 unsigned int max_fast_reg_page_list_len
;
364 u8 local_ca_ack_delay
;
367 struct ib_odp_caps odp_caps
;
368 uint64_t timestamp_mask
;
369 uint64_t hca_core_clock
; /* in KHZ */
370 struct ib_rss_caps rss_caps
;
372 u32 raw_packet_caps
; /* Use ib_raw_packet_caps enum */
373 struct ib_tm_caps tm_caps
;
374 struct ib_cq_caps cq_caps
;
385 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu
)
388 case IB_MTU_256
: return 256;
389 case IB_MTU_512
: return 512;
390 case IB_MTU_1024
: return 1024;
391 case IB_MTU_2048
: return 2048;
392 case IB_MTU_4096
: return 4096;
397 static inline enum ib_mtu
ib_mtu_int_to_enum(int mtu
)
401 else if (mtu
>= 2048)
403 else if (mtu
>= 1024)
417 IB_PORT_ACTIVE_DEFER
= 5
420 enum ib_port_cap_flags
{
422 IB_PORT_NOTICE_SUP
= 1 << 2,
423 IB_PORT_TRAP_SUP
= 1 << 3,
424 IB_PORT_OPT_IPD_SUP
= 1 << 4,
425 IB_PORT_AUTO_MIGR_SUP
= 1 << 5,
426 IB_PORT_SL_MAP_SUP
= 1 << 6,
427 IB_PORT_MKEY_NVRAM
= 1 << 7,
428 IB_PORT_PKEY_NVRAM
= 1 << 8,
429 IB_PORT_LED_INFO_SUP
= 1 << 9,
430 IB_PORT_SM_DISABLED
= 1 << 10,
431 IB_PORT_SYS_IMAGE_GUID_SUP
= 1 << 11,
432 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP
= 1 << 12,
433 IB_PORT_EXTENDED_SPEEDS_SUP
= 1 << 14,
434 IB_PORT_CM_SUP
= 1 << 16,
435 IB_PORT_SNMP_TUNNEL_SUP
= 1 << 17,
436 IB_PORT_REINIT_SUP
= 1 << 18,
437 IB_PORT_DEVICE_MGMT_SUP
= 1 << 19,
438 IB_PORT_VENDOR_CLASS_SUP
= 1 << 20,
439 IB_PORT_DR_NOTICE_SUP
= 1 << 21,
440 IB_PORT_CAP_MASK_NOTICE_SUP
= 1 << 22,
441 IB_PORT_BOOT_MGMT_SUP
= 1 << 23,
442 IB_PORT_LINK_LATENCY_SUP
= 1 << 24,
443 IB_PORT_CLIENT_REG_SUP
= 1 << 25,
444 IB_PORT_IP_BASED_GIDS
= 1 << 26,
454 static inline int ib_width_enum_to_int(enum ib_port_width width
)
457 case IB_WIDTH_1X
: return 1;
458 case IB_WIDTH_4X
: return 4;
459 case IB_WIDTH_8X
: return 8;
460 case IB_WIDTH_12X
: return 12;
476 * struct rdma_hw_stats
477 * @timestamp - Used by the core code to track when the last update was
478 * @lifespan - Used by the core code to determine how old the counters
479 * should be before being updated again. Stored in jiffies, defaults
480 * to 10 milliseconds, drivers can override the default be specifying
481 * their own value during their allocation routine.
482 * @name - Array of pointers to static names used for the counters in
484 * @num_counters - How many hardware counters there are. If name is
485 * shorter than this number, a kernel oops will result. Driver authors
486 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
487 * in their code to prevent this.
488 * @value - Array of u64 counters that are accessed by the sysfs code and
489 * filled in by the drivers get_stats routine
491 struct rdma_hw_stats
{
492 unsigned long timestamp
;
493 unsigned long lifespan
;
494 const char * const *names
;
499 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
501 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
503 * @names - Array of static const char *
504 * @num_counters - How many elements in array
505 * @lifespan - How many milliseconds between updates
507 static inline struct rdma_hw_stats
*rdma_alloc_hw_stats_struct(
508 const char * const *names
, int num_counters
,
509 unsigned long lifespan
)
511 struct rdma_hw_stats
*stats
;
513 stats
= kzalloc(sizeof(*stats
) + num_counters
* sizeof(u64
),
517 stats
->names
= names
;
518 stats
->num_counters
= num_counters
;
519 stats
->lifespan
= msecs_to_jiffies(lifespan
);
525 /* Define bits for the various functionality this port needs to be supported by
528 /* Management 0x00000FFF */
529 #define RDMA_CORE_CAP_IB_MAD 0x00000001
530 #define RDMA_CORE_CAP_IB_SMI 0x00000002
531 #define RDMA_CORE_CAP_IB_CM 0x00000004
532 #define RDMA_CORE_CAP_IW_CM 0x00000008
533 #define RDMA_CORE_CAP_IB_SA 0x00000010
534 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
536 /* Address format 0x000FF000 */
537 #define RDMA_CORE_CAP_AF_IB 0x00001000
538 #define RDMA_CORE_CAP_ETH_AH 0x00002000
539 #define RDMA_CORE_CAP_OPA_AH 0x00004000
541 /* Protocol 0xFFF00000 */
542 #define RDMA_CORE_CAP_PROT_IB 0x00100000
543 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
544 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
545 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
546 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
547 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000
549 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
550 | RDMA_CORE_CAP_IB_MAD \
551 | RDMA_CORE_CAP_IB_SMI \
552 | RDMA_CORE_CAP_IB_CM \
553 | RDMA_CORE_CAP_IB_SA \
554 | RDMA_CORE_CAP_AF_IB)
555 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
556 | RDMA_CORE_CAP_IB_MAD \
557 | RDMA_CORE_CAP_IB_CM \
558 | RDMA_CORE_CAP_AF_IB \
559 | RDMA_CORE_CAP_ETH_AH)
560 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
561 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
562 | RDMA_CORE_CAP_IB_MAD \
563 | RDMA_CORE_CAP_IB_CM \
564 | RDMA_CORE_CAP_AF_IB \
565 | RDMA_CORE_CAP_ETH_AH)
566 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
567 | RDMA_CORE_CAP_IW_CM)
568 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
569 | RDMA_CORE_CAP_OPA_MAD)
571 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
573 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
575 struct ib_port_attr
{
577 enum ib_port_state state
;
579 enum ib_mtu active_mtu
;
599 enum ib_device_modify_flags
{
600 IB_DEVICE_MODIFY_SYS_IMAGE_GUID
= 1 << 0,
601 IB_DEVICE_MODIFY_NODE_DESC
= 1 << 1
604 #define IB_DEVICE_NODE_DESC_MAX 64
606 struct ib_device_modify
{
608 char node_desc
[IB_DEVICE_NODE_DESC_MAX
];
611 enum ib_port_modify_flags
{
612 IB_PORT_SHUTDOWN
= 1,
613 IB_PORT_INIT_TYPE
= (1<<2),
614 IB_PORT_RESET_QKEY_CNTR
= (1<<3),
615 IB_PORT_OPA_MASK_CHG
= (1<<4)
618 struct ib_port_modify
{
619 u32 set_port_cap_mask
;
620 u32 clr_port_cap_mask
;
628 IB_EVENT_QP_ACCESS_ERR
,
632 IB_EVENT_PATH_MIG_ERR
,
633 IB_EVENT_DEVICE_FATAL
,
634 IB_EVENT_PORT_ACTIVE
,
637 IB_EVENT_PKEY_CHANGE
,
640 IB_EVENT_SRQ_LIMIT_REACHED
,
641 IB_EVENT_QP_LAST_WQE_REACHED
,
642 IB_EVENT_CLIENT_REREGISTER
,
647 const char *__attribute_const__
ib_event_msg(enum ib_event_type event
);
650 struct ib_device
*device
;
658 enum ib_event_type event
;
661 struct ib_event_handler
{
662 struct ib_device
*device
;
663 void (*handler
)(struct ib_event_handler
*, struct ib_event
*);
664 struct list_head list
;
667 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
669 (_ptr)->device = _device; \
670 (_ptr)->handler = _handler; \
671 INIT_LIST_HEAD(&(_ptr)->list); \
674 struct ib_global_route
{
683 __be32 version_tclass_flow
;
691 union rdma_network_hdr
{
694 /* The IB spec states that if it's IPv4, the header
695 * is located in the last 20 bytes of the header.
698 struct iphdr roce4grh
;
702 #define IB_QPN_MASK 0xFFFFFF
705 IB_MULTICAST_QPN
= 0xffffff
708 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
709 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
716 IB_RATE_PORT_CURRENT
= 0,
717 IB_RATE_2_5_GBPS
= 2,
725 IB_RATE_120_GBPS
= 10,
726 IB_RATE_14_GBPS
= 11,
727 IB_RATE_56_GBPS
= 12,
728 IB_RATE_112_GBPS
= 13,
729 IB_RATE_168_GBPS
= 14,
730 IB_RATE_25_GBPS
= 15,
731 IB_RATE_100_GBPS
= 16,
732 IB_RATE_200_GBPS
= 17,
733 IB_RATE_300_GBPS
= 18
737 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
738 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
739 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
740 * @rate: rate to convert.
742 __attribute_const__
int ib_rate_to_mult(enum ib_rate rate
);
745 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
746 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
747 * @rate: rate to convert.
749 __attribute_const__
int ib_rate_to_mbps(enum ib_rate rate
);
753 * enum ib_mr_type - memory region type
754 * @IB_MR_TYPE_MEM_REG: memory region that is used for
755 * normal registration
756 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
757 * signature operations (data-integrity
759 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
760 * register any arbitrary sg lists (without
761 * the normal mr constraints - see
766 IB_MR_TYPE_SIGNATURE
,
772 * IB_SIG_TYPE_NONE: Unprotected.
773 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
775 enum ib_signature_type
{
781 * Signature T10-DIF block-guard types
782 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
783 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
785 enum ib_t10_dif_bg_type
{
791 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
793 * @bg_type: T10-DIF block guard type (CRC|CSUM)
794 * @pi_interval: protection information interval.
795 * @bg: seed of guard computation.
796 * @app_tag: application tag of guard block
797 * @ref_tag: initial guard block reference tag.
798 * @ref_remap: Indicate wethear the reftag increments each block
799 * @app_escape: Indicate to skip block check if apptag=0xffff
800 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
801 * @apptag_check_mask: check bitmask of application tag.
803 struct ib_t10_dif_domain
{
804 enum ib_t10_dif_bg_type bg_type
;
812 u16 apptag_check_mask
;
816 * struct ib_sig_domain - Parameters for signature domain
817 * @sig_type: specific signauture type
818 * @sig: union of all signature domain attributes that may
819 * be used to set domain layout.
821 struct ib_sig_domain
{
822 enum ib_signature_type sig_type
;
824 struct ib_t10_dif_domain dif
;
829 * struct ib_sig_attrs - Parameters for signature handover operation
830 * @check_mask: bitmask for signature byte check (8 bytes)
831 * @mem: memory domain layout desciptor.
832 * @wire: wire domain layout desciptor.
834 struct ib_sig_attrs
{
836 struct ib_sig_domain mem
;
837 struct ib_sig_domain wire
;
840 enum ib_sig_err_type
{
847 * struct ib_sig_err - signature error descriptor
850 enum ib_sig_err_type err_type
;
857 enum ib_mr_status_check
{
858 IB_MR_CHECK_SIG_STATUS
= 1,
862 * struct ib_mr_status - Memory region status container
864 * @fail_status: Bitmask of MR checks status. For each
865 * failed check a corresponding status bit is set.
866 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
869 struct ib_mr_status
{
871 struct ib_sig_err sig_err
;
875 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
877 * @mult: multiple to convert.
879 __attribute_const__
enum ib_rate
mult_to_ib_rate(int mult
);
881 enum rdma_ah_attr_type
{
882 RDMA_AH_ATTR_TYPE_UNDEFINED
,
883 RDMA_AH_ATTR_TYPE_IB
,
884 RDMA_AH_ATTR_TYPE_ROCE
,
885 RDMA_AH_ATTR_TYPE_OPA
,
893 struct roce_ah_attr
{
903 struct rdma_ah_attr
{
904 struct ib_global_route grh
;
909 enum rdma_ah_attr_type type
;
911 struct ib_ah_attr ib
;
912 struct roce_ah_attr roce
;
913 struct opa_ah_attr opa
;
921 IB_WC_LOC_EEC_OP_ERR
,
926 IB_WC_LOC_ACCESS_ERR
,
927 IB_WC_REM_INV_REQ_ERR
,
928 IB_WC_REM_ACCESS_ERR
,
931 IB_WC_RNR_RETRY_EXC_ERR
,
932 IB_WC_LOC_RDD_VIOL_ERR
,
933 IB_WC_REM_INV_RD_REQ_ERR
,
936 IB_WC_INV_EEC_STATE_ERR
,
938 IB_WC_RESP_TIMEOUT_ERR
,
942 const char *__attribute_const__
ib_wc_status_msg(enum ib_wc_status status
);
953 IB_WC_MASKED_COMP_SWAP
,
954 IB_WC_MASKED_FETCH_ADD
,
956 * Set value of IB_WC_RECV so consumers can test if a completion is a
957 * receive by testing (opcode & IB_WC_RECV).
960 IB_WC_RECV_RDMA_WITH_IMM
965 IB_WC_WITH_IMM
= (1<<1),
966 IB_WC_WITH_INVALIDATE
= (1<<2),
967 IB_WC_IP_CSUM_OK
= (1<<3),
968 IB_WC_WITH_SMAC
= (1<<4),
969 IB_WC_WITH_VLAN
= (1<<5),
970 IB_WC_WITH_NETWORK_HDR_TYPE
= (1<<6),
976 struct ib_cqe
*wr_cqe
;
978 enum ib_wc_status status
;
979 enum ib_wc_opcode opcode
;
993 u8 port_num
; /* valid only for DR SMPs on switches */
999 enum ib_cq_notify_flags
{
1000 IB_CQ_SOLICITED
= 1 << 0,
1001 IB_CQ_NEXT_COMP
= 1 << 1,
1002 IB_CQ_SOLICITED_MASK
= IB_CQ_SOLICITED
| IB_CQ_NEXT_COMP
,
1003 IB_CQ_REPORT_MISSED_EVENTS
= 1 << 2,
1012 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type
)
1014 return srq_type
== IB_SRQT_XRC
||
1015 srq_type
== IB_SRQT_TM
;
1018 enum ib_srq_attr_mask
{
1019 IB_SRQ_MAX_WR
= 1 << 0,
1020 IB_SRQ_LIMIT
= 1 << 1,
1023 struct ib_srq_attr
{
1029 struct ib_srq_init_attr
{
1030 void (*event_handler
)(struct ib_event
*, void *);
1032 struct ib_srq_attr attr
;
1033 enum ib_srq_type srq_type
;
1039 struct ib_xrcd
*xrcd
;
1054 u32 max_inline_data
;
1057 * Maximum number of rdma_rw_ctx structures in flight at a time.
1058 * ib_create_qp() will calculate the right amount of neededed WRs
1059 * and MRs based on this.
1071 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1072 * here (and in that order) since the MAD layer uses them as
1073 * indices into a 2-entry table.
1082 IB_QPT_RAW_ETHERTYPE
,
1083 IB_QPT_RAW_PACKET
= 8,
1087 /* Reserve a range for qp types internal to the low level driver.
1088 * These qp types will not be visible at the IB core layer, so the
1089 * IB_QPT_MAX usages should not be affected in the core layer
1091 IB_QPT_RESERVED1
= 0x1000,
1103 enum ib_qp_create_flags
{
1104 IB_QP_CREATE_IPOIB_UD_LSO
= 1 << 0,
1105 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK
= 1 << 1,
1106 IB_QP_CREATE_CROSS_CHANNEL
= 1 << 2,
1107 IB_QP_CREATE_MANAGED_SEND
= 1 << 3,
1108 IB_QP_CREATE_MANAGED_RECV
= 1 << 4,
1109 IB_QP_CREATE_NETIF_QP
= 1 << 5,
1110 IB_QP_CREATE_SIGNATURE_EN
= 1 << 6,
1111 /* FREE = 1 << 7, */
1112 IB_QP_CREATE_SCATTER_FCS
= 1 << 8,
1113 IB_QP_CREATE_CVLAN_STRIPPING
= 1 << 9,
1114 IB_QP_CREATE_SOURCE_QPN
= 1 << 10,
1115 IB_QP_CREATE_PCI_WRITE_END_PADDING
= 1 << 11,
1116 /* reserve bits 26-31 for low level drivers' internal use */
1117 IB_QP_CREATE_RESERVED_START
= 1 << 26,
1118 IB_QP_CREATE_RESERVED_END
= 1 << 31,
1122 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1123 * callback to destroy the passed in QP.
1126 struct ib_qp_init_attr
{
1127 void (*event_handler
)(struct ib_event
*, void *);
1129 struct ib_cq
*send_cq
;
1130 struct ib_cq
*recv_cq
;
1132 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
1133 struct ib_qp_cap cap
;
1134 enum ib_sig_type sq_sig_type
;
1135 enum ib_qp_type qp_type
;
1139 * Only needed for special QP types, or when using the RW API.
1142 struct ib_rwq_ind_table
*rwq_ind_tbl
;
1146 struct ib_qp_open_attr
{
1147 void (*event_handler
)(struct ib_event
*, void *);
1150 enum ib_qp_type qp_type
;
1153 enum ib_rnr_timeout
{
1154 IB_RNR_TIMER_655_36
= 0,
1155 IB_RNR_TIMER_000_01
= 1,
1156 IB_RNR_TIMER_000_02
= 2,
1157 IB_RNR_TIMER_000_03
= 3,
1158 IB_RNR_TIMER_000_04
= 4,
1159 IB_RNR_TIMER_000_06
= 5,
1160 IB_RNR_TIMER_000_08
= 6,
1161 IB_RNR_TIMER_000_12
= 7,
1162 IB_RNR_TIMER_000_16
= 8,
1163 IB_RNR_TIMER_000_24
= 9,
1164 IB_RNR_TIMER_000_32
= 10,
1165 IB_RNR_TIMER_000_48
= 11,
1166 IB_RNR_TIMER_000_64
= 12,
1167 IB_RNR_TIMER_000_96
= 13,
1168 IB_RNR_TIMER_001_28
= 14,
1169 IB_RNR_TIMER_001_92
= 15,
1170 IB_RNR_TIMER_002_56
= 16,
1171 IB_RNR_TIMER_003_84
= 17,
1172 IB_RNR_TIMER_005_12
= 18,
1173 IB_RNR_TIMER_007_68
= 19,
1174 IB_RNR_TIMER_010_24
= 20,
1175 IB_RNR_TIMER_015_36
= 21,
1176 IB_RNR_TIMER_020_48
= 22,
1177 IB_RNR_TIMER_030_72
= 23,
1178 IB_RNR_TIMER_040_96
= 24,
1179 IB_RNR_TIMER_061_44
= 25,
1180 IB_RNR_TIMER_081_92
= 26,
1181 IB_RNR_TIMER_122_88
= 27,
1182 IB_RNR_TIMER_163_84
= 28,
1183 IB_RNR_TIMER_245_76
= 29,
1184 IB_RNR_TIMER_327_68
= 30,
1185 IB_RNR_TIMER_491_52
= 31
1188 enum ib_qp_attr_mask
{
1190 IB_QP_CUR_STATE
= (1<<1),
1191 IB_QP_EN_SQD_ASYNC_NOTIFY
= (1<<2),
1192 IB_QP_ACCESS_FLAGS
= (1<<3),
1193 IB_QP_PKEY_INDEX
= (1<<4),
1194 IB_QP_PORT
= (1<<5),
1195 IB_QP_QKEY
= (1<<6),
1197 IB_QP_PATH_MTU
= (1<<8),
1198 IB_QP_TIMEOUT
= (1<<9),
1199 IB_QP_RETRY_CNT
= (1<<10),
1200 IB_QP_RNR_RETRY
= (1<<11),
1201 IB_QP_RQ_PSN
= (1<<12),
1202 IB_QP_MAX_QP_RD_ATOMIC
= (1<<13),
1203 IB_QP_ALT_PATH
= (1<<14),
1204 IB_QP_MIN_RNR_TIMER
= (1<<15),
1205 IB_QP_SQ_PSN
= (1<<16),
1206 IB_QP_MAX_DEST_RD_ATOMIC
= (1<<17),
1207 IB_QP_PATH_MIG_STATE
= (1<<18),
1208 IB_QP_CAP
= (1<<19),
1209 IB_QP_DEST_QPN
= (1<<20),
1210 IB_QP_RESERVED1
= (1<<21),
1211 IB_QP_RESERVED2
= (1<<22),
1212 IB_QP_RESERVED3
= (1<<23),
1213 IB_QP_RESERVED4
= (1<<24),
1214 IB_QP_RATE_LIMIT
= (1<<25),
1239 enum ib_qp_state qp_state
;
1240 enum ib_qp_state cur_qp_state
;
1241 enum ib_mtu path_mtu
;
1242 enum ib_mig_state path_mig_state
;
1247 int qp_access_flags
;
1248 struct ib_qp_cap cap
;
1249 struct rdma_ah_attr ah_attr
;
1250 struct rdma_ah_attr alt_ah_attr
;
1253 u8 en_sqd_async_notify
;
1256 u8 max_dest_rd_atomic
;
1268 /* These are shared with userspace */
1269 IB_WR_RDMA_WRITE
= IB_UVERBS_WR_RDMA_WRITE
,
1270 IB_WR_RDMA_WRITE_WITH_IMM
= IB_UVERBS_WR_RDMA_WRITE_WITH_IMM
,
1271 IB_WR_SEND
= IB_UVERBS_WR_SEND
,
1272 IB_WR_SEND_WITH_IMM
= IB_UVERBS_WR_SEND_WITH_IMM
,
1273 IB_WR_RDMA_READ
= IB_UVERBS_WR_RDMA_READ
,
1274 IB_WR_ATOMIC_CMP_AND_SWP
= IB_UVERBS_WR_ATOMIC_CMP_AND_SWP
,
1275 IB_WR_ATOMIC_FETCH_AND_ADD
= IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD
,
1276 IB_WR_LSO
= IB_UVERBS_WR_TSO
,
1277 IB_WR_SEND_WITH_INV
= IB_UVERBS_WR_SEND_WITH_INV
,
1278 IB_WR_RDMA_READ_WITH_INV
= IB_UVERBS_WR_RDMA_READ_WITH_INV
,
1279 IB_WR_LOCAL_INV
= IB_UVERBS_WR_LOCAL_INV
,
1280 IB_WR_MASKED_ATOMIC_CMP_AND_SWP
=
1281 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP
,
1282 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD
=
1283 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD
,
1285 /* These are kernel only and can not be issued by userspace */
1286 IB_WR_REG_MR
= 0x20,
1289 /* reserve values for low level drivers' internal use.
1290 * These values will not be used at all in the ib core layer.
1292 IB_WR_RESERVED1
= 0xf0,
1304 enum ib_send_flags
{
1306 IB_SEND_SIGNALED
= (1<<1),
1307 IB_SEND_SOLICITED
= (1<<2),
1308 IB_SEND_INLINE
= (1<<3),
1309 IB_SEND_IP_CSUM
= (1<<4),
1311 /* reserve bits 26-31 for low level drivers' internal use */
1312 IB_SEND_RESERVED_START
= (1 << 26),
1313 IB_SEND_RESERVED_END
= (1 << 31),
1323 void (*done
)(struct ib_cq
*cq
, struct ib_wc
*wc
);
1327 struct ib_send_wr
*next
;
1330 struct ib_cqe
*wr_cqe
;
1332 struct ib_sge
*sg_list
;
1334 enum ib_wr_opcode opcode
;
1338 u32 invalidate_rkey
;
1343 struct ib_send_wr wr
;
1348 static inline struct ib_rdma_wr
*rdma_wr(struct ib_send_wr
*wr
)
1350 return container_of(wr
, struct ib_rdma_wr
, wr
);
1353 struct ib_atomic_wr
{
1354 struct ib_send_wr wr
;
1358 u64 compare_add_mask
;
1363 static inline struct ib_atomic_wr
*atomic_wr(struct ib_send_wr
*wr
)
1365 return container_of(wr
, struct ib_atomic_wr
, wr
);
1369 struct ib_send_wr wr
;
1376 u16 pkey_index
; /* valid for GSI only */
1377 u8 port_num
; /* valid for DR SMPs on switch only */
1380 static inline struct ib_ud_wr
*ud_wr(struct ib_send_wr
*wr
)
1382 return container_of(wr
, struct ib_ud_wr
, wr
);
1386 struct ib_send_wr wr
;
1392 static inline struct ib_reg_wr
*reg_wr(struct ib_send_wr
*wr
)
1394 return container_of(wr
, struct ib_reg_wr
, wr
);
1397 struct ib_sig_handover_wr
{
1398 struct ib_send_wr wr
;
1399 struct ib_sig_attrs
*sig_attrs
;
1400 struct ib_mr
*sig_mr
;
1402 struct ib_sge
*prot
;
1405 static inline struct ib_sig_handover_wr
*sig_handover_wr(struct ib_send_wr
*wr
)
1407 return container_of(wr
, struct ib_sig_handover_wr
, wr
);
1411 struct ib_recv_wr
*next
;
1414 struct ib_cqe
*wr_cqe
;
1416 struct ib_sge
*sg_list
;
1420 enum ib_access_flags
{
1421 IB_ACCESS_LOCAL_WRITE
= 1,
1422 IB_ACCESS_REMOTE_WRITE
= (1<<1),
1423 IB_ACCESS_REMOTE_READ
= (1<<2),
1424 IB_ACCESS_REMOTE_ATOMIC
= (1<<3),
1425 IB_ACCESS_MW_BIND
= (1<<4),
1426 IB_ZERO_BASED
= (1<<5),
1427 IB_ACCESS_ON_DEMAND
= (1<<6),
1428 IB_ACCESS_HUGETLB
= (1<<7),
1432 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1433 * are hidden here instead of a uapi header!
1435 enum ib_mr_rereg_flags
{
1436 IB_MR_REREG_TRANS
= 1,
1437 IB_MR_REREG_PD
= (1<<1),
1438 IB_MR_REREG_ACCESS
= (1<<2),
1439 IB_MR_REREG_SUPPORTED
= ((IB_MR_REREG_ACCESS
<< 1) - 1)
1442 struct ib_fmr_attr
{
1450 enum rdma_remove_reason
{
1451 /* Userspace requested uobject deletion. Call could fail */
1452 RDMA_REMOVE_DESTROY
,
1453 /* Context deletion. This call should delete the actual object itself */
1455 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1456 RDMA_REMOVE_DRIVER_REMOVE
,
1457 /* Context is being cleaned-up, but commit was just completed */
1458 RDMA_REMOVE_DURING_CLEANUP
,
1461 struct ib_rdmacg_object
{
1462 #ifdef CONFIG_CGROUP_RDMA
1463 struct rdma_cgroup
*cg
; /* owner rdma cgroup */
1467 struct ib_ucontext
{
1468 struct ib_device
*device
;
1469 struct ib_uverbs_file
*ufile
;
1472 /* locking the uobjects_list */
1473 struct mutex uobjects_lock
;
1474 struct list_head uobjects
;
1475 /* protects cleanup process from other actions */
1476 struct rw_semaphore cleanup_rwsem
;
1477 enum rdma_remove_reason cleanup_reason
;
1480 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1481 struct rb_root_cached umem_tree
;
1483 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1484 * mmu notifiers registration.
1486 struct rw_semaphore umem_rwsem
;
1487 void (*invalidate_range
)(struct ib_umem
*umem
,
1488 unsigned long start
, unsigned long end
);
1490 struct mmu_notifier mn
;
1491 atomic_t notifier_count
;
1492 /* A list of umems that don't have private mmu notifier counters yet. */
1493 struct list_head no_private_counters
;
1497 struct ib_rdmacg_object cg_obj
;
1501 u64 user_handle
; /* handle given to us by userspace */
1502 struct ib_ucontext
*context
; /* associated user context */
1503 void *object
; /* containing object */
1504 struct list_head list
; /* link to context's list */
1505 struct ib_rdmacg_object cg_obj
; /* rdmacg object */
1506 int id
; /* index into kernel idr */
1508 atomic_t usecnt
; /* protects exclusive access */
1509 struct rcu_head rcu
; /* kfree_rcu() overhead */
1511 const struct uverbs_obj_type
*type
;
1514 struct ib_uobject_file
{
1515 struct ib_uobject uobj
;
1516 /* ufile contains the lock between context release and file close */
1517 struct ib_uverbs_file
*ufile
;
1521 const void __user
*inbuf
;
1522 void __user
*outbuf
;
1530 struct ib_device
*device
;
1531 struct ib_uobject
*uobject
;
1532 atomic_t usecnt
; /* count all resources */
1534 u32 unsafe_global_rkey
;
1537 * Implementation details of the RDMA core, don't use in drivers:
1539 struct ib_mr
*__internal_mr
;
1543 struct ib_device
*device
;
1544 atomic_t usecnt
; /* count all exposed resources */
1545 struct inode
*inode
;
1547 struct mutex tgt_qp_mutex
;
1548 struct list_head tgt_qp_list
;
1552 struct ib_device
*device
;
1554 struct ib_uobject
*uobject
;
1555 enum rdma_ah_attr_type type
;
1558 typedef void (*ib_comp_handler
)(struct ib_cq
*cq
, void *cq_context
);
1560 enum ib_poll_context
{
1561 IB_POLL_DIRECT
, /* caller context, no hw completions */
1562 IB_POLL_SOFTIRQ
, /* poll from softirq context */
1563 IB_POLL_WORKQUEUE
, /* poll from workqueue */
1564 IB_POLL_UNBOUND_WORKQUEUE
, /* poll from unbound workqueue */
1568 struct ib_device
*device
;
1569 struct ib_uobject
*uobject
;
1570 ib_comp_handler comp_handler
;
1571 void (*event_handler
)(struct ib_event
*, void *);
1574 atomic_t usecnt
; /* count number of work queues */
1575 enum ib_poll_context poll_ctx
;
1578 struct irq_poll iop
;
1579 struct work_struct work
;
1581 struct workqueue_struct
*comp_wq
;
1585 struct ib_device
*device
;
1587 struct ib_uobject
*uobject
;
1588 void (*event_handler
)(struct ib_event
*, void *);
1590 enum ib_srq_type srq_type
;
1597 struct ib_xrcd
*xrcd
;
1604 enum ib_raw_packet_caps
{
1605 /* Strip cvlan from incoming packet and report it in the matching work
1606 * completion is supported.
1608 IB_RAW_PACKET_CAP_CVLAN_STRIPPING
= (1 << 0),
1609 /* Scatter FCS field of an incoming packet to host memory is supported.
1611 IB_RAW_PACKET_CAP_SCATTER_FCS
= (1 << 1),
1612 /* Checksum offloads are supported (for both send and receive). */
1613 IB_RAW_PACKET_CAP_IP_CSUM
= (1 << 2),
1614 /* When a packet is received for an RQ with no receive WQEs, the
1615 * packet processing is delayed.
1617 IB_RAW_PACKET_CAP_DELAY_DROP
= (1 << 3),
1631 struct ib_device
*device
;
1632 struct ib_uobject
*uobject
;
1634 void (*event_handler
)(struct ib_event
*, void *);
1638 enum ib_wq_state state
;
1639 enum ib_wq_type wq_type
;
1644 IB_WQ_FLAGS_CVLAN_STRIPPING
= 1 << 0,
1645 IB_WQ_FLAGS_SCATTER_FCS
= 1 << 1,
1646 IB_WQ_FLAGS_DELAY_DROP
= 1 << 2,
1647 IB_WQ_FLAGS_PCI_WRITE_END_PADDING
= 1 << 3,
1650 struct ib_wq_init_attr
{
1652 enum ib_wq_type wq_type
;
1656 void (*event_handler
)(struct ib_event
*, void *);
1657 u32 create_flags
; /* Use enum ib_wq_flags */
1660 enum ib_wq_attr_mask
{
1661 IB_WQ_STATE
= 1 << 0,
1662 IB_WQ_CUR_STATE
= 1 << 1,
1663 IB_WQ_FLAGS
= 1 << 2,
1667 enum ib_wq_state wq_state
;
1668 enum ib_wq_state curr_wq_state
;
1669 u32 flags
; /* Use enum ib_wq_flags */
1670 u32 flags_mask
; /* Use enum ib_wq_flags */
1673 struct ib_rwq_ind_table
{
1674 struct ib_device
*device
;
1675 struct ib_uobject
*uobject
;
1678 u32 log_ind_tbl_size
;
1679 struct ib_wq
**ind_tbl
;
1682 struct ib_rwq_ind_table_init_attr
{
1683 u32 log_ind_tbl_size
;
1684 /* Each entry is a pointer to Receive Work Queue */
1685 struct ib_wq
**ind_tbl
;
1688 enum port_pkey_state
{
1689 IB_PORT_PKEY_NOT_VALID
= 0,
1690 IB_PORT_PKEY_VALID
= 1,
1691 IB_PORT_PKEY_LISTED
= 2,
1694 struct ib_qp_security
;
1696 struct ib_port_pkey
{
1697 enum port_pkey_state state
;
1700 struct list_head qp_list
;
1701 struct list_head to_error_list
;
1702 struct ib_qp_security
*sec
;
1705 struct ib_ports_pkeys
{
1706 struct ib_port_pkey main
;
1707 struct ib_port_pkey alt
;
1710 struct ib_qp_security
{
1712 struct ib_device
*dev
;
1713 /* Hold this mutex when changing port and pkey settings. */
1715 struct ib_ports_pkeys
*ports_pkeys
;
1716 /* A list of all open shared QP handles. Required to enforce security
1717 * properly for all users of a shared QP.
1719 struct list_head shared_qp_list
;
1722 atomic_t error_list_count
;
1723 struct completion error_complete
;
1724 int error_comps_pending
;
1728 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1729 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1732 struct ib_device
*device
;
1734 struct ib_cq
*send_cq
;
1735 struct ib_cq
*recv_cq
;
1738 struct list_head rdma_mrs
;
1739 struct list_head sig_mrs
;
1741 struct ib_xrcd
*xrcd
; /* XRC TGT QPs only */
1742 struct list_head xrcd_list
;
1744 /* count times opened, mcast attaches, flow attaches */
1746 struct list_head open_list
;
1747 struct ib_qp
*real_qp
;
1748 struct ib_uobject
*uobject
;
1749 void (*event_handler
)(struct ib_event
*, void *);
1754 enum ib_qp_type qp_type
;
1755 struct ib_rwq_ind_table
*rwq_ind_tbl
;
1756 struct ib_qp_security
*qp_sec
;
1761 struct ib_device
*device
;
1767 unsigned int page_size
;
1770 struct ib_uobject
*uobject
; /* user */
1771 struct list_head qp_entry
; /* FR */
1776 struct ib_device
*device
;
1778 struct ib_uobject
*uobject
;
1780 enum ib_mw_type type
;
1784 struct ib_device
*device
;
1786 struct list_head list
;
1791 /* Supported steering options */
1792 enum ib_flow_attr_type
{
1793 /* steering according to rule specifications */
1794 IB_FLOW_ATTR_NORMAL
= 0x0,
1795 /* default unicast and multicast rule -
1796 * receive all Eth traffic which isn't steered to any QP
1798 IB_FLOW_ATTR_ALL_DEFAULT
= 0x1,
1799 /* default multicast rule -
1800 * receive all Eth multicast traffic which isn't steered to any QP
1802 IB_FLOW_ATTR_MC_DEFAULT
= 0x2,
1803 /* sniffer rule - receive all port traffic */
1804 IB_FLOW_ATTR_SNIFFER
= 0x3
1807 /* Supported steering header types */
1808 enum ib_flow_spec_type
{
1810 IB_FLOW_SPEC_ETH
= 0x20,
1811 IB_FLOW_SPEC_IB
= 0x22,
1813 IB_FLOW_SPEC_IPV4
= 0x30,
1814 IB_FLOW_SPEC_IPV6
= 0x31,
1816 IB_FLOW_SPEC_TCP
= 0x40,
1817 IB_FLOW_SPEC_UDP
= 0x41,
1818 IB_FLOW_SPEC_VXLAN_TUNNEL
= 0x50,
1819 IB_FLOW_SPEC_INNER
= 0x100,
1821 IB_FLOW_SPEC_ACTION_TAG
= 0x1000,
1822 IB_FLOW_SPEC_ACTION_DROP
= 0x1001,
1824 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1825 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1827 /* Flow steering rule priority is set according to it's domain.
1828 * Lower domain value means higher priority.
1830 enum ib_flow_domain
{
1831 IB_FLOW_DOMAIN_USER
,
1832 IB_FLOW_DOMAIN_ETHTOOL
,
1835 IB_FLOW_DOMAIN_NUM
/* Must be last */
1838 enum ib_flow_flags
{
1839 IB_FLOW_ATTR_FLAGS_DONT_TRAP
= 1UL << 1, /* Continue match, no steal */
1840 IB_FLOW_ATTR_FLAGS_RESERVED
= 1UL << 2 /* Must be last */
1843 struct ib_flow_eth_filter
{
1852 struct ib_flow_spec_eth
{
1855 struct ib_flow_eth_filter val
;
1856 struct ib_flow_eth_filter mask
;
1859 struct ib_flow_ib_filter
{
1866 struct ib_flow_spec_ib
{
1869 struct ib_flow_ib_filter val
;
1870 struct ib_flow_ib_filter mask
;
1873 /* IPv4 header flags */
1874 enum ib_ipv4_flags
{
1875 IB_IPV4_DONT_FRAG
= 0x2, /* Don't enable packet fragmentation */
1876 IB_IPV4_MORE_FRAG
= 0X4 /* For All fragmented packets except the
1877 last have this flag set */
1880 struct ib_flow_ipv4_filter
{
1891 struct ib_flow_spec_ipv4
{
1894 struct ib_flow_ipv4_filter val
;
1895 struct ib_flow_ipv4_filter mask
;
1898 struct ib_flow_ipv6_filter
{
1909 struct ib_flow_spec_ipv6
{
1912 struct ib_flow_ipv6_filter val
;
1913 struct ib_flow_ipv6_filter mask
;
1916 struct ib_flow_tcp_udp_filter
{
1923 struct ib_flow_spec_tcp_udp
{
1926 struct ib_flow_tcp_udp_filter val
;
1927 struct ib_flow_tcp_udp_filter mask
;
1930 struct ib_flow_tunnel_filter
{
1935 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1936 * the tunnel_id from val has the vni value
1938 struct ib_flow_spec_tunnel
{
1941 struct ib_flow_tunnel_filter val
;
1942 struct ib_flow_tunnel_filter mask
;
1945 struct ib_flow_spec_action_tag
{
1946 enum ib_flow_spec_type type
;
1951 struct ib_flow_spec_action_drop
{
1952 enum ib_flow_spec_type type
;
1956 union ib_flow_spec
{
1961 struct ib_flow_spec_eth eth
;
1962 struct ib_flow_spec_ib ib
;
1963 struct ib_flow_spec_ipv4 ipv4
;
1964 struct ib_flow_spec_tcp_udp tcp_udp
;
1965 struct ib_flow_spec_ipv6 ipv6
;
1966 struct ib_flow_spec_tunnel tunnel
;
1967 struct ib_flow_spec_action_tag flow_tag
;
1968 struct ib_flow_spec_action_drop drop
;
1971 struct ib_flow_attr
{
1972 enum ib_flow_attr_type type
;
1978 /* Following are the optional layers according to user request
1979 * struct ib_flow_spec_xxx
1980 * struct ib_flow_spec_yyy
1986 struct ib_uobject
*uobject
;
1992 enum ib_process_mad_flags
{
1993 IB_MAD_IGNORE_MKEY
= 1,
1994 IB_MAD_IGNORE_BKEY
= 2,
1995 IB_MAD_IGNORE_ALL
= IB_MAD_IGNORE_MKEY
| IB_MAD_IGNORE_BKEY
1998 enum ib_mad_result
{
1999 IB_MAD_RESULT_FAILURE
= 0, /* (!SUCCESS is the important flag) */
2000 IB_MAD_RESULT_SUCCESS
= 1 << 0, /* MAD was successfully processed */
2001 IB_MAD_RESULT_REPLY
= 1 << 1, /* Reply packet needs to be sent */
2002 IB_MAD_RESULT_CONSUMED
= 1 << 2 /* Packet consumed: stop processing */
2005 struct ib_port_cache
{
2007 struct ib_pkey_cache
*pkey
;
2008 struct ib_gid_table
*gid
;
2010 enum ib_port_state port_state
;
2015 struct ib_event_handler event_handler
;
2016 struct ib_port_cache
*ports
;
2021 struct ib_port_immutable
{
2028 /* rdma netdev type - specifies protocol type */
2029 enum rdma_netdev_t
{
2030 RDMA_NETDEV_OPA_VNIC
,
2035 * struct rdma_netdev - rdma netdev
2036 * For cases where netstack interfacing is required.
2038 struct rdma_netdev
{
2040 struct ib_device
*hca
;
2043 /* cleanup function must be specified */
2044 void (*free_rdma_netdev
)(struct net_device
*netdev
);
2046 /* control functions */
2047 void (*set_id
)(struct net_device
*netdev
, int id
);
2049 int (*send
)(struct net_device
*dev
, struct sk_buff
*skb
,
2050 struct ib_ah
*address
, u32 dqpn
);
2052 int (*attach_mcast
)(struct net_device
*dev
, struct ib_device
*hca
,
2053 union ib_gid
*gid
, u16 mlid
,
2054 int set_qkey
, u32 qkey
);
2055 int (*detach_mcast
)(struct net_device
*dev
, struct ib_device
*hca
,
2056 union ib_gid
*gid
, u16 mlid
);
2059 struct ib_port_pkey_list
{
2060 /* Lock to hold while modifying the list. */
2061 spinlock_t list_lock
;
2062 struct list_head pkey_list
;
2066 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2067 struct device
*dma_device
;
2069 char name
[IB_DEVICE_NAME_MAX
];
2071 struct list_head event_handler_list
;
2072 spinlock_t event_handler_lock
;
2074 spinlock_t client_data_lock
;
2075 struct list_head core_list
;
2076 /* Access to the client_data_list is protected by the client_data_lock
2077 * spinlock and the lists_rwsem read-write semaphore */
2078 struct list_head client_data_list
;
2080 struct ib_cache cache
;
2082 * port_immutable is indexed by port number
2084 struct ib_port_immutable
*port_immutable
;
2086 int num_comp_vectors
;
2088 struct ib_port_pkey_list
*port_pkey_list
;
2090 struct iw_cm_verbs
*iwcm
;
2093 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2094 * driver initialized data. The struct is kfree()'ed by the sysfs
2095 * core when the device is removed. A lifespan of -1 in the return
2096 * struct tells the core to set a default lifespan.
2098 struct rdma_hw_stats
*(*alloc_hw_stats
)(struct ib_device
*device
,
2101 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2102 * @index - The index in the value array we wish to have updated, or
2103 * num_counters if we want all stats updated
2105 * < 0 - Error, no counters updated
2106 * index - Updated the single counter pointed to by index
2107 * num_counters - Updated all counters (will reset the timestamp
2108 * and prevent further calls for lifespan milliseconds)
2109 * Drivers are allowed to update all counters in leiu of just the
2110 * one given in index at their option
2112 int (*get_hw_stats
)(struct ib_device
*device
,
2113 struct rdma_hw_stats
*stats
,
2114 u8 port
, int index
);
2115 int (*query_device
)(struct ib_device
*device
,
2116 struct ib_device_attr
*device_attr
,
2117 struct ib_udata
*udata
);
2118 int (*query_port
)(struct ib_device
*device
,
2120 struct ib_port_attr
*port_attr
);
2121 enum rdma_link_layer (*get_link_layer
)(struct ib_device
*device
,
2123 /* When calling get_netdev, the HW vendor's driver should return the
2124 * net device of device @device at port @port_num or NULL if such
2125 * a net device doesn't exist. The vendor driver should call dev_hold
2126 * on this net device. The HW vendor's device driver must guarantee
2127 * that this function returns NULL before the net device reaches
2128 * NETDEV_UNREGISTER_FINAL state.
2130 struct net_device
*(*get_netdev
)(struct ib_device
*device
,
2132 int (*query_gid
)(struct ib_device
*device
,
2133 u8 port_num
, int index
,
2135 /* When calling add_gid, the HW vendor's driver should
2136 * add the gid of device @device at gid index @index of
2137 * port @port_num to be @gid. Meta-info of that gid (for example,
2138 * the network device related to this gid is available
2139 * at @attr. @context allows the HW vendor driver to store extra
2140 * information together with a GID entry. The HW vendor may allocate
2141 * memory to contain this information and store it in @context when a
2142 * new GID entry is written to. Params are consistent until the next
2143 * call of add_gid or delete_gid. The function should return 0 on
2144 * success or error otherwise. The function could be called
2145 * concurrently for different ports. This function is only called
2146 * when roce_gid_table is used.
2148 int (*add_gid
)(struct ib_device
*device
,
2151 const union ib_gid
*gid
,
2152 const struct ib_gid_attr
*attr
,
2154 /* When calling del_gid, the HW vendor's driver should delete the
2155 * gid of device @device at gid index @index of port @port_num.
2156 * Upon the deletion of a GID entry, the HW vendor must free any
2157 * allocated memory. The caller will clear @context afterwards.
2158 * This function is only called when roce_gid_table is used.
2160 int (*del_gid
)(struct ib_device
*device
,
2164 int (*query_pkey
)(struct ib_device
*device
,
2165 u8 port_num
, u16 index
, u16
*pkey
);
2166 int (*modify_device
)(struct ib_device
*device
,
2167 int device_modify_mask
,
2168 struct ib_device_modify
*device_modify
);
2169 int (*modify_port
)(struct ib_device
*device
,
2170 u8 port_num
, int port_modify_mask
,
2171 struct ib_port_modify
*port_modify
);
2172 struct ib_ucontext
* (*alloc_ucontext
)(struct ib_device
*device
,
2173 struct ib_udata
*udata
);
2174 int (*dealloc_ucontext
)(struct ib_ucontext
*context
);
2175 int (*mmap
)(struct ib_ucontext
*context
,
2176 struct vm_area_struct
*vma
);
2177 struct ib_pd
* (*alloc_pd
)(struct ib_device
*device
,
2178 struct ib_ucontext
*context
,
2179 struct ib_udata
*udata
);
2180 int (*dealloc_pd
)(struct ib_pd
*pd
);
2181 struct ib_ah
* (*create_ah
)(struct ib_pd
*pd
,
2182 struct rdma_ah_attr
*ah_attr
,
2183 struct ib_udata
*udata
);
2184 int (*modify_ah
)(struct ib_ah
*ah
,
2185 struct rdma_ah_attr
*ah_attr
);
2186 int (*query_ah
)(struct ib_ah
*ah
,
2187 struct rdma_ah_attr
*ah_attr
);
2188 int (*destroy_ah
)(struct ib_ah
*ah
);
2189 struct ib_srq
* (*create_srq
)(struct ib_pd
*pd
,
2190 struct ib_srq_init_attr
*srq_init_attr
,
2191 struct ib_udata
*udata
);
2192 int (*modify_srq
)(struct ib_srq
*srq
,
2193 struct ib_srq_attr
*srq_attr
,
2194 enum ib_srq_attr_mask srq_attr_mask
,
2195 struct ib_udata
*udata
);
2196 int (*query_srq
)(struct ib_srq
*srq
,
2197 struct ib_srq_attr
*srq_attr
);
2198 int (*destroy_srq
)(struct ib_srq
*srq
);
2199 int (*post_srq_recv
)(struct ib_srq
*srq
,
2200 struct ib_recv_wr
*recv_wr
,
2201 struct ib_recv_wr
**bad_recv_wr
);
2202 struct ib_qp
* (*create_qp
)(struct ib_pd
*pd
,
2203 struct ib_qp_init_attr
*qp_init_attr
,
2204 struct ib_udata
*udata
);
2205 int (*modify_qp
)(struct ib_qp
*qp
,
2206 struct ib_qp_attr
*qp_attr
,
2208 struct ib_udata
*udata
);
2209 int (*query_qp
)(struct ib_qp
*qp
,
2210 struct ib_qp_attr
*qp_attr
,
2212 struct ib_qp_init_attr
*qp_init_attr
);
2213 int (*destroy_qp
)(struct ib_qp
*qp
);
2214 int (*post_send
)(struct ib_qp
*qp
,
2215 struct ib_send_wr
*send_wr
,
2216 struct ib_send_wr
**bad_send_wr
);
2217 int (*post_recv
)(struct ib_qp
*qp
,
2218 struct ib_recv_wr
*recv_wr
,
2219 struct ib_recv_wr
**bad_recv_wr
);
2220 struct ib_cq
* (*create_cq
)(struct ib_device
*device
,
2221 const struct ib_cq_init_attr
*attr
,
2222 struct ib_ucontext
*context
,
2223 struct ib_udata
*udata
);
2224 int (*modify_cq
)(struct ib_cq
*cq
, u16 cq_count
,
2226 int (*destroy_cq
)(struct ib_cq
*cq
);
2227 int (*resize_cq
)(struct ib_cq
*cq
, int cqe
,
2228 struct ib_udata
*udata
);
2229 int (*poll_cq
)(struct ib_cq
*cq
, int num_entries
,
2231 int (*peek_cq
)(struct ib_cq
*cq
, int wc_cnt
);
2232 int (*req_notify_cq
)(struct ib_cq
*cq
,
2233 enum ib_cq_notify_flags flags
);
2234 int (*req_ncomp_notif
)(struct ib_cq
*cq
,
2236 struct ib_mr
* (*get_dma_mr
)(struct ib_pd
*pd
,
2237 int mr_access_flags
);
2238 struct ib_mr
* (*reg_user_mr
)(struct ib_pd
*pd
,
2239 u64 start
, u64 length
,
2241 int mr_access_flags
,
2242 struct ib_udata
*udata
);
2243 int (*rereg_user_mr
)(struct ib_mr
*mr
,
2245 u64 start
, u64 length
,
2247 int mr_access_flags
,
2249 struct ib_udata
*udata
);
2250 int (*dereg_mr
)(struct ib_mr
*mr
);
2251 struct ib_mr
* (*alloc_mr
)(struct ib_pd
*pd
,
2252 enum ib_mr_type mr_type
,
2254 int (*map_mr_sg
)(struct ib_mr
*mr
,
2255 struct scatterlist
*sg
,
2257 unsigned int *sg_offset
);
2258 struct ib_mw
* (*alloc_mw
)(struct ib_pd
*pd
,
2259 enum ib_mw_type type
,
2260 struct ib_udata
*udata
);
2261 int (*dealloc_mw
)(struct ib_mw
*mw
);
2262 struct ib_fmr
* (*alloc_fmr
)(struct ib_pd
*pd
,
2263 int mr_access_flags
,
2264 struct ib_fmr_attr
*fmr_attr
);
2265 int (*map_phys_fmr
)(struct ib_fmr
*fmr
,
2266 u64
*page_list
, int list_len
,
2268 int (*unmap_fmr
)(struct list_head
*fmr_list
);
2269 int (*dealloc_fmr
)(struct ib_fmr
*fmr
);
2270 int (*attach_mcast
)(struct ib_qp
*qp
,
2273 int (*detach_mcast
)(struct ib_qp
*qp
,
2276 int (*process_mad
)(struct ib_device
*device
,
2277 int process_mad_flags
,
2279 const struct ib_wc
*in_wc
,
2280 const struct ib_grh
*in_grh
,
2281 const struct ib_mad_hdr
*in_mad
,
2283 struct ib_mad_hdr
*out_mad
,
2284 size_t *out_mad_size
,
2285 u16
*out_mad_pkey_index
);
2286 struct ib_xrcd
* (*alloc_xrcd
)(struct ib_device
*device
,
2287 struct ib_ucontext
*ucontext
,
2288 struct ib_udata
*udata
);
2289 int (*dealloc_xrcd
)(struct ib_xrcd
*xrcd
);
2290 struct ib_flow
* (*create_flow
)(struct ib_qp
*qp
,
2294 int (*destroy_flow
)(struct ib_flow
*flow_id
);
2295 int (*check_mr_status
)(struct ib_mr
*mr
, u32 check_mask
,
2296 struct ib_mr_status
*mr_status
);
2297 void (*disassociate_ucontext
)(struct ib_ucontext
*ibcontext
);
2298 void (*drain_rq
)(struct ib_qp
*qp
);
2299 void (*drain_sq
)(struct ib_qp
*qp
);
2300 int (*set_vf_link_state
)(struct ib_device
*device
, int vf
, u8 port
,
2302 int (*get_vf_config
)(struct ib_device
*device
, int vf
, u8 port
,
2303 struct ifla_vf_info
*ivf
);
2304 int (*get_vf_stats
)(struct ib_device
*device
, int vf
, u8 port
,
2305 struct ifla_vf_stats
*stats
);
2306 int (*set_vf_guid
)(struct ib_device
*device
, int vf
, u8 port
, u64 guid
,
2308 struct ib_wq
* (*create_wq
)(struct ib_pd
*pd
,
2309 struct ib_wq_init_attr
*init_attr
,
2310 struct ib_udata
*udata
);
2311 int (*destroy_wq
)(struct ib_wq
*wq
);
2312 int (*modify_wq
)(struct ib_wq
*wq
,
2313 struct ib_wq_attr
*attr
,
2315 struct ib_udata
*udata
);
2316 struct ib_rwq_ind_table
* (*create_rwq_ind_table
)(struct ib_device
*device
,
2317 struct ib_rwq_ind_table_init_attr
*init_attr
,
2318 struct ib_udata
*udata
);
2319 int (*destroy_rwq_ind_table
)(struct ib_rwq_ind_table
*wq_ind_table
);
2321 * rdma netdev operation
2323 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2324 * doesn't support the specified rdma netdev type.
2326 struct net_device
*(*alloc_rdma_netdev
)(
2327 struct ib_device
*device
,
2329 enum rdma_netdev_t type
,
2331 unsigned char name_assign_type
,
2332 void (*setup
)(struct net_device
*));
2334 struct module
*owner
;
2336 struct kobject
*ports_parent
;
2337 struct list_head port_list
;
2340 IB_DEV_UNINITIALIZED
,
2346 u64 uverbs_cmd_mask
;
2347 u64 uverbs_ex_cmd_mask
;
2349 char node_desc
[IB_DEVICE_NODE_DESC_MAX
];
2355 struct ib_device_attr attrs
;
2356 struct attribute_group
*hw_stats_ag
;
2357 struct rdma_hw_stats
*hw_stats
;
2359 #ifdef CONFIG_CGROUP_RDMA
2360 struct rdmacg_device cg_device
;
2366 * The following mandatory functions are used only at device
2367 * registration. Keep functions such as these at the end of this
2368 * structure to avoid cache line misses when accessing struct ib_device
2371 int (*get_port_immutable
)(struct ib_device
*, u8
, struct ib_port_immutable
*);
2372 void (*get_dev_fw_str
)(struct ib_device
*, char *str
);
2373 const struct cpumask
*(*get_vector_affinity
)(struct ib_device
*ibdev
,
2376 struct uverbs_root_spec
*specs_root
;
2381 void (*add
) (struct ib_device
*);
2382 void (*remove
)(struct ib_device
*, void *client_data
);
2384 /* Returns the net_dev belonging to this ib_client and matching the
2386 * @dev: An RDMA device that the net_dev use for communication.
2387 * @port: A physical port number on the RDMA device.
2388 * @pkey: P_Key that the net_dev uses if applicable.
2389 * @gid: A GID that the net_dev uses to communicate.
2390 * @addr: An IP address the net_dev is configured with.
2391 * @client_data: The device's client data set by ib_set_client_data().
2393 * An ib_client that implements a net_dev on top of RDMA devices
2394 * (such as IP over IB) should implement this callback, allowing the
2395 * rdma_cm module to find the right net_dev for a given request.
2397 * The caller is responsible for calling dev_put on the returned
2399 struct net_device
*(*get_net_dev_by_params
)(
2400 struct ib_device
*dev
,
2403 const union ib_gid
*gid
,
2404 const struct sockaddr
*addr
,
2406 struct list_head list
;
2409 struct ib_device
*ib_alloc_device(size_t size
);
2410 void ib_dealloc_device(struct ib_device
*device
);
2412 void ib_get_device_fw_str(struct ib_device
*device
, char *str
);
2414 int ib_register_device(struct ib_device
*device
,
2415 int (*port_callback
)(struct ib_device
*,
2416 u8
, struct kobject
*));
2417 void ib_unregister_device(struct ib_device
*device
);
2419 int ib_register_client (struct ib_client
*client
);
2420 void ib_unregister_client(struct ib_client
*client
);
2422 void *ib_get_client_data(struct ib_device
*device
, struct ib_client
*client
);
2423 void ib_set_client_data(struct ib_device
*device
, struct ib_client
*client
,
2426 static inline int ib_copy_from_udata(void *dest
, struct ib_udata
*udata
, size_t len
)
2428 return copy_from_user(dest
, udata
->inbuf
, len
) ? -EFAULT
: 0;
2431 static inline int ib_copy_to_udata(struct ib_udata
*udata
, void *src
, size_t len
)
2433 return copy_to_user(udata
->outbuf
, src
, len
) ? -EFAULT
: 0;
2436 static inline bool ib_is_udata_cleared(struct ib_udata
*udata
,
2440 const void __user
*p
= udata
->inbuf
+ offset
;
2444 if (len
> USHRT_MAX
)
2447 buf
= memdup_user(p
, len
);
2451 ret
= !memchr_inv(buf
, 0, len
);
2457 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2458 * contains all required attributes and no attributes not allowed for
2459 * the given QP state transition.
2460 * @cur_state: Current QP state
2461 * @next_state: Next QP state
2463 * @mask: Mask of supplied QP attributes
2464 * @ll : link layer of port
2466 * This function is a helper function that a low-level driver's
2467 * modify_qp method can use to validate the consumer's input. It
2468 * checks that cur_state and next_state are valid QP states, that a
2469 * transition from cur_state to next_state is allowed by the IB spec,
2470 * and that the attribute mask supplied is allowed for the transition.
2472 int ib_modify_qp_is_ok(enum ib_qp_state cur_state
, enum ib_qp_state next_state
,
2473 enum ib_qp_type type
, enum ib_qp_attr_mask mask
,
2474 enum rdma_link_layer ll
);
2476 void ib_register_event_handler(struct ib_event_handler
*event_handler
);
2477 void ib_unregister_event_handler(struct ib_event_handler
*event_handler
);
2478 void ib_dispatch_event(struct ib_event
*event
);
2480 int ib_query_port(struct ib_device
*device
,
2481 u8 port_num
, struct ib_port_attr
*port_attr
);
2483 enum rdma_link_layer
rdma_port_get_link_layer(struct ib_device
*device
,
2487 * rdma_cap_ib_switch - Check if the device is IB switch
2488 * @device: Device to check
2490 * Device driver is responsible for setting is_switch bit on
2491 * in ib_device structure at init time.
2493 * Return: true if the device is IB switch.
2495 static inline bool rdma_cap_ib_switch(const struct ib_device
*device
)
2497 return device
->is_switch
;
2501 * rdma_start_port - Return the first valid port number for the device
2504 * @device: Device to be checked
2506 * Return start port number
2508 static inline u8
rdma_start_port(const struct ib_device
*device
)
2510 return rdma_cap_ib_switch(device
) ? 0 : 1;
2514 * rdma_end_port - Return the last valid port number for the device
2517 * @device: Device to be checked
2519 * Return last port number
2521 static inline u8
rdma_end_port(const struct ib_device
*device
)
2523 return rdma_cap_ib_switch(device
) ? 0 : device
->phys_port_cnt
;
2526 static inline int rdma_is_port_valid(const struct ib_device
*device
,
2529 return (port
>= rdma_start_port(device
) &&
2530 port
<= rdma_end_port(device
));
2533 static inline bool rdma_protocol_ib(const struct ib_device
*device
, u8 port_num
)
2535 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IB
;
2538 static inline bool rdma_protocol_roce(const struct ib_device
*device
, u8 port_num
)
2540 return device
->port_immutable
[port_num
].core_cap_flags
&
2541 (RDMA_CORE_CAP_PROT_ROCE
| RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP
);
2544 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device
*device
, u8 port_num
)
2546 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP
;
2549 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device
*device
, u8 port_num
)
2551 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_ROCE
;
2554 static inline bool rdma_protocol_iwarp(const struct ib_device
*device
, u8 port_num
)
2556 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_IWARP
;
2559 static inline bool rdma_ib_or_roce(const struct ib_device
*device
, u8 port_num
)
2561 return rdma_protocol_ib(device
, port_num
) ||
2562 rdma_protocol_roce(device
, port_num
);
2565 static inline bool rdma_protocol_raw_packet(const struct ib_device
*device
, u8 port_num
)
2567 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_RAW_PACKET
;
2570 static inline bool rdma_protocol_usnic(const struct ib_device
*device
, u8 port_num
)
2572 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_PROT_USNIC
;
2576 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2577 * Management Datagrams.
2578 * @device: Device to check
2579 * @port_num: Port number to check
2581 * Management Datagrams (MAD) are a required part of the InfiniBand
2582 * specification and are supported on all InfiniBand devices. A slightly
2583 * extended version are also supported on OPA interfaces.
2585 * Return: true if the port supports sending/receiving of MAD packets.
2587 static inline bool rdma_cap_ib_mad(const struct ib_device
*device
, u8 port_num
)
2589 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_MAD
;
2593 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2594 * Management Datagrams.
2595 * @device: Device to check
2596 * @port_num: Port number to check
2598 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2599 * datagrams with their own versions. These OPA MADs share many but not all of
2600 * the characteristics of InfiniBand MADs.
2602 * OPA MADs differ in the following ways:
2604 * 1) MADs are variable size up to 2K
2605 * IBTA defined MADs remain fixed at 256 bytes
2606 * 2) OPA SMPs must carry valid PKeys
2607 * 3) OPA SMP packets are a different format
2609 * Return: true if the port supports OPA MAD packet formats.
2611 static inline bool rdma_cap_opa_mad(struct ib_device
*device
, u8 port_num
)
2613 return (device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_OPA_MAD
)
2614 == RDMA_CORE_CAP_OPA_MAD
;
2618 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2619 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2620 * @device: Device to check
2621 * @port_num: Port number to check
2623 * Each InfiniBand node is required to provide a Subnet Management Agent
2624 * that the subnet manager can access. Prior to the fabric being fully
2625 * configured by the subnet manager, the SMA is accessed via a well known
2626 * interface called the Subnet Management Interface (SMI). This interface
2627 * uses directed route packets to communicate with the SM to get around the
2628 * chicken and egg problem of the SM needing to know what's on the fabric
2629 * in order to configure the fabric, and needing to configure the fabric in
2630 * order to send packets to the devices on the fabric. These directed
2631 * route packets do not need the fabric fully configured in order to reach
2632 * their destination. The SMI is the only method allowed to send
2633 * directed route packets on an InfiniBand fabric.
2635 * Return: true if the port provides an SMI.
2637 static inline bool rdma_cap_ib_smi(const struct ib_device
*device
, u8 port_num
)
2639 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SMI
;
2643 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2644 * Communication Manager.
2645 * @device: Device to check
2646 * @port_num: Port number to check
2648 * The InfiniBand Communication Manager is one of many pre-defined General
2649 * Service Agents (GSA) that are accessed via the General Service
2650 * Interface (GSI). It's role is to facilitate establishment of connections
2651 * between nodes as well as other management related tasks for established
2654 * Return: true if the port supports an IB CM (this does not guarantee that
2655 * a CM is actually running however).
2657 static inline bool rdma_cap_ib_cm(const struct ib_device
*device
, u8 port_num
)
2659 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_CM
;
2663 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2664 * Communication Manager.
2665 * @device: Device to check
2666 * @port_num: Port number to check
2668 * Similar to above, but specific to iWARP connections which have a different
2669 * managment protocol than InfiniBand.
2671 * Return: true if the port supports an iWARP CM (this does not guarantee that
2672 * a CM is actually running however).
2674 static inline bool rdma_cap_iw_cm(const struct ib_device
*device
, u8 port_num
)
2676 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IW_CM
;
2680 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2681 * Subnet Administration.
2682 * @device: Device to check
2683 * @port_num: Port number to check
2685 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2686 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2687 * fabrics, devices should resolve routes to other hosts by contacting the
2688 * SA to query the proper route.
2690 * Return: true if the port should act as a client to the fabric Subnet
2691 * Administration interface. This does not imply that the SA service is
2694 static inline bool rdma_cap_ib_sa(const struct ib_device
*device
, u8 port_num
)
2696 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_IB_SA
;
2700 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2702 * @device: Device to check
2703 * @port_num: Port number to check
2705 * InfiniBand multicast registration is more complex than normal IPv4 or
2706 * IPv6 multicast registration. Each Host Channel Adapter must register
2707 * with the Subnet Manager when it wishes to join a multicast group. It
2708 * should do so only once regardless of how many queue pairs it subscribes
2709 * to this group. And it should leave the group only after all queue pairs
2710 * attached to the group have been detached.
2712 * Return: true if the port must undertake the additional adminstrative
2713 * overhead of registering/unregistering with the SM and tracking of the
2714 * total number of queue pairs attached to the multicast group.
2716 static inline bool rdma_cap_ib_mcast(const struct ib_device
*device
, u8 port_num
)
2718 return rdma_cap_ib_sa(device
, port_num
);
2722 * rdma_cap_af_ib - Check if the port of device has the capability
2723 * Native Infiniband Address.
2724 * @device: Device to check
2725 * @port_num: Port number to check
2727 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2728 * GID. RoCE uses a different mechanism, but still generates a GID via
2729 * a prescribed mechanism and port specific data.
2731 * Return: true if the port uses a GID address to identify devices on the
2734 static inline bool rdma_cap_af_ib(const struct ib_device
*device
, u8 port_num
)
2736 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_AF_IB
;
2740 * rdma_cap_eth_ah - Check if the port of device has the capability
2741 * Ethernet Address Handle.
2742 * @device: Device to check
2743 * @port_num: Port number to check
2745 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2746 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2747 * port. Normally, packet headers are generated by the sending host
2748 * adapter, but when sending connectionless datagrams, we must manually
2749 * inject the proper headers for the fabric we are communicating over.
2751 * Return: true if we are running as a RoCE port and must force the
2752 * addition of a Global Route Header built from our Ethernet Address
2753 * Handle into our header list for connectionless packets.
2755 static inline bool rdma_cap_eth_ah(const struct ib_device
*device
, u8 port_num
)
2757 return device
->port_immutable
[port_num
].core_cap_flags
& RDMA_CORE_CAP_ETH_AH
;
2761 * rdma_cap_opa_ah - Check if the port of device supports
2762 * OPA Address handles
2763 * @device: Device to check
2764 * @port_num: Port number to check
2766 * Return: true if we are running on an OPA device which supports
2767 * the extended OPA addressing.
2769 static inline bool rdma_cap_opa_ah(struct ib_device
*device
, u8 port_num
)
2771 return (device
->port_immutable
[port_num
].core_cap_flags
&
2772 RDMA_CORE_CAP_OPA_AH
) == RDMA_CORE_CAP_OPA_AH
;
2776 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2779 * @port_num: Port number
2781 * This MAD size includes the MAD headers and MAD payload. No other headers
2784 * Return the max MAD size required by the Port. Will return 0 if the port
2785 * does not support MADs
2787 static inline size_t rdma_max_mad_size(const struct ib_device
*device
, u8 port_num
)
2789 return device
->port_immutable
[port_num
].max_mad_size
;
2793 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2794 * @device: Device to check
2795 * @port_num: Port number to check
2797 * RoCE GID table mechanism manages the various GIDs for a device.
2799 * NOTE: if allocating the port's GID table has failed, this call will still
2800 * return true, but any RoCE GID table API will fail.
2802 * Return: true if the port uses RoCE GID table mechanism in order to manage
2805 static inline bool rdma_cap_roce_gid_table(const struct ib_device
*device
,
2808 return rdma_protocol_roce(device
, port_num
) &&
2809 device
->add_gid
&& device
->del_gid
;
2813 * Check if the device supports READ W/ INVALIDATE.
2815 static inline bool rdma_cap_read_inv(struct ib_device
*dev
, u32 port_num
)
2818 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2819 * has support for it yet.
2821 return rdma_protocol_iwarp(dev
, port_num
);
2824 int ib_query_gid(struct ib_device
*device
,
2825 u8 port_num
, int index
, union ib_gid
*gid
,
2826 struct ib_gid_attr
*attr
);
2828 int ib_set_vf_link_state(struct ib_device
*device
, int vf
, u8 port
,
2830 int ib_get_vf_config(struct ib_device
*device
, int vf
, u8 port
,
2831 struct ifla_vf_info
*info
);
2832 int ib_get_vf_stats(struct ib_device
*device
, int vf
, u8 port
,
2833 struct ifla_vf_stats
*stats
);
2834 int ib_set_vf_guid(struct ib_device
*device
, int vf
, u8 port
, u64 guid
,
2837 int ib_query_pkey(struct ib_device
*device
,
2838 u8 port_num
, u16 index
, u16
*pkey
);
2840 int ib_modify_device(struct ib_device
*device
,
2841 int device_modify_mask
,
2842 struct ib_device_modify
*device_modify
);
2844 int ib_modify_port(struct ib_device
*device
,
2845 u8 port_num
, int port_modify_mask
,
2846 struct ib_port_modify
*port_modify
);
2848 int ib_find_gid(struct ib_device
*device
, union ib_gid
*gid
,
2849 enum ib_gid_type gid_type
, struct net_device
*ndev
,
2850 u8
*port_num
, u16
*index
);
2852 int ib_find_pkey(struct ib_device
*device
,
2853 u8 port_num
, u16 pkey
, u16
*index
);
2857 * Create a memory registration for all memory in the system and place
2858 * the rkey for it into pd->unsafe_global_rkey. This can be used by
2859 * ULPs to avoid the overhead of dynamic MRs.
2861 * This flag is generally considered unsafe and must only be used in
2862 * extremly trusted environments. Every use of it will log a warning
2863 * in the kernel log.
2865 IB_PD_UNSAFE_GLOBAL_RKEY
= 0x01,
2868 struct ib_pd
*__ib_alloc_pd(struct ib_device
*device
, unsigned int flags
,
2869 const char *caller
);
2870 #define ib_alloc_pd(device, flags) \
2871 __ib_alloc_pd((device), (flags), __func__)
2872 void ib_dealloc_pd(struct ib_pd
*pd
);
2875 * rdma_create_ah - Creates an address handle for the given address vector.
2876 * @pd: The protection domain associated with the address handle.
2877 * @ah_attr: The attributes of the address vector.
2879 * The address handle is used to reference a local or global destination
2880 * in all UD QP post sends.
2882 struct ib_ah
*rdma_create_ah(struct ib_pd
*pd
, struct rdma_ah_attr
*ah_attr
);
2885 * rdma_create_user_ah - Creates an address handle for the given address vector.
2886 * It resolves destination mac address for ah attribute of RoCE type.
2887 * @pd: The protection domain associated with the address handle.
2888 * @ah_attr: The attributes of the address vector.
2889 * @udata: pointer to user's input output buffer information need by
2892 * It returns 0 on success and returns appropriate error code on error.
2893 * The address handle is used to reference a local or global destination
2894 * in all UD QP post sends.
2896 struct ib_ah
*rdma_create_user_ah(struct ib_pd
*pd
,
2897 struct rdma_ah_attr
*ah_attr
,
2898 struct ib_udata
*udata
);
2900 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2902 * @hdr: the L3 header to parse
2903 * @net_type: type of header to parse
2904 * @sgid: place to store source gid
2905 * @dgid: place to store destination gid
2907 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr
*hdr
,
2908 enum rdma_network_type net_type
,
2909 union ib_gid
*sgid
, union ib_gid
*dgid
);
2912 * ib_get_rdma_header_version - Get the header version
2913 * @hdr: the L3 header to parse
2915 int ib_get_rdma_header_version(const union rdma_network_hdr
*hdr
);
2918 * ib_init_ah_from_wc - Initializes address handle attributes from a
2920 * @device: Device on which the received message arrived.
2921 * @port_num: Port on which the received message arrived.
2922 * @wc: Work completion associated with the received message.
2923 * @grh: References the received global route header. This parameter is
2924 * ignored unless the work completion indicates that the GRH is valid.
2925 * @ah_attr: Returned attributes that can be used when creating an address
2926 * handle for replying to the message.
2928 int ib_init_ah_from_wc(struct ib_device
*device
, u8 port_num
,
2929 const struct ib_wc
*wc
, const struct ib_grh
*grh
,
2930 struct rdma_ah_attr
*ah_attr
);
2933 * ib_create_ah_from_wc - Creates an address handle associated with the
2934 * sender of the specified work completion.
2935 * @pd: The protection domain associated with the address handle.
2936 * @wc: Work completion information associated with a received message.
2937 * @grh: References the received global route header. This parameter is
2938 * ignored unless the work completion indicates that the GRH is valid.
2939 * @port_num: The outbound port number to associate with the address.
2941 * The address handle is used to reference a local or global destination
2942 * in all UD QP post sends.
2944 struct ib_ah
*ib_create_ah_from_wc(struct ib_pd
*pd
, const struct ib_wc
*wc
,
2945 const struct ib_grh
*grh
, u8 port_num
);
2948 * rdma_modify_ah - Modifies the address vector associated with an address
2950 * @ah: The address handle to modify.
2951 * @ah_attr: The new address vector attributes to associate with the
2954 int rdma_modify_ah(struct ib_ah
*ah
, struct rdma_ah_attr
*ah_attr
);
2957 * rdma_query_ah - Queries the address vector associated with an address
2959 * @ah: The address handle to query.
2960 * @ah_attr: The address vector attributes associated with the address
2963 int rdma_query_ah(struct ib_ah
*ah
, struct rdma_ah_attr
*ah_attr
);
2966 * rdma_destroy_ah - Destroys an address handle.
2967 * @ah: The address handle to destroy.
2969 int rdma_destroy_ah(struct ib_ah
*ah
);
2972 * ib_create_srq - Creates a SRQ associated with the specified protection
2974 * @pd: The protection domain associated with the SRQ.
2975 * @srq_init_attr: A list of initial attributes required to create the
2976 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2977 * the actual capabilities of the created SRQ.
2979 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2980 * requested size of the SRQ, and set to the actual values allocated
2981 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2982 * will always be at least as large as the requested values.
2984 struct ib_srq
*ib_create_srq(struct ib_pd
*pd
,
2985 struct ib_srq_init_attr
*srq_init_attr
);
2988 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2989 * @srq: The SRQ to modify.
2990 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2991 * the current values of selected SRQ attributes are returned.
2992 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2993 * are being modified.
2995 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2996 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2997 * the number of receives queued drops below the limit.
2999 int ib_modify_srq(struct ib_srq
*srq
,
3000 struct ib_srq_attr
*srq_attr
,
3001 enum ib_srq_attr_mask srq_attr_mask
);
3004 * ib_query_srq - Returns the attribute list and current values for the
3006 * @srq: The SRQ to query.
3007 * @srq_attr: The attributes of the specified SRQ.
3009 int ib_query_srq(struct ib_srq
*srq
,
3010 struct ib_srq_attr
*srq_attr
);
3013 * ib_destroy_srq - Destroys the specified SRQ.
3014 * @srq: The SRQ to destroy.
3016 int ib_destroy_srq(struct ib_srq
*srq
);
3019 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3020 * @srq: The SRQ to post the work request on.
3021 * @recv_wr: A list of work requests to post on the receive queue.
3022 * @bad_recv_wr: On an immediate failure, this parameter will reference
3023 * the work request that failed to be posted on the QP.
3025 static inline int ib_post_srq_recv(struct ib_srq
*srq
,
3026 struct ib_recv_wr
*recv_wr
,
3027 struct ib_recv_wr
**bad_recv_wr
)
3029 return srq
->device
->post_srq_recv(srq
, recv_wr
, bad_recv_wr
);
3033 * ib_create_qp - Creates a QP associated with the specified protection
3035 * @pd: The protection domain associated with the QP.
3036 * @qp_init_attr: A list of initial attributes required to create the
3037 * QP. If QP creation succeeds, then the attributes are updated to
3038 * the actual capabilities of the created QP.
3040 struct ib_qp
*ib_create_qp(struct ib_pd
*pd
,
3041 struct ib_qp_init_attr
*qp_init_attr
);
3044 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3045 * @qp: The QP to modify.
3046 * @attr: On input, specifies the QP attributes to modify. On output,
3047 * the current values of selected QP attributes are returned.
3048 * @attr_mask: A bit-mask used to specify which attributes of the QP
3049 * are being modified.
3050 * @udata: pointer to user's input output buffer information
3051 * are being modified.
3052 * It returns 0 on success and returns appropriate error code on error.
3054 int ib_modify_qp_with_udata(struct ib_qp
*qp
,
3055 struct ib_qp_attr
*attr
,
3057 struct ib_udata
*udata
);
3060 * ib_modify_qp - Modifies the attributes for the specified QP and then
3061 * transitions the QP to the given state.
3062 * @qp: The QP to modify.
3063 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3064 * the current values of selected QP attributes are returned.
3065 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3066 * are being modified.
3068 int ib_modify_qp(struct ib_qp
*qp
,
3069 struct ib_qp_attr
*qp_attr
,
3073 * ib_query_qp - Returns the attribute list and current values for the
3075 * @qp: The QP to query.
3076 * @qp_attr: The attributes of the specified QP.
3077 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3078 * @qp_init_attr: Additional attributes of the selected QP.
3080 * The qp_attr_mask may be used to limit the query to gathering only the
3081 * selected attributes.
3083 int ib_query_qp(struct ib_qp
*qp
,
3084 struct ib_qp_attr
*qp_attr
,
3086 struct ib_qp_init_attr
*qp_init_attr
);
3089 * ib_destroy_qp - Destroys the specified QP.
3090 * @qp: The QP to destroy.
3092 int ib_destroy_qp(struct ib_qp
*qp
);
3095 * ib_open_qp - Obtain a reference to an existing sharable QP.
3096 * @xrcd - XRC domain
3097 * @qp_open_attr: Attributes identifying the QP to open.
3099 * Returns a reference to a sharable QP.
3101 struct ib_qp
*ib_open_qp(struct ib_xrcd
*xrcd
,
3102 struct ib_qp_open_attr
*qp_open_attr
);
3105 * ib_close_qp - Release an external reference to a QP.
3106 * @qp: The QP handle to release
3108 * The opened QP handle is released by the caller. The underlying
3109 * shared QP is not destroyed until all internal references are released.
3111 int ib_close_qp(struct ib_qp
*qp
);
3114 * ib_post_send - Posts a list of work requests to the send queue of
3116 * @qp: The QP to post the work request on.
3117 * @send_wr: A list of work requests to post on the send queue.
3118 * @bad_send_wr: On an immediate failure, this parameter will reference
3119 * the work request that failed to be posted on the QP.
3121 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3122 * error is returned, the QP state shall not be affected,
3123 * ib_post_send() will return an immediate error after queueing any
3124 * earlier work requests in the list.
3126 static inline int ib_post_send(struct ib_qp
*qp
,
3127 struct ib_send_wr
*send_wr
,
3128 struct ib_send_wr
**bad_send_wr
)
3130 return qp
->device
->post_send(qp
, send_wr
, bad_send_wr
);
3134 * ib_post_recv - Posts a list of work requests to the receive queue of
3136 * @qp: The QP to post the work request on.
3137 * @recv_wr: A list of work requests to post on the receive queue.
3138 * @bad_recv_wr: On an immediate failure, this parameter will reference
3139 * the work request that failed to be posted on the QP.
3141 static inline int ib_post_recv(struct ib_qp
*qp
,
3142 struct ib_recv_wr
*recv_wr
,
3143 struct ib_recv_wr
**bad_recv_wr
)
3145 return qp
->device
->post_recv(qp
, recv_wr
, bad_recv_wr
);
3148 struct ib_cq
*ib_alloc_cq(struct ib_device
*dev
, void *private,
3149 int nr_cqe
, int comp_vector
, enum ib_poll_context poll_ctx
);
3150 void ib_free_cq(struct ib_cq
*cq
);
3151 int ib_process_cq_direct(struct ib_cq
*cq
, int budget
);
3154 * ib_create_cq - Creates a CQ on the specified device.
3155 * @device: The device on which to create the CQ.
3156 * @comp_handler: A user-specified callback that is invoked when a
3157 * completion event occurs on the CQ.
3158 * @event_handler: A user-specified callback that is invoked when an
3159 * asynchronous event not associated with a completion occurs on the CQ.
3160 * @cq_context: Context associated with the CQ returned to the user via
3161 * the associated completion and event handlers.
3162 * @cq_attr: The attributes the CQ should be created upon.
3164 * Users can examine the cq structure to determine the actual CQ size.
3166 struct ib_cq
*ib_create_cq(struct ib_device
*device
,
3167 ib_comp_handler comp_handler
,
3168 void (*event_handler
)(struct ib_event
*, void *),
3170 const struct ib_cq_init_attr
*cq_attr
);
3173 * ib_resize_cq - Modifies the capacity of the CQ.
3174 * @cq: The CQ to resize.
3175 * @cqe: The minimum size of the CQ.
3177 * Users can examine the cq structure to determine the actual CQ size.
3179 int ib_resize_cq(struct ib_cq
*cq
, int cqe
);
3182 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3183 * @cq: The CQ to modify.
3184 * @cq_count: number of CQEs that will trigger an event
3185 * @cq_period: max period of time in usec before triggering an event
3188 int rdma_set_cq_moderation(struct ib_cq
*cq
, u16 cq_count
, u16 cq_period
);
3191 * ib_destroy_cq - Destroys the specified CQ.
3192 * @cq: The CQ to destroy.
3194 int ib_destroy_cq(struct ib_cq
*cq
);
3197 * ib_poll_cq - poll a CQ for completion(s)
3198 * @cq:the CQ being polled
3199 * @num_entries:maximum number of completions to return
3200 * @wc:array of at least @num_entries &struct ib_wc where completions
3203 * Poll a CQ for (possibly multiple) completions. If the return value
3204 * is < 0, an error occurred. If the return value is >= 0, it is the
3205 * number of completions returned. If the return value is
3206 * non-negative and < num_entries, then the CQ was emptied.
3208 static inline int ib_poll_cq(struct ib_cq
*cq
, int num_entries
,
3211 return cq
->device
->poll_cq(cq
, num_entries
, wc
);
3215 * ib_peek_cq - Returns the number of unreaped completions currently
3216 * on the specified CQ.
3217 * @cq: The CQ to peek.
3218 * @wc_cnt: A minimum number of unreaped completions to check for.
3220 * If the number of unreaped completions is greater than or equal to wc_cnt,
3221 * this function returns wc_cnt, otherwise, it returns the actual number of
3222 * unreaped completions.
3224 int ib_peek_cq(struct ib_cq
*cq
, int wc_cnt
);
3227 * ib_req_notify_cq - Request completion notification on a CQ.
3228 * @cq: The CQ to generate an event for.
3230 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3231 * to request an event on the next solicited event or next work
3232 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3233 * may also be |ed in to request a hint about missed events, as
3237 * < 0 means an error occurred while requesting notification
3238 * == 0 means notification was requested successfully, and if
3239 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3240 * were missed and it is safe to wait for another event. In
3241 * this case is it guaranteed that any work completions added
3242 * to the CQ since the last CQ poll will trigger a completion
3243 * notification event.
3244 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3245 * in. It means that the consumer must poll the CQ again to
3246 * make sure it is empty to avoid missing an event because of a
3247 * race between requesting notification and an entry being
3248 * added to the CQ. This return value means it is possible
3249 * (but not guaranteed) that a work completion has been added
3250 * to the CQ since the last poll without triggering a
3251 * completion notification event.
3253 static inline int ib_req_notify_cq(struct ib_cq
*cq
,
3254 enum ib_cq_notify_flags flags
)
3256 return cq
->device
->req_notify_cq(cq
, flags
);
3260 * ib_req_ncomp_notif - Request completion notification when there are
3261 * at least the specified number of unreaped completions on the CQ.
3262 * @cq: The CQ to generate an event for.
3263 * @wc_cnt: The number of unreaped completions that should be on the
3264 * CQ before an event is generated.
3266 static inline int ib_req_ncomp_notif(struct ib_cq
*cq
, int wc_cnt
)
3268 return cq
->device
->req_ncomp_notif
?
3269 cq
->device
->req_ncomp_notif(cq
, wc_cnt
) :
3274 * ib_dma_mapping_error - check a DMA addr for error
3275 * @dev: The device for which the dma_addr was created
3276 * @dma_addr: The DMA address to check
3278 static inline int ib_dma_mapping_error(struct ib_device
*dev
, u64 dma_addr
)
3280 return dma_mapping_error(dev
->dma_device
, dma_addr
);
3284 * ib_dma_map_single - Map a kernel virtual address to DMA address
3285 * @dev: The device for which the dma_addr is to be created
3286 * @cpu_addr: The kernel virtual address
3287 * @size: The size of the region in bytes
3288 * @direction: The direction of the DMA
3290 static inline u64
ib_dma_map_single(struct ib_device
*dev
,
3291 void *cpu_addr
, size_t size
,
3292 enum dma_data_direction direction
)
3294 return dma_map_single(dev
->dma_device
, cpu_addr
, size
, direction
);
3298 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3299 * @dev: The device for which the DMA address was created
3300 * @addr: The DMA address
3301 * @size: The size of the region in bytes
3302 * @direction: The direction of the DMA
3304 static inline void ib_dma_unmap_single(struct ib_device
*dev
,
3305 u64 addr
, size_t size
,
3306 enum dma_data_direction direction
)
3308 dma_unmap_single(dev
->dma_device
, addr
, size
, direction
);
3312 * ib_dma_map_page - Map a physical page to DMA address
3313 * @dev: The device for which the dma_addr is to be created
3314 * @page: The page to be mapped
3315 * @offset: The offset within the page
3316 * @size: The size of the region in bytes
3317 * @direction: The direction of the DMA
3319 static inline u64
ib_dma_map_page(struct ib_device
*dev
,
3321 unsigned long offset
,
3323 enum dma_data_direction direction
)
3325 return dma_map_page(dev
->dma_device
, page
, offset
, size
, direction
);
3329 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3330 * @dev: The device for which the DMA address was created
3331 * @addr: The DMA address
3332 * @size: The size of the region in bytes
3333 * @direction: The direction of the DMA
3335 static inline void ib_dma_unmap_page(struct ib_device
*dev
,
3336 u64 addr
, size_t size
,
3337 enum dma_data_direction direction
)
3339 dma_unmap_page(dev
->dma_device
, addr
, size
, direction
);
3343 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3344 * @dev: The device for which the DMA addresses are to be created
3345 * @sg: The array of scatter/gather entries
3346 * @nents: The number of scatter/gather entries
3347 * @direction: The direction of the DMA
3349 static inline int ib_dma_map_sg(struct ib_device
*dev
,
3350 struct scatterlist
*sg
, int nents
,
3351 enum dma_data_direction direction
)
3353 return dma_map_sg(dev
->dma_device
, sg
, nents
, direction
);
3357 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3358 * @dev: The device for which the DMA addresses were created
3359 * @sg: The array of scatter/gather entries
3360 * @nents: The number of scatter/gather entries
3361 * @direction: The direction of the DMA
3363 static inline void ib_dma_unmap_sg(struct ib_device
*dev
,
3364 struct scatterlist
*sg
, int nents
,
3365 enum dma_data_direction direction
)
3367 dma_unmap_sg(dev
->dma_device
, sg
, nents
, direction
);
3370 static inline int ib_dma_map_sg_attrs(struct ib_device
*dev
,
3371 struct scatterlist
*sg
, int nents
,
3372 enum dma_data_direction direction
,
3373 unsigned long dma_attrs
)
3375 return dma_map_sg_attrs(dev
->dma_device
, sg
, nents
, direction
,
3379 static inline void ib_dma_unmap_sg_attrs(struct ib_device
*dev
,
3380 struct scatterlist
*sg
, int nents
,
3381 enum dma_data_direction direction
,
3382 unsigned long dma_attrs
)
3384 dma_unmap_sg_attrs(dev
->dma_device
, sg
, nents
, direction
, dma_attrs
);
3387 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3388 * @dev: The device for which the DMA addresses were created
3389 * @sg: The scatter/gather entry
3391 * Note: this function is obsolete. To do: change all occurrences of
3392 * ib_sg_dma_address() into sg_dma_address().
3394 static inline u64
ib_sg_dma_address(struct ib_device
*dev
,
3395 struct scatterlist
*sg
)
3397 return sg_dma_address(sg
);
3401 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3402 * @dev: The device for which the DMA addresses were created
3403 * @sg: The scatter/gather entry
3405 * Note: this function is obsolete. To do: change all occurrences of
3406 * ib_sg_dma_len() into sg_dma_len().
3408 static inline unsigned int ib_sg_dma_len(struct ib_device
*dev
,
3409 struct scatterlist
*sg
)
3411 return sg_dma_len(sg
);
3415 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3416 * @dev: The device for which the DMA address was created
3417 * @addr: The DMA address
3418 * @size: The size of the region in bytes
3419 * @dir: The direction of the DMA
3421 static inline void ib_dma_sync_single_for_cpu(struct ib_device
*dev
,
3424 enum dma_data_direction dir
)
3426 dma_sync_single_for_cpu(dev
->dma_device
, addr
, size
, dir
);
3430 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3431 * @dev: The device for which the DMA address was created
3432 * @addr: The DMA address
3433 * @size: The size of the region in bytes
3434 * @dir: The direction of the DMA
3436 static inline void ib_dma_sync_single_for_device(struct ib_device
*dev
,
3439 enum dma_data_direction dir
)
3441 dma_sync_single_for_device(dev
->dma_device
, addr
, size
, dir
);
3445 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3446 * @dev: The device for which the DMA address is requested
3447 * @size: The size of the region to allocate in bytes
3448 * @dma_handle: A pointer for returning the DMA address of the region
3449 * @flag: memory allocator flags
3451 static inline void *ib_dma_alloc_coherent(struct ib_device
*dev
,
3453 dma_addr_t
*dma_handle
,
3456 return dma_alloc_coherent(dev
->dma_device
, size
, dma_handle
, flag
);
3460 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3461 * @dev: The device for which the DMA addresses were allocated
3462 * @size: The size of the region
3463 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3464 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3466 static inline void ib_dma_free_coherent(struct ib_device
*dev
,
3467 size_t size
, void *cpu_addr
,
3468 dma_addr_t dma_handle
)
3470 dma_free_coherent(dev
->dma_device
, size
, cpu_addr
, dma_handle
);
3474 * ib_dereg_mr - Deregisters a memory region and removes it from the
3475 * HCA translation table.
3476 * @mr: The memory region to deregister.
3478 * This function can fail, if the memory region has memory windows bound to it.
3480 int ib_dereg_mr(struct ib_mr
*mr
);
3482 struct ib_mr
*ib_alloc_mr(struct ib_pd
*pd
,
3483 enum ib_mr_type mr_type
,
3487 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3489 * @mr - struct ib_mr pointer to be updated.
3490 * @newkey - new key to be used.
3492 static inline void ib_update_fast_reg_key(struct ib_mr
*mr
, u8 newkey
)
3494 mr
->lkey
= (mr
->lkey
& 0xffffff00) | newkey
;
3495 mr
->rkey
= (mr
->rkey
& 0xffffff00) | newkey
;
3499 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3500 * for calculating a new rkey for type 2 memory windows.
3501 * @rkey - the rkey to increment.
3503 static inline u32
ib_inc_rkey(u32 rkey
)
3505 const u32 mask
= 0x000000ff;
3506 return ((rkey
+ 1) & mask
) | (rkey
& ~mask
);
3510 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3511 * @pd: The protection domain associated with the unmapped region.
3512 * @mr_access_flags: Specifies the memory access rights.
3513 * @fmr_attr: Attributes of the unmapped region.
3515 * A fast memory region must be mapped before it can be used as part of
3518 struct ib_fmr
*ib_alloc_fmr(struct ib_pd
*pd
,
3519 int mr_access_flags
,
3520 struct ib_fmr_attr
*fmr_attr
);
3523 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3524 * @fmr: The fast memory region to associate with the pages.
3525 * @page_list: An array of physical pages to map to the fast memory region.
3526 * @list_len: The number of pages in page_list.
3527 * @iova: The I/O virtual address to use with the mapped region.
3529 static inline int ib_map_phys_fmr(struct ib_fmr
*fmr
,
3530 u64
*page_list
, int list_len
,
3533 return fmr
->device
->map_phys_fmr(fmr
, page_list
, list_len
, iova
);
3537 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3538 * @fmr_list: A linked list of fast memory regions to unmap.
3540 int ib_unmap_fmr(struct list_head
*fmr_list
);
3543 * ib_dealloc_fmr - Deallocates a fast memory region.
3544 * @fmr: The fast memory region to deallocate.
3546 int ib_dealloc_fmr(struct ib_fmr
*fmr
);
3549 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3550 * @qp: QP to attach to the multicast group. The QP must be type
3552 * @gid: Multicast group GID.
3553 * @lid: Multicast group LID in host byte order.
3555 * In order to send and receive multicast packets, subnet
3556 * administration must have created the multicast group and configured
3557 * the fabric appropriately. The port associated with the specified
3558 * QP must also be a member of the multicast group.
3560 int ib_attach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
3563 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3564 * @qp: QP to detach from the multicast group.
3565 * @gid: Multicast group GID.
3566 * @lid: Multicast group LID in host byte order.
3568 int ib_detach_mcast(struct ib_qp
*qp
, union ib_gid
*gid
, u16 lid
);
3571 * ib_alloc_xrcd - Allocates an XRC domain.
3572 * @device: The device on which to allocate the XRC domain.
3574 struct ib_xrcd
*ib_alloc_xrcd(struct ib_device
*device
);
3577 * ib_dealloc_xrcd - Deallocates an XRC domain.
3578 * @xrcd: The XRC domain to deallocate.
3580 int ib_dealloc_xrcd(struct ib_xrcd
*xrcd
);
3582 struct ib_flow
*ib_create_flow(struct ib_qp
*qp
,
3583 struct ib_flow_attr
*flow_attr
, int domain
);
3584 int ib_destroy_flow(struct ib_flow
*flow_id
);
3586 static inline int ib_check_mr_access(int flags
)
3589 * Local write permission is required if remote write or
3590 * remote atomic permission is also requested.
3592 if (flags
& (IB_ACCESS_REMOTE_ATOMIC
| IB_ACCESS_REMOTE_WRITE
) &&
3593 !(flags
& IB_ACCESS_LOCAL_WRITE
))
3599 static inline bool ib_access_writable(int access_flags
)
3602 * We have writable memory backing the MR if any of the following
3603 * access flags are set. "Local write" and "remote write" obviously
3604 * require write access. "Remote atomic" can do things like fetch and
3605 * add, which will modify memory, and "MW bind" can change permissions
3606 * by binding a window.
3608 return access_flags
&
3609 (IB_ACCESS_LOCAL_WRITE
| IB_ACCESS_REMOTE_WRITE
|
3610 IB_ACCESS_REMOTE_ATOMIC
| IB_ACCESS_MW_BIND
);
3614 * ib_check_mr_status: lightweight check of MR status.
3615 * This routine may provide status checks on a selected
3616 * ib_mr. first use is for signature status check.
3618 * @mr: A memory region.
3619 * @check_mask: Bitmask of which checks to perform from
3620 * ib_mr_status_check enumeration.
3621 * @mr_status: The container of relevant status checks.
3622 * failed checks will be indicated in the status bitmask
3623 * and the relevant info shall be in the error item.
3625 int ib_check_mr_status(struct ib_mr
*mr
, u32 check_mask
,
3626 struct ib_mr_status
*mr_status
);
3628 struct net_device
*ib_get_net_dev_by_params(struct ib_device
*dev
, u8 port
,
3629 u16 pkey
, const union ib_gid
*gid
,
3630 const struct sockaddr
*addr
);
3631 struct ib_wq
*ib_create_wq(struct ib_pd
*pd
,
3632 struct ib_wq_init_attr
*init_attr
);
3633 int ib_destroy_wq(struct ib_wq
*wq
);
3634 int ib_modify_wq(struct ib_wq
*wq
, struct ib_wq_attr
*attr
,
3636 struct ib_rwq_ind_table
*ib_create_rwq_ind_table(struct ib_device
*device
,
3637 struct ib_rwq_ind_table_init_attr
*
3638 wq_ind_table_init_attr
);
3639 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table
*wq_ind_table
);
3641 int ib_map_mr_sg(struct ib_mr
*mr
, struct scatterlist
*sg
, int sg_nents
,
3642 unsigned int *sg_offset
, unsigned int page_size
);
3645 ib_map_mr_sg_zbva(struct ib_mr
*mr
, struct scatterlist
*sg
, int sg_nents
,
3646 unsigned int *sg_offset
, unsigned int page_size
)
3650 n
= ib_map_mr_sg(mr
, sg
, sg_nents
, sg_offset
, page_size
);
3656 int ib_sg_to_pages(struct ib_mr
*mr
, struct scatterlist
*sgl
, int sg_nents
,
3657 unsigned int *sg_offset
, int (*set_page
)(struct ib_mr
*, u64
));
3659 void ib_drain_rq(struct ib_qp
*qp
);
3660 void ib_drain_sq(struct ib_qp
*qp
);
3661 void ib_drain_qp(struct ib_qp
*qp
);
3663 int ib_get_eth_speed(struct ib_device
*dev
, u8 port_num
, u8
*speed
, u8
*width
);
3665 static inline u8
*rdma_ah_retrieve_dmac(struct rdma_ah_attr
*attr
)
3667 if (attr
->type
== RDMA_AH_ATTR_TYPE_ROCE
)
3668 return attr
->roce
.dmac
;
3672 static inline void rdma_ah_set_dlid(struct rdma_ah_attr
*attr
, u32 dlid
)
3674 if (attr
->type
== RDMA_AH_ATTR_TYPE_IB
)
3675 attr
->ib
.dlid
= (u16
)dlid
;
3676 else if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3677 attr
->opa
.dlid
= dlid
;
3680 static inline u32
rdma_ah_get_dlid(const struct rdma_ah_attr
*attr
)
3682 if (attr
->type
== RDMA_AH_ATTR_TYPE_IB
)
3683 return attr
->ib
.dlid
;
3684 else if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3685 return attr
->opa
.dlid
;
3689 static inline void rdma_ah_set_sl(struct rdma_ah_attr
*attr
, u8 sl
)
3694 static inline u8
rdma_ah_get_sl(const struct rdma_ah_attr
*attr
)
3699 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr
*attr
,
3702 if (attr
->type
== RDMA_AH_ATTR_TYPE_IB
)
3703 attr
->ib
.src_path_bits
= src_path_bits
;
3704 else if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3705 attr
->opa
.src_path_bits
= src_path_bits
;
3708 static inline u8
rdma_ah_get_path_bits(const struct rdma_ah_attr
*attr
)
3710 if (attr
->type
== RDMA_AH_ATTR_TYPE_IB
)
3711 return attr
->ib
.src_path_bits
;
3712 else if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3713 return attr
->opa
.src_path_bits
;
3717 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr
*attr
,
3720 if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3721 attr
->opa
.make_grd
= make_grd
;
3724 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr
*attr
)
3726 if (attr
->type
== RDMA_AH_ATTR_TYPE_OPA
)
3727 return attr
->opa
.make_grd
;
3731 static inline void rdma_ah_set_port_num(struct rdma_ah_attr
*attr
, u8 port_num
)
3733 attr
->port_num
= port_num
;
3736 static inline u8
rdma_ah_get_port_num(const struct rdma_ah_attr
*attr
)
3738 return attr
->port_num
;
3741 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr
*attr
,
3744 attr
->static_rate
= static_rate
;
3747 static inline u8
rdma_ah_get_static_rate(const struct rdma_ah_attr
*attr
)
3749 return attr
->static_rate
;
3752 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr
*attr
,
3753 enum ib_ah_flags flag
)
3755 attr
->ah_flags
= flag
;
3758 static inline enum ib_ah_flags
3759 rdma_ah_get_ah_flags(const struct rdma_ah_attr
*attr
)
3761 return attr
->ah_flags
;
3764 static inline const struct ib_global_route
3765 *rdma_ah_read_grh(const struct rdma_ah_attr
*attr
)
3770 /*To retrieve and modify the grh */
3771 static inline struct ib_global_route
3772 *rdma_ah_retrieve_grh(struct rdma_ah_attr
*attr
)
3777 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr
*attr
, void *dgid
)
3779 struct ib_global_route
*grh
= rdma_ah_retrieve_grh(attr
);
3781 memcpy(grh
->dgid
.raw
, dgid
, sizeof(grh
->dgid
));
3784 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr
*attr
,
3787 struct ib_global_route
*grh
= rdma_ah_retrieve_grh(attr
);
3789 grh
->dgid
.global
.subnet_prefix
= prefix
;
3792 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr
*attr
,
3795 struct ib_global_route
*grh
= rdma_ah_retrieve_grh(attr
);
3797 grh
->dgid
.global
.interface_id
= if_id
;
3800 static inline void rdma_ah_set_grh(struct rdma_ah_attr
*attr
,
3801 union ib_gid
*dgid
, u32 flow_label
,
3802 u8 sgid_index
, u8 hop_limit
,
3805 struct ib_global_route
*grh
= rdma_ah_retrieve_grh(attr
);
3807 attr
->ah_flags
= IB_AH_GRH
;
3810 grh
->flow_label
= flow_label
;
3811 grh
->sgid_index
= sgid_index
;
3812 grh
->hop_limit
= hop_limit
;
3813 grh
->traffic_class
= traffic_class
;
3817 * rdma_ah_find_type - Return address handle type.
3819 * @dev: Device to be checked
3820 * @port_num: Port number
3822 static inline enum rdma_ah_attr_type
rdma_ah_find_type(struct ib_device
*dev
,
3825 if (rdma_protocol_roce(dev
, port_num
))
3826 return RDMA_AH_ATTR_TYPE_ROCE
;
3827 if (rdma_protocol_ib(dev
, port_num
)) {
3828 if (rdma_cap_opa_ah(dev
, port_num
))
3829 return RDMA_AH_ATTR_TYPE_OPA
;
3830 return RDMA_AH_ATTR_TYPE_IB
;
3833 return RDMA_AH_ATTR_TYPE_UNDEFINED
;
3837 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
3838 * In the current implementation the only way to get
3839 * get the 32bit lid is from other sources for OPA.
3840 * For IB, lids will always be 16bits so cast the
3841 * value accordingly.
3845 static inline u16
ib_lid_cpu16(u32 lid
)
3847 WARN_ON_ONCE(lid
& 0xFFFF0000);
3852 * ib_lid_be16 - Return lid in 16bit BE encoding.
3856 static inline __be16
ib_lid_be16(u32 lid
)
3858 WARN_ON_ONCE(lid
& 0xFFFF0000);
3859 return cpu_to_be16((u16
)lid
);
3863 * ib_get_vector_affinity - Get the affinity mappings of a given completion
3865 * @device: the rdma device
3866 * @comp_vector: index of completion vector
3868 * Returns NULL on failure, otherwise a corresponding cpu map of the
3869 * completion vector (returns all-cpus map if the device driver doesn't
3870 * implement get_vector_affinity).
3872 static inline const struct cpumask
*
3873 ib_get_vector_affinity(struct ib_device
*device
, int comp_vector
)
3875 if (comp_vector
< 0 || comp_vector
>= device
->num_comp_vectors
||
3876 !device
->get_vector_affinity
)
3879 return device
->get_vector_affinity(device
, comp_vector
);
3883 #endif /* IB_VERBS_H */